Language People Numbers Corpus Linguistics and Society

Чтобы посмотреть этот PDF файл с форматированием и разметкой, скачайте его и откройте на своем компьютере.
No 64
edited by
Christian Mair
Charles F. Meyer
Nelleke Oostdijk
Cover design: Pier Post
Online access is included in print subscriptions:
On the occasion of Michael W. Stubbs’ 60
Table of Contents
Oliver Mason and Andrea Gerbig
Contributing authors
Michael W. Stubbs – a select bibliography
Michael Stubbs: a theoretician of applied linguistics
Susan Hunston
Borrowed ideas
John Sinclair
How 'systemic' is a large corpus of English
? ...................................................
Robert de Beaugrande
Some notes on the concept of cognitive linguistics
Wolfgang Teubert
Developing language education policy in Europe –
and searching for theory
Michael Byram
The semiotic patterning of Cædmon's Hymn as a 'hypersign'
Wolfgang Kühlwein
Traditional grammar and corpus linguistics
David A. Reibel
Travelogues in time and space:
a diachronic and intercultural genre study
Andrea Gerbig
An extended view of extended lexical units:
tracking development and use
Naomi Hallan
I don't know- differences in patterns of collocation and
semantic prosody in phrases of different lengths
Bettina Starcke
with critical notes
Stubbing your toe against a hard mass of facts:
corpus data and the phraseology of
Hans Lindquist
Stringing together a sentence:
linearity and the lexis-syntax interface
Oliver Mason
'Sailing the islands or watching from the dock':
the treacherous simplicity of a metaphor: How we handle
'new (electronic) hypertext'
'old (printed) text'
Wolfram Bublitz
Linking the verbal and visual:
new directions for corpus linguistics
Ronald Carter
Svenja Adolphs
The novel features of text. Corpus analysis and stylistics
Henry G. Widdowson
Hocus pocus or God's truth:
The dual identity of Michael Stubbs
Guy Cook

Oliver Mason and Andrea Gerbig
The papers in this volume are all concerned with structural aspects of language in
relation to its users in a variety of socio-cultural situations. All papers are based
on the assumption that only the use of authentic language data can inform us
about the role and function of language, its structure and use. The papers give
qualitative and mostly also quantitative analyses of their data, always with respect
to the specific 'work' the language does for its users.
The contributors to this volume have all been in critical exchange with
Michael Stubbs' leading, corpus-based work on theories of language structure and
use and its applications to education, cognition and culture, literature, and
politics. Such discussion is of course always ongoing; it involves innovative
theoretical thinking, looks at the most recent data which have become available
through new technology, and puts in perspective and evaluates established
theories and findings. The present papers are all original work providing such
new insights. They are all written in honour of Michael Stubbs' outstanding
contributions to this discussion which are often of a programmatic nature, paving
the way for more detailed work, following up his theoretical leads.
Susan Hunston's
short contribution evaluates and honours the work and
influence of Michael Stubbs in major fields of linguistics, thereby providing an
outlook on directions of future research. This is followed by three papers of a
more theoretical orientation, starting with an article by
John Sinclair
, who had a
profound influence on the field of corpus linguistics and on Michael Stubbs
himself. Sinclair discusses how imports from other disciplines have shaped
linguistics, and points out a number of pitfalls: sometimes imports do not make
sense, or are simply misapplied, so that statistical significance tests for example
provide a false sense of security where they should not have been applied in the
first place. As always, Sinclair is not afraid to be controversial, casting doubt on
many assumptions that are usually taken for granted by many researchers.
Continuing in the realm of theory,
Robert de Beaugrande
attempts to find
an answer to how systemic a corpus of English is, investigating a number of
systems in the process. He investigates the relationship between text and
language, and comes to the conclusion that both of them are in fact systemic.
Wolfgang Teubert
then looks at the mental lexicon. Meaning has often
been ignored by corpus research, as it is much harder a problem to tackle than
lexis or even grammar/phraseology. Teubert argues that there could be a fruitful
dialogue between cognitive linguists and corpus linguists, and that Michael
Stubbs would be the one person who could facilitate such a dialogue.
Michael Byram
discusses issues of language and politics. Commonly, the
national identity is based on a common (national) language, but how does it work
with a supranational entity such as the European Union? Questions arise here of
language policies, both regarding communication and cultural issues. And who
Mason & Gerbig
draws up those European policies? In his contribution, Byram echoes early work
by Michael Stubbs on the National Curriculum.
The collection further continues with a set of papers tracing linguistic
progress in the description and investigation of diachronic language data, again in
relation to the role such language had at the time it was used and what influences
result for our present views on language. The first of these papers, by
, draws connecting lines from lexicological to intertextual to semiotic
research, and demonstrates how these can be applied to a language for which we
have something we can only dream of for modern languages: a corpus of all
known utterances of the language.
Still within the theme of historical work,
David Reibel
looks into
empiricism among early grammarians, working on 'traditional grammar', which is
often used by modern corpus linguists as something to distance themselves from.
Reibel shows that issues such as what constitutes 'proper English' have been
around for a long time.
Andrea Gerbig
then shows that both sides of the Saussurean dualism
'synchronic/diachronic' can be studied with a diachronic corpus, touching on
issues such as language and the representation of reality/shared cultural
knowledge, and language change. Gerbig here exploits the fact that her corpus of
travel writing is controlled for topic, and captures the experience of a closely
defined sub-group. Gerbig's contribution also provides a link from the historical
studies to the phraseological ones.
The following four papers emphasise the phraseological aspect of
language, the field in which Michael Stubbs has most recently set new standards
in a collection of publications. Function words are routinely being ignored by
corpus linguists, on the grounds that they are too frequent or have no meaning.
John Sinclair looked at
Corpus Concordance Collocation
(1991), and
here analyses the uses of
, with a special focus on the use by children
of different age bands and the differences there are compared to adults.
Unsurprisingly, the picture is more complex as one would have expected; this
again shows that there is no substitute for looking at real data.
In the following paper,
Bettina Starcke
investigates changing discourse
prosodies of phrases based around the same nucleus. Starcke finds that the
prosody of a phrase relates to its length (effectively its specificity) because the
contexts in which the longer phrase is used are more restricted. The shorter
variants are also more often used in a literal sense, whereas longer phrases tend to
be non-compositional.
Hans Lindquist
then compares varieties of English, British and American,
and discovers that there are differences between literal and metaphorical uses of a
formulaic expression. Even though originating from America, the particular
phrase under investigation (
to stub one's toe
) is now about equally frequent in
British English, but is predominantly used literally. Studies such as this can
provide useful insights into the development and change of language.
Oliver Mason
concludes the phraseology section with a new approach to
the description of grammatical structure. Mason uses multi-word units (partly
based on n-grams as used by other authors in this volume) to investigate the
linear sequence of the sentence (in a way also pursued by Sinclair in the initial
paper). The degree of success in the analysis of a particular sentence can be
linked to parameters such as creativity and naturalness.
In future, (corpus) linguists will have to deal with language beyond the
written and spoken word. New communication technology and multi-media
computing allow us to look at other manifestations of language, and at the same
time they also have an impact on the further development and evolution of
language itself. These aspects of language study are discussed by Bublitz and by
Carter and Adolphs.
Wolfram Bublitz
discusses the impact of new communication media
on the traditional view of communication as a dyadic process, characterised by
the participants speaker/hearer or writer/reader. In current online chats there is no
longer a simple relationship between the two roles, and the exchange structure is
usually no longer comprised of adjacent pairs of utterances. However, Bublitz
concludes that this is not exclusively a feature of computer-mediated discourse
Ronald Carter
Svenja Adolphs
take the notion of a corpus a step
further, including not only actual speech, but also non-verbal gestures. In their
paper, they describe the experimental set-up for collecting such a corpus, and the
problems and issues that one needs to take into account with multiple streams of
different kinds of data. Just as computers enabled the use of electronic corpora
initially, now advances in video processing allow us to extend the object of study
to include the visual dimension.
Henry Widdowson
then covers another area in which Michael Stubbs has
pushed forward the boundaries: stylistics. Widdowson argues for a distinction of
two different entities when looking at Stubbs' work on Conrad's
Heart of
, namely the text, with all its words and textual patterns, and the novel,
with its characters and plot. The issue at stake is then how elements of these two
correlate and can be linked, thus escaping the criticism that stylistics is entirely
circular in its nature.
Guy Cook
investigates Stubbs himself: where is his place in the
hocus pocus/God's truth dualism? Reflecting on a wide range of Stubbs' work and
on Stubbs' criticism of his (Cook's) own work, Cook summarises the traits that
make Michael Stubbs such an influential scholar.
Contributing Authors
Svenja Adolphs
Associate Professor
School of English Studies
The University of Nottingham, UK
Robert de Beaugrande
Professor of English language
Wolfram Bublitz
Universität Augsburg, Germany
Michael Byram
Professor of Education
Durham University, UK
Ronald Carter
Professor of Modern English Language
The University of Nottingham, UK
Guy Cook
Professor of Language and Education
The Open University, UK
Bettina Fischer-Starcke
Research Associate
Universität Trier, Germany
Andrea Gerbig
Associate Professor
Universität Bochum, Germany
Naomi Hallan
Universität Trier, Germany
Contributing Authors
Susan Hunston
Professor of English Language
Department of English
The University of Birmingham, UK
Wolfgang Kühlwein
Emeritus Professor
Universität Trier, Germany
Hans Lindquist
Associate Professor
School of Humanities
Växjö Universitet, Sweden
Oliver Mason
Department of English
The University of Birmingham, UK
David Reibel
Emeritus Professor of English Language
Eberhard-Karls-Universität, Tübingen, Germany
John McH. Sinclair
Emeritus Professor of Modern English Language
The University of Birmingham, UK
The Tuscan Word Center, Siena, Italy
Wolfgang Teubert
Professor of Corpus Linguistics
Department of English
The University of Birmingham, UK
Henry Widdowson
Emeritus Professor of English Linguistics
Universität Wien, Austria
Michael Stubbs – A Select Bibliography
2001: Words and Phrases: Corpus Studies in Lexical Semantics. Oxford:
1996: Text and Corpus Analysis: Computer-assisted Studies of Language and
Culture. Oxford: Blackwell.
1986: Educational Linguistics. Oxford: Blackwell.
1983: Discourse Analysis: The Sociolinguistic Analysis of Natural Language.
Oxford: Blackwell.
1980: Language and Literacy: The Sociolinguistics of Reading and Writing.
London: Routledge & Kegan Paul.
1976: Language, Schools and Classrooms. London: Methuen. (2nd ed. 1983).
Edited Books
1983: (coedited with H. Hillier)
Readings of Language, Schools and Classrooms
London: Methuen.
1976: (coedited with S. Delamont)
Explorations in Classroom Observation
London: Wiley.
Articles in Refereed Journals
2005: Conrad in the computer: examples of quantitative stylistic methods.
Language and Literature
, 14, 1: 5-24.
2003: [with I. Barth] Using recurrent phrases as text-type discriminators: a
quantitative method and some findings.
Functions of Language
. 10, 1: 65-
2002: Two quantitative methods of studying phraseology in English.
International Journal of Corpus Linguistics
. 7, 2: 215-44.
2001: On inference theories and code theories: corpus evidence for semantic
, 21/3: 437-65.
2001: Texts, corpora and problems of interpretation. A Response to Widdowson.
Applied Linguistics
, 22, 2: 149-72.
1995: Collocations and semantic profiles: on the cause of the trouble with
quantitative studies.
Functions of Language
, 2, 1: 23-55.
1995: Collocations and cultural connotations of common words.
Linguistics and
, 7, 4: 379-90.
1994: Grammar, text and ideology: computer-assisted methods in the linguistics
of representation.
Applied Linguistics
, 7, 1: 1-25.
Stubbs - Bibliography
1989: The state of English in the English state: reflections on the Cox Report.
Language and Education
, 3, 4: 235-50.
1986: A matter of prolonged fieldwork: notes towards a modal grammar of
Applied Linguistics
, 7, 1: 1-25.
1984: (with G. Keck) Koschmieder on speech act theory: a historical note.
Journal of Pragmatics
, 8, 3: 305-10.
1983: Can I have that in writing, please? Some neglected topics in speech act
Journal of Pragmatics
, 7: 479-94.
1982: The sociolinguistics of the English writing system: or why children aren't
Australian Journal of Reading
, 5, 1: 30-36.
1981: Oracy and educational linguistics: the quality (of the theory) of listening.
First Language
, 2:21-30.
Other Main Articles
In press 2007: On texts, corpora and models of language. In M. Hoey, M.
Mahlberg, M. Stubbs & W. Teubert.
Text, Discourse and Corpora
London: Continuum.
In press 2007: Quantitative data on multi-word sequences in English: the case of
the word 'world'. In M. Hoey, M. Mahlberg, M. Stubbs & W. Teubert.
Text, Discourse and Corpora
. London: Continuum.
2006: Corpus analysis: the state of the art and three types of unanswered
questions. In G. Thompson & S. Hunston eds
System and Corpus
. London:
Equinox, 15-36.
2004: Language corpora. In A. Davies & C. Elder eds
Handbook of Applied
. Oxford: Blackwell. 106-32.
2001: Computer-assisted text and corpus analysis: lexical cohesion and
communicative competence. In D. Schiffrin et al. eds
Handbook of
Discourse Analysis
. Oxford: Blackwell. 304-20.
1997: Language and the mediation of experience: linguistic representation and
cognitive orientation. In F. Coulmas ed.
Handbook of Sociolinguistics
Oxford: Blackwell. 358-73.
1997: Whorf's children: critical comments on critical discourse analysis. In A.
Wray & A. Ryan eds
Evolving Models of Language
. Clevedon:
Multilingual Matters. 100-16.
1991: Educational language planning in England and Wales: multi-cultural
rhetoric and assimilationist assumptions. In F. Coulmas ed.
A Language
Policy for the European Community
. Berlin & NY: de Gruyter. 215-39.
1990: Language in education. In N. E. Collinge ed.
An Encyclopedia of
. London & NY: Routledge. 551-89.
1989: (co-author)
English for Ages 5 to 16
. [Cox Report]. Dept of Education and
Science and Welsh Office.
1984: Discourse analysis and educational linguistics. In P. Trudgill ed.
. London: Academic Press. 203-43.
Stubbs - Bibliography
Other Articles in Conference Proceeding, Festschriften, Edited Books, Etc.
(in prep): Technology and phraseology: with notes on the history of corpus
linguistics. In U. Römer & R. Schulze eds
The Lexis-Grammar Interface
Amsterdam: Benjamins
2007: On very frequent phraseology in English: structures, distributions and
functions. In R. Facchinetti ed.
Corpus Linguistics Twenty-five Years On
Amsterdam: Ropdopi.
2007: Inferring meaning: text, technology and questions of induction. In A.
Mehler & R. Köhler eds
Aspects of Automatic Text Analysis
. Berlin:
Springer. [Festschrift for Burghard Rieger] 233-53.
2006: On teaching critical rationalism: reconciling linguistic and literary text
analysis. In A. Gerbig & A. Müller-Wood eds
How Globalization Affects
the Teaching of English
. Lampeter: Mellen. 15-29.
2006: Exploring
with computational methods. In S. Goodman & K. A.
O'Halloran eds
The Art of English: Literary Creativity
. Basingstoke:
Palgrave Macmillan / Open University. 138-44.
2005: Using language corpora to study pragmatic meaning. In A. Schuth, K.
Horner & J. J. Weber eds
Life in Language
. Trier: Wissenschaftlicher
Verlag. [Festschrift for Wolfgang Kühlwein] 3-15.
2004: A quantitative approach to collocations. In D. Allerton et al. eds
Phraseological Units: Basic Concepts and their Application
. Basel:
Schwabe. 107-19.
2004: Conrad, concordance, collocation: heart of darkness or light at the end of
the tunnel? The Third Sinclair Open Lecture. University of Birmingham.
2003: [with A. Pusch] Frequent terms in linguistics: a pilot corpus study for a
pedagogical word-list. In C. Tschichold ed.
English Core Linguistics
Bern: Peter Lang. [Festschrift for David Allerton] 247-67.
2002: Human and inhuman geography: a comparative analysis of two long texts
and a corpus. In M. Toolan ed.
Critical Discourse Analysis
. Routledge.
98-130. [Reprint: original publication 1996 in M. Stubbs Text and Corpus
Analysis Oxford: Blackwell. Also reprinted in: C. Coffin et al. eds
Applying English Grammar
. London: Arnold.247-74.]
2000: Using very large text collections to study semantic schemas: a research
note. In C. Heffer and H. Saunston eds
Words in Context
. Birmingham
University. [CD-ROM.]
2000: Text and corpus linguistics. In M. Byram ed.
Encyclopedia of Language
Teaching and Learning
. London: Routledge.
2000: Society, education and language: the last 2000 (and the next 20?) years of
language teaching. In H. Trappes-Lomax ed.
Continuity and Change in
Applied Linguistics
. Clevedon: Multilingual Matters.
1998: German loanwords and cultural stereotypes.
English Today
, 53 (14, 1, Jan
1998): 19-26.
1998: A note on the phraseological tendencies in the core vocabulary of English.
Studia Anglica Posnaniensia
, 23: 399-410.
Stubbs - Bibliography
1998: Judging the facts: an example of applied discourse analysis. In J. Cheshire
& P. Trudgill eds
The Sociolinguistics Reader
. Vol 2. London: Arnold.
[Reprint: original publication 1991, in C. Uhlig & R. Zimmermann, eds
Anglistentag 1990 Marburg: Proceedings. Tübingen: Niemeyer. 312-331.]
1997: Eine Sprache idiomatisch sprechen: Computer, Korpora, Kommunikative
Kompetenz und Kultur. K. J. Mattheier ed.
Norm und Variation
Frankfurt: Peter Lang. 151-67.
1996: The English writing system. In H. Günther & O. Ludwig, eds
Schrift und
Schriftlichkeit / Writing and Its Use
. Vol 2. Berlin & NY: de Gruyter.
1995: Corpus evidence for norms of lexical collocation. In G. Cook & B.
Seidlhofer, eds
Principle and Practice in Applied Linguistics
. Oxford:
Oxford University Press. 245-56. [Festschrift for H. G. Widdowson].
1995: Educational language planning in England and Wales: multi-cultural
rhetoric and assimilationist assumptions. In O. García & C. Baker, eds
Policy and Practice in Bilingual Education
. Clevedon: Multilingual
Matters.25-39. [Reprint: original publication 1991, in F. Coulmas, ed.
Language Policy for the European Community
. Berlin & NY: de Gruyter.
1993: (with A. Gerbig) Human and inhuman geography: on the computer-assisted
analysis of long texts. In M. Hoey ed.
Data, Description, Discourse
London: HarperCollins. 64-85. [Festschrift for John Sinclair].
1993: British traditions in text analysis: from Firth to Sinclair. In M. Baker et al.
Text and Technology
. Amsterdam: Benjamins. 1-33. [Festschrift for
John Sinclair].
1992: English teaching, information technology and critical language awareness.
In N. Fairclough ed.
Critical Language Awareness
. London: Longman.
1992: Institutional linguistics: language and institutions, linguistics and
sociology. In M. Pütz ed.
Thirty Years of Linguistic Evolution
Amsterdam: Benjamins. 189-211. [Festschrift for René Dirven].
1992: Spelling in society: forms and variants, uses and users. In R. Tracy ed
Climbs the Grammar Tree
. Tübingen: Niemeyer. 221-34. [Festschrift for
David Reibel].
1991: Knowledge about Language: Grammar, Ignorance and Society. University
of London: Institute of Education / Kogan Page. [Professorial Lecture].
1991: Judging the facts: an example of applied discourse analysis. In C. Uhlig &
R. Zimmermann eds
Anglistentag 1990 Marburg
. Tübingen: Niemeyer.
[Reprinted in J. Cheshire & P. Trudgill eds The Sociolinguistics Reader.
Vol 2. London: Arnold. 1998: 348-66].
Stubbs - Bibliography
1989: On ivory towers and the market place: everyday and specialist knowledge
in applied linguistics. In C. S. Butler et al. eds
Language and Literature:
Theory and Practice.
University of Nottingham. 27-38. [Festschrift for
Walter Grauberg].
1987: An educational theory of (written) language. In J. Norrish & T. Bloor eds
Written Language. London: CILT. 3.38.
1986: Language development, lexical competence and nuclear vocabulary. In K.
Durkin ed.
Language Development in the School Years
. London: Croom
Helm. 57-76.
1986: Lexical density: a computational technique and some findings. In M.
Coulthard ed.
Talking about Text
. University of Birmingham: English
Language Research. 27-42. [Festschrift for David Brazil].
1983: Understanding linguistic diversity: what teachers should know about
educational linguistics. In M. Stubbs & H. Hillier eds 1983: 11-35.
1982: Stir until the plot thickens. In R. Carter & D. Burton eds
Literary Text and
Language Study
. London: Edward Arnold. 56-85.
1982: Written language and society: some particular cases and general
observations. In M. Nystrand ed.
What Writers Know
. New York:
Academic Press. 31-55.
1981: Scratching the surface: linguistic data in educational research. In C.
Adelman ed.
Uttering, Muttering
. London: Grant McIntyre. 114-33.
1981: Motivating analyses of exchange structure. In M. Coulthard & M.
Montgomery eds
Studies in Discourse Analysis
. London: Routledge &
Kegan Paul. 107-19.
1981: What's the relationship between sociolinguistics and language teaching,
please? In H. Eichheim & A. Maley eds
Fremdsprachenunterricht im
Spannungsfeld zwischen Gesellschaft, Schule und Wissenschaften
Munich: Goethe Institut. 27-45.
1980: What is English? Modern English language in the curriculum.
English in
, 51:3-20.
1980: (with M. Berry) The Duke of Wellington's gambit: notes on the English
verbal group.
Nottingham Linguistic Circular
, 9,2. 143-62.
1980: The sociolinguistics of literacy. In T. Bessell-Browne et al. eds
into the Eighties.
Adelaide: Australian Reading Association. 99-113.
1979: (with B. Robinson) Analysing classroom language. In M. Stubbs, B.
Robinson & S. Twite
Observing Classroom Language
, Block 5, PE232.
Milton Keynes: Open University Press. 5-59.
1976: Keeping in touch: some functions of teacher-talk. In M. Stubbs & S.
Delamont eds 1976: 151-72.
1975: Teaching and talking: a sociolinguistic approach to classroom interaction.
In G. Chanan & S. Delamont eds
Frontiers of Classroom Research
Slough: NFER. 233-47.
Michael Stubbs: a theoretician of applied linguistics
Susan Hunston
University of Birmingham
My brief for this short paper is to consider Michael Stubbs' influence on the field
of linguistics. This is not something to be undertaken lightly, partly because
Michael's erudition vastly exceeds my own, and partly because consideration of
the role he plays raises questions concerning the relationship of theoretical and
applied linguistics, between theory and practice. Michael Stubbs' work has
always been located within Applied Linguistics, in the sense that many of his
concerns, especially his abiding interest in education (e.g. Stubbs 1976; 1980;
1986b; 1995b), would normally come under the heading of 'applied' research. His
contributions to Critical Discourse Analysis and to literary stylistics come into
this category too. However, his work forces us to recognise that 'applied' in this
sense in no way implies 'theory-impoverished' or 'intellectually inconsequential'.
What it does imply is the observation and analysis of naturally-occurring
language in its social context, and a demand for a linguistic theory that takes such
language as its starting point.
A case in point is one of Stubbs' papers from the 1980s, ' "A matter of
prolonged field work": notes towards a modal grammar of English' (Stubbs
1986a). This takes as its data naturally-occurring language from a large number
of sources and focuses on those aspects of spoken or written language that
express speaker/writer attitude, in particular those that express commitment to or
detachment from the truth of a proposition. The paper brings together a number of
linguistic topics: attribution, speech act theory, vague language, verb aspect, verb
process types, and connectors among them. Stubbs points out that although some
acts performed through words (such as 'excommunication') are non-negotiable,
illocutionary acts proper are open to hedging, emphasis and verbal prevarications
of many kinds. Speakers exploit the resources of language to give their assertions
the weight of authority or to remove from themselves any responsibility for the
truth-value of propositions. The unifying theme in the paper is that speaker
attitude is central to language description, and that consideration of attitudinal and
interactional factors can challenge assumptions or solve problems in fields such
as speech act theory and syntax. The focus on commitment / detachment and on
the averred source of propositions finds echoes in Sinclair (1988), Cooper (1981),
and Tadros (1993), and in much subsequent work on written discourse in
particular. It is central to considerations of how knowledge is constructed and
transmitted (Hunston 1993). The centrality of the attitudinal and the interactional
to language theory and description is a theme taken up by numerous writers
including Martin and White (2005) and the papers in Hunston and Thompson
(2000). What remains typical of Stubbs' approach is the insistence that intuition-
dependent theories and naturally-occurring data be placed alongside each other,
the latter both informing and challenging the former.
Like John Sinclair, Stubbs was led by his interest in the patterns of lexis
and grammar in naturally-occurring discourse to exploit the growing power of
computers to analyse large quantities of text. Much of Stubbs' work since 1995
shares some assumptions with Sinclair's approach (e.g. Sinclair 1991, 2004), and
he is personally associated with at least four major contributions to the field. The
first two follow the practice of placing large amounts of data and particular
analytical techniques at the service of traditions that typically draw on smaller
amounts and different methods (Critical Discourse Analysis and literary
stylistics). The second two offer critical and visionary accounts of some corpus
linguistic practices themselves.
Text and Corpus Analysis
(Stubbs 1996) was among the first publications
to unite the insights of corpus investigation techniques with those of more
traditional discourse analysis and with the use of text in investigating cultural
practices. In the book, Stubbs presents examples in three sets of contextual
parameters. His study of
in two Baden-Powell texts uses
concordance lines taken from only those texts. On the other hand, the study
undertaken in collaboration with Andrea Gerbig of ergative verbs in two sets of
textbooks uses a much larger dataset and a greater degree of statistical processing
(see also Stubbs and Gerbig 1993). Finally, the study of cultural keywords and
their most frequent collocates uses frequency information from a number of very
large reference corpora (see also Stubbs 2001). Mautner (2007: 8) describes such
work as uniting qualitative and quantitative methodologies. Put simply, Stubbs
notes that frequency is important because it reflects the prevalence in a given
cultural context of a particular formulation. For example, he observes (1996: 184)
that frequently-occurring formulations such as
child care, care in the community
care and resettlement of offenders
reflect (and perpetuate) a cultural context
in which diverse groups of individuals are construed as constituencies requiring a
common institutional response. Another theme is the conveying of covert
evaluative meaning through intertextuality: a given instance has a 'hidden
meaning' because of the way a word or phrase in it is used in a number of other
texts. Stubbs offers many examples of this, one being the collocations associated
the streets
that connote danger and menace (Stubbs 2001: 203-206; cf
Channell 2000). By using frequency to interrogate intertextuality, Stubbs
demonstrates that a concern for numbers does not condemn the researcher to a
level of abstraction that precludes interpretation or sensitivity to context.
Many researchers have used methods similar to Stubbs' to further the
agenda of Critical Discourse Analysis, to some extent in response to Stubbs' own
critique of CDA methods (Stubbs 1997). Mautner (2007), for example, uses the
'collocational profile' of the word
to argue that the term can be regarded as
ageist. Baker (2006) offers a number of studies using different techniques,
including a concordance-based study of
in British newspapers which
shows the predominant discourse contexts of the word. Coffin and O'Halloran
(2006) take as their starting point a single text – an article from the Sun
Michael Stubbs: a theoretician of applied linguistics
newspaper about migrants to Britain from Eastern Europe – and explore it
alongside a corpus of editions of the same newspaper. They argue that a number
of phrases that are evaluatively ambiguous in the context of the single text alone
(such as
poverty-stricken former Soviet states
desperate for…any job at all
can be shown to resonate with unpleasant connotations when their phraseologies
are examined in the larger corpus. Coffin and O'Halloran draw on the concepts of
logogenesis, ontogenesis and phylogenesis to model the relations between a
single text and the reader's longer-term experience of texts from the same
journalistic source. Partington (2004: 19) gives a name to the union of discourse
and corpus techniques: Corpus-Assisted Discourse Studies.
A second area of applied work is literary stylistics. Stubbs' interest in the
interaction of language and literature (evident in his sociolinguistic analyses of
literary texts in Stubbs 1983) has culminated in his studies of Conrad's
Heart of
(2004; 2005). In Stubbs 2005 he offers a number of studies, using word
frequency, collocation, keyword analysis and word distribution analysis, to
corroborate and extend observations about Conrad's novel that have been made
by literary critics. For example, critics have observed that much of the novel is
vague and imprecise: descriptions are hazy, actions are indeterminate. It is not
possible even to know what Kurtz means by his (infamous) cry
The horror, the
. Whereas critics have drawn attention to the motif of indistinctness, and to
the noticeable frequency of lexical words such as
murky, blurred, darkness
Stubbs goes further and establishes that grammatical words indicating vague
reference (
something, someone, somewhere
and so on) are significantly more
frequent in
Heart of Darkness
than in a general Fiction corpus or in written
English in general. Critics have also commented, not always positively, on the
repetitious nature of Conrad's prose. Stubbs establishes wherein that repetition
lies, not just in individual words but in patterns such as 'the noun of (a) neg-prefix
adjective noun' (
the darkness of an impenetrable night, the sea of inexorable time,
the stillness of an implacable force
and so on). He also comments on spatial-
reference phrases such as
to the end of the
in the middle of the
, which appear
to be frequent in
Heart of Darkness
, but which Conrad in fact uses with the same
frequency as is general in English, though with less specific referents. With this
work, Stubbs adds to a growing body of research using corpus techniques to
study literary works (e.g. Semino and Short 2002; Culpepper et. al. forthcoming;
Toolan, forthcoming). What is notable about Stubbs' work is that he takes as his
starting point the critical literature on his chosen writer, which he then
supplements and sometimes challenges.
To some extent, then, Stubbs acts as an ambassador for corpus linguistics
in the wider Applied Linguistics community. What is noticeable, however, is his
insistence on the need both to question and to develop methodological
assumptions, his refusal to take easy routes of interpretation. These concerns are
most apparent in his warnings on the interpretation of statistics (Stubbs 1995a)
and in his work on both systematising and diversifying the study of phraseology
(Stubbs 2001; 2002; 2006). We find echoes of this caution in Coffin and
O'Halloran's (2006) and Mautner's (2007) caveats about the limitations of corpus
techniques and the need for triangulation in methodology. And Stubbs' concern to
extend methods of examining recurrent phraseology and collocation / colligation
is reflected in Mason (this volume); Fletcher (2006); Rayson (2006); Groom
(2007) among others.
For me, Michael Stubbs' most profound legacy is probably his theory-
oriented writing that integrates corpus investigation techniques, and indeed
discourse analytical methods, with the 'bigger questions' in linguistics. Although
like many corpus linguists he illustrates his arguments with specific examples –
the use in English of the word
, for instance (Stubbs 2001: 156-159), or
Conrad's use of
(Stubbs 2005), or the way that judges in Britain use
phrases such as
you may think
(Stubbs 1996: 113-117) – these carefully-observed
phenomena are never ends in themselves, but a step towards a wider vision. In a
number of wide-ranging papers (e.g. 1993; 2000; 2006) Stubbs contextualises
corpus studies within a number of intellectual traditions, making him one of the
leading theorists of corpus linguistics today, as well as one of its most respected
practitioners. I shall take his 2006 paper 'Corpus analysis: the state of the art and
three types of unanswered questions' as a case in point. The paper, as is typical,
recounts a number of sample analyses: the distribution of number of collocates
across words, the pragmatic import of the phrase
ripe old age
(which might be
contrasted with Mautner's
!), the relation of collocates and schemata, with
as exemplars. More profoundly, though, he relates such findings
to questions about language that have been asked by researchers from very
different traditions, theorists whose work is often by-passed as irrelevant by other
corpus linguists. He raises Chomsky's distinction between description and
explanation (cf Meyer 2002: 2-4), not to dismiss it but to argue for its
applicability to observations such as collocation. Returning to an earlier concern
(cf Stubbs 1986a) about the discipline of pragmatics and naturally-occurring
language, he argues that a corpus-inspired view of the consistency between form
and function 'rescues pragmatics from the notion that it is condemned to deal with
idiosyncratic, one-off, context-bound interpretations' (Stubbs 2006: 27). Finally,
he notes that work such as his own on 'cultural keywords' (e.g.
above) can be
used to complement the work of philosophers on the construction of social reality
(ibid: 30-32). Taking Searle's example of
as a socially constructed entity,
he demonstrates that a corpus can be interrogated to discover how people talk
about money, and therefore how such entities come to be construction and
transmitted. In other words, he offers corpus investigation techniques, and the
theories about language that have arisen from them, as a way of answering
questions about language from outside the corpus field, but he also places a
demand upon corpus work to meet the challenges of those questions and not to
dismiss them as irrelevant.
Michael Stubbs is a modest and accessible writer, scrupulous always in
recognising influences. His own influence on other researchers is huge. To take
(somewhat flippant) quantitative data, a survey of the sixteen extant volumes in a
series of books on corpus linguistics reveals that he is referenced extensively in
no fewer than twelve of them. Interpreting that data more qualitatively, we can
Michael Stubbs: a theoretician of applied linguistics
see that many of the key ideas, the most apt examples, the most profound
questions, are to be found in his writing.
Baker P. 2006.
Using Corpora in Discourse Analysis
. London: Continuum.
Channell J. 2000. 'Corpus-based analysis of evaluative lexis'. In Hunston and
Thompson (eds.) 39-55.
Coffin C. and K. O'Halloran 2006. 'The role of appraisal and corpora in detecting
covert evaluation'
Functions of Language
13: 77-110.
Cooper M. 1981. 'Aspects of the structure of written academic discourse and
implications for the design of reading programmes'. In Hoedt et al. (eds.)
Pragmatics and LSP: proceedings of the 3
European symposium on LSP,
Copenhagen, August 1981
. Copenhagen: Copenhagen School of
Economics. 403-433.
Culpepper J., Hoover, D., Louw, B. & Wynne M. forthcoming.
Approaches to
Corpus Stylistics
. Routledge.
Fletcher W. 2006. 'Real-Time Identification of MWE Candidates in Data from the
BNC and the Web'. BAAL Corpus SIG meeting, Oxford, April 2006.
Groom N. 2007
Phraseology and epistemology in humanities writing.
Unpublished PhD thesis, University of Birmingham.
Hoey M. (ed.) 1993.
Data, Description, Discourse: paper on the English
Language in honour of John McH Sinclair
. London: HarperCollins.
Hunston S. 1993. 'Evaluation and ideology in scientific discourse'. In M.
Ghadessy (ed.)
Register Analysis: theory and practice
. London: Pinter. 57-
Hunston S. and G. Thompson (eds.) 2000.
Evaluation in Text: Authorial stance
and the construction of discourse
. Oxford: Oxford University Press.
Martin J. and P. White 2005.
The Language of Evaluation: appraisal in English
London: Palgrave.
Mautner G. 2007. 'Mining large corpora for social information: the case of
Language in Society
Meyer C. 2002.
English Corpus Linguistics: an introduction
. Cambridge:
Cambridge University Press.
Partington A. 2004. 'Corpora and discourse, a most congruous beast'. In A.
Partington, J. Morley & L. Haarman (eds.)
Corpora and Discourse
. Bern:
Peter Lang. 11-20.
Rayson P. 2006. 'Right from the word go: identifying multi-word-expressions for
semantic tagging'. BAAL Corpus SIG meeting, Oxford, April 2006.
Semino E. and M. Short 2002.
Corpus Stylistics: speech, writing and thought
presentation in a corpus of English writing
. London: Routledge.
Sinclair J. 1988. 'Mirror for a text'
Journal of English and Foreign Languages
Sinclair J. 1991.
Corpus Concordance Collocation
. Oxford: Oxford University
Sinclair J. 1994.
Trust the Text: language, corpus and discourse
. London:
Stubbs M. 1976.
Language, Schools and Classrooms
. London: Methuen.
Stubbs M. 1980.
Language and Literacy
. London: Routledge.
Stubbs M. 1983.
Discourse Analysis
. Oxford: Blackwell.
Stubbs M. 1986a. A matter of prolonged field work: notes towards a modal
grammar of English.
Applied Linguistics
7: 1-25.
Stubbs M. 1986b.
Educational Linguistics
. Oxford: Blackwell.
Stubbs M.1993. 'British traditions in text analysis: from Firth to Sinclair'. In M.
Baker, G. Francis & E. Tognini-Bonelli (eds.)
Text and Technology: in
honour of John Sinclair
. Amsterdam: Benjamins. 1-36.
Stubbs M. 1995a. 'Collocations and semantic profiles: on the cause of the trouble
with quantitative studies'
Functions of Language
2: 23-55.
Stubbs M. 1995b. 'Educational language planning in England and Wales: multi-
cultural rhetoric and assimilationist assumptions'. In O. García & C. Baker
Policy and Practice in Bilingual Education
. Clevedon: Multilingual
Matters. 25-39.
Stubbs M. 1996.
Text and Corpus Analysis
. Oxford: Blackwell.
Stubbs M. 1997. 'Whorf's children: critical comments on critical discourse
analysis'. In A. Wray & A. Ryan (eds.)
Evolving Models of Language
Clevedon: Multilingual Matters. 100-16.
Stubbs M. 2000. 'Society, education and language: the last 2000 (and the next
20?) years of language teaching'. In H. Trappes-Lomax (ed.)
and Change in Applied Linguistics.
Clevedon: Multilingual Matters. 15-
Stubbs M. 2001.
Words and Phrases: Corpus Studies in Lexical Semantics
Oxford: Blackwell.
Stubbs M. 2002. 'Two quantitative methods of studying phraseology in English'
International Journal of Corpus Linguistics
7: 215-44.
Stubbs M. 2004. 'Conrad, concordance, collocation: heart of darkness or light at
the end of the tunnel?'
The Third Sinclair Open Lecture
. University of
Stubbs M. 2005. 'Conrad in the computer: examples of quantitative stylistic
Language and Literature
14: 5-24 .
Stubbs M. 2006. 'Corpus analysis: the state of the art and three types of
unanswered questions'. In G. Thompson & S. Hunston (eds.)
System and
Corpus: exploring connections
. London: Equinox. 15-36.
M. Stubbs & A. Gerbig 1993. 'Human and inhuman geography: on the computer-
assisted analysis of long texts'. In M. Hoey (ed.) 64-85.
Tadros A. 1993. 'The pragmatics of text averral and attribution in academic texts'.
In Hoey (ed.) 115-138.
Toolan M. forthcoming.
Narrative Progression in the Short Story
. London:
Borrowed ideas
John Sinclair
Tuscan Word Center, Siena
Linguistics fits uneasily in the panoply of academic disciplines, with many links to the soft
sciences, more reliance on measurement as the examinable body of language grows in the
electronic domain, and with its roots still firmly in the humanities; the unique property of
intuitive access to meaning keeps it apart. It is thus in a position to adopt conventions,
concepts, terms and working practices from a wide range of sources, and while this is a
fundamental strength, there are dangers. This paper suggests that some influences have
been uncritically imported, obscuring fundamental properties of language that are not
shared by other data types.
Of the four instances discussed the first and third deal with sampling language,
and the other two with the textual property of linearity. The first is a cautionary tale about
how a general working practice in many data types became almost inviolate in early
corpus linguistics although quite irrelevant; the second emphasises that terminology
should be chosen with care; the importation of hierarchical terms based on the prefix
meta- is in danger of obscuring the key property of the linearity of text. The third point
returns to sampling issues and wonders why chance plays such a large role in our
statistical analyses of large corpora when we know that the words of a text are chosen and
arranged to create meanings. The final mini-study argues that the adoption of conventions
of logical notation in formal grammars suppresses an essential sub-property of linearity –
the directionality of text – and unnecessarily cuts down the structural options that should
be available in description.
Academic disciplines have a tendency to be insular. They shape their arguments,
terminology and experimental methods in order to make the best descriptions of
their data, and thus their categories and methods often do not lend themselves to
applications to other data, and they can act as deterrents to interdisciplinary work.
However, younger disciplines can often profit by importing models and
techniques from established areas of investigation, either from specific subject
areas or from general experience. For instance "scientific method" is an imprecise
but powerful attitude to the handling of data, experiment and argument that
encompasses such valuable notions as objectivity, replicability of procedure and
the role of falsification attempts; these notions are not identified with any single
subject but are widely adopted in the experimental sciences and beyond.
Some disciplines are so important to others that they are in part shaped by
their applications; statistics is one of these; a number of disciplines are in essence
applications of statistics to particular bodies of data, and many more disciplines
rely heavily on statistical tools. Logic is so basic that every discipline is supposed
to take it for granted and apply it without comment; in addition the descriptive
techniques of formal and mathematical logic are widely used in an explicit form,
with formalisms and conventions imported in bulk; mathematics has similar
status. These are not conventions to be trifled with.
Computer science, as a very young discipline, has borrowed most of its
procedures and practices and a lot of its arguments from these bedrock
disciplines, but with its own status as a central service discipline it is also a
remarkably productive source of models, notions and procedures. Binary
mathematics existed before computers but has been developed to a point where
some of its concepts are becoming part of the general vocabulary of academics
and even extend to the public arena. "Eighty gigs" means "lots of storage" in
domestic computing.
Linguistics is also in many ways a young discipline, though the study of
language has been with us for a long time. In the last half-century it has been
boosted by pressure from applications such as language learning, in turn deriving
from aspects of globalisation, and it has matured with the help of modern
technology. The study of the spoken language was speculative and anecdotal until
the invention of cheap recording devices, and the meaningful patterns of both
spoken and written text are only just being uncovered through the computational
analysis of large corpora.
Linguistics is not a "pure" science because its touchstone is meaning, and
meaning is partly determined by the perception of individuals, and accessed via
their reports. The intuition, as it is somewhat misleadingly called, is a decision-
making mental facility which is non-negotiable, differs from one person to
another, and offers no reasons for its decisions; any reasons advanced by an
informant are bogus. Intuition has a delphic status in appearing to be quite
arbitrary, mysterious and impenetrable, leaving the scholar to sort out how to
interpret its "proclamations".
While the intuition maintains an element of subjectivity which is right in
the centre of any linguistic argumentation, linguistics as a subject does not sit
easily among the humanities because of its heavy reliance on experimental
methods and, nowadays, computing. Nor is it more than peripheral among the
social sciences. It sits uneasily at a disciplinary crossroads. Contributing further
to the unease is the poor standard of applications that has been achieved. In the
present state of the world, good-quality applications of linguistics are much
sought after and would be highly prized. If a reliable means of deriving meaning
from texts, comprehensively and automatically, could be achieved it would
constitute the launching pad of a major improvement in the efficiency of social
institutions, media services, international understanding and security. Despite
very large investment over decades, this goal seems only to recede.
The present position is fairly desperate, and linguists are losing credibility
because practitioners of other disciplines seem to achieve better results in the
solving of language-oriented tasks, speech recognition being the most notorious
Borrowed Ideas
of these in recent years. Gradually scholars are beginning to face a most
unpalatable prospect – that their models and theories are faulty and are leading
them astray and contributing to the routine failure of applications. To remedy this
will be a lengthy process, however, because of the large investments and the
threats to the careers of thousands of people. Also constructive criticism at this
level of profundity requires not just attacks on existing beliefs, but the emergence
of more reliable alternatives to put in their place.
This paper is far too limited in scope to tackle such a major problem, and
devotes itself to a minor piece of ground-clearing. It makes a claim that some
notions and procedures have been imported into linguistics rather uncritically
from other disciplines or the wider academic scene, and are perhaps contributing
to the inadequacy of models and arguments. These only compound the problem
rather than causing it, but the introduction of even a small amount of clarity in a
few areas could highlight the larger need for development and accelerate its
We will consider four areas where concepts and/or routines have been
imported into linguistics; three we will deal with rather briskly as aperitifs, and
the last in more detail. The first and third deal with sampling language, and the
other two with the textual property of linearity. One is the question of data
sampling conventions for language text, the second raises the matter of the
linearity of text and how it can be obscured by unfortunate terminology; the third
concerns the relation between linguistic patterns and chance, and the fourth
returns to linearity and directionality and considers the use of logical notation to
represent structure.
Some of the concepts that will be discussed below are, in their natural habitat,
complex and sophisticated, and are the product of advanced research in their
parent disciplines. Several would take more than one paper even to outline
satisfactorily; to assess them critically would require skills well beyond those of
the present author. I would like to make it quite clear that I do not intend to
engage with such matters, for which I am ill-equipped, but only to take up the
way in which the concepts and routines associated with them are applied within
the discipline of linguistics. A philosopher of science, a statistician, a logician
will probably wince at the superficiality of the arguments put forward, and my
only defence is that the way in which the imported ideas are dealt with here
comes from my own experience in linguistics.
Nor should it be assumed that I am trying to protect linguistics from
outside influences, to be promoting the view that external imports are
undesirable. Far from it – one of the shaping influences on my attitudes to
language study has been the realisation that the really mould-breaking ideas have
come from outside the subject, and not from developing notions and observations
derived from inside. As an example of this, perhaps the most innovative and far-
reaching development in linguistic perception in the last fifty years was the
philosopher Austin's idea of illocutionary force (Austin 1962). By arguing that
sentences did other things than just
, Austin opened up the prospect of
structures above the sentence, including interactive constructions. Before Austin,
linguists had been embarrassingly short of relevant comments to make on the
nature, structure, direction and results of language interaction.
Austin's work was subjected to much criticism from linguists just because
it enabled the study of discourse to get going, and thus isolated the many
grammarians who still see the sentence as the boundary of the organisation that
they seek to describe. The kind of influence that I will draw attention to in this
paper, however, is not of this "breakthrough" variety such as Austin, but it is
much lower key; it is the kind of model, concept or practice that is adopted with
little or no criticism, but just imported as an apparently self-evident concept or a
procedure; one that is well established in several disciplines, so why not
linguistics as well?
Sample size
The issue of the number of words that constitutes a proper sample of text is one
that, happily, is now a matter of history. I begin with it because, since it is no
longer a burning issue, we can attain a certain objectivity in retrospect, which
may help us when we engage with matters which are currently accepted
Some forty-five years ago the compilation of corpora began in earnest.
There were pioneers even before 1960, in particular Father Busa, and there were
by 1970 several models of corpus architecture available. Father Busa was
engaged in the huge task of indexing the whole of the works of St Thomas
Aquinas (, Bernard Quemada had
compiled the Trésor de la language Français (Imbs and Quemada, 1988), a team
in UK had prepared the first corpus of transcribed spoken language
(Krishnamurthy, ed. 2004), and a team at Brown University in USA published
and made available a million-word corpus of selected English from American
publications of the year 1961 (
Of these, only the Brown was readily accessible, offered with characteristic
American generosity, and it rapidly established itself as the archetypal corpus
(Léon 2005). For more than twenty years it was the reference point for anyone
wanting to know what a corpus was, and its architecture was still being replicated
in corpora in the 1990s, compiled specifically to be compared with Brown and its
UK clone, LOB (for all the Brown clones see
One million words, five hundred samples, each of around two thousand
words, in a range of sixteen roughly-described genres classified as either
informative or imaginative prose; the samples were of a uniform size but the
genres differed greatly in the number of samples in each; so for example Learned
Borrowed Ideas
Prose had eighty samples, around 160,000 words, while Science Fiction had only
six samples, totalling around 12,000 words. Although the many clones varied
these proportions quite a lot, the steady element was the 500 samples of 2000
words in each.
All of these figures reflected what was possible in 1963 or so, achieved
despite the puny capacity and power of computers, the ghastly problems of data
entry, the ponderous and unsuitable programming languages and the difficulty of
handling the operating systems. Yet strangely, with the technology advancing at
breakneck speed, even ten or fifteen years later the dimensions and character of
Brown seemed to restrict people's vision of what could be achieved. When the
corpus that became The Bank of English (
group=153) was designed in 1980, aiming initially at a modest five million words
and rising to twenty million, it was difficult to persuade potential backers that
such dimensions were not ridiculously extravagant. And in the summer of 2006, a
billion-word collection was announced by Oxford University Press, and data from
a trillion words from the Internet is available from the Linguistic Data
Consortium (
In this context the 2000-word sample became an untenable limitation, and
had to rise, and rise sharply. Unfortunately one of its least valuable features was
retained and defended stoutly for some years. This was the idea that all text
samples should be of the same size, to facilitate comparisons. It was claimed that
a uniform size of sample was an essential feature of sampling technique in all
serious science, and a fixture of "scientific method".
This was patently ridiculous. Back in the days of Brown, the sample size
had to be small in order to include sufficient variety in a general corpus. Since
few published documents are only 2000 words in length, the corpus was made up
of fragments. This was felt to be the lesser evil, because if whole documents had
been selected there would not be nearly enough variety in a million words – ten
smallish books would fill the corpus to the brim. The penalty, however, was
serious. One fairly obvious feature of a text is that it is not the same all the way
through. In barest outline it has a beginning, middle and end, but it is likely to
have a much more elaborate structure than that, and each aspect of its internal
structure leads to different phraseology, different vocabulary and different
When the dimensions of corpora increased and it became possible to
handle many millions of words, one of the first restrictions to be dropped was the
fixed sample size. Corpora like The Bank of English consist of whole documents
and transcribed speech encounters, of widely varying sizes. It was possible to get
the permission of rights holders for this corpus because there was never any
intention to make and distribute copies of the Bank of English, only to provide
scholars with access to it via the emerging Internet. Other large corpora have
suffered from copyright restrictions, preventing them from including whole
documents; so the British National Corpus, for example (http://www.natcorp.ox.- restricts itself to 40,000 words or ninety per cent
of any text, whichever is the smaller figure. There the question arises, which
ninety per cent gets in?
It may be difficult nowadays to look back and consider a time when a
corpus consisting of texts of varying dimensions was called unscientific and of
little use to serious researchers, who apparently were thought to lack the
techniques of comparison of datasets of differing sizes. The great variation in size
of the genres in the Brown corpus, which perhaps should have been a cause for
concern, attracted little comment.
Poor taste in terminology
The second concept that has been uncritically imported is the idea of "meta…". I
can deal with it briefly here because of recently having published a short paper
devoted to it (Sinclair 2005a), to which this is a footnote. See also Ädel (2006:
"Meta…" is a prefix which comes from philosophy, and is very old, as
shown, for example, by the word
. Originally meaning something
beyond or outside the word it prefixes, it nowadays is more specific. To quote
Wikipedia, it is "used to indicate a concept which is an abstraction from another
concept, used to analyze the latter." "
Meta language
", it instances, "refers to a
type of language or system which describes language."
I do not wish to question the use of "metalanguage" in this sense, though
there is a risk of misunderstanding that can arise from this usage. The
professional terminology may quite reasonably be called metalanguage, but the
sentences that are composed using that terminology are sentences in the ordinary
language. Here, the prefix
may be considered appropriate, but
with its implication of being used in communication, is misleading. Only the
terminology can be considered meta, not their deployment in sentences.
That is to say, in the sentence "A
consists of a single
main clause" (Trask 2000: 24), the phrases
simple sentence
main clause
be designated terms in the metalanguage, but the remainder is just ordinary
English, and could appear again in something like "A light lunch consists of a
single main dish."
This distinction is essential if we are to avoid confusion. The stream of
speech is linear and no segments of it assume a superior or abstracted position
with reference to the rest of it. Words and phrases whose reference is to aspects of
the language system behave just like other words – their relationship with the
language system is purely semantic.
Nor do I want to spend time on the use of the term
to describe
the information about a text that is often gathered to aid its classification when it
is added into a corpus. It is a very silly and pretentious use of such a term, but
innocuous except in one context, to which I will return. To be sure, "metadata"
about something is external to it, but it is not a set of abstractions from it; in fact
it is the opposite – it is external information that cannot be derived in any direct
Borrowed Ideas
way from the text, such as the date of birth of the author. But merely to be data
about an object but external to it is hardly sufficient reason to be in a "meta"
relationship to it, in the way the word is used nowadays. Or else you could
consider your passport meta-you.
Where "metadata" can be pernicious is when it is not kept separate from
the text to which it is external; indeed I suspect that the origin of this use of the
term is the old-fashioned practice of mixing data
a text with the text itself,
in such strange constructs as "DTDs" (Document Type Definitions) or "headers"
that were enthusiastically promoted as enrichments to ordinary text, e.g. in the
Text Encoding Initiative ( Despite warnings that such
insertions were in danger of corrupting the text, because they can never be
reliably removed, many corpora of earlier periods are irretrievably damaged. The
term chosen, "metadata", should have been sufficient warning in itself, because
textual metadata has by definition no place in text; nevertheless it was regularly
stuffed in, with predictable consequences.
The term to which I took, and take, exception is
, which is
and its cotext. The term is particularly associated with comments about the talk
that is evolving and of which the comments are part. Other words for the same
category are
, and I will use the former in this paper.
From a linear point of view, the role of discourse self-reference is different
according to where the referent lies in relation to the referring item. If it is in text
of the future (even in the immediately succeeding text) then the self-reference is a
, and acts as a preface,
advance label
(Tadros 1985) or something
similar, introducing what follows. If the referent is in the immediately preceding
text, then the self-reference
it, cancelling its interactive function.
Encapsulation is the normal function of most cohesive devices, but it is unlikely
that proponents of the "metadiscourse" category would accept all anaphoric and
homophoric references as belonging to this category (it would be hard to exclude
and, but
from a full list of words and phrases whose occurrence entailed
the existence of previous states of the text).
In understanding prospection and encapsulation it is not necessary to
postulate that some segment of a text is in a conceptually superior position to its
surroundings. Spoken text as a physical event is momentary, and so only its
meaning-trace is available for reference. The meaning-trace remains relevant for
a short time, but its relevance decays unless it is encapsulated. Written text is
physically longer-lasting, and the conventions of writing stress that care is taken
to make the cohesive references clear and precise; however, it seems that readers
do not normally take advantage of the option, always open, to return to an earlier
place in the writing to check an encapsulation; they prefer to rely on the
immediate text to give them enough to continue looking ahead rather than behind.
So in practice the permanence of writing has little effect upon the process of
engaging with text.
The key event for "metadiscourse" is where a self-reference cancels a
prospection, in an utterance like:
indistinguishable from ordinary ("object-") language except that its topic is itself
1. Interviewer: How do you intend to achieve this
Politician: That's a very interesting question

(Francis & Hunston 1987 p.127)
A question prospects an answer, and since text is linear the expectation is set up
that the next utterance will begin an answer. This is a strong prospection, and
participants will act on it and interpret the next utterance as some kind of answer
unless it is clearly one of the few types of move that challenge, defer or otherwise
divert the discourse from its immediate goal. One of the recognised diversions is
self-reference, where, instead of answering a question, you talk
it. The
question becomes part of the subject-matter, its meaning-trace is removed from
the linear organisation and its interactive function is thereby cancelled.
In the example above the politician now has the option of ignoring the
question altogether, because the prospection no longer applies.
All utterances prospect at the very least that they will be treated as
contributions to the linear discourse, requiring attention. So in Ädel's example
(op.cit. p 1) the word
presumably refers to an attributed statement.
2. I never said that!
Rather than (2) being somehow elevated from the surrounding discourse, a better
representation of the textual relations is that the referent of
is "demoted"
from participation in the ongoing discourse, which of course remains inalienably
linear. But it is difficult to see the relations in this way when the technical term
suggests the opposite.
For these reasons I believe that the term
is a barrier to clear
thinking on the way in which language refers to itself as it goes along. The
alternative that I proposed some time ago is
plane change,
where referents are,
by being referred to, moved away from the plane on which the interaction is
taking shape. In this representation the linearity of the discourse is not threatened,
but no other aspect of the description is disturbed.
What to count, what not to count
I was not looking forward to writing this section, because the intricacies of
statistical calculations and argument are not an interest of mine, but in corpus
study statistics seems to be unavoidable. However a recent paper by Kilgarriff
(2005) does all the difficult bits, leaving me with only the need to comment
briefly. He starts from the self-evident premise that language text does not occur
at random. Despite this, he points out, all the statistical measures we use are
based on the possibility of randomness, so if our results cannot be distinguished
from random then our only possible conclusion is that our test data were not
extensive enough. This unsatisfactory position arises because (a) randomness is
Borrowed Ideas
not an option and (b) randomness gets less likely as the data set gets more
Kilgarriff distinguishes four types of association that might pertain, for
example, between co-occurring words in a text. They could be "Random,
Arbitrary, Motivated or Predictable (R, A, M, P)." He continues (p. 263) "The
bulk of linguistic questions concern the distinction between A and M."
Unfortunately, the computer has no way of distinguishing A from M because this
distinction depends on the meaning, so the whole point of the practice is called
into question. Later in the paper (p.264), Kilgarriff tries to make a virtue out of
the fact that the primary test – of randomness – tells the linguist nothing of
interest. "Making false assumptions is often an ingenious way to proceed."
Making assumptions that
be false – even assumptions that are very
likely to be false – is an accepted component of scientific method, but making
assumptions that are
to be false is not likely to be a prelude to acquiring
useful information.
Most of Kilgarriff's points are perceptive, however, and could be the basis
of a variety of interpretations; in particular they could lead to more drastic
conclusions than his own concerning the comparison of corpus patterns with
random or chance occurrences. If we follow through his arguments, they could
lead to the whole practice of applying chance-based calculations being questioned
and revised. After all, the only statistically-relevant fact that is known about a
corpus is that its distribution does
occur by chance, so why use chance as a
criterion of relevance? Whether the occurrence of a pattern beats or does not beat
chance predictions tell us nothing about the meaningful units and their relations.
Before suggesting alternatives, I should make it clear that there are a
number of active areas of textual research to which the remaining arguments in
this section are not relevant. These areas of investigation make full use of
conventional statistical techniques and develop their own penetrating routines
without focusing at all on the meaning of the texts under study. There lies an
important distinction; studies of language variety, of authorship – any study
which examines solely the numerical properties of a body of textual material has
no interest in the communicative events of which the texts are the physical record.
For example one of the earliest studies, Zipf's Law, established a relationship
between the number of times a word form occurs and its rank in a wordlist
ordered by frequency. What the words mean, and what sort of meaningful
relationships they contract with each other, are not directly relevant to their
frequency distribution or their conformity to arithmetic laws.
Returning to consideration of the meaning-oriented investigations, we note
that chance-based measures are in constant use, notwithstanding the reservations
we might have about their relevance. They have for many years given linguists
general pointers towards which usage patterns are worth consideration, and no
doubt they will continue to do so for some time. However, in so far as they
essentially report on whether or not textual patterns fall inside or outside the
range indicated by chance occurrence, then they are stop-gap at best, and could be
misleading. Their prevalence is probably a result of a misunderstanding about the
nature of a corpus as a sample of a language, to which we now turn.
The distinctive character of a corpus as against other collections of texts is
that it claims to be reasonably representative of a language or a language variety
(Sinclair 2005b). Since language text is a population without limit, and a corpus
is necessarily finite at any one point; a corpus, no matter how big, is not
guaranteed to exemplify all the patterns of the language in roughly their normal
proportions. But since there is no known alternative method for finding them all,
we use corpora in full awareness of their possible shortcomings.
We have no reason to believe that two corpora drawn from the same
population using the same methods should show the same distribution of
phenomena. In fact we have every reason to believe the opposite; we know that
the characteristic patterns of a text are dependent on what the text means, and that
unless the meanings in the two corpora are replicated, the texts will not exhibit
the same patterns.
If we take a shovelful of sand from a sandpit and examine its constituency,
and then take another shovelful, we can postulate that the two shovelfuls will
share many constituent features, because we have no reason to believe that there
are orderings within the sandpit. But in the case of language text we
there are orderings, in fact we know that the reason that the text
is because
it is ordered for communication, and so it is meaningless to take shovelfuls of text
and expect them to have similar constituency.
It is not clear to me why we deal in probabilities when analysing corpora.
In front of us are not probabilities but actualities, and those should be the focus of
our attention. Any actual set of events can be trivially converted into a
probability, so that if A and B co-occur
times in a corpus, we can assert that
there is a probability that they will co-occur approximately
times in a similar
poor prediction anyway. The aim of studying language in corpora is to describe
might contain. In any case, Kilgarriff establishes firmly that probabilities between
corpora are not appropriate, no matter how similar the corpora.
All the familiar measures of significance, the chi-squared, the log-
simply descriptive they would not need to invoke probabilities. Corpus linguistics
needs its own methods of statistical analysis, which should be purely descriptive
and which should quantify linguistic concepts and categories.
The confusion may lie in the word
itself. A corpus is a carefully
selected collection of texts, involving a great deal of expert human judgement. A
statistical sample is "expected to be selected in such a way as to avoid presenting
a biased view of the population" (Wikipedia). These are diametrically opposed
concepts; nothing could be more "biased" in its selection methods than a corpus.
So perhaps no warning bells sounded when a corpus began to be treated as if it
were the kind of sample which is amenable to statistical analysis.
corpus. To which the obvious response is "So what?" – and we know that it is a
and explain the observed phenomena, not to predict what some other corpus
likelihood, the t-score, z-score and the poor MI are predictive; if they were
Borrowed Ideas
Once the tangentiality of our present measures of the relative importance
of text patterns is accepted in the corpus community, it will not be long before
alternatives emerge. It is too early to say in detail what is likely to happen, but a
concentration on the collocations and other co-occurring features, rather than on
their components, is long overdue and is a productive entry tactic.
Little boxes
The final section of the paper considers the use of labelled bracketing as a
notational device. In the section above on metadiscourse, it was stressed that
language text is linear and directional. Whether in speech or writing, only one
event is happening at once. Position in a linear string is potentially significant, in
that it can contribute to the realisation of meaning. Text is directional in that it
does not usually mean the same thing if reversed, if indeed it means anything at
all. The meaning does not necessarily survive any changes in the positioning of
the units of realisation.
All this almost goes without saying; Saussure ascribes two fundamental
qualities to the linguistic sign, its arbitrariness and its linearity, in successive
sections of the
(Saussure 1916). While much has been made of the first
quality over the years, very little mention is made of the second although it is at
least as important a quality; perhaps the fact that it applies to text –
parole –
rather than to
has relegated it to a secondary status.
Back in the early days of experimentation in computational analysis of
language, linguists led by Geoffrey Leech at Lancaster University were the UK
pioneers (and are still leaders twenty years later). During a presentation in 1984
Leech made a passing remark that their software worked better if the sentences
were input backwards.
It was light-hearted enough, but I could not get it out of
my head, and it is one of the starting points of the argument of this section. I
repeat the question I asked at the time – Why, if this is the case, do we not speak
For the present discussion, we need only note that Leech had detected
some imbalance of signs in the sequencing, such that in one direction the text was
easier to divide into meaningful segments. That in turn suggests some imbalance
at the boundaries of such segments. Text structure is not symmetrical, although
the descriptions tend to depict it as a construction of neatly nesting units. Several
years passed before I found a possible explanation for this phenomenon.
Labelled bracketing
First of all, let us study an example of the kind of notation that is at issue. Formal
grammars are often displayed as tree diagrams or similar networks, for ease of
understanding, but these are equivalent to the notational conventions of "labelled
bracketing". They conform to constraints like those of Halliday's "taxonomic
hierarchy" (1985) and so can be represented by strings of symbols in several
levels of bracketing. Symbols attached to the brackets themselves indicate the
structural value of the contents. As an example, let us consider Figure 1, which
applies a labelled bracketing to the following sentence.
The boy stood on the burning deck, whence all but he had fled.
[ [ [ [ the ][ boy ] ] [ [stood] ]
S MC NP det /det n /n /NP V v /v /V
[ [ [ on ] [ [ the ][ burning ][ deck] ] ] ] ]
A PP prep /prep NP det /det adj /adj n /n /NP /PP A /MC
[ [ [whence ] ] [ [all] [ [but] [ [ he] ] ] ]
SC A conj /conj /A NP pron /pron PP prep /prep NP pron /pron /NP /PP /NP
[ [ [had] ] [ fled] ] ] ]
V aux
/aux vb /vb /V /SC /S
Figure 1: labelled bracketing
: S= Sentence, MC= Main Clause, NP= Noun Phrase, det= determiner, n=
noun, V= verb, A= Adjunct, PP= Prepositional Phrase, prep= preposition, adj=
adjective, SC= Subordinate Clause, conj= conjunction, pron= pronoun, aux=
auxiliary verb
The structural description above is intended to be non-controversial, the
application of a consensus grammar to an often-cited specimen sentence. Notice
that between
, there are five brackets closed and three opened. I
presume that this is intended to represent in some way the structural perceptions
of the person originating the sentence and any person who experiences it as a
The question arises, how does the text signal to the reader these eight
pieces of structural information? The appearance of the word
tells us that, as
a noun, it is capable of carrying the role of headword of the noun phrase that we
know started with
two words earlier – but it does not settle the matter, because
could just as well be a modifier of another noun, say
tarpaulin, winch
, in which case the noun phrase [NP], the prepositional phrase [PP] and the
main clause [MC] would not be terminated immediately after
So the
information must largely arrive with the occurrence of
. This word is a
fairly reliable structural marker, telling us of the beginning of a subordinate
clause of place and therefore by implication the end of the previous one; if a new
Borrowed Ideas
clause has begun, and is not embedded in another clause, then it follows that the
previous clause must be terminated. And if the clause is terminated, then the
restrictions of the taxonomic hierarchy mean that all its component structures are
also terminated – the PP and its NP in this case.
The important point to note from Fig.1 is that there is no actual signal of
the end of the main clause or of its subsidiary structures, but there is a clear signal
of the beginning of the subordinate clause, which triggers all the closing brackets.
As to the opening brackets, since
is a conjunction we can open a
bracket for a clause and another for an initial adjunct, since it is a subordinator we
can open a bracket for a subordinate clause, and of course since it is a word we
can open a bracket for a word. The amount of structural information supplied by
the choice of
is very substantial.
The point where boundary information is the next most concentrated is
, where there are four closing brackets and three opening
ones. As a pronoun,
is almost certain to stand as a NP on its own, and this
would explain the closure of the first two brackets. The outer structures, however,
are not definitely closed by the occurrence of this word, because it could easily be
followed by
and the captain
, thus prolonging the prepositional phrase and
therefore the original NP starting with
The appearance of
makes it clear that the original NP and all its
components are terminated. The combination
he had
signals fairly definitely that
they are separated by a subject-predicate boundary, and thus justifies the last two
closing brackets, as well as opening a bracket for the predicator [V]. The
assignment of
to the status of auxiliary is not at this juncture clear, and is
only confirmed by the occurrence of
. The sentence could have continued "all
but he had guns."
Complete and finished
It is helpful at this point to draw a distinction between structures which are
and whose which are
A complete structure is one which is
well-formed and thus has meaning-potential, in Halliday's terms – it is an abstract
concept, not without problems but in many cases specification is straightforward.
On the other hand a finished structure is a segment of text which is actually
terminated. So in our specimen sentence,
the boy
is both complete and finished as
a nominal element, but in another text it could occur as a component of an
indefinitely large number of other nominal structures beginning
the boy…
The general point to be made here is that in understanding text the
movement from recognising a structure as complete to appreciating that it is
finished is largely a matter of hindsight – settled by the appearance of a word
which is incompatible with the evolving structure. The possibilities are:
1. complete and finished
2. complete but unfinished
3. incomplete but finished
There are no examples of possibility (3) in the test sentence, but they are common
enough in conversation; here is one:
A: I mean you know its not important its just er
B: What do you mean ……

(Francis & Hunston 1987 p.144)
A case can be made for
I mean
you know
to be freestanding discourse units
that do not require anything to follow them, but
its just
is clearly an unfinished
unit, cut off by the superimposition of B's question (though the tell-tale
suggests that A was glad not to have to finish the structure. Sinclair and
Mauranen (2006) offer an up-to-date analysis of such phenomena.
Of the thirteen words in our specimen sentence, only two include in their
meaning the notion that they are final in the phrase which they wholly or partly
realise. These are the conjunction
and the pronoun
. That is to say, it is
difficult to imagine any way in which the conjunction
could be
on the other hand can be qualified in limited ways (e.g.
he who…
but extensions such as these are very rare in normal texts. So the analyst can be
fairly confident in placing a boundary after
indicating that it is a full
elements of clause structure despite being a single word. This reflects the
expectations of readers of the original sentence. Note that this built-in boundary
marking applies only to single words and to a very small number of them, some
of the subordinating conjunctions perhaps. In the case of the other eleven words it
is the occurrence of the following word that establishes the boundary.
We can now revise Fig. 1 by suppressing all boundaries except those that
can be predicted before the next word occurs. To avoid clutter I am also removing
the word brackets, which are redundant structurally and only serve as positions
for the word class to be identified.
[ [ [the boy [stood [ [on [the burning deck
[ [whence] [ all [ but [he] [ [ had fled
SC A /A NP PP NP /NP V aux
Figure 2: reduced labelled bracketing
There are now fourteen opening brackets and two closing ones; twelve of the
structures are not explicitly signalled as being finished; they are simply replaced
by another structure. Therefore twelve of the closing brackets in the original
analysis must be inferred from hindsight, albeit momentary hindsight. There is
thus a large imbalance between the way in which the openings and closings of
Borrowed Ideas
structures are signalled, and I presume that Leech's team, back in the eighties,
became aware of this.
This finding can be supported and partly explained by theory. Chomsky
(1957: 23-25) pointed out a long time ago that the set of all well-formed
sentences of a language could not be limited. This property arises from the
declaration of iterative rules in the grammar; an iterative rule is one which has the
same symbol on both sides of the arrow of derivation. This kind of rule can be
applied over and over again, generating longer and longer sentences and never
reaching an end-point. Just one iterative rule would give the property of
limitlessness to the set of sentences generated by a grammar, but in a natural
language there is a range of iterative rules which extend sentences in a rich
variety of ways.
It follows from this property, also, that there is no such thing as the longest
sentence in a language, since any candidate could simply have an iterative rule
applied to it, or reapplied if it was present already in the phrase structure. Below
the sentence, indefinite extension is not guaranteed, but a moment's reflection will
confirm that in all cases where a complete structure contains more than one word
the same feature will apply; there are, as we have seen, a few candidates for
"terminators" among the conjunctions and perhaps pronouns, but those are
operating at the lowest level of structure.
Figure 2 is a fairly acceptable analysis of the sentence, in that it assigns
labelled brackets where these are clearly indicated in the text. However, its lack
of symmetry in bracketing would cause it to be rejected by any parser of the usual
variety; in fact one of the first checks of the parser is to count the number of
opening and closing brackets and ensure there is the same number of each. There
is a huge discrepancy here. Options for resolving the situation include:
a) introduce an automatic "bracket equaliser", which regularises the
notation. So after
(fig 2) a closing bracket would be added, for
example. I expect that this would present few problems, but it transforms
the structure into something that is not justified by the facts.
b) revise the underlying model, the conventions of bracketing, in such a
way that structures which do not signal their finishedness within
themselves, are simply left open.
Option b) is the closest to the facts, and it allows for the structure to be
determined further by punctuational or prosodic features.
The boy
, in the abstract
world of completeness, carries forever the potential, as a noun group, for being
continued in a number of diverse ways. In the actual sentence, this potential is
over-ridden by the appearance of
. Such a model will bring out the dynamics
of text, which especially in writing is often neglected.
It is an untidy model compared with the cool symmetry of logical
structures, but we are often made aware of how untidy language is when it is in
use in communication; so there is no surprise here. It will be interesting to see
how such a model will compare with the traditional forms of analysis; despite the
lack of any disagreement about the structure of the text, we have seen that there
are seriously different ways of representing that structure.
Punctuation and prosodic features
The lack of signalling finishedness does not appear to detract much from people's
ability to speak and write effectively, and to understand both modes of
communication. In writing there has developed a system of punctuation which
sometimes helps in boundary assignment, the comma especially, though it is
often ambiguous. We are accustomed to using fairly settled conventions of
punctuation nowadays, but these were gradually stabilised by the printing
industry, are thus of very recent origin, and cannot be considered essential
components of the language system.
The three levels of boundary that we have in focus are that of sentence,
clause and group/phrase. The tricky boundary is the lowest one, and the system of
punctuation while helpful, is not decisive. Also it does not explain why a reader
only rarely needs punctuation marks. In understanding a text a reader must make
some assignment of boundaries, perhaps subliminally.
Some punctuation marks signal that a structure is finished; so a full stop
would allow us to close the predicate, the subordinate clause, and the
sentence as a whole. It would be quite natural, though not obligatory, to place a
comma after
so that
did not have to carry all the large informative
load which is otherwise placed on it. The redundancy of punctuation allows a
sharing of the information load.
From the point of view of boundary marking, punctuation marks in
English support closings. If the initial capital letter of a sentence is considered
part of the punctuation system, then it is the only one that marks an opening.
According to present-day practice, punctuation is not permitted between elements
that form a syntactic unit below clause level, e.g. between subject and predicator
of a clause, or between a preposition and its object. The distinction between
defining and non-defining relative clauses follows this practice, in that the
syntactic unity formed by the defining clause cannot be interrupted by
The full stop, question mark and exclamation mark all indicate clearly the
end of a sentence. The colon and semi-colon indicate clearly the finish of all units
below the sentence. But the occurrence of a comma, while excluding some
options, does not unequivocally terminate all structural units that are open when it
occurs. It acts as a resolution of tension between the demands of linearity, which
includes some elastic limit on the size of meaningful segments, and the
realisations of the complex hierarchies of the grammar. The physical size and
length of the realisation of each abstract category is one factor in the balance; the
demands of the structure of which it is a part form another; the possibilities for
enhancement, extension, elaboration etc. ad infinitum form a third mechanism for
concatenating words and phrases together. Balancing this is the need to keep the
Borrowed Ideas
discourse in bite-sized pieces, so that a listener or reader will not get confused in
real time with the complexity of the message. So after five words or so, the
pressure will increase to place a comma at a boundary that in other circumstances
might not merit such a formal mark.
In the spoken mode, there are patterns of tone contour which again give
clues to when a speaker is finishing a structural unit. Brazil's (1997) description
gives a clear picture of the harmonious co-ordination of choices that is natural
Having got thus far, it is worth a few moments' attention to the details of
completeness. It is an intuitive decision, whether an evolving structure is
complete or not, and so may not always be clear or logical, but always indicative
of something in the structure that is worth consideration. An analyst is too
detached to offer a reliable commentary on intuitive matters, but we have to do
the best we can. It is a rough-and-ready decision-making process, far removed
from the "well-formedness" that formal grammars envisage, but using similar
Let us look at cases. Recall the first instance of a complete but not finished
structure above:
The boy stood
is an intransitive verb, this phrase is well-formed at some level of
abstraction, but it looks unlikely with these actual realisations. The incidence in
The Bank of English of a punctuation mark following
in a thousand
instances chosen without bias is 42, less than 5%. As used in this sentence, the
seems to require some adverbial element to terminate it. So the
argument for the structure being complete at this point is weak and not
The boy stood on
This is much less likely; although
now has an adverb, our intuitive feel for
the actual phrase is that
is a preposition, so we await the object; it is
incomplete. With a similar verb,
stayed on
, there is clear evidence that the
structure is complete, and of course
walked on, drove on
frequently as an
adverb. In the spoken language there would be a stress distinction marking the
difference, with the preposition unstressed, and there is a hint of irony in this
piece of doggerel in that
occurs on a stressed syllable in an iambic rhythm, a
trap for inexperienced reciters. There are no plausible instances in the Bank of
English of
stood on
finishing an active verb structure.
This is complete as an NP, but it is not finished. There are plenty of attestations in
the corpus for
realising the subject position in a subordinate clause of place, so
all but…
is also found it is much less frequent.
There are good reasons why a set of notation conventions based on Figure
2 should be developed; the directionality of text is prominent, structures like
discontinuity will be better described and there will be much more flexibility in
describing the way a sentence can develop. The symmetrical labelled bracketing
that we are accustomed to cuts off, quite unnecessarily, many developmental
options for the structures. With asymmetric bracketing we are much closer to
symbolising the textual signals.
The late, brilliant Nick Lafitte, who took up linguistics in flight from his previous
career in Econometrics, used to wonder (pers. comms.) why linguistic
descriptions, statements about language, were so different in their nature from
language text. His interest was aroused by redundancy, which was routinely said
to be a major feature of language text. Nick presumed from this that either the
structural representations of text would exhibit this feature, or the descriptive
categories. But there is no treatment of redundancy at all in descriptions or the
theories that lie behind them; it just disappears. Presumably this discrepancy is
not just a mistake or oversight, but rather that the ground rules of the theories and
descriptions preclude a feature like redundancy, despite its noticeable presence in
text. These ground rules are imported from outside, and perhaps they are not quite
right for the job; we frequently say that language is unique because its theories
are also written in language (see "meta-" above) but we then describe language
phenomena without taking advantage of this coincidence.
Something of the same can be said of linearity, which is so obviously a
major constructional feature of text that it is usually taken for granted. It is a pity
that we lose sight of it, because it could act as a brake against over-indulgence in
abstractions and over-complex representations of simple phenomena. In most
descriptions linearity all but disappears in favour of multi-layered hierarchies,
which do not always seem to be strongly motivated. A recent study (Sinclair and
Mauranen op.cit.) seeks to describe language text while maintaining linearity for
as long as possible.
My framework for comment on the above issues is the network of
relations between text and meaning, because those relations form the apex of
language description. I find that it is all too easy to mix up conventions of
practice and properties of the data (e.g. sample size), and we must be vigilant in
protecting research against vested interests. I find that we sometimes import
terminology from other disciplines without sufficient care, especially if it sounds
good (e.g. meta-); ill-fitting terminology can certainly distort your thinking. I find
that it is not appropriate to measure the salience of patterns of combination in
texts by means of predictions concerning the distribution of the component word
forms (conventional corpus statistics). New ideas are needed here. And I find that
the notation of labelled bracketing, almost universally accepted in formal
grammars, is a faulty representation of text structure; like poor terminology, it
Borrowed Ideas
impedes clear thinking, and has inhibited grammarians from developing notations
which better represent the texts.
There is a well-established branch of computational linguistics whose
techniques are developed for determining authorship from internal textual
evidence; these produce genuine metadata (see JLLC,
). However
their results are of course only postulates, and can be discussed and
disputed with arguments that are quite different from the way in which
arguments about external evidence are conducted.
This is a big topic, too big to pursue in this paper. The written form of the
language has two distinct functions; one is to be read, in which case I
would claim that readers behave quite like speakers in that they do not rely
on the possibility of returning to an earlier state of the text. The other
function is to make and keep records, where the text must carry all
necessary detail and must cohere, avoid ambiguity and the like. While
these are distinct functions, a user in reading mode who gets into
difficulties can switch to record-keeping mode in order to resolve the
problem. In ordinary, everyday reading this does not seem to be a common
An early treatment of this point is to be found in Sinclair (2004 (1982):
Op.cit.: 268-270. Actually, Kilgarriff does not conduct a straightforward
comparison of his two selections from the same large corpus, but
associates the words with POS tags, which complicates the issue
considerably, and makes it much less likely that the samples will match.
He does not discuss his motivation for doing this.
I say "poor MI" because, in my limited experience of significance
measures, MI is the only one which has to have both its head and tail
chopped off before it makes sense. It is not difficult to understand that the
tail of a significance measure gets less and less interesting as it goes down,
and that a cut-off point is desirable in practical applications; however, to
remove the items that are shown to be the
significant is bizarre. I put
this point to Kilgarriff at TALC in Lancaster, 1994, but he did not address
At the ICAME conference, May 1984, in Windermere, hosted by
The conventions are that below each opening bracket is a symbol
designating the structural value of what is inside the bracketed segment;
below each closing bracket is the appropriate symbol prefaced by a
diagonal slash. S=sentence, MC=main clause, NP=noun phrase,
det=determiner, n=noun, V=verbal element, v=verb, A=adjunct,
PP=prepositional phrase, prep=preposition, adj=adjective, SC=subordinate
clause, conj=conjunction, pron=pronoun, aux=auxiliary verb.
This contrast cuts across the hallowed distinction between langue and
parole, competence and performance. It first came to my notice as the
origin of textual effects in the analysis of Wordsworth's poetry (Sinclair
1972). In being exploited for stylistic effect, I used the terms
(structure incomplete, new structure initiated) and
complete, more material added without initiation of a new structure). Very
recently, in working on linearity in grammar, I have returned to the
distinction because it plays an important role in
(see Sinclair and
Mauranen 2006).
The Bank of English offers 1755 instances of "he who" or "he whom", of
which quite a number show the pronoun followed by a defining relative
clause. But these are characteristic of certain marked text types, from the
domain of religion, or faked antiquity, gnomic utterances or just a
particular pomposity. In cleft structures like "It was he who answered." a
conventional analysis would have "he" as the complement on its own.
Metrists might reasonably argue that the placement of a line-end after
is equivalent to an auxiliary punctuation mark.
Ädel, A. 2006
Metadiscourse in L1 and L2 English
. Amsterdam and Philadelphia,
John Benjamins.
Austin, J. 1962
How to do things with words
; ed. J.Urmson, Oxford, Clarendon
Brazil, D. 1997
The Communicative Value of Intonation in English.
Edition]. Cambridge: Cambridge University Press.
Chomsky, N. 1957
Syntactic structures
, The Hague, Mouton.
Francis, G. and S. Hunston 1987 "Analysing everyday conversation" in
Coulthard, R. (ed.)
Discussing Discourse
, Birmingham, ELR Monograph
no. 14.
Halliday, M. 1985
Introduction to functional grammar
, London, Edward Arnold.
Imbs, P. and B. Quemada 1988 ,
Trésor de la langue française
, Paris, Gallimard.
Kilgarriff, A. 2005 "Language is never, ever, ever random" in
Corpus Linguistics
and Linguistic Theory
1-2, 263 – 275.
Krishnamurthy, R. ed. 2004 (1970)
English Collocational Studies
by J. Sinclair,
S. Jones and R Daley; London, Continuum.
Borrowed Ideas
Léon, J. 2005. "Claimed and unclaimed sources of Corpus Linguistics",
Sweet Society Bulletin
. N°44. pp. 36-50.
Saussure, F. 1916
Cours de linguistique générale
compiled by C. Bally and A.
Sechehaye, Paris: Payot.
Sinclair, J. 1972 'Lines about Lines', in B. Kachru, B. Stahlke, F. W. Herbert
Current Trends on Stylistics
, Edmonton, Linguistic Research Inc.
251-61, reprinted in R. Carter (ed.)
Language and Literature
, London:
AlIen & Unwin 1982, 163-76.
Sinclair, J. 2004 "Planes of discourse" in
Trust the Text
, London, Routledge, 51-
66 (reprinted from S. N. A. Rizvil (ed.)
The Two-fold Voice: Essays in
honour of Ramesh Mohan
, Pitambar Publishing Co., India, 1982).
Sinclair, J. 2005a "Language as a string of beads: Discourse and the M-word" in
E. Tognini-Bonelli and G. Del Lungo Camiciotti (eds.)
Strategies in
Academic Discourse
Amsterdam and Philadelphia, John Benjamins;
Studies in Corpus Linguistics 19, 163-8.
Sinclair, J. 2005b "Corpus and Text – Basic Principles" in M. Wynne (ed.)
Developing Linguistic Corpora: a Guide to Good Practice
Oxbow books. Also available in electronic form:
Sinclair, J. and A. Mauranen 2006
Linear unit grammar – integrating speech and
, Amsterdam and Philadelphia, John Benjamins.
Tadros, A. 1985
Prediction in Text
; Birmingham, ELR Monograph no.10.
Trask, R. 2000
The Penguin Dictionary of English Grammar
, London, Penguin
Bank of English, The
Brown corpus
Index Thomisticum
LOB corpus
How 'systemic' is a large corpus of English?
Robert de Beaugrande
That both the text and the discourse are 'systemic' entities, is, I believe, a principle of
general consensus within Systemic Functional Linguistics, although in a sense related yet
distinct from the 'systemic status of a 'language'' plausibly following the distinction
between 'actual' versus 'virtual' (or 'potential', to use a less overloaded term). Now that
very large corpora of authentic text and discourse are readily accessible, we can take up
the question of whether such a corpus may in turn be 'systemic' in a sense mediating
between the poles of this distinction: both 'intersystemic' and 'intertextual' at once. The
present investigation adduces newly extracted corpus data to answer this question in the
affirmative, notably by demonstrating how these data project and confirm 'systemic'
tendencies exerting pressures that can modify, expand, or alter the language system itself.
Indeed, such demonstrations might pass unnoticed if these very factors did not guide the
methods of search and retrieval I deployed.
Language and text as 'system' and 'systemic'
Among the foremost achievements of 'systemic functional linguistics' (SFL) has
been to expound an alternative view of the relation between
(e.g. Halliday 1992; Martin 1992); intentionally interrelated texts can be said to
constitute a
, the most common of course being a conversation.
A 'language' is a
potential system
; a 'text' is an
actual system
. Thus, a
complex process of
is implicated in the production or reception of
any text. By operations of selection and combination, a set of
choices become a set of
choices, and the relation between these two
sets is
. Along the way, the
systemic function
of any given
expression can be reset: adjusted, specified, weighted, colligated, collocated, and
so on.
Prior to SLF, the trend in language studies and linguistics often was to
take the process of actualisation for granted and begin the 'investigation' with a
handful of samples invented by the investigator, doubling, one might say as
actualiser. Since the language was assumed to be uniform (heterogeneous) across
an entire language community, the identity of the actualiser was judged
immaterial for the 'analysis'. This expedient logic was turned back to front and
made circular: those aspects or features of language were "investigated" which
were judged the most uniform, the rest being airily relegated to pragmatics,
stylistics, rhetoric, sociolinguistics, or whatever seemed most opportune for the
De Beaugrande
evasion. The "speaker" was "idealised" and "homogenous" into the blandest
possible human, who never said anything worth hearing.
I shall argue that the gap between language and text might be most
effectively relaxed by a
very large corpus of texts
, whose ability to widen the
coverage of the actualisation moving from potential to actual, and the
representative status of the population of actualisers, increases and diversifies
with size and balance. Admittedly, no corpus of any size can lead us to a
complete display, but we have repeatedly learned that as our corpora grow, the
picture becomes sharper and more facetted, often unveiling curious surprises:
nouns whose plural does not match the meaning of the singular, verbs whose
passive does not follow from the active, and so on. In essence, these insights
contribute to the deconstruction of the premature hypothesis of uniformity and its
centrality I have cited. They demonstrate the vital interaction between grammar
and lexicon well beyond how they are conventionally conceived.
However, to my knowledge, the complex process of actualisation has not
been thoroughly analysed as such — and not just its initiation as a system
property (e.g. speech enunciation) and its termination as a text property (e.g.
sentence analysis). The outstanding question remains how and how far
participants in communication — whose knowledge of the potential system
be uniform — perform the process with sufficient (though hardly
complete) uniformity to arrive at the condition reassuringly known as
'understanding'. The answer I have repeatedly suggested in my books is that the
process is so devised as to favour a temporary 'tuning' which can not only occur
during the overall process of actualisation, but must also be prefigured and
supported in the design strategies of both language and text. To revise the familiar
terms, the 'process' aims at an interactive and convergent constellation of
outcomes in the 'product', especially by the multifarious ways in which some
choices or sets of choices raise or lower the probability and suitability of other
choices (cf. Halliday 1992).
Intersystemic processes between language and discourse
Provisionally, then, I shall propose ten
processes of actualisation
note some data where dynamic pressure and evolution seem to be still operative,
taken from my own English Prose Corpus (EPC, 100 million words of 'classic'
texts), the British National Corpus (BNC, 100 million words of contemporary
texts), and the Internet accessed via AltaVista (uncountable words in on-line
texts, annotated with
). These processes have been implicated in the
educational goal of
of English, mandated in the Tories' cloud-
National Curriculum English
(1988) to quite disparate degrees, if
at all (Beaugrande 2004), which is like a road map grandly displaying only the
goals but not the roads for going there.
How 'Systemic' is a Large Corpus of English?
Among the most capital errors of the "generative" approach and its
epigones has been to situate
as prefigured in the system,
whereas in fact it can only be the textual product of
'Grammaticality' thus can only be accessed via texts, however bland and banal,
after the process has been achieved. I have never seen or heard a convincing
example of 'grammaticality' or 'ungrammaticality' in the potential system. The
flurry of 'sentences' conjured up to demonstrate the ostensible borderline were
simply instances where grammaticalisation has been either trivialised [1] or
vandalised [2].
[1] John knew what Mary was doing (Annie Tremblay)

[2] *Which room is there in a very strange beast with enormous antlers and
five arms? (Joseph Sabbagh and Lotus Goldberg)

The jammy sun-drenched students at the University of Manoa in Hawaii were not
supposed to draw the immanent inference that Mary is up to something naughty
or sneaky (or worse) [1]. Still less were the frost-bitten students at MIT and
McGill supposed to notice that the freak anatomy of the 'beast' smuggles in non-
grammatical unacceptability [2].
The ongoing dynamics of grammaticalization can be documented by the at
times wilful conversion and confabulation among word classes:
[3] The principle of inverse irreversibility. An inquest into scientific
methodology, from the Popperian hypotheticodeductive perspective, with
Kuhnian paradigmatic nonreconstructionism to Feyerabendian
counterinductivism (New Scientist)
[4] Cyclosporin A, a cyclic undecapeptide, is a potent immunosuppressant
that binds to a peptidyl-prolyl cis-trans isomerase, cyclophilin.The
cyclosporin A/cyclophilin complex inhibits the calcium- and calmodulin-
dependent phosphatase, calcineurin. (Nature)
Although the marked items are lexical items in an ordinary sense, the only way I
can see to understand them is to run a grammatical back-analysis into the
A quite different clue of dynamics is the rise of items whose grammatical
functions are not accountable in grammar-books. The adverb 'so' is placed and
given stress as an intensifier in positions it would not usually occupy:
De Beaugrande
[5] Well I'm just so going to bed now (Xanga)WWW
[6] And I'm just so going guy crazy and I want one (MySpace)WWW
One wonders how you 'so
go to bed' – taking a ferocious running jump, perhaps.
But how 'so
going guy crazy' differs from 'going so
guy crazy' remains for me a
mystic rite of teenagerhood.
Another grammatically displaced item recently emerging is the negative
'not' postponed after an (ironically flavoured) positive statement, viz.:
[7] why do these old men discover they can get new careers lumbering ape-
like through rock history. One awaits the geriatric Techno of the next
century with interest. Not. (New Musical Express)
[8] James Baker III and the seven dwarfs of the "Iraq Study Group" have
come up with some simply brilliant recommendations. Not. (Columbus
Free Press)
Yet the risk of being shouted down as a gormless berk before you get to utter
your Negative may not be a trivial one.
Though it is equally essential, the process of
, which
engenders the
of the text-system, has been relatively neglected,
presumably because it would not fit well either into the tinker-toy formalisms of
trees and networks of "formal grammar" or into the cut-and-dried "lesson plans"
of the "standardisers".
Yet even the most freeze-dried "grammarians" could not deny that their frail
constructions are like hollow eggshells without Lexicalisation to bring them to
Another disquieting factor is that Lexicalisation is far less stable and more
dynamic than Grammaticalisation. Discourses about computer construction and
high-tech race-cars are rife with lexical items hardly one would have understood
some years back and perhaps few enough would today:
[9] The nForce 680i SLI board is eVGA's flagship motherboard with support
for the Core 2 Duo and Quad that uses our new chipset (Morry
How 'Systemic' is a Large Corpus of English?
[10] Enhance the under-hood appearance of your V [V-8-engine block]. A
lightweight carbon-fiber tower-to-tower brace replaces the standard steel
brace. In addition, the radiator and engine cover features an attractive
carbon-fiber appearance.
The dynamics are also readily manifest in the spreading of non-technical
expressions. One amusing source can be regional varieties of English with a
phonetically ironic side-effect, such as Irish English 'gackawacka' for a tiresome
fool [11] or African-American English (or Ebonic) 'badonkadonk' for a
curvaceous female behind [12], reputedly cloned from 'honky-tonk', a music to
which such an, erm, asset can be paraded to eye-catching advantage
[11] How could she admit how silly Sarah always made her feel? "Oh, let me
write her a reply, oh do!" Aislin said, "We'll pay back this gackawacka
for all her stupid blather." (
Maureen Monahan

[12] How J-Lo [Jennifer Lopez] squeezed her badonkadonk
into this Zum
Zum [trendy fabric] wonder is damn impressive (
Carpet Burn

University student discourse is renowned for droll contributions like 'borassic' for
broke [13], and 'trollied' for drunk beyond control [14].
[13] Saz is borassic
and will be until she gets a job after graduation. She needs
money for essentials like a Kylie programme on Saturday. (
[14] Due to being so trollied
on his birthday in Copenhagen, Axel asked folks
how to get to Amsterdam Centraal (not only wrong city... wrong
country!!!). (
Team Plastique

So intricate is the interaction between Grammar and Lexicon that it would often
be more apt to use the systemic term
that engenders
of the text-system. But consistent use would be
cumbersome, and we might be content to deploy the other terms as 'short cuts',
keeping in mind we are doing so (Halliday 1994).
The vibrant dynamics of the Lexicogrammar are best shown in its creation
(habitual grammatical combinations), as in [15], and
(habitual lexical combinations), as in [16].
[15] Author of Bush Biography Commits Suicide. If
, I've
some ocean front property in Arizona I'd like to sell (
Wake Up

[16] Sometimes I come here by myself, just to relax, think things over. By
yourself?" Kelly narrowed her eyes with mock suspicion. 'The eligible
bachelor on
?' (
Stone Cold
These combinations lack the fixity of traditional "idioms", which all too easily
mutate into clichés; and some elements are more variable than others. The jibe of
De Beaugrande
clueless credulity in [15] is fairly stable in the conditional "if"-clause, but the
variety of worthless or jocular scams, usually land or bridges, may vary
considerably, even to "property on the moon".
Complex systems generate a certain chance margin, where Collocations
may seem to abruptly and unwantedly revert to their basic lexicality:
[17] Your face can open doors (
BBC Online

[18] Jury is still out on composting toilets (
Salem Statesman-Journal
As factors within the doughty project of the standardisation of English as a
foreign language, Grammaticality is plainly more cultivated than Lexicality in the
discourse of learners; probably their routine dissociation during "instruction" is
responsible. Some data from my students at the United Arab Emirates University
seemed Grammatically regular but Lexically bizarre, as when I was struggling to
convey the Medial Transitivity in English whilst they struggled in turn to model it
on the Passive as the familiar alternative to the Active in Arabic:
my sample text
: Mrs Bennet fidgeted about in her chair, got up, and sat
down again. (
Pride and Prejudice

student responses:
The chair was fidgeted up and down by Mrs Bennet.
Getting up and sitting down got fidgeted in her chair.
Mrs Bennet's chair was fidgeted about, was got up, and was sat down
By contrast, other Arab student data seemed both Grammatically and Lexically
[20] Miss Raymond looks smelly [smiley] face but speaks in pride ways. She
collects her hair in the back. Her teeth look when she talks, and she owns
angry tone. She is a liar person who lied to disappear her ignorant.
[21] If anyone dressed by the name footman he will be shame that they don't
even want to wear their clothes. In the US was not respect and tricker
man and swindle person.
is a vital process for engendering the
of the Text-
System, whose neglect – aside for the pronunciation of individual sounds and
words -- at nearly all levels of education and research would be astonishing, were
there not doctrinaire motives for eschewing it.
How 'Systemic' is a Large Corpus of English?
It is far harder than Grammar and Lexicon, however construed, to standardise by
'rules' or judge for 'correctness' or 'rightness', much less fix in a 'grade'.
Depending on personal interpretation, either [22-22a] or [23-23a] would be quite
acceptable to me. (Hollow arrows show pitch contour; thick filled arrows show
strong stress; thin filled arrows show weak stress; upright bars indicate a pause
[cf. Beaugrande 2004].)
Some sequences allowed by the Lexicogrammar seem Prosodically unappealing
[24-25] (Slovene student data):
[24] The sea floor is in closer to the shore solid.
[25] On the slope of Cape Roenk, typical sub-Mediterranean species, despite
the fact that it has northern position and that the substratum is less
flyshy, live.
This is because Slovene tends to complete the Subject-Predicate connection at the
end of an Utterance.
The least studied and taught is the process of
that engenders
of the Text-System.
Written language is after all designed to be looked at, from the ornate parchments
of the high middle ages and the renaissance to the flashy websites of today.
Internet browsers also allow for the easy transmission of photographs that
complement and expand the significance of the written text, as in this report from
Der Spiegel
(English edition, June 2006) on the 'gigantic orb' placed slap in front
De Beaugrande
of the famous Brandenburg Gate to commemorate the World Football Cup hosted
by Germany:
[26] The massive sphere glows an eerie blue at night and its gaping maw
seems to swallow up people as they march inside. But what looks like a
scene from a cheap sci-fi flick is something more nefarious than aliens
enslaving mankind.
Unfortunately, the organisers went over the top by launching a barrage of
fireworks whose smoke enveloped the square within seconds and reduced
visibility to nothing, offering a test of what Berlin was probably like when the
Red Army came for a visit in 1945.
How 'Systemic' is a Large Corpus of English?
The very symbols of modernity and unity are discursively metamorphised by this
Visuality into an atavistic and sinister ambience with harsh historical overtones.
These, then, are the four systemic processes of Actualisation I hold to be
indispensable for the creation of Texts. The fact that they have been so unevenly
explored or addressed in projects of standardisation suggests why some
significant issues have barely been raised; and why so many non-natives well
versed in Grammar and Lexicon somehow still do not sound or write like native
Moreover, further processes (or meta-processes) are implicated which
apply to the organisation and evolution of those expounded so far and which, to
my knowledge, are largely missing from both research and pedagogy, because
they are generically inimical to the static notions at the centre of their enterprises.
The process of
engenders the
in or among Text-
Systems, that is, the extent or reach of regularities such as a large corpus can
On the whole, once this process takes up some domain of a system, it seems set to
run its full course, sometimes altering whole patterns or 'paradigms', as if
speakers were subject to its will rather than the reverse. Grammatically, the
formation of English Plural with '-(e)s' not merely managed to displace older
De Beaugrande
formations with stem vowel shift or with '-(e)n', but continues to emit
spontaneous new Plurals, sometimes more than one, viz.:
[27] Use the criterions of deployment or a readiness exercise […] and
sustainment based on functional inspection criterions (
Air Force
[28] Student was weak in two criterias of the exemplary performance for this
objective. (
Baylor University

[29] Use of datums in product definition is required in order to specify part
features that are used as a basis for functional relationship with other
features. (

[30] I am having a problem to get all datas into one table. I must get datas
from 2 other tables. (
By contrast, the formation of the Past Tense of Verbs with the Ending '-d' or '-ed'
or '-t' did not generalise quite so thoroughly and so has left behind some 'non-
standard' detritus which rouses the ire of schoolteachers:
[31] Mostly, we jes clumb up on the shed top, inna shade of a tree, and passed
the time (
Zeke & the Hoss-Puppy)
[32] That iijit [idiot] Frenchman got tryin some fool trick walking a timber
stick and got upsot into the wet. (
Man from Glengarry)
[33] All of us 'fans' ranned outside and we saw him running to the bathroom!

[34] He was teaching his boy Melvin how to play some baseball so he stolt
this baseball bat off the churchhouse softball team. (
Digging Postholes)
Teachers fail to appreciate that the alternates they call 'wrong' or 'bad English' are
consistent with the system of a regional English and thus resist extirpation. Such
holds especially for the universal Negative 'ain't', e.g., for 'isn't', 'haven't', and
'didn't' [35-36].
[35] She ain't
exactly my girlfriend, but we spend loads of time together. I
asked her if she's my girlfriend (
Billy Bayswater
[36] The police shoot them three fellas, but they ain't
get Alfred. (
Seeing in
the Dark
A converse systemic process is
, which engenders the
in or
among Text-Systems.
How 'Systemic' is a Large Corpus of English?
It operates especially in domains of instability and complexity of a degree not
typical of English in general, such as the subsystem of Pro-Nouns, which alone
sustain formal distinctions in 'Case' and 'Gender'. The flock of variations I
encountered suggests speakers being aware that forms differ but finding them
tiresome to 'use correctly' (cf. Beaugrande 2007). For example, I found
Possessive Pro-Nouns occurring in the Subject or Object Forms, e.g.:
[37] Me
mum and me
dad are separated, like, and me
dad reads
The Sun
[38] I remember this song 'Shaddup You Face' [by Joe Dolce], spawning the
annoying catch-phase as a response to almost anything. (
[39] Dude, he
face is alright, if you just glance at him you can tell right away
it's Sheva [soccer star Andriy Shevchenko]. (
Soccer Gaming

[40] In the illustration she hair looks dark. I think they did a perfect job

[41] I'm not fond of Carlisle. We took us caravan up that way a few years ago
BNC data
[42] And we wont stop till we have 'em puttin' they feet in they mouths
Reflexives, which are doubly coded for number, turned up a veritable zoo of
Variations, e.g.:
[43] I like to think of myselves as a catalyst for innovation (
[44] new members introduce yourselfs here with a bit of info on yourselfs
[45] Stewart found hisself with his back to goal, layed it neatly back to
Edwards who
it into the far top corner of the net
Birmingham City)
[46] she was also surprised to hear it; she had never thought of herselves as
strong (
The Valkyrie)
[47] As the security forces transform and rid itsselves
of the baggages of the
apartheid past, they will be able to sufficiently fight crime (Thabo Mbeki
The Mail and Guardian

[48] We see ourself
as the biggest club in Britain, with a stadium to match
(Mark Hateley of Queen's Park Rangers in

[49] we want to profile ourselfs as 'The Best Grower of Thomson Seedless
Grapes In The Country'. (
[50] Willie Calder wants to know if anyone recognises theirself in this Class
Photo from the Gravesend Sea School, 1957 (
Merchant Navy
51] They have the third biggest city to theirselfs and are the only team within
a 90 mile radius (
Soccer 24/7)
De Beaugrande
Creative Variations in the Lexicogrammar were already briefly mentioned.
Threats of punishment for some mistake or neglect can be expressed as
whimsically misappropriating some piece of the hearer's anatomy:
[52] By the third day I expect third-years to work alone, and if you slip up,
gal, I'll have your guts for garters
! (
Hospital Circlers
[53] That's a nice bit of double-barrelled lying. Quick. Out with it, or I'll have
your skin for a cigar case
. (
First of Midnight
[54] A secretary whips away the remote. 'Keep that handy', I warn him, 'or I'll
have your head for a hat-rack
.' (
The Dyke & the Dybbuk
[55] Matt put in a warning. 'Just let Bill hear you say you're the hostess and
he'll have your ears for horse blinkers
.' (
The 'smash and grab raid' as the most primitive robbery from British shops, as in
[56], is varied for occasions that are sometimes more similar, e.g., a police raid'
with 'sledgehammers' [57], and sometimes less so, e.g., lively sports [58] or
pigging out [59].
[56] The thief […] smashed
a hole in the shop window using a hammer and
about £3,000-worth of gold jewellery before making off on foot.
East Anglian Daily Times
[57] A major heroin dealing ring was believed smashed
today after police
made a series of raids in Liverpool. […] 55 officers, some carrying
sledgehammers, launched their 'smash
' raids
on homes in the
Everton and Kirkdale areas. (
Liverpool Daily Post
[58] At the County Ground, it was daylight robbery; a smash
Charlton. They had 3 attacks and scored 2 goals. (
television news

[59] At British Petroleum's annual meeting last year, there were protests about
a 'smash
' raid
by one group who scoffed too many sandwiches.
Daily Telegraph
Perhaps the most dynamic Variation of our times is occurring in Prosodification,
namely the wholesale spread of so-called
Estuary English
outwards from London,
South East England, and the 'estuary' of the river Thames:
The pronunciation of British English is changing quite rapidly. Estuary
English may now and for the foreseeable future, be the strongest native influence
upon RP [Received Pronunciation]. For large and influential sections of the
young, the new model for general imitation may already be 'Estuary English',
which may become the RP of the future. (Rosewarne 1984)
Significant numbers of young people see Estuary English as modern, up-
front, high on 'street cred', and ideal for image-conscious trendsetters. Others
regard it as projecting an approachable, informal, and flexible image. (Coggle
I can't say if the trend reflects any deliberate defiance against the image of
the 'upperclass twit', but it would be far safer than speaking Received
Pronunciation in urban centres like Soweto or Kingston Jamaica.
How 'Systemic' is a Large Corpus of English?
The systemic process of
engenders the
in a Text-
System where much can be said with quite modest resources.
Among the most economical patterns in the lexicogrammar, which has
nonetheless been snubbed by most grammar books, are Non-Clauses lacking
Subject or Predicate (Beaugrande 2007), e.g.:
[60] John Major is now being exposed for what some of us always warned
that he was. A
. A
. A
. A
. (
Daily Mirror
[61] The wedding would come off right enough but the reception would linger
on night after night. Yeah
. Aye
. Singing
. Drinking
. Oh indeed
. (
History Project
At most, grammar-books treat them as 'elliptical' versions of full clauses, which
for mysterious reasons have not been uttered. (Ignorance? Laziness? Bad
manners? Laryngitis?) In order to avoid recognising Non-Clauses,
Comprehensive Grammar of the English Language
(Quirk, Greenbaum, Leech,
and Svartvik 1985: 889f) cobbled together a Rube-Goldberg conversion device
with no less than seven mechanisms of "ellipsis" with steadily diminishing
certification — "strict", "standard", "situational", "structural", "weak", "virtual",
and "quasi-".
Lexically, spoken English manifests a trendy tendency to shorten lexical
items down to single syllable [62-67], even if the result yields the same item
standing for different sources [65-67].
[62] Okay will you excuse me, I'll be back in a mo
[moment] (
County Council

[63] I got kissed lots these hols
[holidays], how about that? (
The Prince
[64] Poor old Johnnie Ray. They tagged him the Nabob Of Sob, the Prince Of
Wails and the Cry Guy. […] The girls thought he was brill
. [brilliant]
New Musical Express
[65] It was the bottle that first got Frank into diffs [difficulties]. He goes on
binges. (
Diamond Waterfall)

[66] I used a code compare program to see the diffs
[differences] between the
two files. (

[67] Fitting higher gear diffs
[differentials] will improve the top speed of the
slow revving engine. (
Know Your Land Rover
Phrases too get clipped:
De Beaugrande
[68] Iris washed her hands and filled the kettle. 'Might as well have a cuppa
while you're waiting.' (
Finishing Touch
) [cup of tea]
[69] When we lost 5-0 at Liverpool a couple of weeks ago we all got together
and had a heart-to-heart
. We sorted a few things out (Andy Thorn of
Crystal Palace in
) [heart-to-heart talk]
The process of
(despite the brittle term) merely engenders the
in or among Text-Systems.
Frequency data are problematic if not paradoxical in a recurring sense for data-
based research on language and texts: the more straightforward the extraction of
data, the less so is its interpretation. Frequent data should signal Generality, but
may merely be trivial for items like 'of' and 'in' that are so multi-functional.
Queried for the basic Verb forms for the jolly old 'five senses', the BNC
returned 'see' at 115,100 occurrences, 'hear' at 13,079, 'touch' at 2431, 'smell' at
1108, and 'taste' at 672. The disparities are striking, but cannot be uncritically
adduced as evidence that sight is many times over the most important sense.
Numerous attestations of 'see' are barely related to vision, such as 'understand'
[70], 'consider' [71], or 'unmask' [72]. If you 'don't see' something, you can mean
it doesn't or won't happen [73]. And so on.
[70] 'You see
— 'he began. 'I do see
. I have seen
for two years. I see
why you
have come here, penniless.' (
Longest Journey
[71] Stanley sees
Blanche as a threat to his marriage and his affection for
Stella (
School essay

[72] For the first time she saw
his mask of precocious intelligence
[73] He's a nice boy, but I don't
him making it to the top. (
Some frequencies may appear so general as to be unaccountable. I once made an
extensive tally of male and female Pro-Nouns in two corpora installed in
concordance programmes and differing quite markedly in sources, text sizes,
discourse producers, and dates. My EPC was then at 392 complete 'classic' texts
from 253 writers since Shakespeare; the BNC had 4214 partially incomplete
contemporary texts, including a laudable contingent of transcribed spontaneous
Using the subcorpora of 'literary' works in the EPC, back then with
23,936,418 words in 303 texts, I queried the Female and Male Subject Pronouns
'she' and 'he'. I found the proportion of 184,961 for 'she' and 331,595 for 'he' —
55.77% Female to Male. I then ran the same queries on the BNC and found 'she'
How 'Systemic' is a Large Corpus of English?
at 352,872 and 'he' at 640,736 — 56.26%. The close fit was astounding. A bit
unnerved, I calculated the totals for
Female Pronouns and all Male Pronouns.
In the EPC, I got 389,007 Female and 691,387 Male to yield – yes, 55.77%; in
the BNC I got 658,965 Female and 1,204,215 Male to yield – yes again, 54.72%.
Feeling my workstation fading over into a peak in Darien, I did the same
searches on all the comedies of Shakespeare and got 2263 Female and 4090 Male
— 55.33%. A strange attractor?
I experimented with various selections to see if the uncanny consistency
could be disturbed. In a mini-corpus of three Jane Austen novels, Males were
only 76% as frequent as Females. But when I added just three novels by Dickens,
two by P.G. Wodehouse, two by Virginia Woolf, plus one each by Fielding,
Thackeray, and Oscar Wilde, the figures came out 34,978 Female and 64,133
Male — a ratio of 54.53%.
On my next computation, I calculated the ratio for three female authors
programmatically writing about women and their concerns: Margaret Oliphant,
Mary Wollstonecraft, and Charlotte Perkins Gilman (just under 300,000 words
together). There, the Female Pronouns (5817) were 232% as often as Male
(2602). But I only needed to add in the voluminous, mild-mannered Bullfinch and
Pepys and I got figures of 8,985 and 16,600, with the Females smack dab at 54%.
I confess myself wholly at a loss to explain such precise generality among
frequencies; and the mystery is all the deeper, given the sizes and diversity of the
data sets. Any assistance would be highly appreciated.
The systemic process of
in or among
It could plausibly be regarded as the human correlate of
, a factor that
information theory dryly proposed to extract directly out of "transition"
frequencies, which tear to shreds any conception of 'context'. In discourse,
probabilities become active only in so far as they guide predictions.
The Collocation 'I'll see you' is predictably followed by an Adverbial of
Time, the most frequent in the BNC being 'later' (79 occurrences), e.g. [74],
'tomorrow' (34), 'again' (17), 'then' (15), in the morning' (15), 'tonight' (7), and
'next week' (7). As such, it can serve as a convenient salutation when parting [74].
Unpredictable continuations can serve as defiance [75] or threat [76].
[74] 'I'll be off, then.' I said: 'Goodbye.' 'Yes', he said. 'I'll see you later'. (
[75] 'I will trouble you to hand over that purse of gold you had saved to pay
for my head.' 'I'll see you hanged first!' raged the Bishop. (
Robin Hood
De Beaugrande
[76] 'Before I let you foul Walter's memory, I'll see you in hell!' she yelled.
Posthumous Papers
If you encounter the Verb 'remanded', you can fairly well predict the next words
will be 'in custody' — 273 attestations out of 405 in the BNC, e.g. [77]. 'On bail'
lagged behind at 42 attestations, e.g. [78], which says a lot about British justice,
whose prisons even Tony Blair admits are 'full to bursting point' (BBC Online). If
you read someone 'was remanded to Crumlin Road' [79], you assume (even if
you've never been to Belfast) he was packed off to a jail there and not ordered
just to loiter on the same street or camp out there 'until June 11'.
[77] Horace Notice, a former British and Commonwealth heavyweight boxing
champion, was one of four men remanded in custody last night for a
week accused of rioting at an acid house party. (
[78] Three workers from McDonalds restaurant appeared in court yesterday
charged with making a hoax bomb call to rivals Burger King. They were
remanded on bail until March 11. (
Northern Echo
[79] Joseph Walsh Wilkinson was remanded to Crumlin Road until June 11.
Belfast Telegraph
By contrast, 'commanded' is less predictable. If the Definite Article is included in
the query, I still find no specific next words with special frequency, but there is a
definite pattern of military Collocates: 'army, troops, force, militia, regiment,
battalion, corps, division, squadron, cavalry'.
'Commanded by' correlates with a predictable metalworks of military
brass: 'General, Admiral, Marshal, Colonel, Major, Lieutenant, Captain,
Constable' along with their Byzantine combinations (like 'Lieutenant-General'),
plus a gallery of titled nobles and notables and, in exactly one case, a 'king' (but
only that frilly showoff Charles I).
Finally, the systemic process of
or among Text-Systems and concerns the process whereby the diversity of uses
and meanings of individual words and expressions come together in a sharable
use and meaning for a context.
The impact of Convergence is perhaps most tangible when it effortlessly
constructs meanings that are by no means the summing up of 'literal' meanings,
[80] Pancakes to sell for grave flags (
University Herald
[81] Internal memos on tampon introduced (
Washington Post
[82] Insecticide sprayed on judge's oral ruling (
Spokane Chronicle
How 'Systemic' is a Large Corpus of English?
[83] Congress votes for running trains over union workers
Courier and Journal
In exchange, unduly effortless Convergence can seem trivial and patronising:
[84] War dims hope for peace (
Columbus Dispatch
[85] Cold Wave Linked to Temperatures (
Daily Sun
[86] Study says dead patients usually not saved (
Miami Herald
[87] Plane Too Close To Ground, Crash Probe Told (
San Antonio Express-
How Convergence can in fact be achieved is only gradually being explained, and
the best experimental evidence indicates that it is not done in a way common
sense would suggest (Kintsch 1988, 1989).
The dynamics of intersystemic processes in Actualisation
In sum, I would propose a model of four properly textual processes, plus six more
multi-purpose processes that can also apply, say, to the perception of visual
scenes or the audition of symphonic music. I believe that at least these processes
are indispensable for the Actualisation that mediates between language-system
and text-system. Being dynamic by nature, they tend to evolve, despite the
traction of institutional standardisation.
If Walter Kintsch's 'construction-integration model' (cited above) has
'psychological reality', the process of actualisation during reading normally runs
between 5 and 500 milliseconds, and such speed leaves us largely dependent on
indirect evidence. Whereas Walter's lab work sifts traces in super-fast operations
of perception, memory, and response, I have here undertaken to sift traces in
longer-term corpus evidence which might plausibly evince an account for certain
classes of phenomena that stand out from the ordinary.
Back in the 1980s, Walter ruefully quipped he had arrived at the
conclusion that 'reading is too difficult to be done'; we can just cling to the reality
that it
done, and mostly well enough too. I in turn can adduce the reality that
Actualisation is also done; the difficulty lies not in the complex of processes, but
in the patchwork of piecemeal models that often mix determinable facts with
wishful thinking and thin air.
If, as I have suggested, significant and essential processes have gone
mainly unexplored because they don't figure well in such models, then we may be
seeing SFL and corpus research nearing a space of convergence for questions
which can only be properly tackled with very large sets of authentic data.
And so I come back to the question posed in my title. I am confident both
language and text are not merely systemic, but are mutually designed to sustain
systemic actualisation from processes to products. As the corpora continue to
grow, our insights will be deepened and broadened in both directions: toward
language and toward text.
De Beaugrande
If we once imagined a 'dream' of 'lexis as most delicate grammar' (Hasan
1987), we might now imagine a 'dream' of 'text as most delicate corpus'. (Even
dreams inspire.) Actualisation must be intersystemic; a corpus must be
intertextual; and I submit that our most auspicious pathways for exploring these
vastly rolling wordscapes will be with parallel expeditions in lively contact.
Beaugrande, R. de 1991.
Linguistic Theory: The Discourse of Fundamental
London: Longman.
Beaugrande, R. de 1997.
New Foundations for a Science of Text and Discourse
Greenwich, CT: Ablex.
Beaugrande, R. de 1998. Performative speech acts in linguistic theory: The
rationality of Noam Chomsky.
Journal of Pragmatics
29, 765-803.
Beaugrande, R. de July 2004.
A New Introduction to the Study of Text and
. Published on the Internet.
Beaugrande, R. de January 2007.
A Friendly Grammar of English
. Published on
the Internet.
Biber, D., S. Johansson, G. Leech, S. Conrad, and E. Finegan
Grammar of Written and Spoken English
London: Longman.
Coggle, P. 1993.
Do you speak Estuary? The New Standard English
. London:
Bloomsbury Publishing.
Halliday, M.A.K. 2004. Language as system and language as instance: The
corpus as a theoretical construct. In J. Svartvik (ed.),
Directions in Corpus
Berlin: Mouton, 1992, 61-77. Also in
Collected Works of M. A.
K. Halliday
, (ed. J. J. Webster), vol. 6:
Computational and Quantitative
. New York: Continuum International Publishing.
Hasan, R. 1987. The grammarian's dream: Lexis as most delicate grammar. In M.
Halliday and R. Fawcett (eds.),
New Developments in Systemic
. London: Pinter, 184-211.
Kintsch, W. 1988. The role of knowledge in discourse comprehension: A
'construction-integration model'.
Psychological Review
95/2, 163-82.
Kintsch, W. 1989. The representation of knowledge and the use of knowledge in
discourse comprehension. In R. Dietrich & C. Graumann (eds.),
Processing in Social Context
. Amsterdam: North Holland, 185-209.
Martin, J. R. 1992.
English text: systems and structure.
Amsterdam: Benjamins.
Rosewarne, D. 1984. Estuary English.
Times Educational Supplement
, 19,
October 1984.
Quirk, R., S. Greenbaum, G. Leech and J. Svartvik, 1985.
A Comprehensive
Grammar of the English Language
. London: Longman.
Weinberg, G. M. 2001.
An Introduction to General Systems Thinking
. New York:
Dorset House.
Some notes on the concept of cognitive linguistics
Wolfgang Teubert
University of Brimingham
In this contribution, I argue that the cognitive sciences are troubled by some internal
contradictions that seem to me difficult to resolve. Cognitive linguistics is the part of the
cognitive sciences dealing with language, and the philosophy of mind provides its
theoretical underpinnings. Its goal is to describe the language system as a mechanism that
processes thoughts into utterances and utterances into thoughts. While thoughts involve
intentionality, the processing mechanism is thought to operate without our awareness. But
what do we actually know about this mechanism? Is there really a language of thought,
and how innate and how universal would it be? What do we know about the mind as the
locus where cognition is processed? How dependable is the computational model of the
mind? Do the various factions of cognitive linguistics offer scientific evidence or just
possible models of how the mind, if there is one, might work? In the end, cognitive
linguistics cannot account for meaning. We do not have access to our own or anyone else's
mental concepts. Meaning and knowledge, on the other hand, is public; it is what is
exchanged, negotiated, and shared in the discourse. Whatever cognitive linguists may be
able to find out about our mental representations, to the extent that it is effable it can
never be more than a duplication of what we find in the discourse.
What is meaning?
Many linguists have such a respect for meaning that they are careful to avoid the
issue wherever possible. Traditionally, language study has a strong focus on
grammar. Grammar is, if one keeps sorting the elements that make up language
long enough, a land of apparent law and order, in which every part and parcel
finds, in the end, its pigeonhole. The meanings of words, however, behave
disorderly. Words are ambiguous and fuzzy. This is why it always has seemed
prudent to leave them to the poor cousins of linguists, to the lexicographers.
However, whenever the makers of our dictionaries try to make sense of them it is
the linguists who habitually criticise them for all the inconsistencies abounding in
even the best dictionaries.
For two schools of linguistics, this is picture is true no longer. Corpus
linguistics and cognitive linguistics share the fascination for meaning, though not
for much else. Yet they look for meaning in different places. For corpus linguists,
the meaning of a lexical item can only be studied in real language data, in the
texts in which they occur, in the contexts in which they are embedded. Here
miraculously all ambiguity and fuzziness seems to fade. If we read a text we
rarely have the problem of not knowing what the words are supposed to mean.
Only if we look at these strings of alphabetic characters, with a space in front of
them and behind them, in isolation, we start to wonder whether
means 'river
edge' or 'financial institution'.
Michael Stubbs' publications have made corpus linguistics popular almost
all over the world.
Tout le monde
now uses corpus evidence. By itself, however,
working with corpus data does not make one a corpus linguist. More and more
linguists, including cognitive linguists, may underpin their investigations with
examples discovered in corpora. But corpus linguistics is more. As Mike Stubbs
has demonstrated time and again, most convincingly, I believe, in his magisterial
Words and Phrases
(2001), corpus linguistics opens a new perspective on
language. It replaces the traditional practice of analysing and categorising
linguistic phenomena in the sterile conditions of a post-mortem autopsy, by
interpreting these phenomena
in vivo
, in their contexts and with their implicit and
explicit links to other discourse events. For this task corpus linguists use corpora,
principled samples of the discourse, and computers. They correlate the statistical
analysis of lexical correspondences over the corpus to their semantic relevance.
Corpus linguistics understands language in terms of open choice and co-selection,
as John Sinclair has pointed out repeatedly. The discourse, this entirety of texts
that have been and are constantly being exchanged between the members of a
discourse community, is a network of intertextual references. Whenever
something is said, it is said as a reaction to what has been said in other, previous
texts. Only rarely we say something new. Instead, we re-use the phrases and
expressions that we find. Now and then we may add our own touch. What a
phrase, an expression means is how it has been used and paraphrased in its
previous history. Meaning, therefore, is in the discourse. Language has to be
viewed as a social phenomenon. That these ideas are now increasingly accepted is
largely to the credit of Mike Stubbs. He has been untiring in his efforts to develop
a consistent theoretical framework that will let us make sense of the findings the
methodology of corpus research is supplying, always sticking to his characteristic
lucid, straightforward and unpretentious style that is so typical of this great
For cognitive linguists, meaning is not in the discourse; rather, it is in
people's heads. They view traditional linguists, the philologists for instance, as
sitting in a dark cave looking in one direction only, toward the back wall of the
cave. Behind them is an open fire providing light, and between the fire and where
they are sitting there is a catwalk on which the mental concepts move, casting
lexical shadows on the wall. Shadows are all the traditional linguists see. As long
as they stay in this position they take the shadowy words for the stuff meaning
consists of. If they only turned around, they would be confronted with something
more real, with mental concepts or cognitive representations. Regardless whether
these things are metaphysical Platonic forms, i.e. what Plato calls
or whether they are only models of the 'real' things, accepting them in lieu of
words would be a step in the right direction, a step taken by cognitive linguists.
Once it is taken, they can start working on the last obstacle to truth, the division
between brain sciences and mind sciences. Eventually, cognitive linguistics
Some notes on the concept of cognitive linguistics
promises, we will grasp what is really out there in the uncharted territories in our
In this contribution, I want to investigate the key element of cognitive
linguistics, the mental/cognitive concept/representation as the embodiment of
meaning. To the extent there is a common philosophical and theoretical basis of
what is now called cognitive linguistics, we find it in the cognitive sciences, as
they developed after the demise of behaviourism, and in the philosophy of mind.
Many of my arguments refer to these foundations rather than to specific
contemporary schools of cognitive linguistics, of which there are many. From my
outsider's perspective, there seems to be a tendency in several of these schools
towards modelling the linguistic faculty of the mind rather than to demonstrate
the reality of this faculty. Because cognitive linguistics is far from being a
uniform discipline, it will always be possible for one school or another to
maintain that my charges do not apply to them. I am more concerned with the
basic ideas driving cognitive linguistics than with individual schools, namely that
we have to look for meaning in people's heads. My goal is to show that this is a
hopeless enterprise.
Some cognitive linguists draw the main dividing line between one-level
semantics and two-level semantics. Those who subscribe to one-level semantics,
for instance Jerry Fodor and Ray Jackendoff, deny that there is a systematic
difference between the meanings of words (the meaning of the expression) and
the respective mental representations. Two-level semantics, on the other hand,
holds that mental representations are richer and more specific, and certainly not
isomorphic with the meaning expressed in natural language utterances. Stephen
Levinson seems to make a distinction between two-level semanticians like
himself, on the one hand, and 'Cognitive Linguists' on the other hand, and he
applies this label apparently only to one-level semanticians. (Levinson 1996, 24)
There is a multitude of views as to the nature of the mental concept, and there are
different terminologies. Looking at these texts from the perspective of corpus
linguistics, I find myself unable to distinguish between concepts and
representations, and between mental and cognitive. Mental/cognitive concepts
and/or representations all seem to refer to more or less the same idea.
Still much what I present here has been put forward in irreconcilably
different accounts. Dan Sperber and Deirdre Wilson identify three views as to the
number of concepts: There could be less than there are words (the compositional
view, argued by, among others, Anna Wierzbicka and also Stephen Levinson),
there could roughly the same number (a view to which Jerry Fodor and Stephen
Pinker seem to subscribe), and there could be infinitely more concepts than words
(as Sperber and Wilson themselves believe). (Sperber/Wilson 1998, 186f.) The
different camps of cognitive linguistics are not co-extensive with the academic
disciplines of the scholars. We find linguists, cognitive scientists, computer
scientists, neurologists, philosophers of mind and analytic philosophers.
My main claim is that we know too little about what is going on in our
heads to make strong claims about the connection between thinking and language.
Thinking is a mental activity that takes place in the brain. Thinking, for me at
least, involves consciousness, and consciousness involves intentionality. Both
features are highly enigmatic. It is doubtful if they can be related to brain
functions. Then there is the enigmatic nature of thoughts. We do not know the
content of a thought unless it is expressed in language. Is it possible at all to
distinguish between a thought and its expression? Cognitive linguists, like all
linguists, have to start with linguistic expressions. But they contend (and who
could possible hold against it?) that any linguistic expression must be caused by a
thought, and that its intended destination, in the hearer's head, is again a thought.
Can we learn to travel that route? Can we put our fingers on a thought just before
it becomes the expression of a thought? What do we know about the thought
caused by an expression? How can we trace a thought? How can we describe the
content of a thought without using language? If there is more to a thought than
what can be verbalised in the expression of it, how can we access this
extralinguistic content? To me it seems there is no easy way to resolve these
matters. The mental/conceptual concepts/representations I find in cognitive
accounts are, at their very best, a duplication (and sometimes even a triplication)
of what we can find out about the meaning of a lexical item, about the meaning of
what has been said. Indeed, what more could we hope to find in people's heads?
Cognitivists are "
interested ultimately
", in the words of Ray Jackendoff, "
in the
manner in which language is embodied in the human brain
." They are not
interested in language as "an abstract phenomenon or a social artifact". There
"might be properties that language has because of the social context into which it
is embedded". But there are also "some important properties … that can be
effectively studied without taking account of social factors". Is this "mental
stance" a stance concerning the brain or the mind? While Jackendoff calls
language a "mental organ", his ultimate goal seems to locate it in the brain.
"Language, vision, proprioception, and motor control" all are "instantiated" by
neurons of basically similar design. Thus, in the long run, we should study
language as a "physical organ". (Jackendoff 1997, 2ff.; Jackendoff's emphasis)
Should we assume, then, that mental concepts or cognitive representations have a
physical reality? And will our quest for meaning have come to a conclusion once
we have identified the neurons and their nature that correlate with a linguistic
We use language firstly to interact with people and secondly to give our
linguistic and non-linguistic interactions a meaning. Our societies are much more
complex than those of our closest relatives, the non-human primates. This
complexity requires not only a division of labour but also distributed knowledge.
We must negotiate who carries out which task, and we must provide the
necessary knowledge. Language enables us to trade content. For corpus linguists,
language is, first of all, a social phenomenon. Language is public. Language is
observable. Outside of language there is no symbolic content that could possibly
be conveyed. Whatever we think, we have to express it in language so that others
can share it. How we turn our thoughts into language, how we understand what
other people say is a matter of speculation. But what something that has been said
means is not a matter of speculation but of negotiation. When someone has said
Some notes on the concept of cognitive linguistics
something, we, whether we are the addressees of this statement or whether we
learnt about it some other way, can discuss its meaning. We may not necessarily
agree. It can be a discussion without a conclusion. But beyond our various
interpretations of this statement, concordant or discordant as they may be, there is
nothing to be found out about its meaning. To know which neural processes and
which hormonal outpourings led to the statement, and which neural processes and
hormonal outpourings it triggered in the people who were addressed does not lead
us to understand the statement any better. Meaning is, just as language, public. It
belongs to the sphere of social interaction, not to the realm of mental processes.
This is the stance of corpus linguistics, as I see it. Is there any reason why we
should take up the cognitive approach to find out about meaning?
Gisela Harras, who prefers two-level semantics distinguishing between
'semantic content' and cognitive or mental concepts, gives an example to tell us
why language does not tell the full story. She compares the two sentences 'Open
the bottle' und 'Open the washing machine' and concludes that a verb such as
can convey a sheer unlimited amount of concepts. The semantics of the
verb, she maintains, does not tell us what
means when applied to different
things. The bottle may have a screw top or a cork. The plumber opens the
washing machine at a different spot from the normal user. When we, as the
addressees, have to interpret the different utterances containing the verb
, we
have to go beyond its semantic content, Harras says. The hearer has to apply
cognitive mechanisms that will tell her or him which of the many concepts
expressed by the verb
the speaker has intended. For Dan Sperber and
Deirdre Wilson, to whom Harras is indebted for her example, these concepts
seem to be individual rather than universal: "A concept, as we understand the
term, is an enduring elementary mental structure...It is arguable that each of us
has ineffable concepts [e.g. of a special kind of pain] – perhaps a great many of
them… For the time being, we will restrict ourselves to effable concepts:
concepts that can be part of the content of communicable thought…[T]here are a
great many stable and effable mental concepts that do not map onto words."
(Sperber/Wilson 1998, 189) How do we have to understand these strange mental
concepts which are effable, but do not map onto words?
Let us assume the bottle referred to in the utterance 'Open the bottle!' is a
standard bottle of wine of 1998. Then the cognitive representation of the verb
would be the concept of 'uncorking'. If we accept this, the question must be
allowed how this mental concept differs systematically from the natural language
expression. The concept 'uncork' is not only effable, it happens to map
miraculously onto the word
. But how do I move on from the rather
general verb
to the much more specific concept 'uncork'? Do I have to rely
on cognitive mechanisms that make me understand the speaker intended 'uncork'
when she or he said
? Corpus linguists can easily do without such constructs.
They would find citations in their corpus which paraphrase the meaning of
in the case of wine bottles, i.e. which tell us what open means here:
How to
a bottle: It's actually pretty simple to
a bottle of
wine. These steps are for a double-action, or wing, corkscrew, which
has two arms (or wings) that help lever the cork out of the bottle.
Meaning is in the discourse. All we have learnt about the meanings of lexical
items we have learnt from other people's contributions to the discourse. This is
how we learn how to use words when we grow up. Our carers or, a bit later, our
peers, explain what they mean. This is how new lexical items are introduced: The
text introducing them has to explain them, has to paraphrase them. If I have not
been told what
opening a bottle
means, no cognitive mechanism will help me to
understand the speaker's intentions. However, if I know the meaning of
open a
bottle of wine
, I do not need such a mechanism.
Why do we not readily accept such a simple solution? The reason, I think,
is that we attribute too much importance to the meaning of a single word in
isolation. We grow up in the belief that words are the core elements of language,
and that their meanings are registered in dictionaries. Indeed, there we are
confronted with the apparently incontrovertible evidence that words, frequent
words in particular words like
, are fuzzy, polysemous or ambiguous. A
central part of the endeavours of cognitive linguistics are directed at
disambiguation, at the resolution of ambiguity. For in contrast to words, our
thoughts, we feel, are unambiguous. But is language really more ambiguous than
thinking? When we are confronted with a text of common length, do we feel it is
ambiguous? Once we give up the belief that words are the core elements, the
ambiguity starts disappearing. All we have to do is to replace the linguistic
construct 'word' by the linguistic construct 'unit of meaning', defined as a node
word plus all the words in its immediate context that make it unambiguous.
Instead of wondering about the polysemy of
, we now have to deal with a
unit of meaning, a lexical item
open* a bottle of wine,
which is fairly
monosemous. If our speaker tells us 'Last night I opened a bottle of wine
', we
still cannot be absolutely sure that she uncorked it. She might have unscrewed the
top, if she is the kind of person to have wine with a screw top. Unless she tells us,
we will never know. No cognitive mechanism will provide that information. Once
we throw the notion of the word as unit of meaning overboard, ambiguity
recedes. Forty years ago, when corpus research was taking off, John Sinclair
argued for the principle of collocation which gives us complex lexical items
larger than the single word but with only one meaning. (Krishnamurthy 2004, 10)
Corpus linguistics has nothing to tell us about how thoughts are turned into
linguistic expressions, and how linguistic expressions are turned into non-
linguistic thoughts. But corpus linguistics can deal with meaning to the extent that
meaning is public and negotiable. How this is done I have sketched in "My
Version of Corpus Linguistics" (Teubert 2005).
In this contribution, I will have a closer look at two essential aspects of
cognitive linguistics. The first aspect deals with the model of the mind that we
find in the cognitive sciences. The second aspect concerns the innateness and
universality of the mental lexicon. I will conclude by asking if there is any
Some notes on the concept of cognitive linguistics
common ground on which corpus linguists and cognitive linguists can work
The model of the mind in the cognitive sciences
There is a rather disquieting aspect of cognitive linguistics. It concerns the
foundation of the cognitive sciences in general. In the fifties of the last century,
cognitivism has replaced behaviourism with the promise to keep up the claim to
scientificity established by behaviourism while at the same time pledging to
demonstrate and explain directly the working of the mind, without having to
resort to ambiguous stimulus-response situations. But then it is not so easy to
look into people's heads. A model was needed to spell out how the mind is
The computer had just been invented, and its prospects seemed boundless.
In the long run, people believed, it was only a matter of size to compete with
human reasoning. More memory, more operations per second, more complex
programs, and computers would emulate if not surpass human thinking. As they
were believed to be, in principle, functionally equal to the human mind, they were
seen as the perfect model of cognition, offering the additional advantage of
blocking out the erratic impact of emotions. That was the hour of birth of the
computational theory of the mind, a theory quickly becoming a doctrine stating
that the working of the mind can be understood and described in analogy to the
working of the computer. Ray Jackendoff was influential in relating this model to
the study of language, for example in his book of 1987
Consciousness and the
Computational Mind
This model appealed to those within the human sciences who wanted to
look their colleagues in the natural sciences straight into their eyes. The cognitive
sciences enabled them to locate the study of the psyche, the human mind, within
the sciences. What had been an object of mostly philosophical investigation, with
all the arbitrariness of interpretation, could now be based on solid fact. This was a
welcome opportunity for linguistics to bid farewell to the increasingly
embarrassing lodgings within the
and to jump on the
bandwagon of scientific and technological modernity. It was there, and not in the
humanities, where the big research funds were on offer.
The promise was that thanks to the computational model the mystery of
the mind could now be finally solved. Further rewarding consequences soon
began to show. Thanks to the new paradigm, other disciplines began to take the
cognitive sciences seriously. Therefore it came as no big surprise that when
computer scientists set out to invent artificial intelligence, they now, in turn
looked to the cognitive sciences as their inspiration. The cognitivists naturally
were happy to sell back the blueprints they had copied a few years earlier from
the computer engineers, together with some new annotations, to the emerging
artificial intelligence community. The new goal was to transfer the now
established view of how humans perform mental operations and solve problems
of all kinds to computer scientists eager to develop 'intelligent' machines. This
cross-fertilisation continues to the present day, as does the ignorance about the
circularity involved. So when the computer scientists developed the concepts of
parallel processing and connectionism, these new ideas were immediately taken
over by the cognitivists. Thus Pierro Scaruffi tells us: "[A] connectionist structure
such as our brain works in a non-sequential way: many "nodes" of the network
can be triggered at the same time by another node. The result of the computation
is a product of the parallel processing of many streams of information." (Scaruffi
The new developments were greeted by many active in the philosophy of
mind. They seemed to vindicate the computational theory of mind.
"Connectionism", says Daniel Dennett, "is a fairly recent development in
A[rtificial] I[ntelligence] that promises to move cognitive modelling closer to
neural modelling, since the elements that are
bricks are nodes in parallel
networks that are connected up in ways that look
like neural networks in
brains." (Dennett 1993, 269, Dennett's emphasis)
The belief that parallel processing and neural networks are all that it takes
to make our computers truly intelligent was permeating the whole western(ised)
society. This was when everybody was talking about a new generation of
computers believed to be able to emulate human intelligence. The new computers
could learn (program themselves) how to do things instead of just carrying out the
programmer's instructions. They could be trained to reason based on common
sense, thus enabling them to pass reliable judgment on matters too complex to be
thought through by humans. These new 'virtual machines' would employ an
'architecture' of processors operating in a parallel, interactive way. The
relationship between the initial and the resulting states of such a machine would
not be determined by pre-programmed commands, but would develop themselves
on the bases of huge amounts of training data. This is the idea of connectionism.
Once the computer has 'learned' which initial stages lead to which resulting states,
this knowledge can be applied to any new initial state displaying the same
properties, and the computer will deliver the correct final state.
Eric Pederson and Jan Nuyts, the editors of
Language and
, would not disagree:
Thus, while 'classical' cognitive theories would consider
representations to be virtual 'objects' of some type, manipulated by a
'machinery' of procedures or rules which are somehow implemented
in the human brain, connectionist and parallel distributed processing
theories consider representations to be simply the resultant
characteristics of peculiar states of the 'conceptual system' distributed
across the neural networks of the brain… In the latter view, if the
notions of knowledge and representations are to be used at all, any
characterizations of them beyond the vague ones given above are no
longer acceptable as descriptions of actual cognitive mechanisms
creating human behaviour. (Nuyts/Pederson 1997, 1f.)
Some notes on the concept of cognitive linguistics
So can we be sure that the brain/mind is isomorphic with a (connectionist)
computer? Perhaps not quite, Ray Jackendoff thinks, but that does not do harm to
this model. He accepts that "eventually the neuronal basis of mental functioning
is a necessary part of a complete theory". But neuroscience today is "far from
being able to tackle the question of how mental grammar is neurally instantiated".
Therefore "the formal/computational approach is among the best tools we have
for understanding the brain at the level of functioning relevant to language, and
over the years it has proven a pragmatically useful perspective". (Jackendoff
1997, 9)
For the John Searle of the late eighties it became exactly this equation of
computers and human minds which he sees the foundational error of the cognitive
If one looks at the books and articles supporting Cognitivism one
finds certain common assumptions, often unstated, but nonetheless
First, it is often assumed that the only alternative to the view that the
brain is a digital computer is some form of dualism. The idea is that
unless you believe in the existence of immortal Cartesian souls, you
must believe that the brain is a computer. Indeed, it often seems to be
assumed that the question whether the brain is a physical mechanism
determining our mental states and whether the brain is a digital
computer are the same question. Rhetorically speaking, the idea is to
bully the reader into thinking that unless he accepts the idea that the
brain is some kind of computer, he is committed to some weird
antiscientific views. Recently the field has opened up a bit to allow
that the brain might not be an old fashioned von Neumann style
digital computer, but rather a more sophisticated kind of parallel
processing computational equipment. Still, to deny that the brain is
computational is to risk losing your membership in the scientific
community. (Searle 1990)
Yet does it really work, this model used both by the cognitive and the artificial
intelligence community? Some misgivings are allowed. In spite of billions of
dollars invested in artificial intelligence and machine translation systems, results
are far from satisfactory. For forty years we have been hearing that success is 'just
around the corner'. We are still not there. The epochal endeavour to teach
computers 'common sense', the megalomaniac CYC project of a comprehensive
ontology of knowledge, never came close to the originally envisaged results
(, nor did EUROTRA, the European project for automatic
translations from and into all the EU languages, which cost the taxpayers
hundreds of millions of euros ( Could it
be that the whole approach was faulty?
Are concepts inherited and universal?
The second question: Are the mental concepts of cognitive linguistics universal as
it has been claimed? For Jerry Fodor there is no doubt that the 'language of
thought' (what comes to be called mentalese by Stephen Pinker) is really
universal, and that it is in this language of thought, and not in natural languages,
that meaning resides: "English
no semantics. Learning English isn't learning a
theory about what its sentences mean, it's learning how to associate its sentences
with the corresponding thoughts." (Fodor 1998, 9; Fodor's emphasis) How literal
should we take this claim? In Stephen Pinker's
The Language Instinct
we read:
People do not think in English or Chinese or Apache; they think in a
language of thought. This language of thought probably looks a bit
like all these languages; presumably it has symbols for concepts, and
arrangements of symbols. … [C]ompared with any given language,
mentalese must be richer in some ways and simpler in others. It must
be richer, for example, in that several concepts must correspond to a
given English word like stool or stud. … On the other hand, mentalese
must be simpler than spoken languages; conversation-specific words
and constructions (like a and the) are absent, and information about
pronouncing words, or even ordering them, is unnecessary. (Pinker
94, pp 81-2)
This quote shows nicely how cognitive linguistics multiplies entities without
need, and thus violates Ockham's razor. Natural languages have words (
), mentalese has symbols representing the various meanings of
and these symbols stand for concepts. Does that mean that first we have to
translate a natural language sentence into mentalese, and then link this mentalese
sentence to its cognitive representation, and then we have understood what the
sentence means? This would amount to a triplication of our semantic apparatus.
More interesting in our current context is that these mentalese symbols
grosso modo
to the meanings of natural language words. Indeed there
has not been a lack of endeavours over the last decade to build multilingual
conceptual 'ontologies' on the basis of this hypothesis. There has been, for
instance, the large project financed by the European Commission and the member
states involved with the title EuroWordNet (
which uses a slimmed-down version of the Princeton WordNet, the largest
American English language online dictionary/database (www.wordnet.princeton.
edu/). It relates the senses (formerly called 'synsets') given in WordNet to senses
of words in a large number of European languages. There is further offspring in
form of 'localisations' of WordNet for some of these and some other languages,
such as GermaNet ( What is
called an ontology is, of course, in reality nothing more than a taxonomy. It is not
a reality-inherent classification of whatever we find in reality. Neither is it a
language-inherent classification of all the senses (or concepts) that we find in
language. Rather it is a classification of English words and their senses as they
Some notes on the concept of cognitive linguistics
have been found fit for lexicographic purposes. There you look at single words in
isolation and not at words embedded in a text. However, numerous attempts to
have people or computers assign the 'proper' sense as found in a given dictionary
to a word embedded in the text have yielded rather disappointing results. How
else could it be? The entry for the noun
has eight senses in WordNet; in the
New Oxford Dictionary of English (1998) it has only two senses, and in the
Collins Dictionary (fourth edition, 1998) it has 13 senses (excluding idioms).
Assigning senses to words in isolation is, to a large extent, arbitrary, once we
leave aside obvious cases such as the two meanings of the commonly invoked
, a homonym conflating two etymons. There is little evidence that
the word senses we find listed in dictionaries relate to some natural language
'reality', and there is even less evidence that we find these senses, as they are
listed in dictionaries of one language, in dictionaries of a different language.
For once we proceed from a monolingual to a multilingual perspective, we
are bound to recognise quickly that as long as we look at words in isolation they
hardly seem to map onto each other from one language to the next. Whatever
there might be, in terms of mental concepts in our heads, there is hardly any
evidence to call it genetically inherited, innate or universal. Take the German
. Bilingual dictionaries tell us that their English
equivalents are
. There is no one-to-one relationship,
however. The English
translates into
, when a young girl is left
by her lover, it translates into
, if an old man cannot accept his fate, and it
translated into
if someone post puberty suffers the loss of someone dear
and seemingly irreplaceable. German-English dictionaries invariably offer
as the equivalents of
, but never provide a distinction between the
two equivalents. But there must be a distinction. In eight out of ten sentences
featuring either
, native speakers insist that one cannot replace the
other. Again we are confronted with the fact that there is no way to deal with the
meaning of single words in isolation. It is the context, the situation and an infinity
of peripheral conditions which have to me matched.
in Thomas Mann's
novels is different from
as we find it in tabloids. Words out of context,
in isolation, are, to a large extent, empty, waiting to acquire a specific meaning
from the wider and the narrow context, in particular from the collocates they co-
occur with. How should we imagine the mental concepts corresponding to single
words? Would they not have to be similarly indeterminate? Or do they have a
specified meaning, regardless of the context in which they occur, just as H2O is
always the same substance wherever it occurs? Are we expected find the same
mental concepts in the heads of all people, regardless which language they speak?
How do they get there? Are we born with them?
For Noam Chomsky, it was a well-advised decision to exclude, over
decades, the semantic component from his investigation into the workings of the
language faculty. To him we are indebted for the apophthegm that a visiting
scientist from Mars would conclude that, aside from their mutually
incomprehensible vocabularies, all earthlings speak the same language.
(Chomsky 2000, 118) That the different vocabularies keep me from
understanding people speaking another language is, for the Chomsky of the
Aspects of a Theory of Syntax
, but a surface phenomenon. Just as all languages
share the same syntactic deep structure, Chomsky the mentalist seems to believe,
they also share a common pool of mental concepts. Whenever he discusses the
question of the universality of mental concepts, he invokes the exclusive domain
of the innate language organ, with far-reaching consequences. In his article
"Language and Interpretation" (first published in 1992) he takes up, in modified
form, Jerry Fodor's claim, that concepts are holistic and cannot be decomposed
into more basic primeval, concepts, and he agrees with Fodor that all concepts are
somehow already present in the human language faculty. "There is, it seems
rather clear, a rich conceptual structure determined by the initial state of the
language faculty (perhaps drawing from the resources of other genetically
determined faculties of mind), waiting to be awakened by experience." (Chomsky
2000 [1992], 64) In another contribution "Language as a Natural Object",
reprinted in the same volume, he explains why this must be the case:
The linkage of concept and sound can be acquired [by children] on
minimal evidence… However, the possible sounds are narrowly
constrained, and the concepts may be virtually fixed. It is hard to
imagine otherwise, given the rate of lexical acquisition, which is about
a word an hour from ages two to eight, with lexical items typically
acquired on a single exposure, in highly ambiguous circumstances,
but understood in delicate and extraordinary complexity that goes
vastly beyond what is recorded in the most comprehensive dictionary,
which, like the most comprehensive grammar, merely give hints that
suffice for people who basically know the answers, largely innately.
(Chomsky 2000 [1994], 120)
This is, to say the least, highly speculative. If children around eight years of age
really had a fully working vocabulary of 26280 words (12 words x 365 days x 6
years) we should wonder why they do not put it to better use. Be this as it may,
though, the underlying question is what it means when Chomsky says that
children typically acquire a lexical item in a single exposure which could be only
explained by the fact that they "basically know the answers, largely innate." What
does the sentence "the concepts may be virtually fixed" mean? On the same page,
we also find "There is reason to believe that the computational system [of the
mind] is invariant, virtually." Is it just one of those typical Chomsky-sentences
whose main purpose seems to add rhetorical fervour to his argumentation but
which should not be taken too seriously in terms of their content, such as this
sentence, also on page 120: "But there is evidence that the languages [English,
German, Latin, Greek, Sanskrit and Chinese] have basically the same inflectional
systems, differing only in the way formal elements are accessed by the part of the
computational procedure that provides instructions to articulatory and perceptual
organs." While Chomsky may not have been aware that the first five languages he
mentions belong to the same family, he certainly knew that there are no data
confirming what he calls evidence in the case of Chinese. However, his
Some notes on the concept of cognitive linguistics
employment of hedging adverbs such as
make every
argument invulnerable. "virtually all walls," we could say, "can be painted in
black or in white; black and white are basically the same thing."
Unsurprisingly, Hilary Putnam strongly disagrees when it comes to the
question of mental concepts: "Contrary to a doctrine that has been with us since
the seventeenth century, meanings just aren't in the head." (Putnam 1981, 19;
Putnam's emphasis) This is what Hilary Putnam has been saying consistently over
many decades. If meanings are not in the head, then the idea of innate mental
concepts does not make sense: "A Chomskyan theory of the semantic level will
say that there are 'semantic representations' in the mind-brain; that these are
innate and universal; and that all our concepts are decomposable into such
semantic representations. This is the theory I hope to destroy." (Putnam 1998, 5)
This is how Putnam describes this theory:
Mentalists who follow Fodor's lead are committed to the idea that
there is an innate stock of semantic representations in terms of which
all our concepts can be explicitly defined. …
How could such
concepts as
be possibly innate?
Primitive peoples who
have had no acquaintance with internal combustion engines show no
difficulty in acquiring such concepts. On Fodor's account this means
that their 'language of thought' contained the concept 'carburetor' prior
to their acquiring a word for that concept, even though nothing in their
evolutionary history could account for how the concept 'got there'
(Putnam 1988, 15; Putnam's emphasis)
How is it possible to argue against so much common sense? Can Chomsky
underpin his claim to the contrary? Can he repudiate Putnam's injunction? He
would not explicitly refer to it if he thought he could not: "Some, for example
Hilary Putnam, have argued that it is entirely implausible to suppose that we have
'an innate stock of notions' including carburetor and bureaucrat." (Chomsky 2000
[1992], 65) But Chomsky's repudiation avoids straightforward argumentation.
Instead he reverts to a parable:
Notice that the argument is invalid from the start. To suppose that, in
the course of evolution, humans come to have an innate stock of
notions including
does not entail that
evolution was able to anticipate
future physical and cultural
contingency – only these contingencies. That aside, notice that a very
similar argument had long been accepted in immunology: namely the
number of antigens is so immense, including even artificially
synthesized substances that had never existed in the world, that it was
considered absurd to suppose that evolution had provided 'an innate
stock of antibodies'; rather, formation of antibodies must be a kind of
'learning process' in which the antigens played an 'instructive role'.
But this assumption might well be false. Niels Kaj Jerne won the
Nobel Prize for his work challenging this idea, and upholding his own
conception that an animal 'cannot be stimulated to make specific
antibodies, unless it has already made antibodies before the antigen
arrives' (Jerne 1985: 1059), so that antibody formation is selective
process in which the antigen plays a selective and amplifying role.
(Chomsky 2000 [1992], 65)
Thus in the very moment when we experience a particular stimulus or trigger the
corresponding mental concept comes to our rescue. The trigger can be the
perception of something 'real', as in the case of a carburettor, or it can be an idea,
like that we are burdened by too much admin, as in the case of bureaucracy. Not
all cognitive scientists will be ready to follow him that far. I myself have no idea
how realistic Jerne's antibody theory is. Yet to believe that the whole infinity of
future discourse objects is somehow,
in nuce
, already present in our genes seems
to overstress the suggestiveness of Chomsky's charisma. Fodor himself, who
shares with Chomsky the belief that concepts such as 'bureaucracy' and
'carburettor' have to be regarded holistically, and cannot be decomposed into
semantic primitives, would, I think, hesitate to underwrite this claim, at least
since the publication of his book
Where Cognitive Science Went
. It seems now that he prefers not to become involved. Concerning the
innateness of the concepts 'carburettor' and his other favourite example,
'doorknob', he only tells us: "A lot of people have Very Strong Feelings about
what concepts are allowed to be innate…[T]here is, at present, a very strong
consensus against, as it may be, DOORKNOB or CARBURETTOR. I have no
desire to join this game of pick and choose since, as far as I can tell, it hasn't any
rules." (Fodor 1998, 28) In those few instances in which he is more specific, his
peculiar metaphoric way of speaking makes it hard to pin down his true position.
Concerning the concept 'doorknob' he explains: "[W]hat has to be innately given
to get us locked to doorknobhood is whatever mechanisms are required to come
to strike us as such. Put slightly differently: if the locking story about concept
possession and the mind-dependence story about the metaphysics of
doorknobhood are both true, then the kind of nativism about DOORKNOB that
an informational atomist has to put up with is perhaps not one of
but of
." (Fodor 1998, 142) Thus Fodor implies that the question whether
doorknobs correspond to an innate concept is the wrong question. For him, the
important issue is the innateness of the mechanism that links concept and object.
Thus he leaves himself a door open. He could still agree with Ruth Millikan when
she, as she recently did on a conference, pleads for first-person experiences, and
not innateness, as the precondition for mental concepts to become relevant "[A]ll
concepts, including logical concepts, are tested
for their very having of content
through ongoing experience... What can be gained through conceptual analysis is
then only what has previously been inductively acquired through experience."
(Millikan [2004], 1)
Particularly in continental Europe, the view that complex concepts cannot
be reduced to basic concepts is not very popular. While on the one hand many
cognitive linguists like for instance Anna Wierzbicka take universal, innate or
inherited concepts for granted, they insist that there is only a rather limited
Some notes on the concept of cognitive linguistics
number of inherited basic concepts, called
semantic primes
by her, which are the
building bricks for all more complex concepts. Wierzbicka lists about 50
semantic primes
, among them variables like
verbs such as
, four adjectives:
good, bad, big,
, nouns such as:
part, kind, kind, people
, two pronouns:
, and a
medley of connectors like
where, above, after, if, because
etc. This, then, would
be the translation of the natural language sentence "X felt guilty" into a
representation by semantic primes:
X felt something
sometimes a person thinks something like this:
I did something
because of this, something bad happened
because of this, this person feels something bad
Of course, Anna Wierzbicka is well aware of the problematic nature of such a
theoretical construct. If there 'really' were language-independent concepts, of a
more primitive or even of a complex nature, holistic or not, how would we know
what they 'mean' and how they would translate into natural language? Only in
Texas it may be the common understanding that English is the language of
thought. Inaccessible as language-independent
semantic primes
are for us, we can
only encounter them once they are translated into a natural language, and we will
never be able to control the appropriateness of this translation. Furthermore they
are, in translation, as ambiguous as natural language tends to be. Thus we are left
in doubt whether the translation of the complex mental concept corresponding to
X felt guilty
is correct. And is it 'really' true that
in 'something bad happened'
is the same
as in 'feels something bad'? Is
X felt guilty
really the same as
fühlte sich schuldig
? Does it mean the same as
X had a bad conscience
X had
pangs of conscience
X had a sense of guilt, X felt remorse
X repented
Wierzbicka would point out that she is only sketching a model and that this
model is not materially and perhaps not even functionally equivalent to the 'real'
mental representation. Widely read as she is, she has repeatedly related her
approach to Leibniz; for example in this quote: "Im wesentlichen geht diese Idee
auf Leibniz zurück und auf seine Vorstellung 'eines Alphabets menschlichen
Denkens', das heißt, "einen Katalog der Begriffe, die aus sich selber verstanden
werden können, und aus deren Kombinationen unsere anderen Vorstelllungen
entspringen'" [In all relevant aspects this idea is based on Leibniz and on his
model of an 'alphabet of human thought', i.e. a catalogue of concepts which can
be understood out of themselves, and whose combinations engender our other
ideas.] ( Even if
Leibniz never distanced himself from the youthful folly of his doctoral
Ars Combinatoria
(1666), it was as unsuccessful as all other
endeavours in the last millennium to construct a perfect language. This is the
sobering conclusion we can draw from Umberto Eco's
The Search for the Perfect
Wierzbicka's original contribution is her suggestion of a (universal) syntax
that informs the relationship her semantic primes have with each other when they
are composed into a mental concept. Is this how we have to imagine mentalese,
the language of thought? Do we find there the same categories that we use to
describe the natural languages? Do we find there finite verbs, as we have them in
English, but not in Chinese? Ray Jackendoff seems to know that the system of
cognitive representations, mentalese, or, in his terminology, 'conceptual structure',
does not have parts of speech: "Whatever we know about this system, we know it
is not built out of nouns and verbs and adjectives." (Jackendoff 1997, 31) This is
not how Steven Pinker sees it.
Are there 'really' fifty semantic primes whose meaning is universal but can
only be understood once it is translated into a natural language? Is there 'really' a
'conceptual structure' in which we find concepts but no parts of speech? Is there
'really' something we 'know' about life after death? Or is what we claim to know
about semantic primes, conceptual structures and life after death more a
conjectural hypothesis than empirical knowledge?
Apart from Wierzbicka's mental syntax, we may well compare her
semantic primes
to the semes which were at the core of the mainstream
continental European semantic theories of the sixties and the seventies. Usually
we identify this semantic feature theory with Louis Hjelmslev's
(Hjelmslev 1963 [1943]) His phonological analysis and his concept of the
phoneme became the model for semantic analysis and the concept of the séme.
Bernard Pottier combined Hjelmslev's approach with the Prague school of
structuralism. He was the first to call the 'distinctive semantic features of lexemes'
. This is how he describes the meaning of
{s1, s2, s3, s4} ("to sit on, on legs, for one person, with a
backrest"). Relative to the set containing
easy chair
, chair is defined
the seme s5 ("with armrests") and so on. (Pottier 1978, 86)
Thus meaning can be analysed in term of differences, through the presence or
absence of sémes. It is this focus on difference which grounds this theory in de
Saussure's structuralism.
Algirdas Julien Greimas, too, uses the concept of sémes. (Greimas 1966,
22 ff.). He distinguishes between the presence of a séme, the negation of this
presence ('negative séme') and a state in which a given theme is neither present
nor absent ('neutral séme'). For Pottier and Greimas the séme thus is the smallest
feature (
trait distinctif
) to distinguish meaning that accounts for the difference
between one word such as
and another, semantically related word,
(a word belonging to the same semantic field). Sémes here are understood
as heuristic constructs. In the second edition of Theodor Lewandowski's
Linguistischem Wörterbuch
(1976) we find this entry for
semantisches Merkmal
['semantic feature']:
Some notes on the concept of cognitive linguistics
Bedeutungsatom, Bedeutungskomponente, Element des Begriffs bzw.
Inhalts, der als in sich (mikro)strukturiert aufgefaßt wird, Basis-
Element und Konstrukt einer semantischen Theorie, das sich mit
Konstrukten wie Atom, Gen usw. vergleichen lässt. Bei der
Konzeption des semantischen Merkmas handelt es sich um eine
Übertragung des Prinzips der distinktiven Merkmale auf den Bereich
der Semantik…
Bei Bierwisch (1967, 3) sind semantische Merkmale "certain deep
seated, innate properties which determine the way in which the
universe is conceived, adapted, and worked on." [Atom, component of
meaning, element of the concept or of the content looked at as a
(micro-) structure in a semantic theory that can be compared to
constructs such as atom, gene etc. Conceiving of semantic features in
this way is a transference of the principle of distinctive features onto
the field of semantics… For [Manfred] Bierwisch, semantic features
are "certain deep seated, innate properties which determine the way in
which the universe is conceived, adapted, and worked on."
(Lewandowski 1976, 3, 663)
What is interesting here is the naivety in this entry in which semantic features are
fused with semantic primes. For these features are theoretical constructs within a
linguistic model to which no ontological reality is ascribed. This is the
understanding we find in the entry
['feature'] in Lewandowski's
linguistic glossary:
Begriffliches Konstrukt, ein Begriff, der für das richtige Verstehen der
Sprachstrukturierung unentbehrlich ist (Martinet); für die
Konstruktion und Funktion sprachlicher Einheiten als notwendig
betrachtete begrifflich-hypothetische Mikroelemente. [Theoretical
construct, a concept indispensable for the proper understanding of
structuring language (Martinet); conceptual-hypothetical micro-
elements considered as essential for the construction and function of
linguistic units. (Lewandowski 1976, 2, 446)
Would Bierwisch agree? Developing further the contention forwarded by Jerry
Fodor and Jerold Katz that it were possible to "construe a meta-theory containing
a list of semantic features from which we can take the theoretical vocabulary of
any special semantic theory" (Fodor/Katz 1963, 208), Bierwisch explains:
This does not mean, of course, that the dictionary of each given
language must show exactly the same distinction as that of any other
language. It implies only that, if a distinction is made, that property
can be characterized in a nontrivial way in terms of a universal set of
semantic markers. If we accept this view, then two different questions
immediately arise:
What is the theoretical status of the universal semantic markers; how must they
be interpreted?
What are the elements of the universal set and how can they be
... The question here is: in what way, by what type of phenomena, are
they motivated outside of the structure in the narrower sense? In other
words: what is the interpretation of semantic markers, how are they
connected with thought? [The German version here reads: "welche
Beziehungen bestehen zwischen ihnen und den kognitiven und
perzeptiven Leistungen des Menschen?"] (Bierwisch 1967, 2; 1970,
Bierwisch, it seems, is not troubled by the question whether the semantic features
are theoretical constructs of the linguist which he or she derives from the analysis
of a natural language. For him they are real, ontologically given, located in
human cognition. Are they learnt or inherited? He is obviously very sure: "Not
only is there no reasonable explication of how semantic markers are learned. It is
also very difficult to explain in a natural way such well known facts as displaced
speech, fictitious objects and in general all gaps between meaning and reality."
(Bierwisch 1967, 3) For Bierwisch, there is no alternative to the inheritance
option. This was not so unexpected at a time when the attraction of Chomsky's
model had its first peak in Europe. Particularly linguists in East Germany and
other Eastern European countries embraced it because it located their field
securely within the sciences, outside of the
with their
suspected bourgeois affinities. As real scientists, linguists thus could exempt
themselves from the obligation to justify their approach from a Marxist-Leninist
perspective. If the language organ was real, then semantic features must be real,
as well:
There are good reasons to believe that the semantic markers in an
adequate description of a natural language do not represent properties
of the surrounding world in the broadest sense, but rather certain deep
seated, innate properties of the human organism and the perceptual
apparatus, properties which determine the way in which the universe
is conceived, adapted, and worked on. (Bierwisch 1967, 3)
Bierwisch fully subscribed to this fairly pervasive Anglo-Saxon entrenchment in
a realism born out of common sense which up to this day still determines, to a
very large extent, analytic philosophy in America, turning relativism there almost
into a swearword. The abyss between Anglo-Saxon realism and Continental
constructionism, in the guise of nominalism, hermeneutics or (post)structuralism,
is rarely bridged. Bierwisch's semantic features do no longer correspond to the
sémes hypothesised by Poittier or Greimas. Lewandowski's dictionary passes
over this crucial difference in silence, namely that sémes are viewed as the
linguist's constructs derived from the analysis of a natural language by applying
heuristic procedures, while the cognitivists’ mental concepts are seen as
Some notes on the concept of cognitive linguistics
ontologically real entities. Whether it makes sense to posit a given séme in order
to account for the difference between two semantically closely related words can
be negotiated. But whether there is an innate mental concept meaning 'with
armrest' is not a matter for discussion; it has to be proven in a scientific sense.
It is this insistence on ontological reality as opposed to hypothetical
models that distinguishes the programme of cognitive semantics from that of
structural semantics in continental Europe. Behind innate mental concepts we
find looming the postulation of universality. It is not any more a particular natural
language we are analysing, but the language of thought, the common mother of
all languages. Does this utopian claim really help us to reveal the mystery of
meaning? What do we gain if we set out to investigate some elusive model of the
workings of the mind rather than sticking to the real language data to which we
share access? It does not matter how we have to imagine mental concepts. They
may be holistic as in the case of Chomsky's example of the carburettor, or we
may imagine them as concepts composed out of semantic primitives; there is
never a kind of empirical evidence about them that could be objectified. Perhaps
this is the most plausible explanation for the lack of consensus among cognitive
linguists as to the nature of these mental/cognitive concepts/representations.
Natural languages leave a lot to be desired. They are, as we have learned,
full of vagueness and ambiguity. They are subject to constant change, and under
closer scrutiny they tend to get lost in an almost infinite diversity of regional,
situational, social, genre-specific and domain-specific variation. We are always
encountering language usages which seem foreign to us. For thousands of years
people have been complaining about the decay of language. Often when we
question language use (normally the way other people use language) we look into
the past. Then we ask ourselves what the word in question 'really' means, and for
we could read
. Apparently we long for a Golden Age when there
was still a natural, uncorrupted relationship between the word and what it stood
for. This explains the popularity of etymological dictionaries. But even Plato, in
his dialogue
, could not convincingly answer the question what makes
such a relationship natural. Thus Socrates asks: "For the gods must clearly be
supposed to call things by their right and natural names, do you not think so?", to
which Hermogenes responds: "Why, of course they call them rightly, if they call
them at all." (Translated by B. Jowett;
uploads/R00010/Cratylus.pdf) In which language do the Olympians converse
when they are among themselves?
In this context it might be worth having a look at machine translation and
artificial intelligence. Perhaps there we can learn why concepts seem to be so
much more attractive than natural language expressions. For these are fields in
which concepts have been key features from early on. But in machine translation
we are confronted with a fusion of two theoretical concepts of the concept.
Concepts are not only mental entities, they are also the staple fare of terminology.
Terms are the expressions of concepts, and a concept is how an element, a feature
or a property pertaining to a specific domain has been defined by the experts. In
terminology, a concept is identical with its definition. Context and usage are
irrelevant, and so is the natural language expression by which it is denoted. In
terminology, just as in cognitive linguistics, the concept is universal, while the
terms, the expressions will differ from language to language.
Terminological concepts can be easily processed by computers. Terms can
be translated on a one-to-one basis and they can be used in information retrieval.
General language words, however, pose problems. They are fuzzy and
ambiguous. A word in one language hardly ever maps onto a word in another
language. In each occurrence of a word, its meaning is contaminated by its
context. This is what makes natural language processing do unrewarding. A
solution to this problem which still has a large following in artificial intelligence
and machine translation is to convert words into concepts. Thus everything
disturbing, unclean, fuzzy and ambiguous can be filtered out, so that we are left
with nothing but the true, authentic, uncorrupted meaning of a concept. This
expectation explains the popularity of conceptual ontologies in the artificial
intelligence community. They account for all the concepts of a given domain and
the relationships that obtain between them. Concepts in language engineering
thus are the language-independent, spiritual, angelic natures of natural language
words which have become unclean through their incarnation. This is how the
difference between the word and its concept is commonly described:
Ontologies describe concepts, not the way these concepts are
expressed in words in a natural language. Therefore it is usually
assumed that the ontology is language-independent. (Hans Weigand
(1997): A Multilingual Ontology-based Lexicon for News Filtering.
Concepts represent the abstract meanings of words, and lexical entries
represent the surface realizations of these meanings… Concepts
represent word meanings, whereas the lexical knowledge they have
represents ways to express these meanings with words. (Mattias
Agnesund (1997): Representing culture-specific knowledge in a
multilingual ontology.
Concepts, it seems, are pure meanings, cleansed from the impurities which they
contracted through the contingencies of change afflicting natural languages. In
these 'language-independent' ontologies there is no room for doubt what is a
concept, how it is defined and how it is related to other concepts. Concepts are
neither fuzzy nor ambiguous. Every proposition is either correct, 'grammatical', or
not. If machine translation is still unsatisfactory then only, according to this
claim, because we still have problems in converting natural language sentences
into their conceptual representations.
What ontology engineers such as Agnesund and Weigand tell us about the
relationship between words and concepts surely accounts also for the attraction of
the innateness theory. Language ceases being unruly once it has been transferred
into a language-independent universal representation. As soon as a natural
language sentence is translated into mentalese, we have something similar to a
mathematical equation, an expression that can be decided on the basis of its
Some notes on the concept of cognitive linguistics
formal properties. In addition, it can claim reality and universality. In formal
calculi, we observe the workings of immutable laws. The linguistics of formal
languages is a pure science.
Is understanding really that simple? Leaving aside, for the moment, the
question whether concepts are innate, holistically acquired or composed of
inherited semantic primitives, what then is the 'real' content of the concept
'carburettor'? Is it enough to know that a carburettor is an important part of an
internal combustion engine which has to be repaired or replaced when it stops
working, or does it include a comprehensive representation of its functionality? Is
the meaning of the word
identical with the (content of the) concept,
or, if not, what is the difference? What could we gain from an analysis of the
concept 'carburettor' what we could not gain from the analysis of the word
? What is the use of mental representations of concepts for linguists?
One reason, I believe, why cognitive linguists not normally ask these questions is
that many of them, particularly those belonging to the camp of one-level
semantics, want to avoid the issue of intentionality. They are content with a
purely syntactic processing of a cognitive representation. For them, the ultimate
'meaning' of a natural language expression is the form of the corresponding
mental concept, i.e. its neuronal representation in the brain tissue.
Corpus linguistics and cognitive linguistics: is there a common
As long as we have nothing but conjectures on swampy ground for cognition, as
long as there is no
which correlates the mind to the brain, would we not
do better to focus our investigation not on language as a mental phenomenon, but
on language as a social phenomenon? This is the programme of corpus
linguistics, The discourse, this virtual structure containing the entirety of all the
verbal exchanges between the members of a discourse community, is the market
into which new objects of discussion are introduced, in which the meanings of
discourse objects are negotiated, by acceptance, modification or rejection what
has been said about them before, by explication and paraphrase, in which we are
told what is proper to say and what not. The discourse is the supermarket where
we shop our attitudes, beliefs and ideologies.
Without the discourse, our minds would be blank slates. The content we
have in our minds is the content we have downloaded from the discourse. The
discourse has an answer to our question what it means to open a bottle of wine.
There is no other way for us to know how a hearer understands what the speaker
says than by exchanging texts, by entering the discourse. Mental concepts, if
there are any, are but the residue of the meaning of the lexical items we encounter
in the discourse. We know what
, means, we know
what we are expected to do if we are asked to open a bottle of wine because it
was explained to us in the discourse.
Can we expect that the two paradigms, corpus linguistics and cognitive
linguistics, will eventually form a joint platform? Mike Stubbs has not given up
this hope. He asks "whether it is possible to state causal relations between
linguistic, cognitive and social patterns". (Stubbs 2003, 1) I am not worried about
finding out how social patterns determine the discourse or how the discourse
constructs social patterns. This is well within the realm of corpus linguistics. I am
much more hesitant with a prognosis of a satisfactory account of the interaction
between linguistic and cognitive patterns. Stubbs is right when he states that
corpus linguists "have hardly considered the relevance of corpus evidence to
questions about the mental lexicon" (Stubbs 2003, 2). But would cognitive
linguists be at all interested in what corpus linguistics has to offer? They tend to
use the corpus mostly as a quarry for examples that suit the points they want to
make. They still cling to the notion of single words as the core units of meaning,
and they do not take into consideration that corpus linguistics has rendered the
alleged fuzziness and ambiguity of natural language as a mock problem that
disappears once we shed the belief that meaning can be studied in single words in
isolation. Can we discuss how what we get out of the discourse has an effect on
what we contribute to the discourse? Corpus linguistics is providing the data for
an empirical theory of 'cultural transmission', and more than that, it has its own
theoretical framework, intertextual hermeneutics, to explain how new objects are
entered into the discourse, how they are accepted, modified, changed or rejected,
and how they are compared, across languages, cultures and times, to other
discourse objects. But do we have to understand how the mind works to
understand these discourse processes? Language is public. The mind is private.
Only when first-person experiences are communicated as testimony they become
accessible. But then they are already a part of the discourse. So what is there to
gain from looking into people's heads? Would we not gain more from a serious
project analysing intertextuality in a diachronic corpus?
Mike Stubbs is, I think, the one corpus linguist cognitive linguists would
listen to. He is sympathetic towards their aspirations, while firmly grounded in
the rationality of empirical analysis. He combines steadfastness, common sense
and patience with outstanding scholarship. I hope that he will initiate an open
discussion between the two camps. If anyone can turn such a dialogue into a
success, it will be him.
Bierwisch, M. 1967. Some semantic universals of German adjectivals. In:
Foundations of Language
3, 1-36.
Bierwisch, M. 1970. Einige semantische Universalien in deutschen Adjektiven.
In: Hugo Steger (ed.):
Vorschläge für eine strukturale Grammatik des
. Darmstadt: Wissenschaftliche Buchgesellschaft, 269-318.
Some notes on the concept of cognitive linguistics
Chomsky, N. 2000 [1992]. Language and Interpretation. In: N. Chomsky:
Horizons in the Study of Language and Mind
. Cambridge: Cambridge
University Press, 46-75.
Chomsky, N. 2000 [1994]. Language as a Natural Object. In: N. Chomsky:
Horizons in the Study of Language and Mind
. Cambridge: Cambridge
University Press, 106-134.
Dennett, D. 1993.
Consciousness Explained
. London: Penguin Books
Dennett, D. 2003.
Freedom Evolves
. London: Allen Lane.
Eco, U. 1994.
The Search for the Perfect Language
. Oxford: Blackwell.
Fodor, J. 1998.
Where Cognitive Science Went Wrong
. Oxford:
Clarendon Press.
Fodor, J. No year. Distributed Representations¸ Enough Already.
Fodor, J. and J. Katz 1963. "The structure of a semantic theory". In:
Greimas, A. J. 1966.
Sémantique structurale
. Paris: Larousse.
Grice, H. P. 1989.
Studies in the Way of Words
. Cambridge, Massachussetts:
Harvard University Press.
Harras, G. 2000. Concepts in Linguistics - Concepts in Natural Language In: B.
Ganter and G. M. Mineau (eds.).
Conceptual Structures: Logical,
Linguistic, and Computational Issues
. Heidelberg: Springer, 13-26.
Harras, G. 2001. Sprachproduktion als kollektives Handeln:
Sprachphilosophische Grundlagen. In: T. Herrmann and J. Grabowski
. Göttingen: Hogrefe, 899-930.
Hjelmslev, L. 1963 [1943].
Prolegomena to a Theory of Language
. Madison:
University of Wisconsin Press.
Jackendoff, R. 1987.
Consciousness and the Computational Mind.
Massachusetts: MIT Press.
Jackendoff, R. 1997. The Architecture of the Language Faculty. Cambridge,
Massachusetts: MIT Press.
Levinson, S. C. 1996. From inner to outer space: Linguistic categories and non-
linguistic thinking. In: J. Nuyts and E. Pederson:
Language and
. Cambridge: Cambridge University Press, 13-45.
Lewandowski, T. 1976.
Linguistisches Wörterbuch
. 3 Bände. 2. Auflage.
Heidelberg: Quelle und Meyer.
McGinn, C. 1991.
The Problem of Consciousness
. Oxford: Blackwell.
Millikan, R. 1984.
Language, Thought and other Biological Categories
Cambridge, Massachussetts: MIT Press.
Millikan, R. [2004]. On the Epistemology of Concepts and How It Implicates
Metaphysical Realism. Unpublished manuscript.
Nuyts, J. and E. Pederson 1997. Overview: The relationship between language
and conceptualization. In: J. Nuyts and E. Pederson (eds.):
Language and
. Cambridge: Cambridge University Press, 1-12.
Origgi, G. and D. Sperber [2002]. Evolution, Communication and the Proper
Function of Language
Pinker, S. 1994.
The Language Instinct
. New York: HarperCollins.
Pottier, B. 1964. Vers une sémantique moderne. In:
Travaux de Linguistique et de
II,1, 107-137 (German translation: Entwurf einer modernen
Semantik, in: Geckeler, H. (ed.):
Strukturelle Bedeutungslehre
. Darmstadt,
Wissenschaftliche Buchgesellschaft (1978), 45-89).
Putnam, H. 1988.
Representation and Reality
. Cambridge, Massachussetts: The
MIT Press.
Rose, S. 2005.
The 21
Century Brain: Explaining, Mending and Manipulating
the Mind
. London: Cape.
Scaruffi, P. 2003. Thinking about thought
Searle, J. 1980. Minds, Brains, and Programs. In:
Behavioral and Brain Sciences
3, 417-424.
Searle, J. 1990. Is the Brain a Digital Computer?
Searle, J. 1992.
The Rediscovery of Mind
. Cambridge, Massachusetts: MIT Press.
Searle, J. 1998.
Mind, Language and Society
. New York: Basic Books.
Searle, J. 2005. "Consciousness: What We Still Don't Know". In:
New York
Review of Books,
Vol. LII, Number 1, 36-39.
Sinclair, J., S. Jones and R. Daley 2004.
English Collocation Studies: The OSTI
. Ed. by R. Krishnamurthy. London: Continuum.
Sperber, D. and D. Wilson 1998. The Mapping between the Mental and the
Public Memory. In: P. Carruthers and J. Boucher (eds.):
Language and
Thought: Interdisciplinary Themes
. Cambridge: Cambridge University
Press, 184-200.
Stubbs, M. 2001.
Words and Phrases: Corpus Studies of Lexical Semantics
Oxford: Blackwell.
Stubbs, M. 2003. Corpus Analysis: The State of the Art and Three Types of
Unanswered Questions. Plenary, International Systemic Functional
Linguistic Congress (ISFC 29) on Systemic Linguistics and the Corpus,
Liverpool 15-19 July 2002. Manuscript, version 12 February 2003.
Teubert, W. 2005. My Version of Corpus Linguistics. In:
International Journal of
Corpus Linguistics
10, 1, 2-16.
Toolan, M. 1996.
Total Speech: An Intergrational Linguistic Approach to
. Durham, NC: Duke University Press
Developing language education policy in Europe –
and searching for theory
Michael Byram
University of Durham
Starting from a discussion of work analysing and evaluating language education policy at
European level, conducted under the aegis of the Council of Europe, this chapter shows
that the Council of Europe has a policy position and influences the policy making of
member states. It does so more directly than through the dissemination of ideas via
networks and workshops as has been done for the Common European Framework of
Reference. For in the process of producing Language Education Policy Profiles, the
Council of Europe promotes a specific view of the purposes of langue education and the
preferred objectives for learners and education systems.
The second, more speculative section reflects on the need for explanatory theory
which relates language policy to social conditions and the education environment. There is
a need to go beyond case studies, useful as these may be, to a taxonomy of polices and
then to an explanation of the implementation of policies according to circumstances. The
underlying question is under what circumstances Council of Europe member states might
accept and implement a policy of plurilingualism. Theory which can predict this will be of
significance in Europe but perhaps applicable beyond.
There are two related but separate purposes in this paper:
- to present a critical analysis of current work at the Council of Europe on
the promotion of a policy for language education
- to emphasise the need for a theoretical perspective on this and other
language policy activity which might help to explain and predict the
outcomes of policy-making.
In the main part I will describe one aspect of the current work of the Language
Policy Division at the Council of Europe in Strasbourg, and raise the question
whether this is an example of policy making at supra-national level. In the
conclusion, I will address what seems to me to be a lack of adequate theorising
about language policy, and language education policy in particular. I do not
propose a means of filling this gap, unfortunately, but hope identifying the gap
will be a first step towards this.
Languages and polities
Anderson (1991) in his well-known discussion of the nation as an imagined
community, points to the significance of language and argues that the close
relationship between language and nation was promoted from a European,
Humboldtian perspective, and was part of the 'model' of the nation-state which
was borrowed - or as he says, 'pirated' - in many parts of the world. By referring
not simply to language but to 'print-language' and the power of newspapers and
books to create a sense of community, Anderson also emphasises the significance
of literacy. A nation-state is thus
inter alia
a community of communication which
needs a shared language, and usually this shared language is the one designated as
the national language. Thus, linguistic identity and national identity are closely
connected, wherever there is a formal, institutionalised community of
communication. The connection is reinforced by schools as national institutions
where one learns the national language, whatever one's home or first-acquired
Yet there are also other levels of community within a nation-state which
are not necessarily formalised. The organisations and institutions of civil society
have differing degrees of formality, and where there is freedom of speech, these
communities of communication can challenge the official discourses of the state
(Kennedy and Fairbrother, 2004: 296). Nonetheless, such discourses are likely to
be conducted through the same national, officially recognised language, and again
we see the significance of the national language and the reinforcement of the
relationship between the national language and national identity.
The significance of communication and interaction becomes all the more
evident as the nature of polities changes. For Habermas, the model which should
replace out-dated concepts of 'the classic republican idea of the self-conscious
political integration of a community of free and equal persons', is a model
dependent on communication flows:
a model of deliberative democracy, that no longer hinges on the
assumption of macro-subjects like the 'people' or 'the' community but
on anonymously interlinked discourses or flows of information
(Habermas, 1994: 32).
This applies to the evolution of the nation-state, but all the more to the evolution
of democratic processes in transnational contexts. Communication flows and the
'informal networks of public communication' at a transnational level pre-suppose
favourable conditions for mutual understanding.
The importance of this issue is evident from the evolution of transnational
civil society in response to the trend towards global governance through such
organisations as the World Trade Organisation and the International Monetary
Fund. The exact nature of the organisation and nature of transnational civil
society and of a democratisation of global governance is not yet clear. However,
it can be argued that the present legitimisation based on the notion that experts
can deliberate and come to representative consensus is inadequate and should,
Developing language education policy in Europe - and searching for theory
and will, be replaced by debate in a public sphere, where a public is understood as
'a collectivity of persons connected by processes of communication over
particular aspects of social and political life' (Nanz and Steffek, 2004: 8). Nanz
and Steffek argue that 'organized civil society has a high potential to act as a
'transmission belt' between deliberative processes within international
organisations and emerging transnational public spheres' (ibid: 10).
Perhaps the most likely place for this to happen first is in the political and
cultural space which has been created in Europe over the last half century.
The role of the Council of Europe is important in this, because of its
influence in forty-five European countries. As an inter-governmental
organisation, the Council of Europe does not have a policy-making function
independently of its member States. On the other hand, in practice, proposals
evolve from meetings and conferences and are ultimately endorsed by member
States at Councils of Ministers. As part of this process, the Council of Europe has
developed in the last few decades a clear language education policy position, and
this was recently stated in a draft document to celebrate the 50th anniversary of
the signing of the Cultural Convention. The statement is not a repetition of the
many and various recommendations which have been endorsed by member
States, but rather a summary of the purposes of these recommendations:
Council of Europe language education policies aim to promote:
: all are entitled to develop a degree of communicative
ability in a number of languages over their lifetime in accordance with
their needs
Linguistic diversity
: Europe is multilingual and all its languages are
equally valuable modes of communication and expressions of identity; the
right to use and to learn one's language(s) is protected in Council of
Europe Conventions
Mutual understanding
: the opportunity to learn other languages is an
essential condition for intercultural communication and acceptance of
cultural differences
Democratic citizenship
: participation in democratic and social processes in
multilingual societies is facilitated by the plurilingual competence of
Social cohesion
: equality of opportunity for personal development,
education, employment, access to information and cultural enrichment
depends on access to language learning throughout life
(Council of Europe, 2004a).
This statement can be taken as a policy position and in the sense that it creates a
consensus which is endorsed by member States, the Council of Europe can be
described as a policy-making body.
Turning to the European Union, we can see a more obvious policy-making
function as nation-states gradually give up some of their power and adopt a more
international, or at least European, perspective. In such circumstances, the notion
of a national language and linguistic identity is weakened and there is
encouragement for other, 'foreign', languages to be given a status as part of the
creation of identification with a community. This is made very clear in the EU's
White Paper of 1995:
Languages are also the key to knowing other people. Proficiency in
languages helps to build up the feeling of being European with all its
cultural wealth and diversity and of understanding between the
citizens of Europe.
(….) Multilingualism is part and parcel of both European
identity/citizenship and the learning society.
(European Commission, 1995: 67).
What we have here then is a statement where the word 'European' could be
substituted by the name of almost any nationality, and the parallels with the role
of language in an imagined community are clear. It is also clear that, as in the
nation-state, the levels of communication are not only those which are formal and
institutional, but also include those of civil society.
The subsequent recommendation for practice is that European citizens
should speak their mother tongue(s) plus two other languages, and this implies
that a knowledge of three or more languages – perhaps to different degrees and in
different ways – will create a sense of European identity and citizenship, and a
potential for participation and integration into an international/ European society
and polity.
This position had changed by 2003, when a weaker statement was issued.
Although it still uses the White Paper as one of its sources, the focus now is on
effective participation and social cohesion; the reference to identity no longer
(1) knowledge of language is one of the basic skills which each citizen
needs to acquire in order to take part effectively in the European
knowledge society and therefore facilitates both integration into
society and social cohesion; a thorough knowledge of one's mother
tongue(s) can facilitate the learning of other languages
(2) knowledge of languages plays an important role in facilitating
mobility, both in an educational context as well as for professional
purposes and for cultural and personal reasons
(3) knowledge of languages is also beneficial for European cohesion,
in the light of EU enlargement
(4) all European languages are equal in value and dignity from the
cultural point of view and form an integral part of European culture
and civilisation.
(European Commission 2002)
Here is an emphasis on mobility but the professional / economic purposes are
linked to the personal, and the specific issue of enlargement from 15 to 25
countries is given prominence.
Developing language education policy in Europe - and searching for theory
In summary, it appears that the European Union approach to language
education postulates some, unclear, relationship between national language/
mother tongue learning and foreign language learning; second, a causal
relationship between language learning and identity/ citizenship; and, third, a
conditional relationship between language learning and participation in European
society. Learning several languages is at least a pre-condition and perhaps a
causal factor in the evolution of citizenship in the narrow sense of being an
elector, and in the broader senses of an affective bond with an international
society and a participation in the economic, political and cultural life of the
What makes the European situation different from nation states is that it is
not expected that people should be native speakers of all the languages they might
acquire as part of becoming European citizens, even though there are powerful
forces encouraging people to acquire as high a level of competence as possible.
The success of a European imagined community of communication pre-supposes
plurilingual competence so that discourses at formal level and in civil society can
take place, can be extended beyond the national frontiers, to European level.
Thus, association of native speaker competence with identification with a polity is
put in doubt, and replaced by plurilingual competence.
The alternative, of creating a shared
lingua franca
– which at this point in
history could only be English – is not politically acceptable since there would be
accusations of linguistic imperialism and/or allowing unequal and unfair
dominance to native speakers of English. Whether these are justified or not, a
lingua franca would not be efficient. Transnational discourses cannot rely on a
single, taken-for-granted, shared language and its meanings. The discourse which
is necessary is not simply a matter of establishing an agreement on and/or an
exchange of information such as might be achieved through a
lingua franca
. The
issues which arise in social discourse are shot through with contemporary and
historical nuances, and the relationship between language and thought, between
language and world-view is crucial.
When people engage in cooperation in civil
society, they do so as social beings whose social identities are embodied in the
languages they speak. To use a lingua franca is reductive of their social identities
and diminishes them as human beings.
Policy Profiles - a tool for policy implementation?
We have then in Europe two supra-national bodies with their language education
policy. However, policy without implementation is ineffective, and from this
point I will focus on the Council of Europe and the question of whether it has this
function too.
Over the recent period of two to three years, the Language Policy Division
in Strasbourg has created a new 'service' offered to member States. The Language
Policy Division helps member States which invite it to do so, to review and
develop their policies for language education. This activity has a broad remit, to
include the teaching and learning of all languages in a polity: from a learning
perspective, first languages, second languages, foreign languages; or put it in
sociological terms, minority and regional languages, national or official
languages, immigrant languages, foreign languages. The polity may be a country
but can also be a region or city with its own language education policy.
The role of the Language Policy Division is made clear in this extract from
the guidelines which govern the activity:
Language Education Policy Profiles
The Council of Europe has launched a new activity to
States who so wish in reflecting upon their language education policy.
The aim is to offer member States the opportunity to undertake a
of their policy in a
spirit of dialogue
with Council of
Europe experts, and with a view to focusing on possible future policy
developments within the country. It should be stressed that developing
a language education policy profile
does not mean 'external
. It is a process of reflection by the authorities and
members of civil society, and the Council of Europe experts have the
function of
acting as
in this process.
(Council of Europe, 2004b - my emphasis added)
The purpose is clear: that the Language Policy Division does not interfere in
policy development but facilitates self-analysis.
The process involves several stages:
- after a preliminary organisational visit, a Country Report, by or on
behalf of the authorities, is written to describe whatever issues the
authorities consider important
- a group of three to five experts visits the country for a week,
talking with representatives of stakeholders in education, and
produces its own Experts' Report
- this Experts' Report is circulated within the country to whomever
the authorities wish, including all the stakeholders whom the
experts had met
- a one day round table discussion is held between all stakeholders
invited by the authorities and the experts, where issues of accuracy,
of comprehensiveness are raised, and where the stakeholders can
exchange views
- in a final stage a Language Education Policy Profile is produced by
the expert group (mainly by the group
) in consultation
with the authorities, and published jointly by the authorities and the
Council of Europe.
Throughout this process, the expert group reminds the readers of its report and of
the final profile about the policy position of the Council of Europe, and about the
instruments it has produced which are useful for the implementation of policy.
Developing language education policy in Europe - and searching for theory
These include the
Common European Framework of Reference
, the European
Language Portfolio, and the
Guide for Language Education Policy
. Thus the
experts, as catalysts, bring to the notice of the authorities and other stakeholders
the policy and instruments which all member States have endorsed, and there are
a number of criteria which underpin the Experts' Report:
- that language education must be considered holistically,
overcoming the separation between first, second, foreign languages
- that the promotion of plurilingualism and diversity is axiomatic in
all planning
- that curriculum design and pedagogy must reflect and be
determined by the holistic vision of a language education
- that language education is tied to education for citizenship in all
multilingual polities, of which European countries and Europe as a
whole are clear examples.
Thus far, then, the role of the Council of Europe through its experts involves in
principle a catalytic function. On the other hand, in practice, it is clear that
member States and stakeholders within a country are not as aware as they might
be of Council of Europe policy and instruments. The catalysts in fact bring new
elements to the process, which cause a re-assessment of existing assumptions,
even though the
explicitly say that there is no external evaluation.
These new elements are particularly characterised by the European perspective
and not just a national, regional or local one. This includes for example an
emphasis on the teaching of all languages irrespective of their social status, with a
strong emphasis on diversification of language learning opportunities throughout
life, resistance to the dominance of English as a lingua franca, a transversal,
holistic vision of convergences in the languages curriculum among national,
minority, foreign and other languages.
Further factors in this catalytic process are the role of the Experts' Report,
written independently by the experts bringing both Council of Europe
perspectives and their own expertise to the analysis, and also the role of the
experts, in particular the
, in the authoring of the final profile. Both of
these allow opportunity for a new and ultimately evaluative perspective on the
assumptions of the authorities and other stakeholders. This perspective has in
practice been welcomed and encouraged by the authorities in most cases hitherto.
There is none the less a delicate balance of power to be sought in the final Profile,
since it has to be acceptable to and published by the authorities and the Council of
The Council of Europe as a policy making body?
I pointed out earlier that the Council of Europe is an entity which has language
education policies, even though in principle these are the formulation of the views
of member States and not independent European policies. It also has instruments
which can be used in the implementation of those policies. On the other hand, it
does not have an obvious function in the implementation of policies; it cannot
send out directives to member states in the way that the European Union can.
Does the Council of Europe none the less have the characteristics of a
policy making body? One way of addressing this question is to use Cooper's and
Ager's frameworks for analysing language policy, to see if and how the Council
of Europe fits into them.
Cooper (1989) provides an ordered list of questions which can guide
policy analysis, and I present these here with application to the Council of Europe
and its Policy Profile activity:
Question: What actors?
An inter-governmental organisation acting with national (or regional or
local) authorities. Hitherto, actors have been individuals, groups or
agencies at or below national level, but the two supranational bodies, the
EU and the CoE have now become active. Ager (forthcoming ) argues that
the European Union is a special case where policy is formulated in general
terms and not pursued in detail because of the political sensitivities. The
Council of Europe is more precise in its formulation and, through the
Policy Profiles is seizing the nettle of influencing member States,
Question: attempt to influence what behaviours?
The CoE attempts to influence production of policy at national (etc) level
with respect to planning language acquisition and, indirectly, attitudes
towards plurilingualism,
Question: of which people?
of national (etc) authorities making policy for language acquisition at
national or regional level,
Question: for what ends?
following Ager's (forthcoming) distinctions, at the level of unattainable
but necessary 'ideals', the CoE sets ideals which are
(a) language related: to increase the diversity of languages in society and
the diversification of languages learnt in the curriculum
(b) non-language related: to promote education for democratic citizenship,
and an understanding of linguistic otherness
secondly, with respect to 'attainable, but long-term objectives', the CoE
(a) language related instruments of various kinds to support language
acquisition planning (the
Common European Framework,
the European
Language Portfolio
Guide for the Development of Language
Education Policies in Europe
) and planning for regional and minority
languages (
The European Charter for Regional or Minority Languages
and the
Framework Convention for the Protection of National Minorities
(b) non-language related instruments for planning education for
democratic citizenship (e.g.
Draft Common Guidelines on Education for
Democratic Citizenship
) although at this level there is as yet no proper
Developing language education policy in Europe - and searching for theory
coordination between language and non-language related activities but, at
the third level, the Council of Europe does not become involved in 'short
term objectives' such as curriculum planning,
Question: under what conditions?
against a background of European integration, economic mobility and
human capital theory, and a move towards an international civil society;
the geographical and political space in question is however no longer
limited to the nation-state, but extends to Europe as the totality of all
member States of the CoE,
Question: by what means?
unlike some states in both past and present, the CoE does not use force or
bribery, but its authority as an inter-governmental organisation, and its
reputation gained through language-related work over several decades,
Question: through what decision making process?
through the formulation of Europe-level policy by recommendations – for
long-term objectives and a process of agreement to the
recommendations at meetings of ministers of education and/or Heads of
State and Government; and through reference to these and the instruments
for implementation in the Experts' Report and the Profile,
Question: with what effect?
in completed studies so far:
- in Norway, with impact on current education reform at the level of long-
term objectives and with some impact on details of curriculum planning
- in Hungary, with input to new policies and plans countrywide for
language teaching, and thus at the level of long-term objectives.
It seems therefore that of the Council of Europe, working with member States (or
other polities) fulfils some of the characteristics of a policy making and policy
implementing body, even though it has no direct power over implementation. If
we consider the three levels of 'ends' in Ager's definitions, then the 'ideals' are
present in the discourse at European level and have begun to infiltrate the
discourse of national bodies. The parallel although not identical position of the
European Union no doubt contributes strongly to this. At the level of long-term
objectives, the use of the
Common European Framework
in planning national
curricula is evident in some countries. In the case of the Country Profiles, the
impact will doubtless vary from case to case and is yet to be seen over a number
of forthcoming cases. At the level of short-term and immediate objectives, the
Council of Europe does not expect to have impact but there may be some
evidence that this happens through the Country Profiles; there are as yet too few
cases to draw any conclusions.
It is thus possible to turn now to the Ager's (2001) model for analysing the
motivations behind language policy-making to see if this can throw further light
on the role of the Council of Europe.
Ager focuses on questions of identity and the motivation in policy making
to maintain or develop (national) identity. If we take some of the elements of his
model which deals with identity sequence, attitudes, and purposes, it becomes
evident that the Council of Europe is acting in some respects in a way similar to
the nation states with which Ager is concerned:
Identity promoted in the policy:
- what identity: the construction of European identity underpins
Council of Europe policy
- what ideology: the Council of Europe promotes the equality of all
languages and the correction of inequalities by supporting linguistic
and regional minorities, and by promoting relations among nation
Attitudes in the policy:
- there is emphasis on the attractiveness of plurilingualism
- there is action taken through the policy profiles and the instruments
to promote plurilingualism
Purpose of the policy:
- there is an explicit pursuit of linguistic diversity, of international
citizenship, and of cohesion among member states.
If we consider the sequence of events, however, there is a difference. Nation
states are the realisation of a bottom-up desire of ethnic groups with an existing
identity for a national identity and political power (Edwards, 1994); the
Wilsonian principle of post-1918 and its presence in the Treaty of Versailles was
one very important reflection of this. After 1945, the Council of Europe started
from an ideal of cultural cooperation and mutual respect among nations, as stated
in the summary of the Cultural Convention:
to develop mutual understanding among the peoples of Europe and
reciprocal appreciation of their cultural diversity, to safeguard
European culture, to promote national contributions to Europe's
common cultural heritage respecting the same fundamental values and
to encourage in particular the study of the languages, history and
civilisation of the Parties to the Convention.
This has been formulated since then in terms of social inclusion, citizenship and
mobility for individuals, and in the creation of a European identity: 'The Council
was set up to (…) promote awareness of a European identity based on shared
values and cutting across different cultures' (
COE/). In this it is joined by the European Union, even though the latter had a
different starting point.
There are also other differences between the Council of Europe and the
nation-state as represented in Ager's model, in particular the relationship with
other comparable entities, and the possible integration with other entities. There
are no comparable entities which, in the case of nation states, are the external
'threats' which help maintain internal unity. Moreover, whereas a nation-state will
usually seek to identify and promote one language – so that it becomes or remains
Developing language education policy in Europe - and searching for theory
a national language – this is clearly not the case with the Council of Europe,
which substitutes plurilingualism for monolingualism or, in multilingual
territories, the dominance of one language.
There are then similar but not identical processes and purposes at work
here to those which Ager identifies, and the points in which his model does not
apply are the indications of difference.
The Council of Europe has, in short, the characteristics of a policy-making and a
policy-implementing body. It is comparable to the nation state and other policy
bodies in this respect even though it operates at different levels and has different
means at its disposal. The question which then arises is whether it is possible to
predict the outcomes of this activity and it is here that I fear there is a gap to be
Our need in the case in question is to predict whether a policy of
plurilingualism and diversification as proposed by the Council of Europe will be
accepted and implemented by member States. They have endorsed it in principle
but principle does not necessarily end in practice. Theory which helps to predict
whether a policy will be successfully implemented in a given set of circumstances
might also allow us to identify inhibiting factors, and to change these in order to
facilitate implementation.
One example exists with respect to the teaching of foreign languages, a
necessary but not sufficient aspect of Council of Europe policy. This is a paper by
Trim (1994) in which he identifies a range of different conditions which are more
or less likely to lead to successful policies for foreign language teaching. This
could perhaps be extended to encompass plurilingualism, diversity and
diversification in the curriculum. Implicit in Trim's paper is an attempt to produce
a taxonomy of language situations. A taxonomy is crucial to prediction, but needs
to embrace the multilingualism within a polity in a holistic way if it is to help in
prediction of the success of policies of plurilingualism. This might lead to
predictions of there following kind:
- in language situations of type A, plurilingualism can be attained by
implementation of a policy of type X.
However, language situations need to be theorised in a sociological perspective
too, since policies exist in the interplay of entities holding power. One approach
would be through Bourdieu's theory of social reproduction through education,
which would suggest that language education policies are subject to the efforts of
certain groups in society to maintain their cultural and social capital through
education and education policies.
The potential for successful implementation of language education
policies can also be analysed using economic theory. Grin (2004) has led the way
in applying economic theory of costs and benefits to policies, as a means of
helping authorities and other stakeholders to make decisions. It is also possible to
envisage an analysis of language and education policy from the perspective of the
debate about the marketisation of education, and whether education should be
treated as a public good. This debate has been particularly vehement in
anglophone countries.
These and other approaches need to be explored not only in the context of
Council of Europe policy work, but perhaps the significance of this work makes
the need all the more urgent.
I am very grateful to Dennis Ager for comments on and suggested
additions to a draft of this paper. I remain of course responsible for its
This analysis is deliberately simplified, and has to be modified
for nation states where there are more than one national language
or where the speakers of a minority language are accorded legal rights to
use the language in public discourse.
All the work currently being carried out in the EU on the development of a
'Europass' for languages or in Strasbourg on a 'European Language
Portfolio' is a sign of the recognition by European authorities, and the
national authorities which support them, that plurilingual competence of
some kind is crucial.
I take a 'weak' Whorfian/Humboltian position which cannot be developed
and defended here but for which there is supportive empirical evidence in
Levinson (1997).
See Breidbach (2003) for a further discussion of levels of public fori and
the language combinations which might be required.
Developing language education policy in Europe - and searching for theory
Ager, D. 2001.
Motivation in language planning and language policy
. Clevedon:
Multilingual Matters.
Ager, D. (forthcoming).
Language policy and language planning
. Strasbourg:
Council of Europe.
Anderson, B. 1991.
Imagined communities.
2nd ed. London: Verso.
Breidbach, S. 2003.
Plurilingualism, democratic citizenship in Europe and role of
. Strasbourg: Council of Europe.
Cooper, R.L. 1989.
Language planning and social change.
Cambridge University Press.
Council of Europe, 2004a.
Plurilingual education in Europe
. Draft document
presented at the Language Policy Forum, June 2004.
Council of Europe 2004b.
Language education policy profiles. Guidelines and
. .
Edwards, J. 1994.
London: Routledge.
European Commission 1995.
Teaching and learning: towards the learning
, Brussels: European Commission.
European Commission 2002. Council Resolution of 14 February 2002, Official
Journal C 050, 23. 02.2002.
Grin, F. 2004. Towards a public policy approach to language scenarios for the
EU. Paper at the conference 'Language and the Future of Europe',
Southampton, July 2004.
Habermas, J. 1994. Citizenship and national identity. In B. van Steebergen (ed.)
The Condition of citizenship
London: Sage. 20-35.
Kennedy, K.J. and Fairbrother, G.F. 2004. Asian perspectives on citizenship
education: postcolonial constructions or pre-colonial values? In W.O. Lee
et al.(eds.)
Citizenship education in Asia and the Pacific. Concepts and
. Hong Kong: Comparative Education Research Centre and Kluwer
Academic. 289-301.
Levinson, S.C. 1997. From outer to inner space: linguistic categories and non-
linguistic thinking. In S. Nuyts and E. Pederson (eds.)
Language and
Cambridge: Cambridge University Press. 13-45.
Nanz, P. and J. Steffek 2004. Global governance, participation and the public
Government and Opposition
39, 2, 314-35
Trim, J.L.M. 1994. Some factors affecting national foreign language
policymaking in Europe'. In R. Lambert (ed)
Language planning around
the world: contexts and systemic change.
National Foreign Language
Center Monograph Series. Washington D.C. NFLC. 1-15.
The semiotic patterning of Cædmon's Hymn as a 'hypersign'
Wolfgang Kühlwein
University of Trier & Université du Luxembourg
Cædmon's Hymn has been researched extensively with a view towards its assumed
significance as an early key document of English literary and sociocultural history, and as
an (often questioned) 'masterpiece' of the evolving skills in the use of Anglo-Saxon poetic
devices, such as metre, rhythm, alliteration, variation etc. However, accounts of its overall
structure have remained scarce.
Taking for granted the high intensity of both its emotive and its appellative load as
explicitly commented on by Bede himself, an approach that is methodically based on
(Peircean) semiotics lends itself for a descriptive analysis of its overall patterning.
Basically, the Hymn presents itself as an interlace of the intricate network of the
dyadic interrelationships between 'God and Creation', 'Creation and Mankind' and 'God
and Mankind'. However, it is God's act of GIVING that elevates it above a mere
accumulation of dyadic patterns. It is the explanatory claim of this semiotic approach to
the overall patterning of this 'hypersign' to account for the alleged semiotic thrust of the
Professor Michael Stubbs, whose 60
birthday this Festschrift celebrates, is
enjoying world-wide reputation in many areas of the multifaceted linguistic
landscape; in particular, he may well claim to have been a pioneer for present-day
corpus linguistics and to be one of its prime representatives today.
As to his relationship to the author of this contribution to the Festschrift:
both served as Presidents of national affiliates of the International Association of
Applied Linguistics (AILA), Mike Stubbs of BAAL, the author of GAL; and as
Professors in the Department of English at the University of Trier/Germany both
have in the literal sense of the word been 'next-door' neighbours for these last 17
years – a neighbourhood that enhanced opportunities for informal talks about all
the diverse decision-making procedures in university management and for many
ad hoc discussions of scholarly matters.
What, therefore, seemed to the author to be an appropriate token to show
his gratitude for all that Michael Stubbs has been giving to him, had, of course, to
be a study that is corpus-based and that chooses a "gift" for its theme – the only
difference being that its giver is not Mike Stubbs but God Almighty, and the
receiver is not Wolfgang Kühlwein alone, but all Mankind.
The research object
As opposed to the large size of corpora drawn upon by Professor Stubbs, the
corpus we are going to analyse happens not to exceed 9 lines! It will be set into
relief, however, against the total corpus of Old English poetry wherever
hypotheses based on a work-immanent view call for further evidence, be it for
purposes of their verification or of their falsification. Among scholars of English
and Germanic Philology, Historical English Language Study, Anglo-Saxon
Literature, and Theology of Early Christendom likewise, these 9 lines have
become well known as
Cædmon's Hymn
), '
will have been composed at the Northumbrian abbey of
/ northumbr.
Streunaes Halh
) sometime between 657 A.D.
and 680 A.D., i.e. during the term of office of the reliably recorded abbess Hild,
the ruling Northumbrian King's sister.
has been handed down to posterity about half a century later by
the Anglo-Saxon chronicler Venerable Bede (=
Beda Venerabilis
: b. ca. 672
A.D.; visitor to Yorkshire and in all likelihood to the monastic and scholarly
Streunaes Halh
at least once, in 733 A.D.; d. 735 A.D.). His
Ecclesiastica Gentis Anglorum
(Book IV, Chapter XXIV) offers the by now
famous account of how that
conceived by
the sleeping cowherd
Cædmon at the instigation of a miraculous nocturnal apparition (
sum mon
some 'man' some 'living being': mostly interpreted as 'an angel', or even as 'the
Almighty' [contested]) who induced him to –very reluctantly – sing a song, a
song of the beginning of existence.
Bede offers the Hymn in its Anglo-Saxon version, furnishing a translation
into Latin.
In its Northumbrian version of MS Cambridge University Library Kk,
5.16 the text of
runs as follows:
1 Nu scylun her
an hefaenricaes uard,
2 metudæs maecti end his mod
3 uerc uuldurfadur sue he uundra
4 eci dryctin,
or astelidæ;
5 he aerist scop
aelda barnum
6 eben til hrofe, hale
7 tha middun
eard moncynnæs uard;
8 eci dryctin
æfter tiadæ
9 firum foldu,
frea allmecti
Now we shall praise the Guardian of the heavenly kingdom, the
power of the Almighty, and His spirit and thought, the achievement of
the Father of Glory, because He, the Eternal Lord, set the beginning
for any kind of wonders; in the beginning He, the Holy Creator,
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
Guardian of Mankind, shaped heaven as the roof for the children of
men, and the earth to live on; afterwards the Eternal Lord, the
Almighty Ruler, adorned the world for Mankind
The prime legitimation for our analysis springs from the multitude of scholarly
evaluations of
. Be it from a linguistic, from a literary, from a theological
stance, they do not merely differ, but they are as incompatible as can be,
oftentimes irreconcilably contradictory. To provide a few examples out of many:
(a) "…Cædmon's Hymn appears to display no great originality, for,
though it is technically accurate, nine or more of its eighteen half-
lines can be paralleled in other poems" (Smith 1933: 14f.)
(b) "…it has qualities of balanced and rhythmic grandeur…" (Wrenn
1947: 9)
(c) "The Hymn has hardly enough literary merit to allow of discussing
it at any length as a piece of poetry…" (Kane: 1948 250f.)
(d) "…the
is made up entirely of formulas or systems of
formulas, in a word, …its language is quite traditional" (Magoun
1955: 53)
(e) "a technically miserable performance" (ibid.: 57)
(f) "…his [Bede's] attitude toward poetry is Augustinian…. Since God
is the source of all beauty, it follows that for Bede the
some relation to the divinely inspired poetry of Scripture. Angelic
inspiration implies revelation: The angel brings to a chosen vessel,
characteristically humble, the obligation to receive and to be the
first to communicate God's word in English poetry. In
consequence…. Caedmon's
must for the believer have
seemed as nearly perfect as man's work may be; either the poem
was beautiful to the eyes of faith, or there was no miracle. It is
impossible that God should have inspired what is inferior or merely
workmanlike. Since the demands on the little poem were very
large, Bede must have seen in it much more than the best disposed
modern is likely to allow" (Huppé 1959: 102f.)
(g) "one of the greatest landmarks in the history of our English poetry"
(Wrenn 1968: 57)
(h) "Cædmon's Hymn appears to display no great originality" (Smith
(i) "freshness and originality" (ibid: 15 !)
(j) "a pleonastic tour de force" (Bessinger 1974: 93)
(k) "Cædmon's Hymn…shows an adaptation of traditional style and
Christian content (if the usual view of the poem's heroic diction is
right) so perfect and so comfortable that one cannot help finding
the poem –as Bede did– 'miraculous'" (Gardner 1975: 7).
In view of these discrepancies among both the linguistic and the literary
evaluations, what is indicated is the search for an approach, which is
superordinate to both the particularizing studies of literature and of language.
This cannot but be an approach based on the theory of signs – signs as
encompassing literature, language, and to a large extent, theological exegesis, too:
i.e. the theory of General Semiotics.
The theoretical and methodical toolkit for the analysis
As to the reasons that were responsible for our decision in favour of
semiotic theory (
. competing ones), to base our analysis on, we refer the reader
to Kühlwein (2006a: 105-108). With a view to our specific object of research,
, we opted for Charles Sanders Peirce's semiotic theory as an adequate
research tool. Admittedly, Peirce himself never made use of it to describe and
explain the semioticity of literary texts, let alone of texts from periods as remote
as the Anglo-Saxon one. His own applications are mainly devoted to explain
phenomena from the realms of philosophy, theology, mathematics, natural
sciences, above all from logics. However, Peirce postulates: "Logic, in its general
sense, is… only another name for
semiotic, …
the quasi-necessary, or formal
doctrine of signs" (Peirce 2.227). Both his
and his
Elements of
are concerned with relations – and it is a set of interrelational patterns that
our semiotic analysis intends to cast some light upon, the relations between God
and Creation, between Creation and Mankind, and between God and Mankind, as
well as the intricate relational interlace of these three relations among each other.
As a full presentation of Peircean semiotics would be out of place here, we
shall confine the presentation of our Peircean theoretical-methodical 'toolkit' to a
sketch of those elements that come to bear in the analytical section below:
In his
Peirce distinguishes as the three elements of
phenomena 'quality', 'facts', and 'thought'. His
Elements of Logic
likewise centre
on these three properties, each one being triadic in turn. In 2.233ff. he
distinguishes three general kinds of triadic relations:
Their nature
Triadic relations of comparison
Logical possibilities
Triadic relations of performance
Actual facts
Triadic relations of thought
Any triadic relation has
three correlates
1st correlate (=
Simplest nature Representamen
2nd correlate (=
): Middling
3rd correlate (=
): Most complex Interpretant
Phenomenological definitions and semiotic examples:
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
: a phenomenon, whose essence is determined by mere strength
of itself, e.g. the relationship of a phenomenon to itself as representamen
: a phenomenon, whose essence is determined by strength of its
relation to something other than it is itself, a 'second', e.g. the relationship of a
phenomenon to the object which causes it
: a phenomenon, whose essence is determined by strength of
relating a 'first' to a 'third' via a middling 'second', e.g. a phenomenon as
representing a certain object in such a way as to cause a certain interpretation.
As noticed above, for Peirce logic in a wider sense is semiotic. The
"phenomena" of semiotic are signs. It follows, that signs, too, are subject to these
triadic relations and their correlates. According to 2.243ff. ('
Trichotomy of
'), signs can be subclassified by three categories:
According as the Sign in itself is
a. A mere quality
sign (= a quality which is a Sign) – 'First'
b. An actual existent
sign (= an actual existent thing/event
which is a Sign) –'Second'
c. A general law
sign (= a law that is a Sign) – 'Third'
According as the relation of the Sign to its Object consists in
a. the Sign's having some character in itself
– 'First'
b. the Sign's having some existential relation to
that Object
– 'Second'
c. the Sign's having a relation to its Interpretant
– 'Third'
According as its Interpretant represents it as a Sign of
a. Possibility
(= a Sign of qualitative possibility) – 'First'
b. Fact
(= a Sign of actual existence) – 'Second'
c. Reason
(= a Sign of law) – 'Third'
A few explanatory comments will be required here:
As to (II a), an
Iconic Sign
refers to the Object that it denotes, merely by virtue of
inherent characteristic features of its own
As to (II b), an
Indexalic Sign
refers to the Object that it denotes, by being really
affected by that Object
As to (II c), a
refers to the Object that it denotes, by virtue of a
law, usually an association of general ideas
As to (III a), a
Rhematic Sign
, for its Interpretant, is understood as representing
such and such a kind of possible Object; it represents its Object in its characters
As to (III b), a
Dicentic Sign
is a sign, which, for its Interpretant, represents its
Object in respect to actual existence; therefore it cannot be an Icon, which affords
no ground for an interpretation of it as referring to actual existence; the Dicentic
Sign affords ground to judge whether what it expresses is true or false
As to (III c), an
is a sign which, for its Interpretant, is a sign of law; it
is understood to represent its Object in its character as a Sign; the Interpretant of
an Argument represents it as an instance of a general class of Arguments.
Thus any sign is constituted by the three relationships (I), (II), and (III),
i.e. by its relationship to itself, to its Object, and to its Interpretant.
Each one of these three relationships allows for its further subtle
classification according to the three triads (a), (b), and (c). Theoretically this
would yield 66 classes. However, only 10 actually occur, because both the basic
trichotomy on the one hand (Representamen, Object, Interpretant) and the
respective subclassifications (quali, sin, legi; icon, index, symbol; rheme, dicent,
argument) on the other hand are hierarchically structured: Thirdness involving
Secondness and Secondness, in turn, involving Firstness; e.g. a legisign is
materialized by its occurrence in actual existence, in other words as a sinsign,
which in its turn can only be perceived as a qualisign: In Peircean terms: a
qualisign that embodies a sinsign is a '
' of the latter one; and a sinsign that
manifests a legisign is the latter one's '
' likewise.
an argument sign cannot but be both symbolic and legi
a dicentic sign can neither be iconic nor quali.
Technically speaking, the 'path' in the direction from (I) via (II) to (III) can never
lead 'upwards':
(I) (II) (III)
a. legi symbol argument
b. sin index dicent
c. quali icon rheme
Thus e.g.
a sign with the properties I c – II c – III c can exist: a quali - iconic – rheme sign;
a sign with the properties I b – II c – III c can exist: a sin - icon – rheme sign;
a sign with the properties I b – II b – III b can exist: a sin - index - dicent sign;
but a sign with the properties I c – II b – III c cannot exist;
nor can a sign with the properties I b – II b – III a, etc.
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
Base and aim of the analysis
The Base
We have tried to pave the way towards this analysis by means of two preceding
The first study, Kühlwein 2006a, gave a semiotic in-depth analysis of
Bede's narrative on how Cædmon conceived of the Hymn and of its effects on the
hearers. The concept of 'gift' is the hinge for both Bede's narrative and for the
Hymn likewise: God's gift to Cædmon to praise (according to Bede's narrative)
God's gift of Creation as bestowed upon all Mankind (according to Cædmon's
Hymn); two acts of creative shaping are thus intertwined. Semiotically God's
bestowal of the Hymn to Cædmon reaches beyond random factuality
(=Secondness), the relationship between the giver and the recipient rather turned
out to be a rule-governed one, involving a law, i.e. semiotically the mind. This
elevates that act of giving beyond a sum of two factual dyads , (1) "God gives to
Cædmon" and (2) "Cædmon receives from God" to the level of a semiotic triad,
involving (3) the recipient's responsibility as to his future appropriate use of the
gift, and this is exactly what Bede subsequently expounds: arisen from Cædmon's
pure feeling (Firstness), raised to an appeal (Secondness) to kindle the listeners'
minds to turn to the continuous quest for heavenly bliss (Thirdness).
Finally, the semiotic relationship between the narrative and the Hymn
itself led to the tentative hypothesis of the Hymn being a Dicentic Symbol.
The second study, Kühlwein 2006b, traced that hypothesis from the point-
of-view of the semiotics of the three key concepts of the Hymn. It investigated
the lexemes used in the Hymn to designate God, Creation, and Mankind. The
semiotic relations holding for all lexemes used for each one of these three
concepts were traced separately (i.e. '
conceptually'), not yet, however,
conceptually'. That intraconceptual analysis indicated that the overall
semiotic structure of the Hymn in its entirety might well be much more complex
and considerate than many critics have assumed hitherto – an observation that
would tie in with the previously assumed character of the Hymn as a Dicentic
The Aim
It is the aim of our following analysis (part 5) to ultimately verify or falsify that
hypothesis, which claims that the Hymn in its entirety is o n e Dicentic Symbol.
What is required to achieve that goal, is
- the textstructural semiotic analysis (part 5.1) of the Hymn in its
entirety from the
conceptual stance, i.e. the analyses of
(1) the interrelationship between God and Creation (part 5.1.1)
(2) the interrelationship between Creation and Mankind and (part
(3) the interrelationship between God and Mankind (part 5.1.3) and
- its collation with the results of the above-mentioned lexical analysis
under the auspices of the semioticity of the Hymn when viewed as
one whole entity (part 5.2).
The semiotic analysis
Textstructural semiotic evidence
The Interrelationship between God and Creation
The perspective under which to experience Creation is set by ll. 1 –3a: Cædmon's
exuberant praise of God. These five half-lines comprise as many as four
representamens of God, followed even by a fifth, pronominal one in the sixth
half-line. It is as late as in l. 3 that Creation appears as a theme: strictly speaking,
Cædmon's presentation of process and product of Creation starts with l. 3b. This
thematic divide, that we make between l. 3a and l. 3b seems to be evident on the
sentence-semantic plane, likewise.
The hinge of the matter is the semantics of
- hitherto unresolved in this
passage. In Modern English translations of
it is rendered in three different
(a) Some editors choose 'in this way, thus' for its equivalent. This yields a
reading for ll. 1 – 4 "Let us now praise God,……
In this way
He set the
beginning of all wonders…" i.e. God's doing is described.
(b) Sometimes it is translated as 'how'. This yields a reading for ll. 1 – 4 "Let
us now praise God,…… ,
He set the beginning of all wonders…" i.e.
again God's doing is described.
(c) Despite Bede's translation-equivalent for
as lat.
, e.g. Mitchell
(1967:204) proposes a causal connection as a third possibility, which
would render
as 'because, inasmuch as', thus yielding a reading for ll. 1
– 4 "Let us now praise God,……
he set the beginning of all
wonders…"; i.e. the reason for God's praise is emphasized.
Semiotically, it seems, the reading that implies a certain amount of causality
beyond mere description, can be supported on the basis of the textual structure
Hymn in its entirety:
The semiotic entailment of readings (a) and (b) is that the praise of God in
his unquestionable 'Firstness' would reach far beyond ll. 1 – 3a , and would,
actually, be extended right to the end of the entire text.
The semiotic entailment
of reading (c), on the other hand, equals reading (a) and (b) insofar as it allows
the praise of God in his 'Firstness' in ll. 1- 3a likewise; however, it differs insofar
as it emphatically ('because'!) proceeds from the praise of God's 'Firstness' to His
'Secondness', i.e. to God in his 'factual' role as the agent, the doer, the creator.
Both roles are different but inseparable.
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
As will be become apparent below, the emphasis on that passage from
God's 'Firstness' to His 'Secondness' is a decisive feature of
constitutive theme
in its entirety, contributing towards making it what we call a semiotic
Conceiving of
as a causal connector – be it with causality as the only
and primary sense, be it as a secondary overtone merely -
would mark a
cesure between l. 3a and l. 3b considerably more distinctly than the senses 'thus'
or 'how' would do; and, actually, it is from l. 3b onwards that God is presented in
His 'Secondness'. Very much like
in l. 1a in its function as a
textsemantopragmatic opener to ll. 1 – 3a,
as initiating l. 3b would function as
the textsemantopragmatic opener to all that follows: both ll. 3b – 4 in general and
ll. 5 – 9 in particular. The semiotic reason, why we would like to emphasize a
clear recognition of that passage between God's 'Firstness' to His 'Secondness' –
along with the markedness of the corresponding formal cesure between l. 3a and
l. 3b - will become apparent below.
In any case, our look at the interrelationship between God and Creation
should, definitely, set in with l. 3b. We shall move from form to function.
Formally the part ll. 3b – 9 of the Hymn has to be divided into three
(a) ll. 3b – 4 are a sign for the relationship between God and Creation in
(b) ll. 5 – 7 are a sign for the interrelationship between God and Creation in
(c) ll. 8 – 9 (if not interpreted as a mere follow-up of scop in l. 5) are a sign
All three sections, thus divided, evince an identical basic semiotic pattern. Each
occurrence of Creation in general (see (a)) or of each specific element of Creation
(see (b)) is conjoined with its Creator by a twofold link:
(a) or (
) is caused (
) by
he + eci dryctin
is caused
he + hale
tha middangeard
is caused (
) by
[as resuming
] +
moncynnes uard
is caused (
eci dryctin
frea allmecti
This overall pattern is extremely pervading and, as a consequence, cannot but be
semiotically significant.
The apparent iconicity of this recurrent pattern calls for
an equally iconic visualization, see Figure 1.
for God's making Creation inhabitable for Mankind.

Figure 1: God – Creation
(References to God: bold-face type; to attributes of His: dotted; to
Creation: regular face type)
God's essence
– His 'Firstness'
God the Creator
– His pure 'Secondness'
God: Creator as Giver
– His 'Secondness'
-� His 'Thirdness'
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
These charts bring to light another special feature shared across (a) – (c). Each
one of the references to Creation is framed by one reference to God that precedes
it, and one reference to God, that follows it:
is enclosed between
is enclosed between
(as intensifying
moncynnæs uard
is enclosed between
eci dryctin
frea allmecti
Semiotically we can, legitimately, conceive of this textstructural feature as a sign
of its own right. It is a sign of high iconic load, in other words a very elementary
one: it is in a kind of photographic way that it shows God as the beginning and as
the end of whatever is created. Creation in its entirety (
) is
embedded in the Creator's essence and it is He who literally embraces each
specific constituent of Creation (
eben, middun
eard, folde
But in addition, this 'photographic picture' indicates a relationship of a
higher semiotic order, too. Let us, therefore, still conceive of ll. 5 –9 as being
and ask how it refers to the object it stands for. The continuity of Cædmon's
interlocking pattern between signs denoting God and signs (syntactically:
phrases) denoting the act of Creating indicates, that there is nothing that might
intervene between God's act of creating and the coming into being of Creation
itself. The act of creating pertains to God as such. Semiotically we might say, it is
an emanation of God's Absolute Firstness, i.e. His 'essence'. However, it makes
Him the (active) subject and it makes Creation the immediately affected (=
passive) subject, within a mutual relationship, which holds between Him and
– and this reveals God in His Secondness (= His creatorship).
Linguistically God is topic throughout, Creation is comment. As far as that we are
faced with a prime example of a semiotic
Cædmon cannot but must have felt that dyadic character: quite obviously,
it was impossible for him to think of Creation without referring to God at the
same time. Ll. 1–3a provide further evidence: praising Creation (
) is
equivalent to praising the Creator. Cædmon leaves no doubt either, that the two
subjects, which constitute this dyad 'God/Creation', differ as to their semiotic
status. God is First, Creation Second: even a simple numerical account can be
regarded as a mirror of this gradation: ll.1–3a invoke God as First four times, and
Creation once (
); ll.3b – 9 refer to God as First seven times, to Creation four
The Hymn in its entirety reveals a further highly iconic feature of that
dyad. The signs denoting Creation and its individually mentioned elements are
not only inseparably embedded by those, which denote God, by strength of the
dyadic relationship, but the Hymn sets out with those many invocations of God,
and its last line 9b ends up with a reference to God, too. The appropriate iconic
representation of this will have the shape of a ring. There can only be one object
which this ring will designate: God , who has no beginning and no end, the
e c i
This simile of an encompassing iconicity is indicative of great
compositorial artfulness: its accomplishment is the weaving together of all single
strands (here: all signs) that constitute the Hymn, into a texture of oneness. It is
this very integrity that makes the Hymn what we choose to call a 'hypersign'.
The interrelationship between Creation and Mankind
As to the relation between Creation and Mankind two semiotically relevant
patterns emerge: ( as mirrored in ( below. They are similar to the
ones which reflect Cædmon's conception and presentation of the interrelationship
between God and Creation.
The references to Mankind are syntactically 'enclosed' and semiotically
'embraced' by signs which denote God. What might, therefore, seem to be
without any mention of Mankind at all. However, this is what is semiotically to
be expected qua the character of that introductory part ll. 1- 4: it is a praise God's
Absolute Firstness. It cannot but be along with the references made to non-human
Creation (
), that Mankind, too, enters the Hymn, i.e. as
late as in its second part - because it is this part that reveals God His (creative)
What, nevertheless, seems to be semiotically significant, is the fact that no
specific reference is made as to the sixth day of Creation, i.e. the very Creation as
such of Mankind. Whereas
is satisfied with presenting the mere sequence
of creative acts, ending up with the Creation of Adam and Eve, Cædmon abstains
from merely recapitulating that time- sequence. Instead, he provides a purposeful
underpinning for God's creative doing from the very outset of Creation: that
Creation is intended for the benefit of Mankind is expressed by Cædmon as early
as in l. 5b (
aelda barnum
). Thus this destination initiates his presentation of the
entire act of Creation. In
God gives this purpose to His Creation as late
as on its sixth day:
"dixitque Deus ecce dedi vobis omnem herbam adferentem semen
super terram et universa ligna quae habent in semet ipsis sementem
generis sui ut sint vobis in escam/ et cunctis animantibus terrae
omnique volucri caeli et universis quae moventur in terra et in quibus
est anima vivens ut habeant ad vescendum et factum est ita" (
29-30; Biblia Sacra Vulgata).
["And God said, Behold, I have given you every herb bearing seed,
which is upon the face of all the earth, and every tree, in the which is
the fruit of a tree yielding seed; to you it shall be for meat. / And to
every beast of the earth, and to every fowl of the air, and to every
thing that creepeth upon the earth, wherein there is life, I have given
every green herb for meat: and it was so" (King James Bible)].
Considering that
is a 7
century poem, this difference is amazing.
Semiotically it bears evidence of the great significance, which Cædmon must
incongruent at first sight: the introductory part of the Hymn, i.e. ll. 1 – 4, does
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
have attributed to the idea of Creation as constituting a gift bestowed upon
Mankind (– and thus paralleling the gift which Cædmon himself had received
from the angel in his dream, i.e. the Hymn).
This concept of 'gift' runs through the entire part ll. 5 –9 of the Hymn in a
consequent pattern:
aims at the dative
as its recipient
likewise aims at
aims at the dative
as its recipient.
Figure 2 indicates this iconicity.
Figure 2: Creation – Mankind
(References to Creation: bold-face type; references to Mankind: dotted
In each case the recipients are mentioned first, the gift itself second. In the last
firum foldu
in l. 9, both are 'consociated' with each other to the utmost
density. This consociation between the non-human and the human part of
Creation does not make them a dyad. The contiguity between both rather is
suggestive of a syllogism.
(a) Earth has a certain relation to God (qua having been created and shaped by
(b) Mankind has a similar relation to earth (qua having been created and shaped
from earth): a Biblical reference that will have been well-known at a
religious centre like Whitby, where Cædmon had served as a minor
brother in the monastery)
The result:
(c) Mankind has a similar relation to God (qua having been created and shaped by
Thus, this contiguity that exists between non-human existence and human
existence in the second part of the Hymn, culminating in its last line, provides the
answer to the question, why the Creation of Mankind, i.e. the sixth day of
Creation, is not mentioned as a creative act of its own right. Pragmatically
however, it
mentioned: by means of implicature.
As seen from a semiotic perspective, this fact might well mirror Cædmon's
'hierarchy' in as far as the mere act of creating is concerned, i.e. in as far as God's
Secondness is manifested:
God ranks first, Creation second, Mankind third.
And once again, just like in the case of the interrelationship between God
and Creation, a sheer numerical count of the references to these three concepts
(God, Creation, Mankind) provides further evidence for Cædmon's 'hierarchy':
among these three concepts Mankind is the one, to which the fewest references
are devoted; actually, two altogether. There are
aelda barnum
only; at
moncynnæs uard
can be added. A semiotic look at
at that very place (l. 7 b) in the Hymn might well give further support to our
stance: this syntactic construction consociates God and Mankind to the closest
extent possible – and two lines after that it is in l. 9a, that Creation and Mankind
are consociated in an equally close way:
firum foldu
, i.e. not via the stylistic
device of variation but directly within one half-line; semiotically: God takes care
of Mankind (
God created and adorned (
) the earth (
) as now
existing for the benefit of Mankind.
The interrelationship between God and Mankind
As Mankind does not enter the Hymn until the specific part of Creation as
beginning with l. 5, the text-based analysis of the interrelationship between God
and Mankind should set out from there.
As indicated in 5.1.2 above, semiotic parallels with Cædmon's conception
of the interrelationship between God and Creation are obvious. Like there, both
references to Mankind are joined with its Creator by a twofold link:
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
(reemphasized by
moncynnæs uard
) are the
benefactors for
aelda barnum
eci dryctin
frea allmecti
are the benefactors for
Figure 3: God – Mankind
(References to God: bold-face type; references to Mankind: dotted
To be more specific: each one of the two explicit references to Mankind is framed
by one reference to God, which precedes it, and another one, that follows it:
aelda barnum
is 'embedded' between
is 'embedded' between
eci dryctin
frea allmecti
In addition, as we have seen above, in
moncynnæs uard
both, God and Mankind,
enter into an even closer relationship.
The semiotic conclusions which we drew from the patterning of the
interrelationship between God and Creation of the non-human world, obviously,
hold for God's relation to the human part of Creation, i.e. Mankind, too.
Semiotically Cædmon drew on a high degree of iconicity throughout. We tried to
indicate that pervading feature in our figures above, consistently emphasizing that
pervading pattern of 'embedding' / 'embracing'.
As to the interrelationship between God and Mankind, however, there is
one essential difference:
In His relation to nonhuman Creation God is designated as the 'maker' the
Creator, i.e. in His Secondness.
However, one should mind that in His relation to Mankind His property of
being its Creator is merely
(s.a.). Why?
It rather is God's property as a 'giver', that constitutes His relation to
Mankind in the Hymn.
This difference has quite a strong semiotic impact: 'giving' as a sign is, as has
already been indicated above, not Firstness, nor is it pure Secondness. It is
genuine Thirdness.
Intermediate Conclusions
The semiotically relevant observations as gained from textual structure are:
(a) the conception of God as setting out from His Absolute Firstness
via His Secondness to His Thirdness
(b) the theme of temporality as materializing in the above-mentioned
icon of a ring
(c) the hierarchical sequence 'God first, then Creation in general, then
Mankind specifically'
(d) the Hymn in its textual entirety as constituting
unitary sign,
signifying: The exhortation to God's praise for His gift of Creation
for the benefit of Mankind.
Lexical semiotic evidence
Are these observations supported, questioned or contradicted by the lexical
choices concerning the key concepts of the Hymn?
Ll. 1– 3a: hefaenricaes uard – metud – maecti - mod
idanc – uerc –
The initiating 'opener'
Nu scylun her
('Now ought we to praise') is the verbal
resonance to the strong urge towards making an appeal as felt by Cædmon when
he was stirred by that nightly angelic apparition. It is an imperative, signifying a
command, that is directed to himself and the angel first, to his brethren next, to all
mankind at last; it is an index.
Most other 'openers' met in the corpus of Germanic poetry are quite
different. They would denote circumstances that can – be it historically, be it
fictionally – be either verified (e.g. according to shared experiences, beliefs,
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
assumptions) or falsified. Semiotically, they even come close to being
arguments, employing reasoning, i.e. Thirdness.
The opening to
definitely, is on a considerably lower semiotic level.
As reported by Bede,
arose from mere personal feeling and perception – i.e.
in Peircean semiotic concepts: elementary Firstness – a feeling, that explodes into
an imperative burst of appeal/command. This spontaneous outcry is a Rhematic
Indexical Sinsign. It is indexalic, because this singular utterance directs the
addressees' attention to the very object by which its occurrence is caused, i.e.
God's Creation for Mankind; it is rhematic, because at that initial stage nothing
definite is said yet about that object Creation. As a Rhematic Indexical Sinsign
this opener will have secured the listeners' attention and will have aroused their
desire to hear more about what will have caused the occurrence of a sign of that
To sum up: the opener has its roots in Cædmon's state of pure feeling
(quali – icon – rheme : Firstness), and slightly raises that semiotic level towards
Secondness (sin – index – rheme).
This is the very semiotic level that is maintained in the following ll. 1b –
3a; now, however, projected onto God: (His) Firstness, now likewise slightly
shading over into (His) Secondness.
The phrase
hefaenricaes uard
as it stands is a sign of essence (rheme). If
there were a copula ('he
the warden of the heavenly kingdom'), the statement
would be open to being questioned (dicent). The elementary quality of that sign
hefaenricaes uard
is a safeguard against such –inappropriate – questioning
of God. As such it is a sign of the simplest nature, of mere (unbounded)
possibility - introducing God in his quality as Absolute and Irreducible Firstness.
It is in ll. 2a – 3a that this elementary possibility is to gain a profile: the
essence of God's
, and
is presented gradually and, as there
is no verb that would allow for debate, cautiously. In other words, there cannot be
any questioning of God's might, his mind and thought, his work.
All three essences are associated with each other in a sequence, pointing
out from iconicity towards more indexality in two directions: towards mankind
and by means of their constant association with God (
metudes; his; uuldurfadur
towards God Himself, too: in Anglo-Saxon poetic diction
evokes both
'protection' and 'lordship'!
In many instances
denote 'power' as manifesting itself in
causing, achieving, even creating and providing protection, and the related verb
shades over into 'to help, provide benefit': a strongly intensifying index as
to the
, related to the verb
'to measure', might in the 7
century still
have had that connotative ring of 'measuring out [i.e. the span of life (as the
North-Germanic norns had done)]'. In this capacity it is the proper link between
the component 'Lordship' in the preceding
and its subsequent further
profiling in
as 'planning activity of the
mind', frequently yielding 'pleasure, something to be thanked for', associates the
protective side of God's
The first constituent of the compound
, modifies -
ambivalently. Its semantic centre 'mind, will-power' via 'soul, spirit' extends as far
as into the domain of feelings: both good ones (e.g. 'courage') and bad ones (e.g.
'pride', 'violence') – just as the related adjective
. Thus
signifies a possibility. Here the semiotic correlate of this ambivalence of
the re-emphasis of God's Absolute Firstness: being
m o d
, His
cannot be subjected to any kind of influence. It rather is plain existence and as
such beyond any human evaluation as to being good or bad. Therefore, the
, once again, semiotically links up with
, here, however
with the
's second property, His 'Lordship'.
Textually a mediating position can be attributed to
, denoting a
product, is directly conjoined with the process
as its corporeal
manifestation, and indexically it points back to the
in His capacity as its
'creator/protector', too. At the same time, however, it points forward towards all
the praiseworthy 'wonders' (
), which have come into being. Thus,
on the one hand,
rounds up the first part of the
, i.e. the part that
invokes God in His Absolute Firstness; the opener is not followed by any verb,
i.e. up to here there is no 'action' on God's part; semiotically, this is not at all
surprising; Firstness is
per se
not a matter of cause and effect; Firstness is by its
very nature static. On the other hand,
provides the basis for ll. 3b – 9: the
Hymn can now proceed to the dynamic point-of-view in naming the processes of
is complementary to
. Whereas the latter invoked the
's essence as 'Lord',
denotes the 'Protector'.
As to
, it comes close to the status of a near-synonym of 'God',
implying the association of 'heaven'. Thus it refers back to to the whole
introductory phrase
hefaenricaes uard
. On the other hand, via its use for
'splendour on earth', 'splendour of the universe', 'glorious things to be enjoyed by
man', it points forward towards what is going to be praised in the remaining lines
of the Hymn. The same textual-semiotic status of both pointing backwards and
forwards was diagnosed for the sign
, too. This indexalic equivalence makes
a perfect match for
, with which, in addition, it makes up one
syntactic collocation in the Hymn.
It is beyond the scope of this paper, to set the preceding semiotic findings
into relief with prevailing theological interpretations of
. Therefore, one
reference will have to suffice here: Huppé (1959: 111) argues in favour of a
patristic interpretation of the triad
, and
In the representation of the Trinity through the creation, God the
Father is the Power or Might, the Son is the Shaping Wisdom, the
Holy Ghost, the Perfection of the Work…The three phrases [i.e. triad
] reflect the traditional division of the Persons
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
of the Trinity as they are revealed in Genesis: the Might of God, the
Creator, would represent the Father; the Thought of the Father, His
plan and disposition of creation, the Son; the Work, the Holy Ghost.
In many respects our semiotic analysis above ties in with that theological pattern
of the Fathers' reasoning (with which Bede had, definitely, been familiar).
Conclusion: if this is what had been on the back of the composer of the Hymn's
mind, both his lexical choices and their textstructural/textsemiotic use in the
native Anglo-Saxon tongue deserve unlimited appreciation and cannot but have
been deliberate; if, however, the alleged cowherd–composer had been unaware of
these –in his days- patristic "commonplaces",
Bede's admiration for him is
justified the more.
Ll. 3b – 9: uundor – eci – dryctin – scop – aelda barn – eben – hale

scepen - middun
eard - moncynnæs uard – folde – frea allmecti
In Peircean terms, Creation is a sign of Secondness
par excellence
: it is
constituted by a relation between cause and effect. Here God is cause, is 'agens',
Creation is effect, is 'patiens'.
Ll. 3b – 4 fulfil what has been indicated by l. 3a: they present God as
having 'set the beginning' (
or astelidæ
) of
. Like
that turned
out to be the semiotic hinge for ll. 1 – 3a,
, initiating the second part,
serves as the hinge for all the processes and products of Creation in the remainder
of the Hymn. In its wider sense
can designate both very positive matters
and events and absolutely negative ones, too. In this respect it links up with
i.e. with the first part of the Hymn, where God's essence as being beyond human
categorizing and evaluating is presented. Textsemiotically
is a
continuation of
, pointing both backwards and forward in the text: backwards
via that relation to
, forward by means of its introducing God as creator, i.e.
God in his Secondness. Thus, like
in the same line,
is semiotically
middling between both parts of the Hymn, ll. 1 – 3a on the one hand, and ll. 3b –
9 on the other hand. It follows: God in His Secondness, i.e. as the creator, is
presented under the auspices of God in his Firstness.
It is only as late as in the second half-line of l. 4, that God's Secondness
starts being revealed. This is done under the auspices of temporality:
or astelidæ
Beginning associates ending, too: Creation is transitory by nature.
In l. 5a,
follows suit, and right in l. 5b that notion of temporality and
transitoriness is intensified by the choice of
aelda barnum
to denote Mankind
here; via the word-family of
'period of time, duration of
life, age',
yldran, ældrean
'parents') and via the relation of
bear', the 'born' one, the 'child', too, the cycle of coming into being and passing
away from being, the transitory nature of human being, is associated. It is within
the very same phrase that God is set apart from any transitoriness: the
, the
'Lord of the followers' is
'eternal'.- The re-occurrence of
in l. 8 is a
reminder. In l. 4a
causes the first phase of Creation, in l. 8a
eci dryctin
is rounding up its last phase:
æfter tiadæ
'he finally he adorned it'. The semiotic
impact of this resumption is to reemphasize God's eternality – as opposed to the
old pagan Germanic deities, who had been mortal.
It is only after the general presentation of God as the originator of
everything, as well as after the first appearance of Mankind in the Hymn (l. 5b),
that He can be referred to as
('holy creator'): the quality of
, as in
eci dryctin
, is independent of man, even contrasts with the nature of man. The
attribution of a quality like
, however, admits of human judgment.
And it is human existence, too, that might well account for the seemingly
tautological construction '
+ scepen' in ll. 5f. Textstructually one should mind
the iconicity: '
' is embraced by the process of creating ('
') and
its originator ('
'); once again: the
in His essence of creator as the
protector of Mankind. Furthermore, a corpus-analysis of
many instances where their general meaning 'to create, to produce; Creator,
producer' is specifically qualified as to the notion of 'forming, gaining shape;
shaper'. In early Anglo-Saxon poetic diction
even is occasionally profiled as
far as to, 'to assign', 'to determine', and 'to destine'. If we concede this shade of
meaning here, it greatly impacts on the semioticity of
: with regard to the
close syntactic contiguity of
to Mankind in ll. 5f. and with a view to
the semioticity of 'giving' as the overall pattern of the Hymn, ll. 5f. are the first
explicit indication of what is going to be made clearer and clearer up to the
ending of the Hymn: what the
had created was not mere Creation as such
(='Secondness'), but 'functional Creation' with a view to the weal of Mankind
(and thus for the first time in the Hymn pointing out to 'Thirdness').
However, there are two caveats: the giving
. Wietelmann
(1952: 6-17) convincingly argues that in Cædmon's time its sense oscillated
mysterium fascinosum
mysterium tremendum
– a finding that
semiotically links up with our analysis of
above, i.e. with Cædmon's
basic admonition to praise God in his Firstness. Secondly, in the 7
might still have conveyed the old connotation of 'perfection', here as setting God
apart from the
aelda barnum
, who must needs be the contrary.
The designation of God as
moncynnæs uard
in l. 7b could well be
interpreted as a third reminder (caveat) to be aware of God in His Firstness when
praising Him in His Secondness, i.e. in His capacity as the creator. Evidently, this
phrase refers back to His designation at the very outset of the Hymn as
hefaenricaes uard
. Textsemiotically, however, the two phrases are not at all
simply variations of each other (as purely philological assessments of
ever recognized). L. 1b is a sign of God's Firstness in His incogitable Heavenly
Kingdom. As such it cannot but show up in no interdependence with Mankind
whatsoever. By having come to l. 7b, however, God has by now been revealed in
His Secondness as the creator, and even the perspective to His Thirdness as the
giver, has meanwhile been indicated by
aelda barnum
. And in l. 7b it is this very
specification of His
ship as to that part of Creation that is Mankind, that
makes the syntagma
m o n c y n n e s uard
the semantically and semiotically
appropriate expression at that very place within the entirety of the Hymn. Its
reference to Mankind there (once again) is an index that points both backwards
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
and forwards. Backwards it resumes the recipients of God's gift,
aelda barnum
serving as an intensifier. Forwards it paves the passage towards the immediately
following ll. 8-9. And it is these last lines that finally consummate the passage
from the Creator-God in His Secondness as presented in the middle, semiotically
middling part of the Hymn towards His Thirdness as the Giver-God
employs three lexemes to present those biblically attested products of
God in his pure Secondness:
eben, middun
eard, folde.
What all three of them
stand for is actual existence, is the facticity as attributed to Gen I.1:
In principio
creavit Deus
cælum et terram
. It is verifiable: the Hymn here ascends from being
rhematic to being dicentic.
Due to
the metrical structure of l. 6a
eben til hrofe
strikes the listener's
ear as being a specific one within the Hymn. This insertion of
was required
because of
('roof'). 'Roof' is not part of Gen I.1. Why then did it get into
? Patristic literature - on which Bede could draw- distinguished between
eternal Heaven and Temporal Heaven, the former one being created first and
unseen by man, the latter one being created afterwards for Mankind. It may be the
case, as argued by Huppé (1959:113ff.), that with
the Hymn evokes the association of that primeval eternal Heaven and with
refers to the temporal one. There is semiotic evidence for that view:
conjoined with the first appearance of a sign that denotes Mankind,
, and
is semanto-syntactically squeezed in between
both of which do not merely stand for 'coming into being of something new' but
for functional Creation, i.e. Creation with a view to Mankind. Bede will have
been familiar with metaphors like 'House' or 'Room', applied by the Early Fathers
when expounding 'Heaven'. Thus the Hymn might well have drawn upon that
metaphor of the
: it is an icon that represents Heaven's shielding, protective
quality for Mankind. What God created on the first day of Creation came into
being with the very prospect of events on the 6
day, the Creation of Mankind
The semantic feature of temporality of the non-eternal heaven ties in with
semantic features that qualify the lexeme chosen to designate 'earth' at this very
place in the Hymn:
It implies the feature 'inhabited by Mankind',
which it had denoted in Germanic pagan mythology, and which it may have
connoted still in the 7
century. On that ground
is the perfect
complement for
in its function as
The third reference to elements of Creation that can substantially be
experienced is
, the accusative form of
. It primarily denotes earth in its
material substance, but does not have 'inhabited' as one of its necessary semantic
features (
). But a corpus analysis of Old English poetry indicates
that it shares with
rather more positive than negative connotations.
It lends itself for referring to a dwelling-place, often occurs with a vegetative
component (thus associating Creation), and even lends itself to designate
'paradise' and it is in that final line of the Hymn that Cædmon relates Earth to
Mankind even
expressis verbis
: via the sign
. This latter sign designates
Mankind in a wide sense, and frequently implies the specific relationship between
Mankind and Creation, which grants to man the prime role among all living
beings. In that extolled position as the concluding line of the Hymn undoubtedly
is, Cædmon literally zooms in on the benefactory function of Creation for
Mankind: both lexemes are presented in direct contiguity (
firum foldu
). Each one
of these two signs
per se
is a sign of Secondness, each one refers to objects that
are effects of a cause. Their immediate contiguity, however, makes these two
signs one unique sign of a higher order:
as the gift to
and as pointed
out above, semiotically 'giving' is a genuine sign of Thirdness- and as such the
relationship between the giver, the recipients and the gift itself- is not a mere
existential fact but has the status of a generally holding law.
L. 9, being the architectural keystone of the Hymn, extends this contiguity
by naming the giver without any linguistic filler between him and the preceding
signs for the receivers and for the gift itself
: firum foldu,
frea allmecti
, used here for God, and as being modified in particular by
, evokes the same fundamental values as
, such as leadership and
of loyalty, allegiance, and togetherness. But it reaches beyond them in two
respects. Corpus analysis shows that it puts a stronger emphasis on the supremacy
of God's position – in this line, in relation to both Mankind and Creation. In
addition it may well have connoted semantic features that had still been
denotative ones with contemporaneous cognate lexemes of its word-family, such
, in which two features amalgamated: 'ruler, leader, king' and 'beginning,
origin, cause, creation, creator, founder'.
is a sign, which essentially features supremacy and creativeness, its
is its perfect semiotic match. It is obvious that it links up with
in l. 2a. From what had been observed there about this word family,
highlights the supremacy of
qua giving emphasis to his
and furthermore, it indicates that the
exerts his supreme power to the benefit
and protection of Mankind. In addition, it lends itself for an emphatic conclusion
of the Hymn: in Anglo-Saxon poetry
is reserved to God exclusively!
This final burst of praise in l. 9 thus links up with the exuberant
exhortation in the introductory line 1a. Looking for a metaphor to cover the
semiotic structure of the Hymn in its entirety, once again, a 'ring' would lend
itself as an icon to depict its rhetoric.
is not a 'Song of Creation', not a
'Schöpfungshymnus', but a 'Song of the Creator', a 'Hymnus auf den Schöpfer',
who, like the ring, has no beginning and no ending.
Results: The Hymn as a Dicentic Symbol
It was the aim of the preceding analyses to come to an assessment of
as a
whole when reviewed from a semiotic stance. For that purpose the semioticity of
the interconceptual relationships, that hold between God, Creation, and Mankind
in the Hymn was studied. This was done from a textstructural point-of-view, that
was complemented by a likewise semiotically oriented lexicological analysis.
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
It turned out that the results of the latter investigations supported the ones
gained from the former ones to an unexpected degree. In particular as to the inner
cohesion of the Hymn in its entirety the lexis chosen by the composer reinforced
the textual structure.
Semiotically the structure of the Hymn in its entirety presents God in each
one of the three phenomenologically possible modes:
(1) His simple (though impenetrable) essence in itself, i.e. semiotically God in
His 'Firstness'
(2) His being creator of the actually existing and verifiable objects of creation,
i.e. semiotically His 'Secondness'
(3) His being represented in Creation in such a way as to cause the
interpretation, according to which it lawfully holds that it was God's will
to make Creation serve Mankind, that He intended Creation to be His gift
to Mankind., i.e. God in His 'Thirdness'.
These three modes are hierarchically structured: (1) being the simplest one, (3)
being the most complex one, (2) middling between both. And the textual structure
mirrors this very sequence to a nicety. It ascends from (1) via (2) to (3).
The three modes of God's revelation in the Hymn are not, however, piled
upon each other as isolated bricks. Instead, the transitions from (1) to (2) and
from (2) to (3) are of a processual kind with semiotic markers of overlapping to
mark the gradation.
And this is where the lexical elements as chosen by the composer reveal a
textlinguistic potential that was unexpected by us. All interpretations and
translations of
which we have come across, use to concentrate on its key
lexemes on word-semantic, on literary, on cultural-historical, etc. grounds
exclusively. Hardly any attention has been paid to potential textstructural
significance, let alone from a semiotic point of view. Our projection of the
lexicological analysis onto the preceding textstructural one, however, reveals
right through the entire text that lexemes are chosen in such a way as to ease the
passages from the presentation of God in His Firstness toward God in His
Secondness and from there toward His Thirdness. They are the joints, whose
cohesive power achieves coherence across (1) to (3). This is not always achieved
by means of their denotative meaning, but rather by connotative-associative
components of meaning that might well have still been alive in the 7
In its entirety CH can be seen as a ring. Its conclusion in ll. 8 - 9 shows
God as a 'giver' in His Thirdness, but simultaneously it resumes God in His
Firstness, which is the mode in which ll. 1 –3a introduced Him in the beginning.
Like the ring, God's eternal essence sharply contrasts with the
transitoriness of human being. This contrast is mirrored by the opposition
eternality vs. temporality, that pervades the entire Hymn.
The metaphor of the ring indicates, too, that the overall patterning of the
Hymn is highly iconic. The textual as well as the lexicological-semantic
distribution bear witness of this iconicity.
Semiotically viewing the Hymn as an icon, however, can imply that it is
an instantiation of an index, a
. And textual evidence shows, that, actually,
an appeal, i.e. a sign that directs the listeners' minds (index!) to praise God
section of the Hymn) on the basis of their experience of Creation (2
of the Hymn), which God from the very outset of His act of creating designated to
be His gift to Mankind (3
section of the Hymn). In calling up in the listeners'
minds he image of the Creator which it suggests to the minds, the Hymn as an
entirety acts upon a
already stored in the minds. This very stored symbol
s Creation for Mankind
. Thus the Hymn as an index is a replica of the
Hymn as a symbol, just like the Hymn as an icon is a replica of the Hymn as an
index. In its relation to the interpretant this symbol can be verified on the basis
of the experience of Creation, i.e. it is dicentic.
Summing up, the semiotic analyses above support the semiotic hypothesis
according to which
is a Dicentic Symbol.
With regard to the overall textstructural semiotic patterning, held together
by its lexis in a supporting function, furthermore with regard to the encompassing
iconicity - as a
of its indexality-, in turn being a
of its symbolicity,
with regard to the result of the preceding analysis, that revealed that all of these
properties cause all individual signs of the Hymn to relate to each other is such a
way as to make it one unitary sign –we, therefore, name it a 'Hypersign'- we leave
it to the reader to side with whatever assessment she/he considers as being
appropriate among the assessments quoted to in part 2 above.
Actually, a misnomer as we shall see below.
As a detailed discussion concerning (a) place of origin of
, (b) its exact
date of origin, (c) the relationship among the extant 17 manuscripts, in
has come down to us, and (d) the relationship between the
Anglo-Saxon and the Latin versions is out of place here, we refer the
reader to the state of the art of research as summarized in Kühlwein
(2006a: 101f.): specifically there in detail as to (a): p. 101 FN 1; as to (b)
p. 101 FN 2; as to (c) p. 101 FN 3; as to (d) p.102 FN 4. - As to (d) we
strongly side with the sequence 'Anglo-Saxon version first, translation into
Latin next' (
. vice versa!) for manuscript-based reasons given there.
Due to the limits imposed by translating a text, in particular a poetic one
that is nearly one and a half millennia remote, this translation into Modern
English prose merely serves as a preliminary working- translation; it
should be one supererogatory consequence of the following analysis to lay
bare its shortcomings.
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
The problem with this evaluation and some of the following ones is, that
these parallels are rightly observed; however, as no pre-Cædmonian Old
English Christian literature has come down to us, the question whether
Cædmon' use of poetic language
drew on predecessors or
extant authors of later date drew on him, remains unresolved. Furthermore,
as to the explanatory strength of comparative frequency counts of
Anglosaxon oral-formulaic poetry, the constraints imposed by metre and
alliteration will have exerted much less influence upon thematic and
lexical choice than had previously been asserted (for evidence c.f. Creed
1959, Greenfield 1972, Fry 1974 and 1979, Miletich 1983, O'Keeffe 1987,
Kühlwein 2006b).
Strangely enough, hardly ever quoted in subsequent linguistic and/or
literary critics.
For some less concise outlines cf. Hervey 1982: 17-37, Nöth 2000: 33-46.
As a result, this conception would have to cause editors' punctuation to
have a comma or at best a colon following
, because all that
follows would cognitively merely serve the purpose of a continuation,
which specifies the preceding 'thus' or 'how'.
"I think we must regard Creative Activity as an inseparable attribute of
God" (Peirce 6.506).
Punctuation in this case would better have a semicolon or even a full stop
and thus give special causal emphasis to the acts of
His creative doing as following in ll. 5 – 9.
The adequate punctuation indicating that division is either a semicolon or
even a full stop after
in l. 7.
For the semioticity of
as not simply to be interpreted as a pure
synonym of
'created' but beyond that as 'creatively adorned', as had
been proposed by J. B. Bessinger Jr as early as in 1974 cf. Kühlwein
(2006 b: 76).
This very consistency of that pattern may well be taken as an argument in
favour of the punctuation followed by us. It is on these grounds that we
reject a punctuation that closes l. 6 with a Semicolon, and separates half-
line 7b from 8a by a mere comma, as some editors do, e.g. the Aelfredian
version as taken from the Tanner text (MS Bodleian Tanner 10) as edited
by Krapp in ASPR VI, 105f. or the punctuation offered e.g. by, by
labyrinth/library/oe/texts/a32.1.html (for the Northumbrian version) and (for the West
Saxon version likewise). That latter punctuation would move
to a front position with relation to the two following
references to its Creator (
moncynnes uard
eci dryctin
), i.e. Creation
– and not the Creator - would syntactically become the topic, the two
following references to the Creator (
moncynnæs uard
eci dryctin
would become comment. Such a view would run counter to (1) the overall
semioticity of the Hymn as well as (2) to the semiotic gist of Bede's
narration that encompasses it in the way as semiotically shown in
Kühlwein (2006a: 112ff.), furthermore (3) it would disregard the fact that
in all other cases at least one reference to the Creator, actually, precedes
the respective reference to Creation (or some part of Creation), and (4)
finally, the immediacy of the ensuing conjoining of the two epithets for
God (
moncynnæs uard
, directly followed by
eci dryctin
) within what then
would have to be seen as the same syntactic sequence, would not at all be
in line with the poetic way in which the balance between topic and
comment is achieved in all other parts of the Hymn. [As a result the
syntactic sequence would have been fairly odd: *"… holy creator
inhabited earth,
the protector of Mankind, the eternal Lord
adorned, for Mankind the ground, the almighty Leader"].
By strength of this kind of relationship God and Creation, actually,
determine each other: God enforces the existence of Creation, which in
turn, makes Him the Creator.
Peirce himself provides two examples from thematically related fields for
what he calls a
. One example is
I. 3 "
Dixitque Deus: Fiat
lux. Et facta est l
ux" ('God said, Let there be light, and there was light.'):
"We must simply think of god creating light by fiat. Not that the fiat and
the coming into being of the light were two facts; but that it is in one
indivisible fact. God and light are the subjects. The act of Creation is to be
regarded merely as the suchness of connection of God and light. �The
dyad is the fact…..pure dyadism is an act of arbitrary will" (Peirce:
1.327f.).- For a second example Peirce (2.316ff.) draws on the proposition
'Cain kills Abel', which likewise has two subjects, "Cain" and "Abel";
though it relates to the real existence of either one of them, it nevertheless
"may be regarded as primarily relating to the Dyad composed of Cain, as
first, and of Abel, as second member. This Pair is a single individual
object having this relation to Cain and to Abel, that its existence
in the existence of Cain and in the existence of Abel and in nothing more.
The Pair….. [is] just as truly existent as they severally are."
Even if one includes the two references to 'Mankind' within the ones for
'Creation' the ones to 'God' still outweigh that sum.
For a detailed analysis s.b. 5.1.3.
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
Gen II. 7: "Formavit igitur Dominus Deus hominem de limo terrae…"
(Biblia Sacra Vulgata). [ And the LORD God formed man of the dust of
the ground…] (King James Bible).
Cf., however, the next semiotic level as approached in 5.1.3, where
Creation will be looked at under the perspective of God's intention,
revealing God in His Thirdness.
Space forbids to enlarge upon corpus-based evidence outside
in detail.
For the base of statements concerning the semanticity (including possible
etymological and connotative rings) of the key-lexemes, whose semioticity
will be discussed below in 5.2, we refer to Kühlwein (2006b: 70-85), that
is based on the entire corpus of Old English poetry.
"Because compulsion is essentially
hic et nunc
, the occasion of the
compulsion can only be represented to the listener by compelling him to
have experience of that same occasion. Hence it is requisite that there
should be a kind of sign which shall act dynamically upon the hearer's
attention and direct it to a special object or occasion. Such a sign I call an
Index. It is true that there may, instead of a simple sign of this kind, be a
precept describing how the listener is to act in order to gain the occasion
of experience to which the assertion relates." (2. 334f.)
Cf. e.g. Beowulf, Nibelungenlied, Scaldic Northern sagas.- On the other
hand, an opener in the 1
pers. sing., as met in Old English elegies would
have been inappropriate: Cædmon's personal feeling is the ground, but not
the theme; he merely is the angel's transmitting instrument.
Huppé 1959:111.
This relationship is syntactically mirrored: whereas in ll. 1 – 3a the
references to God are in the genitive or accusative, in ll. 3b – 9 all of them
are in the nominative.
Cf. Wietelmann (1952: 23): "Das Verhältnis zwischen der Ewigkeit und
der Zeit gleicht dem Abstande zwischen dem Schöpfer und seinen
Geschöpfen, d.h. zwischen Gott und den Menschen."
As to intensification (vs. logical precision) as a major characteristic of
Anglo-Saxon poetic diction cf. generally Kühlwein (1967: 42ff.).
Cf. 5.1.2 above.
A certain amount of semiotic significance has to be attributed, too, to the
fact, that amongst all signs which the Hymn uses to designate God, this
final occurrence of such a sign is the only one, where it has a postmodifier
). A metrical analysis that employs the subtlety of John C. Pope's
studies of Anglo-Saxon poetic rhythm reveals, that the collocation in its
form '
' is enjoying a degree of metrical emphasis
unparalleled in the entire hymn.
For evidence cf. the Figures 1 – 3 above.
"…the most perfect of signs are those in which the iconic, indicative, and
symbolic characters are blended as equally as possible." (Peirce 4.448).
Our own vote would be in favour of (i), (f), (g) in that sequence.
Texts and Works of Reference
[Bede]. Miller, T. (ed.) 1891, repr. 1959.
The Old English Version of Bede's
Ecclesiastical History of the English People.
EETS, Original Series No.96.
London, New York, Toronto: Oxford University Press. 2 vols.
[Bede]. Colgrave, B. and R.A.B. Mynors (eds.) 1969.
Bede's Ecclesiastical
History of the English People.
Oxford: Clarendon Press.
Krapp, G. Ph and K. van Dobbie (eds.) 1931 – 1953.
The Anglosaxon Poetic
. 6 vols. New York: Columbia University Press.
Biblia Sacra Vulgata
The Bible, King James Version. Old and New Testaments, with the Apocrypha
The Electronic Text Center, University of Virginia Library
Toller, T. N. (ed.) 1898, repr. 1964.
Bosworth, Joseph. An Anglo-Saxon
Oxford: Oxford University Press/London: Humphrey Milford.
Toller, T. N. (ed.) 1921.
An Anglo-Saxon Dictionary Supplement
Oxford: Oxford
University Press.
Grein, C.W.M. et al. 1912; 2nd ed. repr. 1974.
Sprachschatz der
angelsächsischen Dichter.
Heidelberg: Universitätsbuchhandlung Carl
Further Works of Reference
Bessinger, J. B., Jr. 1974. 'Hommage to Caedmon and Others: A Beowulfian
Praise Song', in Burlin, Robert B. and Edward B. Irving, Jr. (eds.),
English Studies in Honour of John C. Pope,
Toronto and Buffalo:
University of Toronto Press. 91 – 106.
Creed, R. P. 1959. 'The Making of an Anglo-Saxon Poet',
26: 445-454.
Fry, D. K. 1974. 'Cædmon as a Formulaic Poet',
Forum for Modern Language
X : 227-247.
Fry, D. K. 1979. 'Old English Formulaic Statistics',
In Geardagum
, III: 1-6.
The Semiotic Patterning of Cædmon's Hymn as a 'Hypersign'
Gardner, J. 1975.
The Construction of Christian Poetry in Old English
Carbondale and Edwardville: Southern Illinois University Press.
Greenfield, S.B. 1972.
The Interpretation of Old English Poems.
London and
Boston: Routledge and Kegan Paul.
Hervey, S. 1982.
Semiotic Perspectives
. London: George Allen & Unwin.
Huppé, B. F.
Doctrine and Poetry
: Augustine's Influence on Old English
Poetry. New York: State University of New York.
Kane, G. 1948. 'Review of C.L. Wrenn,
The Poetry of
. London.1947',
XLIII : 250-252.
Kühlwein, W. 1967.
Die Verwendung der Feindseligkeitsbezeichnungen in der
altenglischen Dichtersprache
. (Kieler Beiträge zur Anglistik und
Amerikanistik vol. 5). Neumünster: Karl Wachholtz.
Kühlwein, W. 2006a.., 'Bede's Narrative on Cædmon: A Semiotic Analysis', in:
Cho, See-Young – Erich Steiner (eds.)
Information Distribution in English
Grammar and Discourse and Other Topics in Linguistics.
Festschrift for
Peter Erdmann on the Occasion of his 65
Birthday. Frankfurt / M.: Peter
Lang, 99-124.
Kühlwein, W. 2006b. 'The Semioticity of God, Creation, and Mankind in
Cædmon's Hymn', in: Rösel, Peter (ed.).
English in Space and Time.
Englisch in Raum und Zeit.
Forschungsbericht zu Ehren von Klaus Faiß.
Trier: WVT Wissenschaftlicher Verlag Trier, 60 – 89.
Magoun, F. P. Jr. 1955. 'Bede's Story of Cædmon: The Case History of an Anglo-
Saxon Oral Singer',
, XXX: 49-63.
Miletich, J. S. 1983. 'Old English "Formulaic" Studies and Caedmon's Hymn in a
Comparative Context', in: in Mateši
Josip and Erwin Wedel (eds.)
Festschrift für Nikola R. Pribi
. Neuried: Hieronymus: 183-194.
Mitchell, B. 1967. '”Swa” in Cædmon's "Hymn" line 3',
Notes and
Queries n.s.,
XIV: 203-204.
Nöth, W. 1985, 2nd ed. 2000.
Handbuch der Semiotik.
Stuttgart: Metzlersche
O'Keeffe, K. O'Brien 1987. 'Orality and the Developing Text of Cædmon's
, LXII : 1-20.
[Peirce, Charles Sanders] 1931- 1958, 3
printing 1974. Hartshorne, C. and P.
Weiss (eds.)
Collected Papers of Charles Sanders Peirce.
Vols. 1-6;
Hartshorne, Charles – Paul Weiss (eds.) ; A. W. Burks (ed.), vols 7-8.
Cambridge, MA: Belknap Press of Harvard University Press.
Pope, J. C
The Rhythm of Beowulf
. An Interpretation of the Normal and
Hypermetric Verse-forms in Old English Poetry. New Haven: Yale
University Press.
Smith, A.H. (ed.), 1933, 2nd ed. 1968.
Three Northumbrian Poems: Caedmon's
Hymn, Bede's Death Song, and The Leiden Riddle.
London: Methuen.
Wietelmann, I. 1952.
Die Epitheta in den "Caedmonischen" Dichtungen
. PhD
Diss. Göttingen [typescript].
Wrenn, C. L. 1947.
The Poetry of Cædmon
. London.
Wrenn, C. L. 1968. 'The Poetry of Cædmon' in: Bessinger, J. and S. Kahrl
Essential Articles for the Study of Old English Poetry.
Connecticut, 407-427.
Traditional grammar and corpus linguistics
with critical notes
David A. Reibel
Tübingen and York
After Robert Lowth published his
A Short Introduction to English Grammar: With Critical
in 1762, no one who took a serious interest in the subject could not have seen that
he had changed the definition and practice of this subject forever. The purpose of this
study is to show how he did it.
I defend Lowth against the oft-levelled charges of lack of grammatical competence
and acumen, arbitrariness, and disregard for usage; above all, for his desire to 'regulate'
the English language by prescribing arbitrary rules, which would at the same time
proscribe errors.
He is shown as highly competent in the field of grammar and literary criticism,
and displays considerable originality, ingenuity and skill in the fashioning and application
of his rules, based on the meta-principle of Strict Construction. Far from imposing a
Latinate grammar on English, he sought to eliminate, among other constructions, the non-
native Latinisms, imported into English during the English Renaissance (1550-1660)
through the medium of the Periodic Sentence. He also judged improper those native
English syntactic forms which also violated the principle of Strict Construction. In this
regard he represented the 18thC purist view of English that replaced the looser
construction of earlier generations with a more refined, more construable prose,
epitomized by Samuel Johnson.
Lowth is far from perfect, and neither is his English Grammar, but most present-
day critics write about myths and inventions of their own, instead of studying Lowth's life
and works for what they represented to the scholars and educated classes of his day, who
regarded him highly as a respected officer of the Church and a distinguished man of
Ask, and it shall be given you;
seek, and ye shall find;
knock, and it shall be opened unto you.
For everyone that asketh receiveth;
and he that seeketh findeth;
and to him that knocketh it shall be opened.
Sermon on the Mount
Matthew 7:7-8; Luke 11:9-10
I shall only remark here, how easily men take upon trust, how willingly they are
satisfied with, and how confidently they repeat after others, false explanations of
what they do not understand.
Dans les champs de l'observation,
le hasard ne favorise que les esprits préparés.
'In the field of observation, chance favours only the prepared minds. '
"In the empirical sciences, only prepared minds are favoured
by chance discoveries."
— Louis Pasteur (1822-1895), French chemist and biologist. Address
given on the inauguration of the Faculty of Science, University of
Lille, 7 December 1854.
. 16 [June 1770]— … In the afternoon I looked over Dr. Priestley's English
Grammar. I wonder he would publish it after Bishop Lowth's.
In this little
jeu d'esprit
, I defend Robert Lowth against the oft-levelled charges of
lack of grammatical competence and acumen, arbitrariness, and disregard for
usage; above all, for his desire to 'regulate' the language, i.e., set up rules for it
(cf. Latin
'rule'), to prescribe English usage by arbitrary rules, which
would at the same time proscribe errors.
He is shown as highly competent in the field of grammar and literature,
and displays considerable originality, ingenuity and skill in the fashioning and
application of his grammatical rules. Far from imposing a Latinate grammar on
English, he sought to eliminate, among other constructions, the non-native
Latinisms, imported into English during the English Renaissance (1550-1660),
that, as he thought, rightly or wrongly, disfigured the language, especially of the
earlier generation of post-Restoration writers, even the most eminent. He also
judged improper those native English syntactic forms that violated the principles
of Strict Construction. In this regard he represented the 18thC purist view of
English that replaced the looser construction of this and earlier generations with a
more refined, more construable prose. Samuel Johnson epitomizes this carefully
crafted new prose style, based on the periodic sentence.
Lowth is far from perfect, and neither is his
A Short Introduction to
English Grammar: With Critical Notes
(1762), but most present-day critics, from
the depths of their abysmal ignorance of what Lowth actually says and does, and
a priori
prejudices and lack of analytical understanding, write about myths
and inventions of their own,
instead of studying Lowths life and works for what
they represented to the scholars and educated classes of his day, who regarded
him highly as a respected officer of the Church and a distinguished man of letters.
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
Among the many practitioners of Corpus Linguistics, the name of Robert
Lowth (1710-1787) is not likely to be mentioned. But he followed the old-
fashioned time-honoured method of collecting examples from a body of
literature, probably on file slips made from marginal pencil-markings on the
pages of his daily reading-matter, as did Samuel Johnson for his
(1754). 'His temper was generally cheerful, though sometimes irritated by the
vexations of office, and the disappointments and provocations of a life of literary
popularity. It is said that, like George Stevens and Professor Porson, he never
read a book, without a pen or pencil in his hand.' (Hall 1834: 40-41)
Thus the compilation of 'improprieties' or 'inaccuracies' (Preface,
1762:viii) in his
English Grammar
was based just as surely on an open-ended
random corpus of texts as any similar present-day compilation,
with this
important distinctive difference: Lowth had already formulated the general
conclusions to be drawn from the examples in his corpus before he ever started on
this enterprise. As he says in the 'Preface' to the
Short Introduction
The principle design of a Grammar of any Language is to teach us to
express ourselves with propriety ['1. Accuracy; justness.' (Johnson)] in
that Language, and to enable us to judge of every phrase and form of
construction, whether it be right or not. The plain way of doing this, is
to lay down rules, and to illustrate them by examples. But besides
shewing what is right, the matter may be further explained by pointing
out what is wrong.
(1762: x)
In his typical way, when he was commissioned to write this grammar,
he saw at
once that there was a gap in the coverage of all previous works, and came up with
a plan to base the new section on syntax, which he entitles 'Sentences', on a
treatment of the faults of English along with the facts.
Thus the corpus consists virtually exclusively of 'improprieties'. How are
'improprieties' to be identified? They cannot come from the lower orders, who do
not speak or write standard English, nor from writers known to write in an out-
dated style, full of archaisms and similar constructions that have been superseded,
nor from those whose writing is said to contain 'inaccuracies', i.e., grammatical
solecisms. So the corpus is composed of instances from reputable writers which
nevertheless, he says, quoting Swift, ' "offended against every part of Grammar." '
(1762: ii).
The second part of the procedure was to find a grammatical meta-rule
according to which the appropriate 'Rules' might be 'laid down'. For this we need
go no further than Lowth's definition of 'Sentence', at the beginning of the section
on 'Sentences', or syntax:
A SENTENCE is an assemblage of words, expressed in proper form,
and ranged in proper order, and concurring to make a complete sense.
(1762: 94)
To understand all this, we need to have recourse to Johnson's
, which
gives the 18thC senses of the key words.
It should be remembered that the
largest grammatical unit recognized from antiquity down to Lowth's day was the
, or 'periodic sentence', the universally practised classical sentence-form,
from Greek
, 'meandering road' — not a bad description of the feeling
one has when making one's way through one of the longer instances. Here are
other senses of 'Sentence' from Johnson:
Determination or decision, as of a judge, civil or criminal.
It is usually spoken of condemnation pronounced by the judge; doom.
A maxim; an axiom, generally moral.
A short paragraph; a period in writing.
As in most dictionaries, looking up the meaning of the key terms in a definition
can only lead to circularity, as in this from Johnson's list of senses of 'Period':
7. A complete sentence, from one full stop to another.
This is true enough, as long as one knows where and how to place the 'full stops'.
Thus Lowth's definition of 'sentence', taken as a whole, must be considered
wholly new and original, and, as far as can be determined, not paralleled or
repeated by subsequent traditional definitions.
The next two members of the definition, 'expressed in proper form, and
ranged in proper order', probably come from Quintilian's
Institutio Oratoria
('Principles of Oratory'), Book VIII, Chapter ii, § 23, in his definition of
'perspicuity', 'clarity':
propria verba
rectus ordo
. It is clear from the
discussion that follows in the
that Quintilian is thinking of
as 'appropriate diction', and
rectus ordo
as 'straightforward arrangement'.
Lowth has split the sense of propria verba, first, into 'assemblage': '1. A
collection: a number of individuals brought together.' (Johnson); that is, not a
mere fortuitous, random selection or collection; and second, into 'expressed in
proper form'. 'Form' must mean '
form', and 'proper', '6. Exact;
accurate; just.' (Johnson). So the words must have the correct grammatical or
morphological form required by the construction.
Rectus ordo
now means 'ranged
(lined up) in grammatically correct order'. Cf: 'To Range. 2. To be placed in
order; to be ranked properly…' 'To Rank. 3. To arrange methodically.' (Johnson).
So Lowth has taken Quintilian's terms and given them new senses.
Finally, the words must 'concur to make a complete sense.'
This is
usually misunderstood both by later critics of traditional grammar as well as by
its practitioners as meaning that a 'sentence' is
assemblage of words that
makes (a) complete sense. Or else that in order to make complete sense it must be
a grammatically complete sentence. Or that a grammatically complete sentence
makes (a) complete sense. This would be Johnson's tenth and last sense of 'sense':
'Meaning; import'. But Lowth means
sense: cognate parts of cognate
constructions within a sentence must have constituent parts that concur. Forms of
words that fulfil identical functions within cognate constituents of sentences
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
cannot have their grammatical form determined locally, but must agree with each
other in their grammatical — morphological and syntactic — features across
unbounded dependencies. This leads to the principle which I have called 'Strict
Construction', which has very wide-spread applicability.
For example, suppose we have a general rule that if a pronoun is the
grammatical subject or part of the grammatical subject of a sentence, i.e., of the
verb, it must be in the nominative case. Expressions such as 'Us adults are going
to have a party' is ungrammatical because 'us', which is part of the subject of the
verb, is in the objective and not the subjective case of the first person plural
personal pronoun
in English. Selecting the form locally, say by some rule that
says that only when the pronoun is in absolute subject position directly before the
it be in the nominative and not the subjective case. Both versions of
English grammar agree that it must be: 'We are going to have a party'. No one
says 'us are'.
By the same disallowed rule, such expressions as 'Him and me / Me and
him went'; 'Me and my brother / My brother and me are twins' — found in all
forms of non-standard English, not treated by Lowth or other traditional
grammarians until later in the 19thC; cf. the later use of the term 'low expression'
— are by the meta-rule of Strict Construction disallowed in Standard English.
The rule of local determination says that neither
is in
position; the grammatical subject in direct construction with the verb is the
superordinate NP dominating the conjoined
, etc.
Having set up his criteria and found his texts,
Lowth now has to set
about writing his grammar. Of the many criticisms levelled at earlier traditional
grammarians, none is more critical or crucial than the assertion that they had no
qualifications for the job. But Lowth was a 'classic': a man learned in languages:
Latin, Greek, Hebrew. From 1741 to 1750 as Professor of Poetry at Oxford, a
post that was awarded on the candidate's 'Latinity' — being well-versed in the
Latin Language — as much as for any other form of learning. Lowth gave his
Lectures on the Sacred Poetry of the Hebrews
(Latin 1753; English 1787), with
the requisite Ciceronian style that has been independently judged by three
Latinists at the University of York to be very good and typical.
In Lecture XIX,
'The Prophetic Poetry is Sententious' ('Sententious. 2.
Comprising sentences.' Johnson), he finds the solution that had evaded all
previous attempts to find the structural basis of the Hebrew poetry of the Hebrew
Bible. First he asserts that the basic unit is a sentence, and that it is parallelism of
sentences and the (often contrasting) parallelism of their import that is the basic
Without so much as a warning, he now uses the technical term
'sentence' in its present-day sense.
The poetical conformation of the sentences, which has been so often
alluded to as characteristic of the Hebrew poetry,
consists chiefly in
a certain quality, resemblance, or parallelism between the members
[clauses] of each period [complete sentence]; so that in two lines (or
members of the same period) things for the most part shall answer to
things, and words to words, as if fitted to each other by a kind or rule
or measure. This parallelism has much variety and many gradations; it
is sometimes more accurate and manifest, sometimes more vague and
obscure: [Lowth-Gregory 1787.II: 34. Analysis of the three kinds of
parallelism omitted.]
In discussing the first species of the three forms of parallelism that he identifies,
the synonymous parallelism (Lowth 1753: 180; Lowth-Gregory 1787.II: 35) (the
other two are the antithetical parallelism, and the synthetic or constructive
parallelism), Lowth observes:
Saepe deest aliquid in posteriore membro, e priore repetendum ad
explendam sententiam, […] (Lowth 1753: 185)
'There is frequently something wanting in the latter [second] member
[clause], which must be supplied from the former to complete the
sentence [sense and/or clause]:'
"Kings shall see him and shall rise up:"
"Princes [GAP], and they shall worship him;"
[Isaiah XLIX.7] (Lowth-Gregory 1787.II: 41)
In other words, to complete the 'sentence' (Latin
) or 'sense' (NB
equivocation), the VP of the first line, 'shall see him and rise up', — just two
words in the Hebrew — must be interpolated into the second line after (or
perhaps before) the subject NP 'Princes', filling the 'gap'. It may fairly be said that
Lowth discovered gapping, a distinction in the various mechanisms for shortening
consecutive conjoined constituents by deleting repeated terms or constituents,
generally credited to Hudson 1976; see also van Oirsouw 1987. In fact, most
traditional grammars say something about this process, albeit usually in very
general terms.
Lowth is less interested in the grammatical generalization than he is in
accounting for the role that it plays in the structure of successive lines of Hebrew
Asymmetrical Conjunction
The best way to illustrate Lowth's method is to present one of his collections of
instances of an improper construction, and to set the reader the task of setting up a
rule of grammar which, on the face of it, seems an unexceptionably linguistic
commonplace, but which can at the same time be used to rule out the assembled
instances as violations of it, and, therefore, as 'improper', or ungrammatical.
The 'data' are an assemblage of Lowth's own compilations,
taken from
various editions of his
English Grammar
. His square brackets, or 'Crotchets', as
he calls them, enclose the elided word, which he has supplied. Biographical and
bibliographical information has been added in parentheses or square brackets by
DAR as well as occasional editorial clarification.
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
As you read through examples 1 to 11, try your hand at formulating the
rule that Lowth formulated and which excludes these expressions or constructions
from the canon of grammatical sentences or constructions of English. Formulate
also an alternative rule that allows them.
Forasmuch as it hath pleased Almighty God of his goodness to give you
safe deliverance, and [who] hath preserved you in the great danger of
Childbirth:—Liturgy. [
The Book of Common Prayer
(1662); revised
edition of the Prayer Book of Edward VI (1549; 1552), where this
originates. 'The Thanksgiving of women after Childbirth, commonly
called, the Churching of Woman.']
If the calm, in which he was born, and [which] lasted so long, had
continued. Henry Hyde, second earl of Clarendon (1638-1709), Life
(1668-1670; 1672 ff.; published 1759), p. 43.
The Remonstrance which he had lately received from the House of
Commons, and [which] was dispersed throughout the Kingdom.
Clarendon, Hist. (1702-1704) Vol. I. p. 366. 8
These we have extracted from an Historian of undoubted credit, a
reverend bishop, the learned Paulus Jovius; and [they] are the same that
were practised under the pontificate of Leo X. Pope (1688-1744),
Works, Vol. VI, p. 201.
A cloud gathering in the North; which we have helped to raise, and
[which] may quickly break in a storm upon our heads. Jonathan Swift
(1667-1745), Conduct of the Allies (1711).
A man, whose inclinations led him to be corrupt, and [who] had great
abilities to manage and multiply and defend his corruptions. [Swift,]
Gulliver (1726), Part I. Chapt. vi.
My Master likewise mentioned another quality, which his servants had
discovered in many Yahoos, and [which] to him was wholly
unaccountable. Gulliver, Part IV, Chap. vii.
This I filled with the feathers of birds I had taken with springes [snares]
made of horse hairs, and [which] were excellent food. Ibid. Chap x.
Osyrus, whom the Grecians call Dionysius, and [who] is the same with
Bacchus. Swift, Mechan, Oper. of the Spirit, Sect ii (1704).
Two further examples were added in some edition later than The Second Edition,
Corrected (1763):
Which Homer might without a blush rehearse,
And [which] leaves a doubtful palm to Virgil's verse.
Dryden (1631-1700), Fables (1700), Dedication.
["The 'Fables' again show Dryden's energy of thought and language undiminished
by age." Article on Dryden by Sir Erasmus Henry in
Will martial flames for ever fire thy mind,
And [will it, thy mind,] never, never be to heav'n resign'd?
[Pope,] Odyssey, xii. 145.
What would be the first step? Most likely to sort the examples into different
of construction, with a brief piece of observational analysis. They
all seem to involve pairs of conjoined sentences or clauses, with an elided subject
in the second clause whose antecedent is some kind of object, often preposed, in
the first clause.
Type 1. A subject RelPn in the second clause is coreferential with an
object NP in object position in the first clause: Exs. 1, 11. The two
examples are otherwise distinct in construction.

Type 2. A subject NP in the second clause is coreferential with a fronted
object NP in the first clause: Ex. 4.
Pairs of Conjoined Relative Clauses:
Type 3. The subject RelPn in the second RelCl is coreferential with the
object RelPn in the first RelCl: Exs. 3, 5, 7, 8 (with elided object RelPn in
the first RelCl): Exs. 9, 10.
Type 4. A subject RelPn in the second RelCl is coreferential with a RelPn
in a PrepPh in the first RelCl: Ex. 2.
Type 5. A subject RelPn in the second RelCl is coreferential with the
possessive RelPn whose in the subject NP of the first RelCl: Ex. 6.
Here is what Lowth says about Ex. 1 in his 'Critical Note' (footnote) (1762: 122-
The Verb
hath preserved
hath here no Nominative Case; for it cannot
be properly supplied by the preceding word
, which is in the
Objective Case. It ought to be, "
And He hath preserved
you;" or
rather, "
and to preserve
Some of our best Writers have
frequently fallen into this [Swift is represented many times], which I
take to be no small inaccuracy: … [Here follow the examples above.]
By the term 'supplied', Lowth means no more than that the gap or missing or
elided portion of the expression as it stands is to be filled with morpho-
cognate terms (copies) from the preceding cognate
constituents of the overall construction. But this, as he points out, is impossible,
because the gap in the second member of the construction requires a term with
different morpho-syntactic features from those of its cognate term in the first
'Cognate' is to be understood in the appropriate sense: 'coreferential' and/or
'structurally parallel'. Examples 1 and 11 require only that they be coreferential;
the others that they be both coreferential and structurally parallel, that is, initial in
their syntactic category. But in all of these cases, the principle of Strict
Construction has been violated: Morpho-syntactic features of gaps cannot be only
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
locally specified, but must agree with those of their antecedents. The only
solution is to restore the elided elements supplied by Lowth in his 'Crotchets', that
is, supplying them with their overt local morpho-syntactic features, obviating
illegally supplying (copying) them from their antecedents with the wrong
morpho-syntactic features.
A diverse range of grammarians have claimed that these constructions are
nevertheless indeed ungrammatical, despite the fact that they have been in
English ever since OE times. Their tendency to appear almost at random in a
wide variety of historical texts is however well documented. (See Visser 1963-
1973.) Example 12 is from the story of Cædmon in the OE Bede (
History of the English People
). While it is not of the asymmetrical type, it does
show the elision of the relative pronoun in the second of two conjoined relative
clauses. Further, the number on the gapped relative pronoun is determined
locally, singular instead of plural, like its antecedent. Relating how Cædmon
employed his gift of poetry, the following statement appears:
Ond he forþon næfre noht leasunge ne idles leoþes wyrcan ne meahte, ac
efne þa an þa [
neuter plural
] ðe to æfæstnisse belumpon [
], ond
[GAP; supply
ðæt ðe
that which:
] his þa æfæstan tungan
gedeofanade [
] singan.

'And he for this reason [he had not been taught poetry but had received it
as a divine gift] never could compose any falsehoods or idle songs, but
those alone which pertained to piety, and [GAP] suited his pious tongue
to sing.'
That this is an original OE creation is shown by the Latin original, which is
different in construction:
Unde nihil umquam frivoli et supervacui poematis facere potuit, sed ea
] tantummodo quae [
] ad religionem pertinent [
religiosam eius linguam decebant [
Whence he never could compose anything (of) frivolous or vain poetry,
but only those [things] which pertained to religion were suitable for his
pious tongue.
Where the Latin has two conjoined clauses, the second incorporating a relative
clause, the OE splits the second clause into two relative clauses.
What is the explanation for this strange state of affairs? The
disharmonious case relationships and the asymmetry of the types demands some
analysis. In what follows, a very simple form of constituent structure is used
heuristically and a configurational pattern is posited as the explanation, without
case relationships being relevant. The level that is attempted to be attained is
Chomsky's Observational Adequacy.
Let us assume that every time a constituent is preposed to the left of a
sentence, a new superordinate sentence node is created, with a gap left behind
where the moved constituent comes from, thus:
The man S[(whom) S[we invited [GAP] to dinner]]
the man S[we invited the man to dinner]
No. 16. shows that when the object NP
the man
is moved to the left, the S-node
dominating S[we invited the man to dinner] is expanded into another
superordinate S-node with S[whom � the man dominating the lower S-node that
now contains a gap: [we invited [GAP] to dinner].
Let us call each type of S-node a 'projection (of S)'.
An independent or else lowest S-node that does not dominate any other S-
node whether or not it is dominated by another S-node is a
minimal projection
. A
superordinate node that dominates an S-node and is not dominated by another S-
node is a
maximal projection
. S-nodes that dominate S-nodes and in turn are
dominated by S-nodes are
intermediate projections
. In this way chains of
minimal, intermediate, and maximal projections of S can be built up.
(Intermediate projections do not play a role in this analysis.)
Now, in the expression,
the man didn't come
, the S-node dominating it is a
minimal projection of S, because it does not dominate any other S-node.
Gapping of the second relative pronoun in a pair of conjoined relative
clauses occurs when an antecedent relative pronoun invades the second of two
conjoined S-nodes looking for a coreferential node to delete. It is a kind of
search-and-destroy mission. But it can only destroy coreferential nodes that are in
parallel or cognate positions in configurationally similar S-nodes.
These conditions are met in the first, acceptable construction,
The man we
invited to dinner but didn't come
The head NP of the whole NP,
the man
, has a pair of conjoined relative
clauses dominated by an S-node, as a post-modifier. Restoring the elided
preposed object relative pronoun in the first relative clause, we have
the man
whom we invited
to dinner
Whom we invited
to dinner
is a
maximal projection. It has the structure:
S[whom S[we invited to dinner]].
Now the
sets off on its search-and-destroy mission in the second, conjoined
relative clause
who didn't come
, which is a minimal projection. It is a maximal
projection only by default, because it does not dominate any other S-nodes. The
object relative pronoun
can destroy the subject relative pronoun
in the
second relative clause because they are both initial in their syntactic category and
are coreferential. The fact that
who didn't come
is not a maximal projection
(except by default) is irrelevant: the pronouns are in the same initial position with
no superordinate S-node. If this laborious deduction is correct, it confirms that a
configurational account is acceptable.
Now compare this with the situation in the ungrammatical
the man who
came to dinner but
we didn't invite
[GAP]. The first relative clause is, as
we have stated, a minimal projection. It is a maximal projection only by default,
because it does not dominate another S-node. The subject relative pronoun
the first relative clause now sets off on its search-and-destroy mission, looking for
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
a coreferential subject relative pronoun in a minimal projection of S in the second
relative clause S[whom S[we didn't invite [GAP] to dinner] in a minimal
projection of S. The only candidate for a minimal projection of S is [we didn't
invite [GAP] to dinner] which has as its subject
, not
. The
and the
are not coreferential, and the
cannot destroy the
. The mission is aborted.
There seems to be a meta-rule that only one search-and-destroy mission is
allowed. If not, then the
could continue its search in the superordinate S-
node, S[whom S[…]] and successfully destroy the accusative
without any
further conditions, because this syntactic process does not seem to be sensitive to
The crucial difference between the permitted and the proscribed
constructions is their configurational differences. Now all this may seem arbitrary
ad hoc
, but it has at least Formal Adequacy, the level below Observational
Adequacy: it works. It makes use of very simple geometrical configurations that
are not sensitive to case, agreement, or government relations, but only to
positional, that is, configurational, relationships.
For Lowth, it is the syntactic relations and the case relations that matter.
The orphaned [GAP] in the second relative clause could not find a cognate
relative pronoun in the first relative clause, so the construction was improper.
Lowth did not give a complete account of the phenomenon, but must be
credited with its initial discovery. He was interested only in showing that
perfectly unobjectionable self-evident rules of English grammar could be set up
that, using his definition of Sentence and the Principles of Strict Construction,
could eliminate faults in the construction of English sentences.
Given his complete body of data, collected initially quite randomly, and
asked to classify them into fault-types and to provide English grammatical rules
that would judge the acceptable cases to be acceptable, and to show the fault in
the faulty ones, there are very few today who could accomplish this task.
Casus pendens & nominativus pendens
Whereas the first type of construction proscribed by Lowth is a naturally
occurring English construction-type found throughout the known history of the
English language, the second construction, known as
casus pendens
nominativus pendens
('dangling case' or 'dangling nominative'), is one of those
Latinisms that Lowth considered improper in English because it did not construe
according to the interpretation of grammaticality or 'propriety' dictated by the
principles of Strict Construction. Lowth does not offer any definition of this
phenomenon, or name it as such, because he was focusing on the facts of English
and their interpretation according to the precepts and principles that he was
The dangling nominatives in the two first examples offered by Lowth
under the first rule to disallow them are indicated by him by italics. Very briefly,
whereas all the other cases (genitive, dative, accusative, ablative) are
cases, the nominative is the ungoverned case. It is not governed, but governs.
Here are Lowth's instances (1762:123-124):
Which rule
, if it had been observed, a Neighbouring Prince would have
wanted a great deal of that incense, which hath been offered up to him
by his adorers. Francis Atterbury (1662-1732), Vol. I. Serm I.
In some later edition, this additional example was added:
We have no better materials to compound the Priesthood of, than the
mass of Mankind:
, corrupted as it is, those who receive holy
Orders must have some vices to leave behind them, when they enter into
the Church. Swift, Sentiments of a Church of Englandman [with respect
to Religion and Government] (1708)
The following two examples are cited as improper in the Critical Note (footnote)
under the rule for the case of the relative pronoun which has the same form as the
Latin rule, but applies equally to English (1762:134-136).
, instead of going about doing good,
are perpetually intent
upon doing mischief." John Tillotson (1630-1694), Archbishop of
Canterbury (1691-1694), [Works.] Vol. I. Serm. 18.
Lowth's analysis reads: 'The Nominative Case
in this sentence is
superfluous; it was expressed before in the Relative
Also added in some later edition:
Commend me to an argument,
, like a Flail, there's no Fence [sc.
defence] against it." Richard Bentley (1662-1742), Dissert. on
Euripides's Epistles, Sect. i.
Lowth's analysis reads: 'If that be designed for [intended as] a Relative, it ought to
be which, governed by the preposition against, and it is superfluous: thus,
against which
there is no fence:" but if
be a Conjunction, it ought to be in
the preceding member, "
an Argument[,] [that]." ' (1791:122)
The following is from Lowth's own prose (italics added):
The longer
[Hebrew verses], though
admit of every sort of
Parallelism, yet belonging for the most part to the last class, that of
Constructive Parallels, I shall treat of
in this place, and
endeavour to explain the nature, and to point out the marks of them, as
fully and exactly as I can. (
Isaiah. A New Translation
'Preliminary Dissertation')
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
The subscript indices
identify the relevant noun phrase
The longer
and the
anaphoric pronouns
referring back to it. The preposed object
is pleonastically repeated in the resumptive pronoun
. It is evident
that the noun phrase
The longer
has been moved from object position after the
prepositional verb
treat of
and placed in initial position at the front of the
sentence, focusing attention on it as it picks up the previous argument. This is a
common feature of the syntax and pragmatics of the functional sentence
perspective of English style. However, this noun phrase should have left a gap
after its governing verb, but this position has been filled with the resumptive, or
pleonastic, pronoun
, leaving the noun phrase
The longer
dangling at the
front of the sentence, a
casus pendens
, i.e., an accusative without a governing
In addition, the pair of conjoined infinitive phrases, 'to explain the nature,
and to point out the marks of them', with their shared constituent, 'of them', is felt
by some grammarians or rhetoricians to lack 'grace and beauty' at best, and to be
'improper', or ungrammatical, at worst.
After perusing these examples and deciding on their fault and what rule
might be proposed to solve the problem of proscribing them which is at the same
time an unexceptionable rule of English grammar, you may read footnote
The pleonastic resumptive pronoun is superfluous; the accusative has
already been expressed at the beginning of the construction, to which the object
NP has been moved. If the pleonastic resumptive pronoun is retained then the
initial accusative is a
dangling case
without a governing verb, and the pleonastic
must be removed.
That these constructions originate as a Latinism is clearly expressed in the
trenchant critique by Anselm Bayly 1772. There Bayly provides a running
commentary, mostly in the form of quibbles, on Lowth's
English Grammar
. His
critique is interesting as an example of an older idea of the standard of English,
and for his ingenious and well-meaning, if often incoherent or even inept or
wrong-headed analytical proposals, which give some insight into how not only
English but also classical texts must have been construed in order to make sense
out of what were for the scholars of that time inexplicable vagaries of the syntax
of the classical languages compared to English. Here is Bayly's passage on the
nominativus / casus pendens
, where he jumps in at the deep end with quotations
from Cicero:
"Labour to put an end to this horrid war;
if it can be
accomplished, you will do eminent service to your country, and gain
immortal honour yourself; I have been waiting with daily expectation
of receiving messengers from you with letters,
come, I
shall then be able to judge how to act:
which if they
should be written
every one—" [See Bayly's Latin originals below.] In these sentences
the relatives
are certainly the nominatives before the
can be accomplished
be written
, not
, which are
redundant. This manner of expression, though very common, the
author of the short introduction [Lowth] judges to be improper, from a
supposition, that
being the nominatives,
left by themselves without a verb; but I should apprehend he will be
of another opinion upon reflection, that this form of expression is
purely Grecian and Roman, frequently used by Cicero:* And if the
phrase is neat and correct in Greek and Latin without a pleonasm,
certainly that figure cannot make it improper and mean in English.
The elegance of the expression at least will appear from the flatness of
the correction. [With the dangling nominatives removed:] "If it or this
can be accomplished—If they come"—The Latin form, if it must be
excluded by the decisive authority of this literal grammarian [!], may
be expressed by other turns rather than that proposed; "which, if it can
be accomplished, will bring eminent service to your country, and
immortal honour to yourself—So soon as they come, I shall be
able"—"Which rule, had it been observed, would have taken from a
neighbouring prince a great deal of that incense, which hath been
offered up to him by his adorers:” Short Introd. [1762:] 124. (Bayly
1772: 82-83)
[Footnote to p 82:] *
Quod si erit factum
, et rempublicam divino
beneficio affeceris, et ipse æternam gloriam consequere. Cicero Planc.
Fam. 10 4. Nos quotidie tabellarios vestros expectamus;
qui si
, fortasse certiores quid nobis faciendum sit. Fam. 14. 22.
Bayly does not mention that the relatives
have been moved
(extraposed) to the left out of the clauses within which they originate. This is
impossible in English, and explains the resumptive pleonastic pronouns: the
clauses would not construe without them.
The internal evidence is that Bayly's linguistic intuitions are at least a
generation behind Lowth's. He does not see that in Latin, unlike in English, one
can move an item like the subjects
'which' and
'who' out of their clauses
to the left of the complementizer or connective
'if'. The inflection on the verb in
the clauses out of which the
and the
have been moved serves the
function of the overt subject. Why does Bayly not see this?
This account of the method of Traditional Grammar is offered to Mike
Stubbs in recognition of his contribution to the methodology of present-
day linguistics, and to the study of the English Language.
To set the mood for this piece one could do no better than to read, or to
listen to Robert Schuman's setting of, Heinrich Heine's poem,
Die alten,
bösen Lieder
, from
Buch der Lieder
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
John Horne Tooke (1736; 1792),

[Épea Pteróenta
('winged words')].
Or, The Diversions of Purley
. London, 1786-1805. Two
Parts [Volumes]. Cited from: Part I, Chapter V 'and' III, 'Etymology of the
English Conjunctions:
.' [Tooke derives
from the verb 'to add'.]
Tooke here criticises Lowth for stating that: 'THE Conjunction connects or
joins together
Sentences; so as out of two to make one Sentence.'
(1762:92) Tooke points out that in the sentence,
John and Jane are a
handsome couple
, the individual noun phrases
cannot each
appear alone with the predicate,
is a couple
: 'Is John a couple? Is Jane a
couple?' He gives other examples as well. He cites in support the Latin
examples in the
added (1714) by Jacobus Perizonius

(1651-1715) to the edition by Gaspar Scioppius (1576-1649) of the
Minerva sive de causis linguae latinae
(1562) of Franciscus Sanctius
(1523-1601). Tooke cites the examples adduced by Perizonius to refute
Sanctius' assertion also that conjunction results from syllepsis of two
Emi librum .x .drachmis et .iv. obolis.
Saulus et Paulus sunt
iidem. This particular construction was known also to such grammarians
as George Oliver Curme (1860-1948) (
Grammar of the English Language
Part III,
, 1931), and is today termed 'phrasal conjunction',
rediscovered as if for the first time at the beginning of the heyday of the
first era of generative-transformational grammar in the mid 1960s.
The Journal of The Rev. John Wesley
(1703-1791). Edited by The Rev
Nehemiah Curnock (1840-1915). Standard Edition. Eight Volumes.
London: Charles H. Kelly, 1909-1916. Volume V, 1914:370. The
reference is to: Joseph Priestley.(1733-1804) 1761.
The Rudiments of
English Grammar; adapted to the Use of Schools. With Observations on
. London: Printed for R. Griffiths.
Wesley is probably reading a copy of the 'much expanded' second edition
of 1768. By the English Presbyterian minister, schoolmaster, controversial
religious writer, chemist and physicist, and polymath.
Wesley does not seem to have noticed, nor does it matter, that Priestley's
work was published a year before Lowth's. The significance of Wesley's
remark is that Priestley's grammar, while much praised by present-day
students of the history of English traditional school grammar, from Lowth
on, for his support of the primacy of usage over putatively arbitrary rules,
is otherwise very conventional in content and lacks the comprehensiveness
and originality of Lowth's, as Wesley seems to have observed.
The best compilation that I know is by Pullum 1974, in what was
originally one of three essays completed in the academic year 1970-1971
or 1971-1972 as part of the requirements for the three-term course,
'History of the English Language'
'HEL', in the Department of
Language, now Department of Language and Linguistic Science,
University of York (UK).
The task was to take a good, representative traditional grammar from R. C.
Alston's reprint series,
English Linguistics
1600-1800, and to compare it
with the compilation made earlier of typical strictures about such
grammars and their authors in typical textbooks of the History-and-
Structure of English type.
I forbear to quote from Pullum's article lest readers inadvertently conclude
that I concur in the strictures enumerated there.
In a bizarre example of attributing to Lowth not only prescriptive and
proscriptive practices but also the ability to dictate the course of
development of the grammatical usage of a whole generation of Standard
English speakers and writers and their descendants, he is credited with
having introduced into English the rule that 'two negatives make a
positive'. (For a good example of multiple negation in OE see example 12
above.) The
is usually given as the first edition of the
(1762), and a reference to the section on the Adverb in the
Section on Words (Morphology, or Etymology), where it does not ever
appear, with inaccurate page references (1762: 90-91).
In the first edition, and then repeated in later editions as an introductory
statement to the now extended text, all Lowth has is the laconic:
'ADVERBS have no Government.' (1762: 126). There are no illustrative
'critical notes'.
In fact, the rule was added in
The Second Edition, Corrected
, in the
section on 'Sentences' (or syntax), in the passage dealing with Adverbs
(1763: 138-140).
Two Negatives in English destroy one another, or are equivalent to an
Affirmative: as,
did they
perceive the evil plight
In which they were, or [
] the fierce pains
Milton, P. L. i. 335[-136]. (1762-139-140)
There are two further examples (1763: 139-140 from Shakespeare, and
two from Richard Bentley (1662-1742).
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
Lindley Murray, in his version of this rule (Rule XVI) of Syntax in his
English Grammar
, in order to make the import of the rule and the example
from Milton crystal clear, adds the gloss: 'that is, "they did perceive him."'
This suggests very strongly that those critics who give this rule and this
reference have not looked into the 1762 or any other edition of Lowth's
I have gone into this at some length in order to point out that Pullum is the
only person known to me among several generations of linguists who has
actually studied in depth Lowth's English Grammar.
In fact, it is a commonplace of Logic, one of the Seven Liberal Arts, the
Scholastic curriculum, that
duplex negatio affirmat
, 'double negation
affirms'. It is quite ancient and is found in logical systems throughout the
ages, including in texts in Sanskrit, which has double negation. See Mates
1961: 31-32; 95.
Multiple negation had in fact already virtually disappeared from educated
(literate) English by 1600 (Queen Elizabeth's letters show only a few
traces), beginning with the English Renaissance (1550-1660), possibly in
translating legal texts from Latin into English, in order to avoid potential
ambiguity. But this seems to have begun as a natural process, not
motivated by the force of observing arbitrary grammatical strictures.
Wittgenstein has commented (
Philosophische Grammatik
Philosophical Grammar
(1974), both Oxford, Blackwell,
) that the
formula, ~ ~P
P; or: ~ (~P)
P, is not in fact a rule of logic or grammar
at all, but merely a consequence of the behaviour (interaction) of symbols
such as ~, P, and
A pair of complementary assertions often forms part of the uninformed
critiques of so-called traditional grammarians. The first is that they studied
writing instead of speech. So, as it turns out, has nearly everybody else. It
is sufficient to look at the vast majority of descriptive English grammars,
whether by linguists or textbook writers, to see that there are virtually no
English grammars written on the basis of speech alone or in part, except
perhaps Fries 1952, where it is hardly noticeable, or the grammars of
English by Quirk et al., which use the corpus of tagged spoken texts from
the Survey of English Usage in the English Department of University
College, London.
The second is that they did not even examine the language, but rather
some incorporeal idealized abstraction of their own invention, failing to
describe even the actual usage of the written form. This may be true of the
vast majority of modern scientific studies of English grammar, where the
data so often consist of non-attested arbitrarily constructed examples made
ad hoc
for illustrative purposes, often called 'intuitive data', but which
might better be called
. They are not 'data' in any natural language
or natural science sense of the term, obtained by observation or
experiment, and their structural or formal properties are therefore not
'facts'. This circumstance is the rationale for present-day Corpus
However, it is sufficient to look at the long line of compendious English
grammars, often referred to, rather admiringly or affectionately, bordering
on the patronizing, as 'scholarly traditional grammars', from Fiedler and
Sachs (1861-1877), Mätzner (1880-1885), Koch (1878-1891), Poutsma
(1914-1929), Kruisinga (1925), Kruisinga and Erades (1935; 1953-1967),
Jespersen (1909-1949), Zandvoort (1957 ff.), to Curme (1931;1935),
multi al
. (see McKay 1984, which is not complete) to see that the natural
practice of these grammarians was to use a vast corpus of classified
citations from literature, sometimes newspapers and other writing.
Certainly H. W. Fowler's
A Dictionary of Modern English Usage
(1926) is
devoted entirely to real examples, classified and analyzed in detail, from
newspapers and other printed sources. And of course Lindley Murray's
English Grammar
(1795 ff.), based on Lowth's
Short Introduction
, is well-
illustrated with edited quotations of good and bad usage from numerous
good and bad writers that he took over from Lowth and supplemented with
others. Nor does Murray consider only the Standard English of the
educated writer. His
are mainly instances of improper
(ungrammatical) usage from the 'lower orders', what were commonly
called 'low expressions'.
The only English grammar to examine non-standard English in detail is
Fries American English Grammar. The Grammatical Structure of Present-
Day American English with especial Reference to Social Differences or
Class Dialects (1940), based on the corpus of correspondence from the
First World War in the US War Office in Washington, DC.
'Introductory Memoir' (pp. 1-42), pp. 40-41. George Alexander Stevens
(1710-1784). English novelist and humorist; Richard Porson (1759-1808),
Greek classical scholar and regius professor of Greek at Cambridge
(1792), one of the founders of modern classical scholarship; renowned for
his remarkable memory and facility of recall. His ms. Greek hand is the
basis of all present-day Greek typography.
It would be pointless to assemble a finite corpus and study that, as one
cannot be sure that the relevant instance will be represented. It would be
equally pointless to use a promiscuous or random, putatively
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
representative selection or assembly from all the writers of the day. All
that inferior writers could contribute is that they are ignorant of grammar,
by definition. Lowth believed that it is sufficient to show the state of the
language if one uses the language of 'some of our best writers'. These are
men such as Bentley, Clarendon, Tillotson, Swift, and others, all greatly
admired writers of their day. The thought behind this is that the educated
gentleman and scholar, the 'man of taste', embodies the best and most
cultivated form of polite society: in manners, morals, taste, the arts and
sciences, religion, politics, and, of course, in language. If the English
language, as it is written by 'some of our best writers', is not ruled by
grammar, then the language is indeed in need of those rules that will
ensure that the language is so ruled, in other words, so that it does not, as
Swift says, 'offend against every part of Grammar.' Lowth's sources are
therefore selected both to illustrate the present state of the language, and to
illustrate the application of the rules designed to bring that language into
conformity with the precepts of grammar. Lowth's discussion of this point,
like the other matters that he considers in his
, is admirably clear.
Lowth may have initially come across a different version of the idea of
showing the application of a rule by showing not only its application
('what is right') but also its misapplication ('what is wrong') when he was a
scholar at Winchester College from 1722 until he went up to New College,
Oxford in 1729. He must have used the exercises in Latin composition by
translating sentences from English into Latin by John Clark(e) (1687-
1734). An early edition is entitled
An Introduction to the Making
of Latin
, etc., 3rd edition, 1721, by John Clarke [
In three A3 pages of hand-written notes about the curriculum ('Business at
Winton. College 1756-1757') compiled in
1800. amid the plethora of
Greek and Latin authors and the repeated 'Grammar' of a skeleton
timetable, the name 'Clark' appears once. (This information is due to
Suzanne Foster, Winchester College Archivist.)
In the
, the English sentences and a Latin vocabulary are
arranged in parallel columns, English and Latin, under various rules of
grammar and longer texts. The English sentence is provided with a parallel
string of Latin words in the adjacent column in their dictionary entry form
in approximately correct order with which to make a Latin sentence The
Latin words must be converted into the correct inflectional form required
by the Latin construction. An earlier work (details omitted) with this
design, from which Clark must have got the pattern, was published in
1706 by Nathan (
Nathaniel) Bailey (
1742), better known as the
author of
An Universal Etymological English Dictionary
Both Clark and Bayley are mentioned on the synoptic title-page to the
1750 edition of
A New Grammar: Being the most Easy Guide to Speaking
and Writing The English Language Properly and Correctly
… (1745),
which went through at least thirty-four editions to 1800, by Ann Fisher
(1719-1778), maiden name of Mrs Thomas Slack, wife of the Newcastle
printer Thomas Slack: '[Part] IV. Syntax, or the Order of Construction;
which shews how to join Words aright, in a Sentence or Sentences
together. To which are added, [Chap. IV. & V, 5½ pp] Exercises of Bad
English [under all the Rules of Syntax, as recommended by the author of
the before mentioned Letter (the introduction, signed 'A. B.': 'Anselm
Bayly?)], In the Manner of Clark's and Bailey's Examples for the Latin, to
prove [test] our Concord by' (1750: 127).
Fisher states in a footnote on the first page of Chapter IV: 'Some of these
Examples we set right, lest the learner, expecting them always wrong,
should alter them by Guess.' This observation must have been made by an
experienced teacher.
Cf. this entry from Chap. V, 'Promiscuous Exercises: or, examples under
all the Rules': 'Thou and me is both accused of the same Fault. (1750:
Whether Lowth was 'commissioned' to write this grammar, or merely
presented or was presented with the proposition, is immaterial. The facts
are that the publisher Robert Dodsley (1703-1764), of humble origins, but
who was nevertheless accepted and respected by his betters in breeding
and education, had a major hand in its genesis and publication. It could
well have been his initiative that led to Lowth's authorship. The
correspondence on this between Lowth et al. is to be found in Tierney
1988. This work unfortunately ends with Robert's death. There must be
more from Lowth in the subsequent correspondence with Robert's brother
James (1724-1797), his successor, but this has yet to be published. See
also Straus 1910 for details of publishing history; also Solomon 1996.
Trying to retrieve this information from the OED is futile, because all the
data have been pooled, leading to a kind of muddy-brown mass of
information (not unlike what you get if you mix together all the colours of
the paint-box) from which all the relevant chronological information has
been removed except the dates of the citations. It might make more sense
to list them chronologically by birth date of the author. What would be
required is a
dictionary, giving the senses as found in an
historical succession of dictionaries. Illustrative quotations from texts
contemporary with the dictionaries would then have far more illustrative
On this point see Fries 1952, Chapter II, What is a Sentence?, which
discusses a multitude of attempts by 'traditional' grammarians to define
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
Cf. the following, Rule XXII, the last rule of 'Syntax', from Murray's
English Grammar
ALL the parts of a sentence should correspond to each other: a regular and
dependent construction, throughout, should be carefully preserved.
The following sentence is therefore inaccurate: [Example of improper
construction omitted.]
This is as far as I know the first clear statement of the principle of Strict
Construction. The difficulty in applying the rule as seen by Lowth's and
Murray's contemporaries is well expressed in the following note from
West 1953/1996:
This rule, as Murray admits, 'may be considered as comprehending all the
preceding ones', but he justifies its inclusion by giving a large number of
examples which he hopes will 'afford some useful direction, and serve as a
principle to prove [test] the propriety or impropriety of many modes of
expression, which the less general rules cannot determine.' These
examples make up the rest of the observations on this rule. It was quoted
by John Kigan (
Remarks on the Practice of Grammarians
… 1823: 88) as
showing Murray's consciousness of the inadequacy of his own rules; and
Kigan also criticises its vagueness. 'How to resolve or divide a sentence
into those parts that should thus correspond', he says, 'or, in what this
regular and dependent construction consists, he [Murray] has not shown.
So that after the drudgery of committing these rules to memory, and our
endeavours to digest them, we are obliged to learn the true construction of
a sentence from a long continued attention to the practical use of words.'
What Lowth is offering is only the definition of and the procedure for
establishing grammatical propriety. It is not a recipe for defining Standard
English, as he has already taken the decision to collect his data from
reputable writers with a reputation for 'accuracy': grammatical propriety. It
had to wait for George Campbell's
Philosophy of Rhetoric
(1776) for the
additional criteria of Standard English to be established. See Book II, 'The
Foundations and Essential Properties of Elocution', of the doctrine of
'reputable, national, and present use … which gives law to language'
Nothing, however, is always as it seems. In Chapter III, 'Of Grammatical
Purity', Section I, 'The Barbarism'; Section II, 'The Solecism'; Section III,
'The Impropriety', Campbell shows how any use that violates the purity of
the language by containing any one of these three faults, is improper:
The barbarism is an offence against etymology [morphology], the
solecism against syntax, the impropriety against lexicography [diction;
choice of/proper words]. (1776: 190)
This summary statement is sufficient to show that the putative primacy of
use (usage) is, in the view of the normative grammarian, in fact subject to
the laws of grammar.
It is notable that virtually all Lowth's texts come from the previous
generation of post-Restoration authors. Many historians of English
literature say that there was a distinct change in English style around 1700.
A compilation of the authors represented and the number of instances of
improper usage from each cited by Lowth in his 'Critical Notes' shows that
Swift is quoted far more than any other writer. See footnote 23 below.
A more fully developed version of Lowth's proposal will be found in the
'Preliminary Dissertation' to his
Isaiah. A New Translation
(1778: x-
xxxiv). Finding the metrical basis of Hebrew poetry was considered
essential especially to the translating of the Psalms, There was some
considerable correspondence on this matter in the
Gentleman's Magazine
in the 1740s, complete with pointed Hebrew examples, which Lowth
would as a matter of course have read. Lowth saw at once that the metrical
basis of the Psalms and the other poetical books and passages of Hebrew
Scripture could not be reconstructed because the original pronunciation of
Biblical Hebrew had been irretrievably lost. The Masoretic text of the
Hebrew Bible and its system of pointing he dismisses as 'the Jews'
interpretation of the Old Testament'. (Lowth 1778) So his Oxford
'Lectures' could be considered, like his
Short Introduction
, his proposed
solution to a generally recognized problem.
This has misled some enthusiastic but not very observant students of
Biblical poetry to say that it is
parallelism, which had in fact
been noted before. It is the 'sententious' nature of the poetry that is Lowth's
real discovery, whatever later embellishments have flowed from it.
Lowth is being disingenuously generous to his predecessors. It is the
'conformity' i.e., the parallelism that has many times previously been
noted, but not the sentential basis of this 'conformity'. No one until Lowth
had proposed a sentential solution based on this 'conformity' or
parallelism. In particular, his discovery of the need to repeat matter from
one sentence to complete the sense of the next sentence by filling the gap
there was wholly original with him.
It would be an interesting exercise to try to construct an algorithm for
finding these constructions in any finite corpus. There are many reasons
for thinking that this is in fact impossible, because of the infinite variety of
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
the long-range dependencies involved. Even looking for
constructions conjoined to preceding adjectival phrases etc. requires hand-
sorting of the finds into hits and misses. Even then, potential candidates
would fall through the net because the relative pronoun will have been
elided, leaving only the
The elision of a subject NP or Pn in the second of two conjoined clauses
where the antecedent is not the subject of the first clause is also allowed in
earlier forms of English; see Ohlander 1938 and Burnley 1983.
The second emendation preserves the parallelism. A colleague in the
Department of Mathematics at the University of York, with a keen interest
in language, when shown Lowth's example, made the same suggestions,
and with the same reasoning.
The method of correcting or reinterpreting unconstruable or 'faulty'
construction by rearranging the words into a syntactically new or different,
acceptable form, as if that were what was originally or ought to have been
intended, is a common procedure among amateur linguists, who
sometimes tend to treat the original almost as if it were a misprint. This is
what might be called the 'patch-up' procedure of construing.
When DAR was on his way to the University of California, San Diego, to
give a talk on just this topic, he was asked what he was going to talk about
by a person with no special expertise in English Grammar. When given the
expressions, 'The man we invited to dinner, but didn't come'
'The man
who came to dinner but we didn't invite', they immediately exclaimed, 'Oh,
I see — the second is ungrammatical.'
On an earlier occasion, while waiting for a taxi at the railway station on
our way to a meeting, DAR was asked by another waiting colleague what
he was working on at the moment. When he produced the same pair of
contrasting expressions, his interlocutor retorted, 'They're both
ungrammatical.' DAR rejoined: 'Have you ever read any Swift?' The retort
was swift and sharp: 'Oh. — Swift!'
The very wide-spread idea is that in earlier forms of English, anything is
possible, and we are not obliged to take notice of it.
This example is due to Bruce Mitchell, who also supplied references to a
number of other instances of symmetrical and asymmetrical conjunction of
this type in OE.
Cf. this PDE example:
In this context, granting concessions over Cyprus, which the EU is set to
demand, but [
] would be incendiary to the nationalists, may be
practically impossible. [Deleted object relative pronoun in second relative
clause restored in '[ ]'.]
(Ankara's EU project is in danger of collapse.
The Independent
, Europe,
Analysis, by Daniel Howden, Wednesday 24 May 2006 p 18
The distinctions drawn here between the types of projection are probably
what Chomsky has termed an 'epiphenomenon'. It is the automatic
consequence of the operation of the rules of iterative left-dislocation. The
parser automatically recognizes the type of projection from the syntactic
When I gave a talk on this subject at the Neuphilologische Fakultät at
Tübingen, Uwe Mönnich commented that his grandfather used to use this
permitted English-type of conjunction in German, and he had often
wondered about it. It now seems to have died out in favour of a more
construable alternative:
der Mann, den wir zum Abendessen eingeladen
hatten, der aber nicht erschien
. And
vice versa

The very strong sense of case in German does not like local determination
of case, although it is sometimes found, as in the following newspaper
Viele Firmen wurden in die
[accusative singular GAP]
oder an
den Rand der Pleite
[genitive singular]
. 'Many firms were driven
into or to the brink of bankruptcy.'
Local case determination seems to be permissible if the shared item has
the same form, as in:
Wenn sich der Mann überlegte
[takes the dative of
und endlich entschieden
[takes the accusative of
, …
'When the man had reflected and finally decided, …' This example is
quoted from a late 19thC book by an author who styles himself
, 'The Guardian of Language' (cf.
'goalkeeper'), who
condemns it on the grounds that the single
, which he says quite
rightly is dative by its initial position with
, cannot supply the
missing accusative gapped
required by
. Independently of
Lowth, and using only the principle of Strict Construction, he comes to the
same conclusion, and with the same reasoning.
This is typical of his approach in all his work: not to engage in sterile
explication of the obvious or to refer to the work of others as if treating
their views instead of expounding his own. His straightforward expository
style suits this mode of presentation very well, and lends it an authority
and force that Lowth's argument would otherwise not possess.
'[Tillotson] was perhaps the only primate who took first rank in his day as
a preacher, …' (Article on Tillotson by Alexander Gordon (1841-1931) in
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
Every Nominative Case, except the Case Absolute [one use of the ablative
case in Latin, but Lowth says it should be the nominative case in English
(presumably because it is ungoverned)], and when an address is made to a
Person [vocative], belongs to [governs] some verb, either expressed or
implied; … (Lowth 1762: 123-124).
That the nominative governs the verb and not
vice versa
is shown by the
agreement between the person and number on the verb with that of its
nominative case, or subject.
29 This Latinism - extraposed constituents out of relative clauses, to the left
of the RelPn- occasionally appears in English Renaissance verse and
prose. I have not found any discussion of it in Lowth or any contemporary
grammarian. Cf. this example from Shakespeare's Cymbeline, Act 2,
Scene 3, 19-22:
[Musician] (sings)
Hark, hark, the lark at heaven gate sings,
And Phoebus gins arise,
His steeds to water at those springs
On chaliced flowers that lies, ...
The construction of 'at those springs' etc. is:
PrepPh[at NP[those springs RelCl[ S[PrepPh[On chaliced
S[that lies PrepPh[GAP]PrepPh]S ]RelCl ]NP ]PrePh
It should be evident that the PrepPh 'on chaliced flowers' has been
extraposed out of the RelCl to the left of the RelPn 'that', creating an
adjacent S-node, and leaving a GAP behind.
Bayly, A. 1772.
A Plain and Complete Grammar of the English Language; to
which is prefixed The English Accedence: with Remarks and Observations
on a Short Introduction to English Grammar
[Lowth 1762]. London:
Printed by G. Bigg.
Burnley, J, D. 1983.
A Guide to Chaucer
s Language
. Basingstoke: Macmillan
Educational Ltd. The Language of Literature; Norman, Oklahoma:
University of Oklahoma Press. OUP paperback edition 1994. Since 1989
The Language of Chaucer
Campbell, G. 1776.
The Philosophy of Rhetoric
. Two Volumes. London &
Edinburgh: W. Strahan; T. Cadell; W. Creech.
Fries, C. C. 1952.
The Structure of English. An Introduction to the Construction
of English Sentences
. New York: Harcourt Brace & World, Inc.
Gazdar, G. J. M., G. K. Pullum, I. A. Sag. 1985.
Generalized Phrase Structure
. Oxford: Blackwell Publishing; Cambridge, MA: Harvard
University Press.
Hall, P. 1834.
Sermons, and Other Remains, of Robert Lowth, D.D.
… London: J.
G. & F. Rivington.
Hudson, R. A. 1976. 'Conjunction Reduction, Gapping, and Right-Node Raising.'
52: 535-562. ['Right-Node Raising' is also better called 'Shared
Consituent Construction'.]
Lowth, R. 1753.
De Sacra Poesi Hebræorum Prælectiones
. Oxonii: e
Typographeo Clarendoniano.
Lowth, R. 1762.
A Short Introduction to English Grammar: With Critical Notes.
London: A. Millar and R. and J. Dodsley.
Lowth, R.. 1763.
A Short Introduction to English Grammar: With Critical Notes.
The Second Edition, Corrected. London: A. Millar and R. and J. Dodsley.
Lowth, R. 1778.
Isaiah. A New Translation
. London: J. Dodsley and T. Cadell.
Lowth, R. 1787
Lectures on the Sacred Poetry of the Hebrews
. English
Translation from the Latin by G. Gregory, 1787: Two Vols. London: J.
Lowth, R. 1995
Robert Lowth. The Major Works
. D. A. Reibel (ed.). London:
Routledge/Thoemmes Press. [Enumeration of volume titles omitted.]
McKay, J. C. 1984.
A Guide to Germanic Reference Grammars. The Modern
Standard Languages
. Amsterdam/Philadelphia: John Benjamins
Publishing Company. Amsterdam Studies in the Theory and History of
Linguistic Science. Series V. Library and Information Sources in
Linguistics. Volume 15.
Mates, B. 1961.
Stoic Philosophy
. Berkeley and Los Angeles, California:
University of California Press; London, England: Cambridge University
Press. University of California Publications in Philosophy. 1953. Second
Printing (with new Preface).
Oirsouw, R. R. van. 1987.
The Syntax of Coordination
. London; New York;
Sidney: Croom Helm. Croom Helm Linguistics Series.
Ohlander, U. 1938.
Studies in Coordinate Expressions in Middle English
. Lund:
C. W. K. Gleerup; London: Williams & Northgate, Ltd.; Copenhagen:
Levin & Munksgaard—Ejnar Munksgaard. Lund Studies in English V.
Pullum, G. K. 1974. 'Lowth's Grammar: A Re-Evaluation.'
Linguistics: An
International Review
137:63-78. The Hague: Paris: Mouton.
Straus, R. 1910.
Robert Dodsley: Poet, Publisher and Playwright
. London: John
Lane; New York: John Lane Co.
Traditional Grammar and Corpus Linguistics 'With Critical Notes'
Tierney, J. E. (ed.). 1988.
The Correspondence of Robert Dodsley
, 1733-1764.
Cambridge: Cambridge University Press. Cambridge Studies in Publishing
and Printing History.
Visser, F. T. 1963-1973.
An Historical Syntax of the English Language
. Three
Parts in Four Volumes. Leiden: E. J. Brill.
West, C. E. 1953/1996.
Lindley Murray — Grammarian
. Leeds University MA
Thesis, 1953. Reprinted in: D. A. Reibel (ed.). 1996.
Lindley Murray. The
Educational Works
. London: Routledge/Thoemmes Press. [Enumeration
of volume titles omitted.]
Travelogues in time and space:
a diachronic and intercultural genre study
Andrea Gerbig
University of Bochum
On the basis of a recently compiled corpus of travel literature from the 16
to the
century, this paper investigates both synchronic and diachronic variation.
The synchronic investigation uses a subcorpus of texts from the 21
which shows fascinating language choices reflecting shared values and attitudes
among travellers as an intercultural group, negotiating facets of their identity as
independent and adventurous people. The study will show how a local evaluative
schema of use develops for words which were found to be key words in a
statistical sense in the subcorpus. The second case study looks at the pragmatic
extension of the prepositional phrase
'in the middle of'
in the diachronic
development of the language of travel writing covered in the corpus, in a span of
600 years. Some implications of working with such a multifaceted corpus for
researchers and students of linguistics, literary studies and cultural studies are
discussed. Potential applications to language learning and teaching are briefly
covered, suggesting that/showing how a quantitative and qualitative approach to
authentic data in computer-readable format can help language learners to cope
with the phraseological nature of language.
The Corpus – data and potential routes of exploitation
The corpus I use for the following analyses contains travel literature from the 16
to the 21
It comprises about half a million words per century, as evenly
distributed across the centuries as possible. The corpus will be further extended in
order to fill existing gaps. The subcorpus of the 21
century consists of texts
which are all published on the internet, on a well-structured and well-edited web
site, rather than in interactive weblogs, which are mostly of a highly colloquial
The travel corpus offers broad analytical and interpretative potential. At
the macro-level, it is a repository of cultural knowledge and stories about
intercultural encounters. The corpus can tell us about the role of travel in societies
through the centuries; what it has meant for people to travel and to hear about
travels. It provides us with information about the people who were able to travel
and about their status in society, in political as well as economic respects.
Naturally, the regions travelled to as well as the means of travelling have changed
in significant ways throughout time. These factors have of course also influenced
the kind of contact that developed between the traveller and local people.
At the micro-level, the corpus enables researchers to carry out diachronic
and synchronic studies of the language of travel writing. We can aim at a broader
description of the genre, investigate diachronic changes in style and content, and,
more concretely, trace developments on the level of lexis, phraseology and
structure. Two case studies will illustrate some of these possible approaches.
Theory and Methods
The methodological and theoretical approach underlying the present analyses
starts from the assumption that we are interested in how language creates
meaning. This needs to be investigated not in abstract terms of 'language', but in
more concrete terms of 'actual language use', because this is the observable aspect
of language. The regular and frequent way of representing something shapes our
understanding and our way of dealing with it. This mildly constructivist concept
of utterances in relation to social agency assumes that habitual forms of language
use construe topics. The argumentative framework here goes back to Foucault's
concept of discourse formations (e.g. 1980) and work by the language
philosopher Searle (e.g. 1995), who proposes a major role for language in
creating social and institutional facts. (For a more comprehensive discussion of
the relationship between linguistic evidence and socio-cultural conceptualization
see Gerbig 2003. Implications of work in this field for linguistic theory are
discussed by Stubbs 2007).
Take as an example the alleged destruction of the ozone layer. We cannot
perceive the ozone layer – or its absence – with our senses. We are presented with
data which are interpreted for us by scientists. The 'raw' data would not help us
much. Whatever we know about the ozone layer, we know from discourse.
Whether countries reduce their CFC emissions or not is regulated discursively.
Those arguments which are more powerful create realities. Such realities, by
implication, are construed and therefore fragile.
'Representation' is a concept that helps us link people's experience and
cognition to linguistic encoding. Representations construe versions of the world
(Hall 1997), they construe views on how a culture, or in the present context, a
sub-culture, 'functions'. Although this might be a contested view of culture, it can
be revealing to investigate pervasive discourse that shapes – and is shaped by –
the ordinary way a sub-culture and its individual participants function. With
respect to travelling, a rather abstract, though linguistically transmitted, 'way of
life' or category of experience is visible. It involves, for example, backpacks,
buses, uncomfortable sleep, but also an experience of belonging to a group of
independent, adventurous and outgoing people. Travellers semiotically not only
create a system of rules and rituals that other travellers take as a starting point for
their own behaviour, but they also create expectations about, and in a way even
give rise to, those institutions which cater for travellers.
Travelogues in time and space
This is a reciprocal relationship between the cultural-institutional setting
and the individual-cognitive development. Against the background of the
community and its shared language use, the individual develops his or her social
and linguistic competence, including ideas or beliefs and means of
communicating about travel. The individual's language use, within the boundaries
of the language norm, then not only systematically reinforces, but also gradually
modifies existing institutions and the language system; continual use, with its
gradual variation, shows the diachronic perspective (cf. Halliday 1992 and 1993).
This view also explains variation in language use, an issue which often troubles
teachers and learners of a language (cf. Sinclair 2004). The data seem to suggest
that a particular meaning is generated by repeatedly using particular choices
around a node. These choices are then reinforced and thereby set off from other
systematic uses.
Such intricate relations between language, cognition (knowledge) and
culture (in a broad sense) can be accessed via instances of language use in texts.
These are most conveniently handled in the computer-readable format of corpora.
The travel corpus covers a portion of the genre, diachronically and
synchronically. It provides a particular view of variability and regularity of the
language system. In terms of a frequency distribution, we can see cultural
routines and conventions emerging from the collected utterances. The corpus
therefore provides concrete material to investigate conventional linguistic
behaviour that, given its frequency, is presumably significant within the group of
speakers and inseparable from their shared ways of conceptualisation.
Synchronic example – building group identity
This case study is based on a synchronic subcorpus of the travel corpus, made up
of texts published by native English speakers as travelogues on the internet, all
from the 21
century. The areas travelled to are Africa, Australasia, Asia, the
Middle East and the Caribbean. This subcorpus comprises 485,000 words in total.
The texts are the travellers' reports and stories about their daily experiences,
pleasant and unpleasant, and their adventures during their trips. Because the texts
all appear on a web site maintained by a chain of travel equipment, they are
written for like-minded people, to prepare them for what to expect on their trips.
(Although there does not appear any obvious advertising on this web site, the
chain of stores probably also offers this platform to encourage users to return to
displays of their merchandise).
Surprisingly, there are hardly ever descriptions of interaction with the
local people. In some regions, this is clearly due to a language problem. But even
in Australia or New Zealand, the travellers in general spend too little time in one
place to come into meaningful contact with locals. There is no distinctive British,
Australian or other cultural focus; rather, the writers address English-speaking
fellow-travellers, which is the group they want to belong to. This produces in
some important respects a 'local' phraseological use that is different from a 'norm'
of use, as described on the basis of a mix of genres in a large background corpus,
such as the BNC. I want to show how, in the travel corpus, travellers negotiate
their identity as members of groups of like-minded 'globe-trotters' in quite subtle
ways. For this purpose, I will concentrate on the lexical level, by investigating
keywords in their context.
I use keywords here in the sense of Mike Scott's (1997, 2006) approach
keywords are those words that are relatively frequent in a text, compared to their
frequency in a background corpus that contains a representative mix of everyday
language. If words appear significantly more frequently in the text under
investigation than in other texts, they seem to be of some importance. In general,
lexical items are the main hubs in the creation of meaning around which
structures are built; therefore they obviously receive particular emphasis in
analysis. Keywords give information about frequent topics and about their
evaluation. Keywords for the subcorpus were extracted using Wordsmith Tools,
with the BNC as the background corpus.
The 20 most frequent keywords in the texts written by travellers visiting
Africa and Asia respectively are listed below. Function words are ignored. The
occurrences are given in order of frequency:
We, Bus, Our, Us, Tour, Tent, Sleep, Malawi, Trip, Africa, Zimbabwe,
Park, Truck, Camp, Group, Uganda, Tanzania, Food, Zanzibar, Namibia
We, Driver, Trip, Bangkok, Rice, Bus, Thai, Korea, Tourists, Sleep,
Road, Kathmandu, Restaurant, Taxi, Truck, Us, Cambodia, Ride, Lodge,
The observer can quickly recognize from these lists a particular emphasis on the
group rather than the individual; there are no occurrences of
I, Me, My
but many
uses of
We, Our, Us
. Obviously, the regions travelled to appear prominently, but
most significantly, people write in detail about how they travel
which is mainly
by bus or by truck, where they sleep
and where and what they eat.
The keyword
I chose two keywords,
, which capture the main concerns of these
travellers. The following quote from the corpus epitomizes the paramount role of
buses as a means of transport for 'real' travellers, as they are defined within the
"We've rented a jeep." "A jeep? You don't need a jeep. Go by
bus!" she jeered. I guess we weren't REAL travelers going by
As I will show in the following analysis, the keyword
in the texts about
Africa and Asia shows an almost conventional association with 'difficulty',
'unreliability', 'danger' and 'lack of comfort'. Interestingly, in this sub-corpus (21
century, writings about Africa and Asia), these are not necessarily outright
Travelogues in time and space
negative aspects. Rather, the interpretation of most of the occurrences suggests a
voicing of the travellers' in-group status, of having successfully mastered the
local pitfalls of transportation. This points to established in-group conventions of
evaluating situations and behaviours with respect to people's projected image of
being a 'traveller'.
Such conventions can be seen in habitual collocations around the keyword
investigated. The concept of 'collocation' I use here is a statistical one, of two or
more words co-occurring frequently with each other in running text, within a span
of several words. Frequency of co-occurrence is of particular interest, as "corpus
linguistics is based on the assumption that events which are frequent are
significant" (Stubbs 2001, 29). The collocates can often be grouped semantically
into sets. This marks the 'semantic preference' of the word investigated which "is
the relation, not between individual words, but between a lemma or word-form
and a set of semantically related words, and often it is not difficult to find a
semantic label for the set" (Stubbs 2001, 65). Semantic sets very often share a
particular pragmatic attitude and evaluation, visible in "discourse prosodies"
(Stubbs 2001, 65/6), i.e. habitual evaluations marked in sets of semantically
related collocates of a node. A discourse prosody can be understood as the
pragmatic motivation for choosing the particular word/multi-word unit in the first
place. In its pragmatic function, this evaluative element can be compared to the
illocutionary force of speech acts. Therefore, it plays an important part in
structuring the communicative competence of the members of a community.
With the help of WordSmith Tools (mentioned above), I searched the co-
texts in which the keyword
occurs in the texts about Africa and Asia, grouped
the collocates and collocating phrases (i.e. the words and multi-word-units co-
occurring with

in a span of up to ten words to the left and right) into semantic
sets and sorted them according to frequency. Once these sets of semantically
related words were sorted, their particular evaluative directions became visible, as
can be seen in the following examples. For reading convenience, the results
detailed below are here summarized in a table.
Table 1: Keyword
: Collocates and collocating multi-word units
categorization of
% of all
lines of BUS
crammed, full to capacity,
squashed, trying to secure
seats, grab seats, no idea
where we were going, oven-
like conditions, roasting,
chilly and bumpy
soldiers search, weapon,
worried for my life, ravines
looming, dust and fumes,
battered tin can bus,
hazardous, life-threatening
Difficulty managing
the itinerary and other
tour details
intricacies, difficult,
enigma, figure out, late
Noteworthy local
children selling, pig bound,
passed into, through the
windows, jogged alongside,
slapping the sides, vultures
The most frequent set of collocates expresses
uncomfortable situations
22.5% of all concordance lines around
). This can be illustrated again with a
quote capturing a prototypical situation of discomfort on travels:
Malawians have bladders of steel, the only chance to spend a
penny is if the bus stops at an actual bus station, a rare
The most frequent single-and multi-word collocates of the node
are the
crammed, full to capacity, squashed, trying to secure seats, grab
seats, no idea where we were going, oven-like conditions, roasting,
chilly and bumpy

The following selection of concordance lines shows how these and other
collocates combine with the node in wider co-text to form an evaluative semantic
mood expressing the travellers' experience of situations on and around buses. The
node word is in bold, the relevant collocates have been underlined:
- just
kept on selling tickets
until the
burst out of the seams

- already knowing that
too many people
would be on the
- and the
packed to the ceiling with bodies
- and then catch a
to P that was already
heaving with people

watched as the
filled up until it was overflowing
- terrifically cold
speak or look at the scenery;

- squashed
for four. People in the
but were just
Travelogues in time and space
It is clear from these concordances that
in the present texts prefers the
company of expressions from a particular semantic range, i.e.
shows a
semantic preference. The regular negative connotation of the node-collocate
choices forms the discourse prosody. Preference and prosody are thus related
concepts. In comparison, in the BNC, contexts around
provide a different
picture: there are a few delays and minor discomforts, but no substantial
shortcomings. Similarly to the BNC, the texts about Australia and New Zealand
from the travel corpus do not show any unusual collocations. The authors write
about group dynamics within the bus, about the scenery, stops and sightseeing but
there is nothing substantially dramatic.
Overall, in comparative data, the occurrence of
is much less frequent.
would not have been noted as a keyword in the present travel
data. The sheer frequency then suggests that travellers recognize this means of
transportation as highly noteworthy. The space they give to its negative
contextualization, without ever voicing considerations of alternative transport -
because the situation as given would seem unacceptable to the travellers - seems
remarkable for interpretation of the travellers' style. A natural reaction one would
expect to the described dangers and inconveniences would be a change of
behaviour: for example to share cars or to switch to organized tourist transport. At
least, one would expect a word of warning to fellow travellers. But nothing of this
is visible in the data. Instead, judging from other, also frequent co-occurrences of

with contexts of adventure and a certain appreciation of local colour, what is
pragmatically implied, is that 'real' travellers have to endure such hardships.
Telling about them is proof of one's in-group status.
The second most frequent set of collocates around
concerns situations
(20.4% of all concordances), as epitomized in the following quote:
never sit near the front of the bus, it is better not to
witness how close the bus comes to being scrap metal
unless you fancy raising your blood pressure
The most frequent collocates of

capturing such worries about 'danger' are:
soldiers search, weapon, concerns, knee-smashing, worried for my
life, ravines looming, dust and fumes, battered tin can bus,
hazardous, life-threatening
Longer co-texts as in the following concordance lines show the discourse prosody
of fear and worry very clearly:
- truck had nearly been run off the road
by a passing
- iron bars across the glassless windows
and metal shutters on the
- the
had two wheels
still on the road and the other two over
the cliff
- X's
drivers had been reckless
, but Y's drivers are truly on a
suicide business
- the
did sharp turns
around bends where there were no safety
- Our kamikaze
driver was on a mission to get to B. as quickly
as possible;
- this
, an unroadworthy steel wreck
, had been painted on the
front a
The third most frequent set of semantically related collocates around
difficulties managing the itinerary and other tour details
(16.6% of
all concordances), as illustrated nicely by this quote from the corpus:
they assured us that it was no problem to get a bus
there (i.e. there are buses every 20 minutes, and it
only takes half an hour - translation: there are buses
every two hours, which take about one hour and which
stop 5 km from the village, leaving you to walk the rest
of the way)
More concordance lines clearly show contexts of difficulties and people's
apparent unhappiness with such unreliable situations:
- the intricacies
of the transport system.
are often difficult
- Local
are often an enigma
of travel
- I thought I'd figure out
- The expanse of land that served as
station' was teeming
with vultures tou
- The
left at 8:00am (an hour late
), to trundle a mere three
miles down the road
- the
was only two hours late
leaving the station (and by
'station' I mean
If travel arrangements then turn out to be according to schedule or a smooth
experience, this is stressed as noteworthy, as shown in the following set of
semantically related collocates and concordance lines:
amazingly; surprisingly; unexpectedly; left and arrived on time;
smooth; fast

- pleasantly surprised
when a number 3
pulled up only a few
minutes later
- For some reason
I got the most amazing
. It was like flying
first class
, huge
- one of the smoothest
trips we took - left and arrived on time
The last recognizably coherent and still reasonably frequent set of collocates
revolves around
local colour
(12.8% of all concordances), illustrated with a
prominent quote and several concordance lines:
Ordinary life is conducted through the bus windows at the
stops - live chickens, fish, cabbages, onions,…
Travelogues in time and space
- children
come over to the
bags of cool water, peanuts,
baked as
- saw a pig
on the
, bound
to one of the rear seats
- samosas from the street vendors, passed into
through the
- crowd of locals jogged alongside
, slapping the sides
In total, collocates from the above four sets make up 72% of the co-texts around
the keyword. The remaining 28% deal with less spectacular situations. So, the
majority of the occurrences of
are with a restricted set of semantically related
context words or multi-word-units. There are dangerous, overcrowded, unreliable
buses, and curiously 'local' things. The discourse prosody (i.e. the evaluation) is
mixed; more than a third of these collocates are negative, most of the others are
more or less ironically conceding that taking a bus is part of the adventure.
Interestingly, although descriptions abound about how uncomfortable and
dangerous bus trips in various regions obviously are, there is never a serious
suggestion of not
taking the bus.
From these representations, we can observe a local schema emerging.
Obviously, the evaluation of
seen in the presented data is not, or only to a
very small degree, shared in other contexts. The uses shown above do not occur
in any comparable way in the BNC, and not even in the other travel sub-corpora.
In the subcorpus of texts on Africa and Asia, the problems that are reported on are
regularly downtoned by expressions and evaluations foregrounding the
excitement about 'local colour, 'adventure' and – in terms of the travellers' ethos
'doing the appropriate thing'. This makes up the characteristic semantics of the
word in this subcorpus of the travelcorpus. It is interesting to see how widespread
this evaluation is, given that over 70% of all contexts of

could be classified
accordingly, i.e. 'negative' but 'adventurous' and 'appropriate'. In a nutshell, as
these two statements by travellers put it:
- still, I was going to take the
cause it was more adventurous

- The
ride to V. was an experience I will never forget
The keyword
How and where they sleep seems to be as much of an in-group marker for
travellers as taking the bus. People habitually complain, but nobody takes action
against the problems concerned with sleeping. The complaints are always rather
mild, never outraged; there is never a clear signal to other travellers not to go
there, do this, or book that. The principal complaint about lack of, or poor quality
sleep most often goes together with situations implying adventure. This fits with
the group prosody for buses, apparently being part of the same behavioural
pattern. Money is of course an issue
as for example in:

miserable sleep followed, but money was saved

More focused than the financial question, however, is the microcosm of the
community of travellers with their habits and rituals. Foregrounding this 'in-
group' experience, the contexts of
outnumber those of
by far
mainly as part of a group. Not spending money on comfort could be seen as
another group marker.
The majority of collocates around forms of
disruptions of
and hindrances to sleep
, such as shown in the following semantic sets:
- Noise from other people or animals
shouting, rustling, lovemaking, row, argue, music, rabbiting

- Ground and beds
hard/stone) ground, hostel floor, grass, dormitory, outside

- Small animals as nuisance or danger
cockroaches, ants, (tsetse) flies, mosquitoes

- Big animals felt to be a danger, mainly because they come too close to the
sleeping place.
circling, trampling, hippos, elephants, smell

- Natural forces
gale, rain, storm, (howling) wind, water, soak, muddy, cold

- How people describe their sleep
restless, not a wink of, grumpy, deprived, little, irritable
from lack of, fitful
- Wanting to sleep
much needed, need to catch up on, try(ing/tried)
(desperately)) to, manag(ing/ed) to get some, not a chance of
getting, stiff from, unable to, terrible, angry, rough

The semantic profile indicated by the above collocates can be further illustrated
with the following concordance lines:
- Not a chance of being able to get a decent night's sleep. If I
could have slapped them
- I'd had a terrible night's sleep due to the incessant ramblings of
the watchmen.
- Such sleep as could be managed was punctuated by braying from the
- procession of ants marched past over the sheet. Excellent, I had
friends to sleep with.
- miniature scorpion and giant leech. I tossed and turned, finally
drifting off to sleep
The keyword
is not as frequent as
, although both are among the ten
most frequent keywords in each subcorpus. Together with issues of provision
with food, issues such as means of transport and choices of accommodation
usually determine the comfort of travelling. Such comfort generally enables the
pleasure one can gain from visiting and sightseeing. In the present data however,
the collocates around both nodes support the same local schema: 'real' travellers
Travelogues in time and space
gladly undergo the hardships of local transport and makeshift sleeping
arrangements. This marks them as both sharing the travellers' creed of soaking up
the local way of life and as conforming to self-chosen, but apparently
conventionalised in-group-values. The linguistic evidence for this conclusion has
been clearly demonstrated in the above analysis.
Diachronic example - the semantic change of a phrase
It is a trivial truth that words are frequent because they occur in frequent phrases
(Sinclair 1999: 162, Stubbs 2004). And every language learner soon realizes
context around a word is needed to recognize meaning. This leads to phrases as
units of meaning in language beyond the word boundary. Take as an example the
meaning of the word
The explanation given by the Cobuild Dictionary
(1995) is: "The middle of something is the part of it that is furthest from its edges,
ends, or outside surface". This is certainly true. But how does the meaning of the
following uses from the travel corpus fit in?
- as if to highlight the day's absurdities, we got stuck in the
middle of the desert
- he was outraged at being called out in the middle of the night for
basically nothing
It is hard to imagine that the first speaker meant the exact geographical mid-point
in a clearly confined desert area. And when does 'night' start and when does it
end? So, when exactly is the middle of a night? The layer of meaning 'furthest
from its edges / ends' is doubtlessly still present, but rather as a basis for a more
specific, pragmatic meaning. We will come back to this point and more examples
Many researchers have stressed that ordinary language use is to a large
extent made up of more or less pre-constructed chunks (cf. e.g. Pawley and Syder
1983, Cowie 1988, Moon 1998, Hunston and Francis 2000). Such units are
variably called extended lexical units, lexical items, phrases, clusters, and more.
Sinclair (1991: 110 and 1998) showed that, although such stretches of words
"might appear to be analysable into segments", they have to be seen as one
choice, as a form-function unit that is habitually used and that expresses a
conventionalized meaning within a language community.
So while 'middle' itself seems to have a clear, de-contextualized meaning,
the phrase 'in the middle of (+article)' frequently expresses a rather specific
pragmatic meaning, which will become clear from the examples discussed below.
If there is consensus
that our language use is largely made up of such more or less
variable phrases, we naturally want to be able to find them in corpora. Depending
on the way a corpus is annotated, there are two options:
First, as some concordance programs offer searches for n-grams, i.e.
recurrent strings of uninterrupted word-forms stopping at sentence boundaries, we
can check their frequency in a text and their preferred co-texts.
Second, if the words in the corpus are marked for grammatical categories,
i.e. if the corpus is tagged, we can also look for strings of 'part of speech' (POS)-
tags. In the British National Corpus, which is the largest corpus of contemporary
written and spoken British English, Stubbs (2004) has shown that the
prepositional phrase structure 'preposition-determiner-noun-
-determiner' is the
most frequent 5-word string, realized in expressions such as
at the end of the,
at the beginning of the, in the middle of the
. The frequency of this
POS-5-gram can partly be explained on semantic grounds since (as realized in the
noun slot) we often talk about wholes and parts, beginnings and ends, in order to
express spatial and temporal structures in discourse.
The prepositional phrase
in the middle/midst of the/a
occurs 49
times in the subcorpus of weblogs (21
century). Out of these, in only 14
instances does the meaning of
correspond to the description in the
Cobuild Dictionary, given above. In 35 occurrences, the phrase rather has the
pragmatic function of an intensifier, underlining the speaker's surprise or anger
about a situation or about a situation's inappropriateness, as shown in these
examples from the travel corpus:
in the middle of ART
(72% used in a pragmatic function)
- around 4pm and landed in Auckland in the middle of a rain storm. A
- main road was an oasis of a place in the middle of a hot, dusty
dirt track.
- we were served breakfast in the middle of the desert.
- or make mad dashes to toilet blocks in the middle of the night.
- I am not being kicked out of here in the middle of the night and
having to carry
- it's brutally expensive to email in the middle of the jungle. So,
- Out of nowhere appeared a hut in the middle of the ocean. Sticking
The string
the middle of
is actually redundant for the message and functions as
an evaluative marker. The texts from the 21
century do not show any
occurrences of
. Apparently, young people have stopped using this form.
Here are some more examples from the 20
century, for
as well as

in the middle of ART
(64% used with a pragmatic function)

- office buildings and executive flats in the middle of a vast urban
- caravan parks standing in fields in the middle of a lonely,
windbeaten nowhere,
- confusion, like someone wakened in the middle of the night by an
- on their white womanless island in the middle of the sea. As we
- this one in bed (in the middle of the day, remember); and this
Travelogues in time and space
in the midst of ART
( 80% used with a pragmatic function)
- I had scant sense of being in the midst of a rich, proud city
built of
- had the slightest sense that I was in the midst of a lot of
granite, and it was
- pychiatrist, until he turned to me, in the midst of a detailed
explication of the
- new development. It was like being in the midst of an ugly-
building competition.
While the spatial and temporal meanings are both residually present in these
contemporary examples, the pragmatic meaning gains force through time. An
investigation of the diachronic part of the travel corpus shows this interesting
change. Work in construction grammar and grammaticalization (Heine, Claudi
and Hünnemeyer 1991, and Hopper and Traugott 1993) has demonstrated such
processes of semantic weakening of elements in habitual phrases with ensuing
pragmatic strengthening of the entire phrasal use. Typical examples are
metaphorical processes of grammaticalization with terms for body parts
over time come to be used first as locatives, then as temporals and finally with a
more pragmatic function.
I analysed the use of the phrase 'preposition-article-
throughout all centuries in the travel corpus, from the 21
down to the 16
. A
clear development can be seen:
Table 2: Pragmatic use of
from C21 to C16
pragmatic use in % of total
pragmatic use in % of total
(35 of 49) = 72%
no occurrence
(11 of 17) = 64%
(4 of 5) = 80%
(3 of 11) = 27%
(7 of 17) = 41%
(2 of 14) = 14%
no occurrence
(3 of 26) = 12%
no occurrence
(0 of 6) = 0%
(3 of 11) = 27%
The actual numbers are quite small for using percentages. This only serves to
make the proportions clear. In the 21
and 20
century the majority of the
occurrences have a pragmatic function. The numbers for
show that it is
used more often in its evaluative function, increasingly so in more recent
language data. This evaluative expression is an idiomatic form-meaning complex.
Backwards from the 19
to the 16
century however, the use is increasingly
literal. It mainly denotes a specific place or time, as illustrated in two examples
from the 16
century below, where
means 'furthest from the edges / ends'.
- they are reported to have their eyes in their shoulders, and their
mouths in the middle of the breasts, and that a long train of
hair groweth
- that of a buffalo, feet like those of an elephant, and a horn in
the middle of the forehead, which is black and very thick
A comparison with the Early Modern English section of the Helsinki Corpus
(1500-1710), shows that out of in total thirteen occurrences of the phrase
article', there are six uses indicating at least an
approximate spatial or temporal reference. The other five examples indicate
concrete spatial uses of the phrase. There is not a single use of the phrase in the
pragmatic function described above in connection with the travel corpus.
HC, EModE: spatial / temporal approximation
- But this work may be done in the middle of the day, if the heat be
not violent
- London buildings; there is in the middle of the town the Duke of
Norfolks house
- stately building, placed by it selfe about the middle of the
outside of a Town,
- the Fort in the middle of the City is circular; toward
- In the middle of the River we had a pleasant Prospect on both
- in the middle of the Vale we repaired to the
- In the middle of the Munsel i.e. a whole Day's Journy the Butler
HC, EModE: concrete spatial use
- but you must first make Incision alongst wise, vpon the middle of
the foresaid (^Escharre^): Then put in some small quantity (that
- (^cleave^) just through the midst, so as the (^bud^) may be
directly in the middle of the one half; and then snip off a part
of the (^leaf^),
- You may make the (^cross cut^ )in the middle of the downright
(^score^) on the Stock, and lifting up the four
- outhouses very handsome; a coach yard and stables in the middle of
which is large gate into the ground and built over with a high
- that a good space may be left in the middle of the Schoole, so as
six men a breast may walk up and down
Conclusion - and possible applications
The present study has provided access to a local intertextual net where uses of
particular expressions (such as those involving
) are tightly linked.
They derive their meaning partly through delimitation from uses in other data, in
this case other travel data or the BNC, and thus form a local (group-based)
schema in their representation of travellers' preferences and conventions.
Researchers of language and culture tend to agree that linguistic
representations and cultural concepts are related in a non-trivial way (cf. research
in the Whorfian tradition and language philosophy, as briefly discussed above, as
well as work in critical discourse analysis). They do not, however, agree about the
form and extent of this relationship, nor about the kind of research necessary to
document it. Of course no direct link can be presupposed between language use,
cognition and culture. The point made here, however, is that frequently used
Travelogues in time and space
linguistic routines in a particular area of meaning are inseparably linked to both
the cognitive schemata the language users have formed for this area of
meaning/part of their experience, and to institutionalised cultural practices There
are always alternative ways of expression, but if particular forms are habitually
chosen, this points to a cognitive preference. A cultural basis for such shared
preferences seems plausible.
The travel corpus is specific in its topic, that is, stories about travelling.
However, the text types are diverse, from letters, reports, diaries and adventure
novels to publications on an internet platform. This provides for a range of
information on form-meaning relations. In response to questions of ambiguity or
variation, Sinclair (2004: 281) speaks of a "guiding principle", that "each distinct
meaning in language can be associated with a word pattern that is unique to it".
Work in Construction Grammar has come to similar conclusions about the form-
meaning relationship. Kay (2001: 19) states that "pragmatic information … can
be directly associated with linguistic form in irreducible grammatical
constructions – that is, constructions whose form cannot be produced by
combining smaller units of the grammar according to general principles".
Again, the question we are addressing here is that of the size of a unit of
meaning. This question has to take into account structures and word choices in
their semantic co-text, often motivated by a pragmatic intention. By alerting
language learners and teachers to the principles of collocation, language
awareness will grow. A fairly easy start is to investigate sets of semantically
related words (semantic preference) and conventionalized evaluation (discourse
prosody), where these prosodies are often the pragmatic reason for making the
choice at all.

The travel corpus offers insights into the changing role of travelling in
society. Judging from the popularity of travel literature throughout the centuries,
travelling has always been an interesting experience, for the travellers themselves
as well as for the readers of their writings. On the basis of diachronic linguistic
data we can observe shifts in cultural practices; from travelling as a privilege for
the aristocracy and the rich to travelling as a life-style of adventure and
intercultural experience for backpackers, as could be seen in the data from C20
and 21 (for a related study showing such development see Gerbig and Shek
2007). We can also observe, at each historical stage, which role the different
forms of mobility take in the value systems of a culture. The diachronic parts of
the travel corpus are a repository of information about cultural historical events
and changes in the English language and can thus be equally of interest to
students of both culture and linguistics.
As most of the texts in the travel corpus are literary texts, it is a
particularly suitable basis for work in the field of stylistics. In linguistic
descriptions of 'English usage' we need to give more prominence to the place of
literature in our communicative lives. The travel corpus offers a real potential for
such interdisciplinary research (for literary linguistic analyses of a related kind
see e.g. Stubbs 2005, Müller-Wood and Gerbig 2006).
There are many corpora available today
very large, general background
corpora covering language uses in everyday situations
as well as
many smaller,
specialized corpora covering particular topics or genres, like the one used in this
study. It would be desirable of course to have these different corpora accessible in
compatible forms. This would give both researchers and students the opportunity
to search for exactly those uses they are interested in at a particular moment in
their studies, so as to arrive at a more detailed picture of 'norm' and 'variation'.
From the teacher's point of view, variations need to be introduced gradually, so
that the most frequent, usually canonical, forms are prioritized in teaching
(Sinclair 2004: 275), proceeding to sets of more specialized uses, at first at the
receptive level, as the learner's competence increases. Local and specialized uses,
such as those in parts of the travel corpus can then be pointed out to advanced
learners. Furthermore, a corpus in the hands of advanced learners is the ideal
resource for them to increase their phraseological awareness (using material like
the small study of
above). In terms of improving learners' competence, it
is important to draw on authentic language use. This need has been extensively
discussed (see e.g. Sinclair ed 2004); it has been demonstrated in comparative
studies of large background corpora (such as the BNC) with EFL / ESL textbooks
that the latter still misrepresent the distributions and patterns of use as found in
actual language data (cf. e.g. Römer 2004, Conrad 2004).
The analyses in this paper can be viewed as one module of a possible
ethnographic study to discover the meanings people construe, which circulate and
become embedded in their daily experience. Corpus linguistics offers the
possibility of documenting this relationship from the language side. As Allen
(2000: 37) puts it: "Meaning … is always at one and the same time 'inside' and
'outside' the text". The textual basis, however, is the common stock from which
we all draw, which is analysable and therefore accessible.
I would like to thank Patricia Sift, Barry Morley and Ingo Bachmann for
their cooperation in planning and compiling the corpus. I am further
grateful to Patricia Sift for discussions about some of the data and Barry
Morley for writing tailor-made pattern matching software.
I would like to thank Naomi Hallan and two anonymous reviewers for
helpful critical comments on an earlier draft.
See the full list of texts from the 16
to the 21
century in the appendix.
The most recent texts (21
century) are taken from the internet at
Here is an example of 'middle' being used for a body part (Helsinki
Corpus, Early Modern English section)
Travelogues in time and space
and a paire of olde broken slip shooes on his feet, a rope
about his
instead of a girdle, and on his head an old
greasie cap
As these small examples indicate, from a corpus linguistic view, the
autonomy of 'linguistic levels' is not fully tenable.
Allen, G. 2000.
. London and New York: Routledge.
Conrad, S. 2004. "Corpus linguistics, language variation, and language teaching".
In J. Sinclair (ed.) 67-85.
Cowie, A. 1988. "Stable and creative aspects of vocabulary use". In R. Carter and
M. McCarthy (eds).
Vocabulary and Language Teaching,
London: Longman.
Foucault, M. 1980.
, ed. C. Gordon. London: Harvester.
Gerbig, A. 2003.
Korpus und Kultur: Korpuslinguistische Analysen zu
Repräsentationen deutscher und britischer Politik in den Printmedien.
Unpublished manuscript.
Gerbig, A. and A. Shek 2007 "The phraseology of tourism: a central lexical field
and its cultural construction". In P. Skandera (ed.)
Phraseology and
Culture in English
, Amsterdam:
De Gruyter, 303-322.
Hall. S. 1997 (ed.)
Representation: Cultural Representations and Signifying
. London: Sage.
Halliday, M. 1992. "Language as system and language as instance: the corpus as
a theoretical construct". In J. Svartvik (ed.)
Directions in Corpus
61-77. Berlin: Mouton.
Halliday, M. 1993. "Quantitative studies and probabilities in grammar". In
Hoey (ed.)
Data, Description, Discourse
, 1-25. London: HarperCollins.
Heine, B., U. Claudi and F. Hünnemeyer 1991.
Grammaticalization: A
Conceptual Framework
. Chicago: The University of Chicago Press.
Hopper, P. and E. Traugott 1993.
. Cambridge: Cambridge
University Press.
Hoey, M., M. Mahlberg, M. Stubbs, W. Teubert 2007.
Text, Discourse and
Corpora: Theory and Analysis (Corpus and Discourse)
. Continuum
International Publishing Group Ltd.
Hunston, S. and G. Francis 2000.
Pattern Grammar: A corpus-driven approach
to the lexical grammar of English
. Amsterdam / Philadelphia: Benjamins.
Kay, P. 2001. "Pragmatic Aspects of Grammatical Constructions".
Moon, R. 1998.
Fixed Expressions and Idioms in English: A Corpus-Based
. Oxford: Clarendon.
Müller-Wood, A. and A. Gerbig 2006. "A Literary-Linguistic Reading of Graham
Brighton Rock
: Interdisciplinarity in Practice". In A. Gerbig and
A. Müller-Wood (eds).
How Globalization Affects the Teaching of
English: Studying Culture Through Text
. Lampeter: Mellen.
Pawley, A. and Syder, F. 1983. "Two puzzles for linguistic theory". In J. C.
Richards & R. W. Schmidt (eds).
Language and Communication
Longman. 191-226.
Römer, U. 2004. "A corpus-driven approach to modal auxiliaries and their
didactics". In J. Sinclair (ed.) 185-199.
Scott, M. 1997. WordSmith Tools Manual. Oxford: Oxford University Press.
Scott, M. and C. Tribble 2006.
Textual Patterns
. Amsterdam: Benjamins
Searle, J. 1995.
The Construction of Social Reality
. London:
Sinclair, J. 1991.
Corpus, Concordance, Collocation
. Oxford University Press.
Sinclair, J. 1998.
The lexical item". In E. Weigand (ed.)
Contrastive Lexical
. Benjamins. 1-24.
Sinclair, J. 1999. "A way with common words". In H. Hasselgård and S. Oksefjell
Out of Corpora
. Amsterdam: Rodopi. 157-79.
Sinclair, J. (ed.) 2004.
How to Use Corpora in Language Teaching
. Benjamins
Sinclair, J. 2004. "New evidence, new priorities, new attitudes". In J. Sinclair
(ed.) 271-299.
Stubbs, M. 2001. Words and Phrases: Corpus Studies of Lexical Semantics.
Oxford: Blackwell.
Stubbs, M. 2004. "On very frequent phrases in English". http://www.uni-
Stubbs, M. 2005. "Conrad in the computer: examples of quantitative stylistic
Language and Literature
, 14, 1: 5-24.
Stubbs, M. 2007. On texts, corpora and models of language. In M. Hoey, M.
Mahlberg, M. Stubbs & W. Teubert.
Text, Discourse and Corpora
London: Continuum.
List of books included in the travel corpus, 16
to 20
Leland, John,
The Itinerary of Lohn Leland in or About the Years 1535-1543
Torkington, Richard,
Ye Oldest Diarie of Englysshe Travell: Being the Hitherto
Unpublished Narrative of the Pilgrimage of Sir Richard Torkington to
Jerusalem in 1517
Hakluyt, Richard, The Principal Navigations, Voyages, Traffiques and
Discoveries of the English Nation, Vol. I - VI
Coverte, Robert,
A Trve and Almost Incredible Report of an Englishman, 1612
Travelogues in time and space
Taylor, John,
The Penniless Pilgrimage. All the Workes of John Taylor the Water
Poet, 1630
Chardin, John,
Travels in Persia 1673-1677
, Book Two.
Dampier, William,
A Voyage to New Holland
, London, 1702.
Dampier, William,
A Continuation of a Voyage to New Holland
, London, 1709.
Fiennes, Celia,
Through England on a Side Saddle in the Time of William and
Mary. Being the Diary of Celia Fiennes
Fryer, John, A New Account of East India and Persia, Being the Nine Years'
Travels, 1672-1681.
18th century
Defoe, Daniel,
Tour through the Eastern Counties of England, 1722
Fielding, Henry,
The Journal of a Voyage to Lisbon
, 1755.
Johnson, Samuel,
A Journey to the Western Islands of Scotland
, 1775.
Cook, James,
A Voyage Towards the South Pole and Round the World
, Vol. I,
London, 1777.
Piozzi, Hester,
Observations and Reflections Made in the Course of a Journey
through France, Italy, and Germany
, 1789.
Smollett, Tobias,
Travels through France and Italy
Sterne, Lawrence, A Sentimental Journey through France and Italy.
19th century
Stevenson, Robert L.,
Travels with a Donkey in the Cevennes
, New York, 1911.
Franklin, John,
The Journey to the Polar Sea
, London, 1823.
Dickens, Charles,
American Notes
, London, 1842.
Kinglake, Alexander,
, London, 1844.
Burton, Richard F.,
First Footsteps in East Africa
, London, 1856.
20th century
Hudson, William,
Afoot in England
, London, 1909.
Douglas, Norman,
Old Calabria
, 1915.
Douglas, Norman,
, London, 1921.
Chatwin, Bruce,
In Patagonia
, 1977.
Bryson, Bill,
Notes from A Small Island
, 1995.
Fowler, Beth,
Half Baked in Taiwan
, 2000.

An extended view of extended lexical units:
tracking development and use
Naomi Hallan
Trier University
Using corpus data derived from recordings of spontaneous conversations, the article
examines the acquisition and use of various types of extended lexical units involving the
path morpheme
. Data from children acquiring British English is compared with
material from the COLT corpus of teenage English and with adult conversational data
from the British National Corpus and some of the changes in usage that occur during
development are examined.
The astronomical increase of computing power and data storage available to the
ordinary researcher has led to a massive change in the way many scholars
approach the study of language phenomena. It seems no longer necessary to deal
in detail with the objections of the prevailing orthodoxy of the mid-20th century
to the use of empirical data. Scholars such as Stubbs (1996, 2001), Sinclair
(1991), Hunston and Francis (2000) and Moon (1998) have shown how the
application of corpus linguistic methods can lead to new insights into the way
language is used. The methods they describe are being applied ever more widely,
and provide a valuable addition to the study of language acquisition, enabling the
researcher to detect and quantify patterns which are easily overlooked when
simply reading through transcriptions of child speech (c.f. Theakston et al. 2005
on the complexities of the acquisition of auxiliaries).
The investigation discussed in this article forms part of an on-going study,
part of which has already been reported in Hallan (2001). As discussed in that
article, the function of word-forms such as
on, over
, which are traditionally
classed as prepositions, is in fact quite complex. Data on the acquisition of
showed that they are initially acquired not only or even primarily with
prepositional function, such as
on there
over the road
, but rather in adverbial
or particle functions as part of multi-word expressions, such as
over here
fall over
. Following Bowerman (1996) I use the function-neutral term
to refer to the closed class of grammatical words under investigation.
The term is of course not entirely neutral, since it assumes that these forms are
fundamentally spatial terms of some sort, which may not be the whole story. In
the case of
, for example, there is some evidence that the deixis involved
could be as much interpersonal as spatial (Hallan 2001: 100). However, the form
considered in this article is clearly spatial from its very first use.
Linguists have observed for some time that a great deal of language is
produced and understood as relatively unanalysed multi-word building blocks
(e.g. Pawley and Syder 1983, Langacker 1987, 2000, Sinclair 1991, Stubbs 2001,
Bybee 1998). It has become clear that the meaning of such extended lexical units
(ELUs) is not simply built up additively from that of the individual word-forms
they contain. Many of these units are framework constructions with variable slots
(Goldberg 1995, Stubbs 2004). I shall use here Stubbs' (2004) term
to describe such units. As became clear in the earlier work in this project, some
path morphemes are acquired from the beginning as parts of extended lexical
units as well as, or even before, their acquisition as free-standing lexemes.
Different languages vary enormously in the way they encode spatial relations. An
important typological distinction, first described by Talmy (e.g. Talmy 1985,
1991), is between
languages. Verb-frame
languages, such as Spanish or French, encode the path of motion in the verb itself
sortir, entrer, monter, descendre
) and the manner of movement as an optional
adjunct (
sortir en courant
). Satellite-frame languages, such as English, encode the
path of motion using satellites — path morphemes — either as adverbs or
prepositions (
go/come out, in, up, down
); the manner of movement is encoded in
the verb (
run out, crawl in, jump up, roll down
). Basic verbs of motion and
caused motion in English encode some sense of a direction of movement, but
only in terms of changing distance from a reference point (
go, come, bring, take
The path encoded by
is bound up with one of the earliest learned spatial
concepts — that of a container and its contents (see Bowerman 1996 for an
overview of work on conceptual development). In cognitive linguistics this notion
is considered to be one of a number of pre-conceptual
image schemas
Lakoff 1987: 271 ff.), arising naturally out of the configuration of our own bodies
and underlying our interpretation of the physical world. The function of
assumed to be the encoding of the path of something moving from the inside to
the outside of a container of some sort.
As stated above, the encoding of a path is not necessary the only or even
the first function in which young children encounter these word-forms (cf. Hallan
2001: 99, 104). However, in the case of
, the spatial function does seem to be
primary, and has such force that the word can occur independently, in a quasi-
verbal function (cf. Tomasello 2003: 87) with directive force, in the speech of the
youngest learners:
The meanings of out
An extended view of extended lexical units
(1) *GER:
. *GER:
. *MOT: not
. *MOT: although we are
going to the swings.
(Wells age band 2, Gerald 1;6.6 wants to go outside)
The word-form is perceptually salient in continuous speech, as it often carries a
stress, and is rendered more so by the adults, who produce the construction
you *
, where the slot in the phrase-frame is filled by the change-of-state verb
or the basic motion verbs
(2) *MOT:
you get. (Mother lifts him out of the bath). *GER: I'm
cold. *MOT: you what? (Mother dries him). GER: I want. *GER: I
want to be wrapped in a towel.
(Wells age band 5, Gerald 2;3.5)
In theory it ought to be possible for people to say
on you get
down you go
you come
; in practice these phrases are not found in the Wells corpus. The motion
are found with
in, out
, while the change of state verb
occurs with these three and also with
. The reason for this would certainly
repay investigation; in the present context however it is enough to note that the
construction occurs with
and must contribute to the form's salience for the
young language learner. Since it is accompanied by actions, as in (2), this
presumably further reinforces learning.
Containers, frontiers and goals
However, it is important to look carefully at what is actually being learned. Being
lifted out of the bath might be perceived as emerging from a container, and
somehow comparable to removing toy cars from a box:
(3) *DAR: brm brm. *MOT: do you want the brm brm
(Wells age band 2, Darren, 1; 6.2)
It seems improbable however that children as young as 18 months should
perceive an analogy between the removal of the toy car from the box and their
own desired exit from the house into the garden (cf. example (1) above). The path
encoded by this use of
seems to have more to do with the crossing of a
frontier, with all the formalities such transitions can require:
(4) *ELL: goak [coat]. *MOT: you wants your coat on? *ELL: yeh.
*MOT: it's windy
, El.
(Wells age band 3, Ellen 1; 9.0 wants to go outside so asks for her coat)
(5) *ELS: boots. *MOT: you don't need your boots on. *MOT: you can go
with your shoes on.
(Wells age band 4, Elspeth 2; 0.2)
(6) *ABI: (action: goes into the garden with no shoes on and steps in a
puddle) *MOT: goodness sake you've come
in your tights.
MOT: after I've just dressed you.
(Wells age band 2 Abigail 1; 5.28).
\n \f\r



Figure 1: Use of
by different speakers in Wells age band 2
In the early age bands the children themselves say very little: only 5 of 107 uses
in the recordings at 18 months (see figure 1). However they are hearing the
word all the time, as parents, siblings, grandparents and visitors attempt to direct
their movement and actions:
(7) (Betty has shut a door on another child, Joanne) *MOT: Betty let
Joanne come
please because she's a pest on her own. *FAT: she
in the bedroom? *MOT: yeh. *FAT: Betty come on. *FAT: let
. *BET: Joanne
(Wells age band 4, Betty, 2;0.3)
In addition, there seems to be a hierarchy of
-ness within the house, with
bedroom and living room being the most
, and hall and kitchen referred to as
, although the whole house is still on the same side of the greater frontier:
(8) *MOT: Sarah? *MOT: go on out and play. *SAR: go on out and
play? *MOT: that's right. *SAR: well it's raining. *MOT: go out in
the hall then.
(Wells age band 6, Laura, 2;6.3 and her elder sister are annoying the
An extended view of extended lexical units
(9) *MOT: get on your bike then. *MOT: out in the kitchen. (Neville
gets his bike in the kitchen - noises as bike sets off) *NEV: round
corner. (rides his bike from kitchen towards sitting room) *MOT:
coming in here are you?
(Wells age band 5, Neville, 2;3.0)
At the same time, the children's entourage regularly use
to refer to a different
sort of path: going or being absent from home for a relatively short period (not
more than a day) in pursuit of some goal:
(10) *ELS: I want to go
. *MOT: well we'll be going
*MOT: we've got to go up to school shortly.
(Wells age band 6, Elspeth, 2; 6.6)
This use is more frequent than the others in the teenage and general corpora and
will be discussed below. Although it can be related to the others we cannot
assume that it is necessarily perceived as a derived meaning by the young
children who are learning it. The children are confronted with three uses of
which they will doubtless integrate into a more generalised category as they grow
older, but which are at least in the first years perceptibly distinct.
Functions of
Quasi-verbs and adverbs
As we have seen,
encodes a variety of related paths and can do so as an
independent adverb or even with a verb-like function. In addition to the dynamic
can encode a static locative, and in both functions it is frequently
associated with
there, here,
or another locative expression (cf. example (11)
In both these adverbial functions it can be used with either an intransitive
or a transitive verb:
(11) *BET: bubble. *MOT: no, have them
in the garden. *BET:
bubbles. *MOT: because they're no good indoors, because they
don't go up.
(Wells age band 3, Betty, 1; 9.4)
Phrasal verb particles
Other constructions involving
and a verb can be classed as phrasal verbs:
unlike those described above,
functions as a particle rather than an
independent adverb and the constructions convey meanings which can be
described as metaphorical extensions of the path emerging from a container.
Examples include
spread out
clear out, find out.
It is easy to imagine how the
sense of these ELUs might have developed from the primary meaning of
such derivations are a feature of the classic expositions of cognitive linguistics
(for examples, see Hallan 2001, Langacker 2000). With verbs like
find out, write
work out
, where the particle has completive force rather than any locative
sense (or at most a very tenuous one), it is doubtless possible to show how their
meanings can be related to the central uses of
, but questionable whether
language users really make the connection when actually uttering the forms.
Since the phrasal verbs are present in the language they hear from the very
beginning, it seems reasonable to suppose that children acquire the supposedly
derived meanings in parallel to the 'primary' ones. Tomasello (1992: 172) found
that the phrasal verb particle function was acquired later than the verbal/adverbial
one, and this is borne out by the later production of such uses in the Wells corpus:
(12) *JON: I'll show you. *JON: you just stretch it
like that look.
*MOT: let me see then.
(Wells age band 8, Jonathan 2; 11.29 helping his mother to make a paper
The more complex processing of the placement of the direct object between the
verb and particle could be a sufficient explanation for the later production of such
constructions — the children show by their behaviour that they understand them
much earlier.
Any investigation of the prepositional use of
is complicated by the parallel
existence of the complex preposition
out of
, which shares most but not all of he
functions of
. The choice of one form or the other is probably influenced by
variety, register and sociolect.
But there is some meaning difference as well.
There are no examples of this in the Wells corpus, but the following, from the
BNC sub-corpus, will illustrate the difference:
(13) The only thing is, see what I'm scared of now, you go and put the
mower in there and cart it
out the back
and the bits of moss
dropping off the mower onto that patch! (KCN 6153)
(14) I said I want to get my teddy bears
out of the back
. (KBE 1020)
Clearly, in example (13) the simple prepositional phrase encodes the goal of the
movement and, unsurprisingly, is the only one used for the corresponding static
locative — after it has been moved, the mower is
out the back
. In (14) the
compound prepositional phrase encodes the source of the movement, not the
position of the direct object: the corresponding locative would be
in the back
since it appears from the context that the teddy bears are in a back room rather
than outside the building.
An extended view of extended lexical units
The corpora
The material used for this study comes from three corpora of spoken British
English, covering acquisition by children up to five years old, teenage language
and casual conversation among adults. The current lack of a corpus of spoken
language for the primary school period is a real problem for a study of this type,
but there is nevertheless a great deal to be learned even without the intermediate
data such a corpus could provide.
The language acquisition data
The Wells corpus (Wells 1981, 1986) is made available through the CHILDES
databank (MacWhinney 2000), a large and continuously growing collection of
child language data contributed by scholars all over the world. The Wells corpus
contains transcriptions of recordings of spontaneous speech from 32 children, 16
boys and 16 girls, born in the Bristol area in the second half of 1972. The children
were fitted with a radio microphone on a harness and samples were recorded at
random times during the day using a tape recorder in another part of the house.
The children and their entourage were not aware whether or not they were being
recorded at any given moment. Recordings were made during a full day every
three months between ages 18 months and 3 years 6 months and again at almost 5
years, giving ten age bands in all. The material obtained covers the full range of
everyday activities and contains not only the children's own productions but also
anything said in their presence, whether addressed to them or not. Contextual
information provided by parents enables a better interpretation of elliptical
speech. The corpus contains about 395,000 words.
The teenage language
The Bergen Corpus of London Teenage English (COLT) (Stenström et al. 2002)
was recorded in the late spring and early autumn of 1993 by teenagers recruited
from schools in Barnet, Camden, Hackney, and Tower Hamlets, as well as from a
boarding school in the Greater London Metropolitan Area, in Hertfordshire. The
teenagers carried Walkman tape recorders with a lapel microphone, and taped
their conversations in a variety of situations over three to five days, keeping a log
of the situations and participants. The material was first of all transcribed by the
Longman team preparing the British National Corpus, and then corrected and
edited by the Bergen team before being annotated in various ways. For this study
I have made use of the edited orthographic transcription. The whole corpus
contains approximately half a million words, however 13 files (out of 337)
contain extended passages of teacher monologue, so I removed these from the
analysis, reducing the word count by some 31,000 words
The pupils were aged between 13 and 17 at the time of the recordings and
were thus born between 4 and 8 years later than the children in the Wells corpus.
In addition, they were all Londoners, and a number were from ethnic minorities,
whereas the Wells study was carried out in Bristol, and deliberately excluded
ethnic minority families. There is consequently a certain problem with the
comparability of the two corpora. Unfortunately, as is so often the case with
spoken language data, there are no more suitable datasets available — collecting
and transcribing spoken language is difficult and time-consuming, and above all
expensive, so researchers are very often obliged to make the best of what is
already out there. In the case of the child data, the universality of the topics and
situations in a household with a toddler makes for a reduction in the variability of
the child-directed speech. The teenage data also shows a relatively restricted set
of situations, with peer-group interactions predominating, and many of these
being concerned with different types of relationships (Stenström et al. 2002: 28-
29). This pre-occupation has a similar unifying effect, and it seems reasonable to
assume that, had the recordings been made in Bristol, the talk would have been
remarkably similar
The spoken BNC
The spoken section of the British National Corpus (2001) contains about 10
million words. However the material includes a wide variety of speech types,
from more or less scripted monologues such as lectures, sermons or reports in
meetings, to casual conversation. In order to restrict the material to something
more directly comparable with the two other corpora, I made use of David Lee's
Index to the BNC (Lee 2001, 2003). I was able to select the spoken conversations
in Lee's 'demographic' domain, which are spontaneous conversations recorded by
informants belonging to different social classes. Since the recordings used to
compile COLT were also included in the BNC, in the initial transcription made
by the Longman team, it seemed sensible to avoid simply comparing two versions
of the same thing, by identifying and removing the BNC texts containing COLT
material. The remaining texts contain 3741769 words, according to Lee's
spreadsheet. This is clearly a much larger body of data than either of the two
other corpora, which means that a detailed analysis of individual contexts is
scarcely practical. In addition, the material contains speech not only from adults
but also from young children and teenagers, as well as drawing on informants
speaking a wide range of regional varieties of British English. Nevertheless it
seems reasonable to suppose that the material averages out to a snapshot of
spoken English as a whole and can thus serve as a basis for comparison when
attempting to identify specific characteristics of the other datasets.
The data
The corpora were examined using concordancing software, either WordSmith
Tools for the PC or
for the Macintosh, and the results were displayed as a
KWIC (Key Word In Context) concordance. It might be thought that much could
be done by using part-of-speech tagged data to identify particular constructions.
There are two reasons for not doing so, one general and one specific to this
An extended view of extended lexical units
As a number of scholars have pointed out (e.g. Sinclair 1991, Tognini-
Bonelli 2001, Römer 2005), the use of tagged corpora can pre-judge the issue one
is trying to investigate. The categories ascribed by the parser to the different
word-forms in the corpus are based on a pre-existing view of how the language is
structured. Particularly in the case of spoken language, with its fluid structures
and often elliptical style, a category assignment based on an often unconscious
acceptance of quite traditional (and sometimes fundamentally prescriptive) ideas
about grammar will not do justice to the data (cf. Hallan 2001: 91-2, 94-6).
In the case of this particular study, the boundaries between the functional
categories for path morphemes are very fuzzy:
The distinction between verb particles and prepositions is a
problematic one in adult language, and, as usual, that means that it is
even more problematic in early child language. (Tomasello 1992:
The tags in the %mor tier of the Wells corpus (an interlinear tagging level which
is provided in the CHILDES CHAT format, MacWhinney 2000), are a case in
is frequently, but not consistently, tagged as a preposition where the
utterance clearly contains an adverb or particle. A similar problem exists with the
tagging of the BNC (cf. Hallan 2001: 102).
Clearly the practicability of a direct categorization depends on the volume
of material one has. So far as the Wells corpus is concerned, concordances of the
separate age bands can easily be inspected and a functional categorization made
on the basis of context as well as utterance structure. The individual texts can
easily be examined using the CLAN software (MacWhinney 2000), and the
contextual comments are a valuable aid. The COLT corpus has a similar volume
of occurrences (see below) and thus a direct inspection is possible here too. In the
case of the BNC sub-corpus, the volume becomes unmanageable in the short
term. The concordances have therefore been restricted by searching for
interesting constructions rather than the word-form
on its own.
Initial results
There were 1270 occurrences of
in the Wells corpus, between 110 and 139 in
each of the different ages bands, and 1546 occurrences in the COLT corpus. This
is equivalent to approximately 3215 and 3092 occurrences per million words
respectively. The BNC sub-corpus had 12644 occurrences of
giving a figure
of approximately 3379 per million words.
Thus the frequency of
in the three corpora is not radically different.
The important findings lie in the distributional differences across different
constructions. As the analyses proceed, it will ultimately be possible to produce a
distributional profile or
for the different language samples.
As an example, figure 1 shows the distributional profile for
in the first age
band of the Wells corpus. The construction codes are explained below.

" #
" " 
Figure 2: Uses of
in Wells band 2
Some possible constructions are not represented in this sample, which may be an
indicator of the features of child-directed speech in this population (for an
overview of issues related to CDS see Snow 1995).
For the Wells and COLT corpora, the individual examples were examined and
classified according to the constructions they represent. The constructions
identified were as follows:
used as a free-standing locative adverb or in combination with
there, here
, or another locative expression, with or without a form of BE;
: a static verb other than BE, such as STAND or STOP used with
as a locative adverb;
used as a free-standing directional adverb;
: an intransitive verb of motion (
come, go)
or change of state, or a
manner-of-motion verb (jump, dance, climb) used with
as a directional
adverb. This classification is restricted to clear examples of the physical
movement of entities or substances from physical containers;
: as 4., but with transitive verbs of caused motion (
take, bring
, etc)
and a concrete direct object. Passives of such constructions are included
Throw X out
belongs here when it refers to the violent ejection of an
entity from a location, but in 9. below when it means
dispose of/get rid of
: a simple prepositional phrase,
out X
: a compound prepositional phrase,
out of X
: a phrasal verb with
as the particle;
An extended view of extended lexical units
: a transitive phrasal verb with a direct object. Passives of these were
also included here;
: a phrasal verb with a wh-complement;
: a transitive verb and object followed by a compound prepositional
V X out of Y
. This construction is mostly found as a small group of
fixed expressions and is discussed below (6.2.1);
: a participial form (mostly past participle) of a phrasal verb used as
an adjective,
stressed out
: a compound noun formed from a phrasal verb,
anything where a detailed examination of extended context failed to
bring comprehension, often because the speaker was indistinct at a crucial
point in the utterance. Where the verb and particle were clear but the
object not, it was often possible nevertheless to classify the construction.
Some specific cases
Many corpus studies are concerned with the behaviour of specific word-forms
and their collocates. In this analysis, rather than extracting frequent collocates and
attempting to measure the strength of attraction between them (e.g. Gries and
Stepanowitsch 2004), I have considered the behaviour of
in some specific
phrase-frames and constructions.
Going out
It seems obvious that
will frequently be found next to a form of the lemma
GO. What is interesting is how the collocation functions. Table 1 shows some
frequencies for the different corpora.
As discussed in section 2.2 above,
GO out
not only encodes movement
along a path, specifically one exiting from a container, but also the crossing of the
frontier between "indoors" and "outdoors" (see examples 4-6 above), and the
pursuit of social goals outside the home:
(15) No, if you were feeling okay and wanted to
go out
go out
Saturday night somewhere. Let's
go out
for a nice meal
somewhere. (KD3 426)
The construction
GO out with
can function as a subset of this usage, where the
other participants in the social activity are made explicit. This phrasal-
prepositional verb has another, idiomatic meaning, however, particularly (though
not exclusively) for teenagers:
X GO out with Y
means that X is having a
romantic or a sexual relationship with Y (cf. Stenström et al. 2002: 39):
(16) Sarah's going out with someone nice, Sarah. who's she
going out
? someone in the R A F. not Mike? no, this guy called Bob
(KD6 4191)
(17) Why are you and James cuddling up to each other, you
going out
each other, or you just a friendly hug? (b133701)
Table 1:
GO out
in the different corpora
as percentage of total
out with
as percentage of total
as percentage of GO
with something
as social/work outing
percentage of GO
out with
percentage of GO
percentage of all
as romantic relationship
percentage of GO
out with
percentage of GO
percentage of all
Table 1 shows the proportions of
GO out
, and of
GO out with
in their different
meanings, for the three corpora. Although this is not shown in the table, the
progressive forms predominate, as would be expected when referring to an on-
going situation (although all tense and aspectual forms are present). Clearly, who
is, or has been, or will be
going out with
whom occupies a disproportionate place
in the conversation of teenagers compared with speakers in general: the
percentage of all uses of
to be found in this phrase-frame with this meaning is
almost 12 times as great in COLT as in the BNC. In the Wells corpus it is
completely absent, despite the recordings containing a proportion of conversation
between adults on purely adult topics. The only examples of the phrase frame in
An extended view of extended lexical units
Wells refer to
going out with something
, and are about wearing the right or wrong
footwear for crossing the frontier to the outdoors (cf. examples 4-6 above).
It is well known that different text types or genres can be distinguished by the
range and type of grammatical structures they contain (e.g. Biber et al. 1994). It is
trivial that the topic(s) of discourse in a text can be identified from its lexis. What
Table 1 appears to show is that speakers' pre-occupations can be tracked in their
lexicogrammar. Stenström et al. (2002: 32-41) discuss the importance of
'Romance' and 'Sex talk' as topics of conversation, something that is confirmed by
one of the teenagers in no uncertain terms:
(18) Oh God this whole school revolves around snogging people,
out with
people, shagging people. It's just a nightmare. (b142303).
Not only is it a nightmare, but it substantially alters the proportion of phrasal
prepositional verbs in the usage profile of
in COLT!
Prepositional phrases
As discussed in section 3.3 above,
forms part of prepositional phrases both on
its own and in the compound preposition
out of
. An examination of the phrase-
out NP
in the spoken BNC, using the search facility at Mark Davies' VIEW
web-site, suggests that the most common simple prepositional phrases will be
those where NP is definite. I therefore compared
out the X
out of the X
each of the three corpora. Table 2 shows the data for the Wells corpus.
Table 2:
Out the
out of the
in the Wells corpus
out the out of the
percentage of
percentage of
percentage of
out the
percentage of
percentage of
out the/out of the
Frequency of nouns in PP (% of PP)
16 nouns
11 nouns
24 (30.4)
21 (15.7)
8 (11.3)
6 (13.0)
7 (9.9)
Verbs preceding PP (% of PP)
no verb: static
5 (7)
no verb: dynamic
14 (19.7)
4 (8.7)
12 (16.9)
26 (56.5)
13 (18.3)
4 (5.6)
9 (12.7)
4 (8.7)
manner of (caused) motion
The table suggests that both types of prepositional phrases are in fact acquired
and used as ELUs. There are two nouns that are used frequently with both
(since the numbers are so small, the percentages
are calculated only to give an idea of proportions). The frequent verbs are
An extended view of extended lexical units
the basic motion verbs. The stand-alone prepositional phrases are almost all
out (of) the way
, and the phrases with
are instructions to the child to
direct its gaze towards events outside the house (apart from a sentence from a
story being read out, where a cat
jumped out of the window
). A similar situation
was found with
(Hallan 2001: 105-6): although there were many more nouns
occurring in PPs with
, they appear to be common fixed expressions for
referring to everyday situations.
The frequencies in COLT (table 3, below) show how the functions of the
two PPs have changed.
Out the
is still found overwhelmingly with a handful of
nouns: these are clearly the fixed expressions acquired in childhood, which have
doubtless persisted largely unanalysed, especially since they are correspondingly
infrequent with
out of the
. On the other hand the range of nouns found with
out of
is much greater, suggesting that this is the construction of choice for talking
about a range of more specialised situations. There were also some interesting
idioms (see 6.2.1, below) and one occurrence of
out of the question
The verbs used with both constructions have changed too: in particular the
proportion of manner-of-motion verbs has gone up at the expense of the basic
motion verbs. In particular, words for different styles of
are very
frequent with
out the
, and this is primarily due to a notable difference in
behaviour between toddlers and teenagers: whereas young children mostly look
out of windows, teenagers (especially boys) spend more time throwing things out.
This is perhaps another expression of the ebullient energy of the teenage
informants (cf. Stenström et al. 2002), although it might also be a result of the
opportunities for experiment and amusement afforded by first floor classrooms!
In the BNC sub-corpus (table 4, below) the sequence
out the
is much more
frequent than
out of the
. Surprisingly the number of prepositional phrases with
out the
is higher too. However the table makes clear that this is largely due to the
familiar unanalysed fixed expressions we saw in the other corpora. The PPs with
out of the
include additional fixed expressions such as
out of the blue
out of
the ordinary,
which are presumably acquired during schooling. The large number
of static locatives for
out of the
, often referring to the origin of something,
confirms the suggestion made above that
out of the
focuses on the beginning of
the path of motion, while
out the
relates to its end. A closer examination of the
other verbs used with the different preposition forms, and the type of nouns they
govern, would doubtless provide more information about the meaning difference
which enables the two to continue to exist and function alongside one another.
Table 3:
Out the
out of the
out the out of the
percentage of
percentage of
percentage of
out the
percentage of
percentage of
out the/out of the
Frequency of nouns in PP (% of PP)
24 nouns
38 nouns
6 (11.5)
2 (4.5)
19 (36.5)
4 (9.1)
5 (9.6)
vending machine
Verbs preceding PP (% of PP)
no verb: static
8 (18.2)
no verb: dynamic
5 (9.6)
4 (9.1)
3 (5.8)
4 (9.1)
8 (12.7)
5 (1.4)
12 (23.1)
other manner of (caused) motion
12 (23.1)
10 (22.7)
An extended view of extended lexical units
Table 4:
out the
out of the
in the BNC sub-corpus
out the out of the
percentage of
percentage of
percentage of
out the
percentage of
percentage of
out the/out of the
Frequency of nouns in PP (% of PP)
~130 nouns
~130 nouns
110 (22.9)
44 (11.5)
5 (1)
8 (2.1)
36 (7.5)
6 (1.6)
49 (10.2)
5 (1.3)
19 (4)
10 (2.6)
13 (2.7)
11 (2.9)
9 (1.9)
7 (1.8)
Verbs preceding PP (% of PP)
no verb: static
33 (8.6)
no verb: dynamic
25 (5.2)
5 (1.3)
66 (13.7)
113 (29.5)
40 (8.3)
9 (2.3)
36 (7.5)
36 (9.4)
17 (3.5)
10 (2.1)
5 (1.3)
12 (2.5)
other manner of motion
Verbal and physical assault
In the course of examining the prepositional phrases, I observed a number of uses
of a more complex construction,
X V N out of Y
(no. 11 in section 5.2 above). The
vast majority of examples were from two phrase-frames:
(a) X take the mick/mickey/piss out of Y;
(b) X beat/knock/punch/blow the shit/crap/fuck/stuffing out of Y.
Stenström et al. (2002: 200-8) show how important ritual insult is as a part of
teenagers' interactions, particularly among the boys. It is therefore unsurprising
that (a), which is used to talk about what one has said or will say, or in some
cases to defuse a situation, is so frequent in COLT.
(19) laughing&#x-3l1; -2u;g-6;&#xh6i-;n-;g60;I'm just taking the piss out of you Jock and it's working,
for the first time in my life it is working. (b141801)
Phrase-frame (a) occurs 38 times in COLT, all but 4 with
. In contrast, it only
occurs 18 times in the BNC sub-corpus, 9 with
, 4 with
and 5 with
. Phrase-frame (b), used to talk about physical assault, is reassuringly
infrequent in both corpora, occurring only 4 times in COLT and, even less
frequently over all, only 5 times in the BNC sub-corpus. The BNC also has
frighten the life/living daylights out of X
(3 times) and
take it out of X
once. These
are the same construction, and similarly negative in effect, but very infrequent.
The use of phrase-frame (a) in COLT, almost 1 per cent of all occurrences of
(312) in any construction, testifies to the importance of this sort of interaction
in teenage talk.
The cases I have discussed here represent only a small part of the information
which the systematic study of path morphemes is bringing to light. An important
goal is to construct the usage profiles of
for children of different ages, for
teenagers and for a large sample of English conversation. It is clear that even the
interim results presented here can give new insights into the functions of a class
of English words whose role is too often assumed to be unambiguously
prepositional. The range of information, from construction frequency to
sociolinguistic insights, to which the techniques of corpus linguistics give access
makes the use of such empirical methods an essential part of linguistics.
An extended view of extended lexical units
I would like to thank Andrea Gerbig and Oliver Mason for inviting me to
contribute to this volume and for their helpful comments, and above all
Mike Stubbs for all his encouragement and patience.
In this section and section 3 I will mainly give examples from the
language use of young children hearing and acquiring British English, and
from that of their entourage. In the following sections I will also include
data from the COLT and BNC sub-corpora. Examples from the Wells
corpus are identified by the age band, based on the numbering of the Wells
files, and the name of the child, with the child's age given in the format y;
m.d. There is no age band 1. Most of the Wells formatting, such as
intonation codes, has been removed, and more than one utterance has been
placed on one line. The mother and father are designated by *MOT and
*FAT, the target child and other participants by *XXX, where XXX is
represents the first 3 letters of the name. Examples from COLT and the
BNC sub-corpus are identified by the text reference. BNC data cited in
this article have been extracted from the British National Corpus World
Edition, distributed by Oxford University Computing Services on behalf
of the BNC Consortium. All rights in the texts cited are reserved.
Biber et al. (1999) do not mention the prepositional use of simple
. It is
not apparently something taught to second language learners: works such
as Murphy (1994) or the
Longman Language Activator
(1993) contain no
mention of it. The
Collins COBUILD English Dictionary
(1995) states that
"[i]n American English and informal British English,
is often used
instead of
out of.
" The
Collins COBUILD Dictionary of Phrasal Verbs
(1989: 477), in the entry on
in the Particle Index, states that
"[i]n some varieties of English such as American English, and also in
non-standard British English,
can also be used as a preposition
with verbs of movement. […] in standard British English you need to
add the preposition 'of'."
Since "non-standard" is often a euphemism for "low-class", it is worth
examining the distribution of prepositional
in the BNC sub-corpus,
using the socioeconomic class information in Lee's (2003) index. 27% of
the texts were recorded by AB (upper and upper-middle class) speakers,
and produced 17% of the examples. The 33% of texts recorded by C1
(lower-middle class) speakers contained 29% of the examples. C2 (skilled
working class) informants accounted for almost 23% of texts and 37% of
examples, DE (unskilled working class and unemployed) and unclassified
informants provided nearly 18% of texts and almost 17% of examples.
These figures are scarcely conclusive, especially since the class indicators
refer only to the person providing the recording; without identifying the
individual speaker for each example it is difficult to claim that the whole
of any text represents only the language use of a particular social group.
One thing is certain, however: British people are all taught in school that
rather than
out of
is simply "wrong". That so many of them do
so, at least in casual conversation, would doubtless be regarded by many
as yet another example of declining standards!
Biber, D., S. Conrad and R. Reppen 1994. 'Corpus-based approaches to issues in
applied linguistics',
Applied Linguistics,
15 (2), 169-189.
Biber, D., S. Johansson, G. Leech, S. Conrad, E. Finegan 1999.
Grammar of Spoken and Written English
. Harlow, England: Pearson
Education Limited.
Bowerman, M. 1996. 'The origins of children's spatial semantic categories:
cognitive versus linguistic determinants', in: J.J. Gumperz and S Levinson
Rethinking Linguistic Relativity
, (Studies in the Social and Cultural
Foundations of Language No. 17), Cambridge, Cambridge University
Press, 145-76.
The British National Corpus, version 2 (BNC World)
2001. Distributed by Oxford
University Computing Services on behalf of the BNC Consortium. URL:
Bybee, J 1998. 'The emergent lexicon',
CLS 34: The Panels
, Chicago: Chicago
Linguistic Society, 421-35.
Davies, M. 2005. 'VIEW: Variation In English Words and Phrases' URL:
Gries, S. Th. and A. Stefanowitsch 2004. 'Co-varying collexemes in the into-
causative', in: M. Achard and S. Kemmer (eds.),
Language, culture, and
, Stanford: CSLI Publications, 225-236.
Hallan, N. 2001. 'Paths to prepositions? A corpus-based study of the acquisition
of a lexico-grammatical category', in: J. Bybee and P. Hopper (eds.),
Frequency and the Emergence of Linguistic Structure
Amsterdam/Philadelphia: John Benjamins Publishing Company, 91-120.
Hunston, S. and G. Francis 2000.
Pattern grammar: A corpus-driven approach to
the lexical grammar of English
. Amsterdam/Philadelphia: Benjamins.
Lakoff, G. 1987.
Women, fire and dangerous things: What categories reveal
about the mind).
Chicago/London: University of Chicago Press.
Langacker, R.W. 1987.
Foundations of cognitive grammar, 1: theoretical
. Stanford: Stanford University Press.
Langacker, R.W. 2000. 'A dynamic usage-based model', in: M. Barlow and S.
Kemmer (eds.),
Usage-based models of language
, Stanford: CSLI
Publications, 1-63.
An extended view of extended lexical units
Lee, D.Y.W. 2001. 'Genres, registers, text types, domains and styles: clarifying
the concepts and navigating a path through the BNC jungle',
Learning and Technology
, 5 (3): 37-72.
Lee, D.Y.W. 2003.
The BNC Index
(Excel spreadsheet available for download at
MacWhinney, B. 2000.
The CHILDES project: Tools for analysing talk.
Mahwah, NJ: Lawrence Erlbaum Associates.
Moon, R 1998.
Fixed expressions and idioms in English: A corpus-based
. Oxford: Clarendon Press.
Murphy, Raymond 1994.
English Grammar in Use: A self-study reference and
practice book for intermediate students.
Cambridge: Cambridge university
Pawley, A. and F.H. Syder 1983. 'Two puzzles for linguistic theory: Nativelike
selection and native fluency', in:
Language and Communication
, J.C.
Richards and R.W. Schmidt (eds.), London: Longman, 191-226.
Römer, Ute, 2005.
Progressives, patterns, pedagogy: A corpus-driven approach
to English progressive forms, functions, contexts and didactics.
(Studies in
Corpus Linguistics 18). Amsterdam/Philadelphia: Benjamins.
Sinclair, J. 1991.
Corpus, concordance, collocation
. Oxford: Oxford University
Snow, C.E. 1995. 'Issues in the study of input: Finetuning, universality, individual
and developmental differences, and necessary causes', in: P. Fletcher and
B. MacWhinney (eds.),
The Handbook of child language
, Oxford:
Blackwell, 180-193.
Stenström, A.-B, G. Andersen and I.K. Hasund 2002.
Trends in teenage talk:
Corpus compilation, analysis and findings.
Stubbs, M. 1996.
Text and corpus analysis: Computer-assisted studies of
language and culture
. Oxford: Blackwell.
Stubbs, M. 2001.
Words and phrases: Corpus studies of lexical semantics.
Oxford: Blackwell.
Stubbs, M 2004. 'On very frequent phrases in English: distributions, functions
and structures', plenary lecture given at ICAME 25, the 25th anniversary
meeting of the International Computer Archive for Modern and Medieval
English, in Verona, Italy, 19-23 May 2004.
Talmy, L. 1985. 'Lexicalization patterns: semantic structures in lexical forms', in:
T. Shopen (ed.)
Language typology and syntactic description. Vol. III:
Grammatical categories and the lexicon
, Cambridge: C.U.P., 57-149.
Talmy, L. 1991. 'Path to realization: Via aspect and result',
Proceedings of the
17th Annual Meeting of the Berkeley Linguistics Society
, Berkeley:
Berkeley Linguistics Society. 480-520.
Theakston, A.L., E.V.M. Lieven, J.M. Pine and C.F. Rowland 2005. 'The
acquisition of auxiliary syntax: BE and HAVE',
Cognitive Linguistics
, 16-
1, 247-277.
Tognini-Bonelli, E. 2001.
Corpus linguistics at work
(Studies in corpus
linguistics 6). Amsterdam/Philadelphia: Benjamins.
Tomasello, M. 1992.
First verbs: A case study of early grammatical development.
Cambridge: Cambridge University Press.
Wells, G. 1981.
Learning through interaction: The study of language
Cambridge: Cambridge University Press.
Wells, G 1986.
The meaning makers: Children learning language and using
language to learn.
Sevenoaks, England: Hodder and Stoughton.
I don't know – differences in patterns of collocation and
semantic prosody in phrases of different lengths
Bettina Starcke
University of Trier
Different realisations of a lemma have different collocations (cf. Sinclair 1991 and
Sinclair in Moon 1987: 89). This is also true for the different longer variants of a short
Taking the most frequent 3-gram in the BNC, 'i do n't', as a basis or nucleus
phrase for the analysis, 4- to 6-grams which include 'i do n't' are analysed for their
collocations and semantic prosodies. The results reveal that there are distinct differences
in the usage of the n-grams. While the 3-gram 'i do n't' and the 4-gram 'i do n't know'
collocate with hedging expressions and markers of uncertainty, their literal meaning, not
knowing something, frequently disappears as their core meaning. In contrast to that, the 5-
gram 'i do n't know what' and the 6-gram 'i do n't know what you' are mostly used in their
literal sense which is negating knowledge of something. Unlike the 3- and the 4-grams, the
longer phrases also have a distinct semantic prosody, namely that of anger, aggression,
despair and frustration. A second, briefer study of phrases containing 'the end of' toward
the end of the article will support the hypothesis that phrases of different lengths but with a
shared nucleus phrase have different collocational patterns and distinct semantic
In the second part of this article, explanations for these differences in collocations
and semantic prosodies are offered. It is suggested that the longer a phrase, the more
predetermined is its use. This is because semantic, pragmatic and syntactical restraints
increase with the length of a phrase. Finally, implications of these findings for corpus
linguistics in general are discussed.
Language is frequently understood as a system of words which are formed into
strings on the basis of grammatical rules. Words form into phrases, sentences,
paragraphs and texts. The main unit of meaning and therefore also linguistic
analysis has traditionally been the word. Only fairly recently have phrases
become a focus in the analysis of language when linguists have recognised that
they too carry meaning in language.
Since this finding has been relatively recent, basic concepts of the analysis
of language have so far mainly been applied to the study of words and not to the
study of phrases and phrasal meaning. This is for example the case with the
concepts of collocation and semantic prosody.
Collocation and phraseology
The concept of collocation is one of the most essential in corpus linguistics. Firth
discusses it as early as 1951 and later on describes it by saying that "you shall
know a word by the company it keeps" (1958: 179). Leech (1974: 20) takes up
the idea and introduces the concept of
collocative meaning [which] consists of the associations a word
acquires on account of the meanings of words which tend to occur in
its environment. (...) collocative meaning is simply an idiosyncratic
property of individual words.
He then goes on to classify collocative meaning as an associative meaning and
defines it as "What is communicated through association with words which tend
to occur in the environment of another word" (26). Since it is an associative
meaning, it is individual to the speakers of a language depending on whether they
recognise this association or not.
The introduction of corpus linguistic techniques in the analysis of
language has changed this perception of collocation as individual to the different
speakers back to Firth's original notion. Collocation is now defined as the habitual
co-occurrence of words as observable in a corpus. It is not intuitive anymore and
the concept is firmly based on empirical evidence that was not available to Firth
and Leech at the time when they wrote their definitions.
But not only are collocations intersubjective, empirical research has also
revealed that different realisations of one lemma have different collocations.
Sinclair (1991) first demonstrated that with an analysis of the different forms of
YIELD in his discussion of how the lemma should be represented in a dictionary.
What looked like a fairly easy concept – a lemma – turned out to be highly
Partington (2004) confirms that different forms of one lemma, in his case
study HAPPEN, occur in different contexts. While
occurs in
predominantly neutral contexts in his corpus of academic writings,
mostly occurs in a negative context, and
occurs twice as much in negative
or neutral contexts as in positive ones. He also finds that this, as he calls it by
referring to Hoey (2004), semantic priming is "realised within and through
separate and typical phraseologies, characteristic syntactic patterns" (141).
While the differences in collocational patterning between different forms
of lemmata have been discussed in depth, the application of this finding to
phrases of various lengths has not been analysed. Phraseological research has
frequently focussed on multi-word units as units of meaning (for example Sinclair
1996 and Stubbs 2005). Stubbs (2005) approaches this question by asking why
the node
is among the most frequent words in the BNC. He finds that
it is
often part of a longer phrase such as
best in the world
which is extremely frequent
in the language. It is therefore not the single word but the phrasal construction
which is frequent. Consequently, phrases are seen to be the units of meaning in
language and not single words. Words cannot be seen in isolation but have to be
looked at in their contexts.
Differences in patterns of collocation and semantic prosody
This ties in with one of Stubbs' earlier findings (2001), namely that there is
a relationship between the semantic prosody of a word and its syntactic
surroundings. For example ACCOST has a predominantly negative semantic
prosody and consequently frequently occurs with passive constructions in which
someone is accosted by somebody else. It seems unlikely that somebody would
describe one's own behaviour as accosting someone.
It is again Partington (2004) who argues in a similar way. In his analysis
of semantic prosody and semantic preferences in language, he finds that "certain
prosodies/preferences are typically expressed using certain phraseologies" (144).
This means that phrases instead of single words carry prosodic and connotative
meaning, and adds further evidence for Sinclair's proposition (1996: 30) that
"units of meaning are expected to be largely phrasal". Sinclair gives evidence for
this claim by discussing the phrasal meanings of e.g. "naked eye" and "true
feelings" which he both finds to be "the core of a compound lexical item" (21)
with inherent components such as semantic prosody and semantic preference. The
phrases must therefore have inherent meanings.
A second major area of phraseological research has been to look at phrases
as characteristic features of text-types. Studies by for instance Stubbs & Barth
(2003), Cortes (2002) and Biber & Conrad (1999) have revealed that different
phrases are characteristic for different genres. The latter two articles discuss
academic writing by undergraduate students and find that much of it is patterned.
Stubbs & Barth (2003) show that the most frequent lexis and phrases differ
between genres. In this article, I discuss how situation specific frequent phrases
Semantic prosody
The concept of semantic prosody describes the characteristics of a word in terms
of its semantic context. This context has implications for the meaning of a word
since the prosody becomes part of the word-meaning. The term and the concept
of semantic prosody have been introduced by Louw (1993) who defines prosodies
as "reflections of either pejorative or ameliorative [semantic] changes [over a
period of time]" (169) and says that "prosodies based on frequent forms can
bifurcate into 'good' and 'bad' " (171). This means that prosodies of words have
developed diachronically and that words can adopt an either positive or negative
connotation through their contexts.
The unit of analysis of semantic prosodies has mostly been the traditional
unit of meaning – the word. For example Sinclair (1991) analyses the phrasal
set in
and discovers that it mostly occurs in a negative context. Looking at
this finding in retrospect, we can now say that it has a negative semantic prosody.
The analysis
The analysis of semantic prosodies and collocations has, as shown above, mostly
focussed on single words or phrasal verbs. But also phrases of different lengths
which share a core part differ in their collocations and semantic prosodies. This
will be demonstrated in the following analysis.
To do so, I will firstly analyse the semantic prosodies and connotations of
four related phrases of different lengths containing
i do n't
on the basis of
concordance lines. I will demonstrate that their prosodies are distinct and display
a progression from an expression of uncertainty and of hedging to a distinctly
unpleasant semantic prosody of aggression or defiance as a response to an
aggression. Following this analysis, I will give some tentative explanations for
the variation in semantic prosody and connotation with phrases of different
lengths. I will show that semantic, pragmatic and syntactic constrains on the
phrases increase with their lengths. My explanations will be supported by further
analyses of phrases containing
the end of
which tie in with my original findings.
The terminology
By the term
, I understand the core part of a phrase, for example
i do n't
is that part of the phrase which is added to the nucleus in longer
realisations of it, for example
i do n't know
i do n't know what you
. While the
nucleus is the core of the phrase and contains its core meaning, the extension
determines the contextual meaning and the collocational patterning of the phrase.
A phrase, that is an uninterrupted string of n words, is called n-gram.
The data
The data for the analysis is the British National Corpus (BNC) and some of its
most frequent phrases of different lengths that share a nucleus. The phrases for
my primary analysis are
i do n't
i do n't know
i do n't know what
i do n't know what you
(the spacing between do and n't occurs in the original data from the BNC and will
therefore be retained in the entire article, see the Appendix for the frequency lists
of multi-word units in the BNC). The findings from this analysis will be
supported by an analysis of a second, much briefer, analysis of the phrases
the end of
the end of the
at the end of the
at the end of the day
These phrases are also among the most frequent phrases of the BNC of their
respective lengths.
The data for the analyses have been extracted from the BNC by
disregarding the POS markup of the corpus. While the knowledge about parts of
speech (POS) is interesting and important for much research, it is of no
Differences in patterns of collocation and semantic prosody
consequence for the present one. My aim is to look at the most frequent phrases
in the language in absolute terms and independent of grammatical categories.
The phrases I look at are all among the 20 most frequent phrases of their
respective lengths (see table 1). They share a nucleus 3-gram as their basis which
provides adequate data for a comparison of semantic prosodies and collocations
of phrases of different lengths.
I will proceed by firstly analysing concordance lines of the different
phrases and by secondly comparing my findings for the phrases. The concordance
lines have been extracted from the BNC by using the database
Phrases in English
(PIE, Fletcher 2003/04) which displays randomly selected concordance lines for
every query. This is the textual evidence for my findings. Concordance lines
given as evidence for my claims have been chosen so that no concordance line
occurs as evidence for phrases of different lengths.
First findings
When looking at the 20 most frequent 3- to 6-grams of the BNC, one thing to
notice is that only a limited number of phrases occurs on the lists (see the
Appendix for the complete lists). Longer phrases frequently consist of one
nucleus with various extensions of different lengths. This is for example the case
the end of, the end of the, by the end of the, at the end of the day
by the
end of the year
. While the nucleus remains constant, the extensions vary with the
length of the phrase.
The nucleus
i do n't
occurs in the following realisations of the top 20 3- to
6-grams in the BNC:
Table 1: phrasal realisations of
i do n't

occurring mostly in
written or spoken
Position on the
n-gram list
i do n't
i do n't know
i do n't think
i do n't want to
i do n't know what
i do n
t think i
i do n't know how
i do n't think it
i do n't know what you
It is striking that all of the phrases above occur mostly in spoken language as
represented by the BNC while most of the other phrases among the top 20 which
are not listed here mostly occur in written language. The occurrence of the short
do n't
it 's
hints at their use in spoken language.
The frequent occurrence of
within the most frequent phrases indicates
that negations are particularly formulaic in spoken language while written
language seems to be more formulaic in a wider range of aspects. Since,
according to Watt (1960), there are no negatives in nature but only in the human
mind, this finding suggests that expressing one's disappointment of positive
expectations is frequent in spoken language. The fact that the negation mostly
refers to mental processes (
) indicates that these are dominant in
people's discourse.
The data selected for the primary analysis of this article are the various
realisations of
i do n't
in connection with
. Findings for phrases which
in their extensions will be discussed with the implications of the
The analysis
As already mentioned, the different realisations of
i do n't
have distinct
semantic prosodies and are increasingly restricted to being used in particular
kinds of situations the longer the phrases are. They are not necessarily text-
specific but rather specific to situations of usage. In addition, the phrase becomes
increasingly delexicalised the longer it is. While the 3-gram is mostly used in its
literal sense of negating something, speakers of the longer phrases frequently use
them to express indignation, anger or frustration. The semantic and pragmatic
contexts of the phrases differ between phrases of different lengths.
As a first step, I will analyse the connotations and semantic prosodies of
each phrase. Secondly, I will compare my findings for the four phrases and will,
thirdly, give some tentative explanations for the differences in usage found
between the phrases. This explanation suggests that variation in semantic prosody
and connotation between related phrases of different lengths is a pervasive
phraseological phenomenon.
i do n't
The concordance lines of
i do n't
show that the phrase is mainly used in its literal
sense. It occurs in constructions which literally describe something that the
speaker does not, for instance, know, agree to or think. It frequently collocates
with mental verbs, hedging expressions and expressions of uncertainty. This is
exemplified by the following selection of concordance lines:
Differences in patterns of collocation and semantic prosody
"I wanted to kill her.
I don't
know what stopped me.
"Can't do any harm. But
I don't
think it would do any good.”
It was Keith who spoke first, looking red-faced and embarrassed."
want to sound awkward, but -- you said Free People were
devious. Don't you think it might be a hoax?
Don't you have any feelings for what I'm goin' through?” "Waal,
exactly feel too good myself,” he replied morosely.
Fiona Fullerton
I don't
think I was told a great deal about the
facts of life. My mother was a Calvinistic Methodist, so you
can draw your own conclusions.
And why are we telling them this if it's not eligible for
consideration as a possible way of depriving us of an cash?
I mean probably it's just to be to the safe side. Mm.
Doesn't mention her mother, and Leo's whole thesis about was built
on this single phrase in Who's Who. words, some future
researched, that the printer here, missed that bit, you know,
she should have said, and her mother's name, but her mother's
got missed out on the proofs or somethin,
I don't
, this is the
kind of thing that happens, of course. Leo's entire book will
collapse, er, as, as, as perhaps it should.
The analysis of the concordance lines does not tell us anything unexpected. The
phrase is used as a negation and frequently occurs in spoken language. What is
intuitively perhaps not quite expected is the frequent collocation of the phrase
with mental verbs and expressions which denote uncertainty in this particular
context, such as
. They function as hedging expressions and the
negation inherent in the sentence appears tentative. The occurrence of mental
verbs as collocates of the phrase hints at mental processes being frequent topics in
The 3-gram frequently collocates with unpleasant notions, such as a
project collapsing or wanting to kill somebody. It therefore has a semantic
prosody of unpleasantness. This is not entirely surprising with a phrase
containing a grammatical negation, but the fierceness of this unpleasantness could
not be predicted from the grammatical negative alone.
i do n't know
On the first glance, there are again no surprises in the analysis of
i do n't know
following the analysis of the 3-gram. The phrase
includes a mental verb, a
semantic class which has been identified as a frequent collocate of the 3-gram,
and the phrase collocates with hedging expressions so that it appears tentative in
its negation. But unlike the 3-gram,
i do n't know
is frequently delexicalised in its
use and fulfils the function of a filling phrase or a hedging expression itself. It is
used to soften a statement and to hedge it when its literal meaning, not knowing
something, can be inferred from the context. The explicit acknowledgement of
not knowing is only a secondary message of the 4-gram. This is exemplified by
the following concordance lines:
But I would have thought, I mean it
I don't know
if it was No, no
What actually happened there? Well
I don't know
really. All I know
is the reports which the lads filed.
Oh what's the time then? Oh
I don't know
, I don't know Nearly
eleven, this stop
When is it?
I don't know
, not very long. I think so I think we'd
said we'd go.
I don't know
, but I know he wanted it. about the turbo or
It seems so recent. But
I don't know
whether that policy still
prevails but Erm I I think does but er I don't think it er
became quite the successful initiative that er that they hoped
it would be.
I don't know
. I didn't know that, I thought I'd put down what,
some of the people one of the babysit for as one of my
references because I babysit her kids at primary school age.
These concordance lines show that the phrase does not express a definite negative
of a fact or process in all of its occurrences. It is instead used as a hedger or to
conceal uncertainty. The collocation of the phrase with softening expressions
softens the negation
. While the negation of a state or process is one function
of the phrase, the second function is to express uncertainty and to weaken the
message of the surrounding sentence(s).
i do n't know what
While there are similarities in the usage of the 3- and the 4-grams,
i do n't know
shows distinct differences in its connotations and collocations. The 5-gram
is mainly used in its literal sense of not knowing and the connotations of the
phrase include anguish, frustration, anger or failing to understand something but
thinking of its consequences as negative. One distinct connotation is that of
gratefulness for an action or service which prevented something negative
happening to the speaker. This is observable in the following concordance lines:
Make a copy for me while you're at it.” "
I don't know what
it will
do to her when she hears that they're the wrong bodies.” "Does
she have to be told?”
He thought he'd got a job for life when he got his old mate Humphrey
in as master -- they were at school together, you know -- but
all that's backfired pretty badly.
I don't know what
they fell
out over, but it must have been serious.
Eat it while it's hot.” "
I don't know what
I'd have done without you
these last weeks, Carrie. You've been a God-send, and no
Now look here!
I don't know what
you're implying, but, for your
information, I have no idea what those goons wanted. I've done
nothing to put myself in a position where I have… hit men
coming after me!”
Me sister never showed 'er legs in all 'er life, nor me, neither.
don't know what
girls are comin' to. Ain't it shockin',
mister?” she said to Joe.
Last night, Johanna's father broke down as he revealed his daughter
had mysteriously taken down her 50 Christmas cards from her
bedroom wall only hours before disappearing. "
I don't know
Differences in patterns of collocation and semantic prosody
happened,” said Robert, 40, a self-employed TV repairer
and electrician. "The last time I saw her she was bubbly and
full of life.
Oh be quiet Er
I don't know what
we'd do without you Paul I'm gonna,
I'm gonna Potatoes are nearly ready
The concordance lines show a semantic prosody of conflict and problem which
differs from that of the 3- and 4-grams. The semantic prosody for the shorter
phrases was that of hesitation and uncertainty. While the 5-gram collocates with
conflict, violence or aggressions do not seem to be involved.
i do n't know what you
Again, the collocations of the 6-gram slightly differ from those of the previously
discussed n-grams. While the previous phrases had semantic prosodies and
connotations that were mostly peaceful, the connotation of the 6-gram is often
that of either aggression or defiance with the literal meaning of not knowing
prevailing. Despair or frustration do not anymore collocate with the phrase. This
is exemplified by the following concordance lines:
The question made her flinch. "
I don't know what you
mean.” "I mean
-- have you been avoiding me like the very plague simply
because of who I am?”
Sweat from the washing-up misted her forehead and nose.
I don't know
what you
're talking about, Léonie lied: I don't know what's
the matter with you. Thérèse clasped the biscuit tin in the
crook of her arm.
"So you can't tell me much?” "
I don't know what you
're after,” she
said. "I don't reckon you know yourself.
He hadn't backed down on that. Her throat constricted and she
swallowed hard before stammering, "I --
I don't know what you
mean. I'm not acting.”
She'd never felt so weak, so helpless. "
I don't know what you
Ruth said huskily. "Let me tell you.”
Court had decided to bluff. He said, "
I don't know what you
talking about. What's more, you now that I don't.
Surely we should try and see
I don't know what you
're talking about,
we of course we're interested in peace, we want peace. We
The semantic prosody of the phrase is, in a large number of occurrences, either
explicitly or implicitly that of open and strong conflict. Usually it is the person in
the weaker position, that is the person being accused of something, who uses the
phrase. It therefore takes on the characteristics of a defence against verbal
aggression. It is frequently used as a response to a statement, often an accusation
or unwelcome question, and the statement either preceding or following the
utterance frequently includes an aggression against the speaker of the phrase. The
phrase itself functions as a defence against this aggression. Also a gender bias in
the use of the phrase is observable: it seems to be used by women mostly. This
can either hint at a gender bias in terms of who sender and receiver of aggression
in society are or at a female communicative strategy.
An attempt at explanation and systematisation
The analysis above has shown that collocations and semantic prosodies of the
phrases seem to fall into two groups. The 3- and the 4-grams
i do n't
i do n't
frequently function as hedges and softeners in discourse. They express
uncertainty but without having distinct semantic prosodies. The 5- and 6-grams
do n't know what
i do n't know what you
frequently collocate with
expressions of aggression, despair or frustration. Their semantic prosody is that of
conflict. The two groups of n-grams seem to differ significantly in their functions.
But on closer analysis, a gradation rather than a division into two groups
of collocations and semantic prosodies of the four phrases becomes visible. The
gradation of usage ranges from a literal use of the 3-gram to an increasing
restriction of context and semantic prosody of the phrases. The 3-gram is often
used in its literal sense to negate something and frequently collocates with
hedging and softening expressions. The 4-gram is already more restricted in its
usage: negating knowledge of something becomes secondary to hedging a
statement and to softening its propositional force. The phrase is partly
delexicalized and its semantic prosody is more restricted than that of the 3-gram.
This development away from the literal meaning of the phrase is reversed
with the 5-gram:
i do n't know what
is again used mainly in its literal sense of not
knowing something – even though it is not always visible whether the speaker
really does not know or whether he pretends not to – , but it is used in different
kinds of situations than the 3- and 4-grams. While the latter frequently occur in
contexts where the speaker aims at avoiding conflict, the 5-gram occurs in
contexts where conflict has already arisen. The function of the phrase appears to
be to deescalate conflict.
Also the 6-gram
i do n't know what you
is mainly used in its literal sense
and it also collocates with expressions of conflict such as
Shut up, bloody, flinch
. Again, conflict is the dominant semantic prosody of the phrase.
Compared to the 5-gram, there is an increase in fierceness in tone of the
utterances in which the phrase occurs. Its frequent use as a defence against
aggression distinguishes it from the 5-gram.
An increase in length of the phrase leads to an increasingly fixed semantic
context in which the phrase occurs. While the 3- and the 4-grams do not have
distinct semantic prosodies, the 5- and the 6-grams do. This hints at a relationship
between the length of a phrase and possible restrictions in its usage with
restrictions increasing with its length.
Also the functions of the phrases become more distinct with an increase in
length. While the 3- and the 4-grams are mostly used in their literal senses as
responses to queries, the 5- and 6-grams are mostly used as responses to verbal
aggression. They are used as a defence against accusations and, by not giving in
to these accusations, the phrases function as markers of hidden aggression or
defiance on the part of their speakers. This becomes more pronounced the longer
the phrases are. The functions of the shorter phrases are included in those of the
longer ones even though they occur in increasingly restricted contexts. The longer
Differences in patterns of collocation and semantic prosody
phrases therefore have two pragmatic functions: They firstly function as a
defence against aggression or as an expression of ignorance on being faced with
aggression. Secondly, they fulfil a hedging function in so far as not the actual
content of the utterance is of primary importance but rather the expression of
These findings entail that the longer phrases are, the more situation-
specific they become. Therefore also the semantic and pragmatic contexts of the
phrases become increasingly restricted and predetermined. This results in a
distinctly negative semantic prosody of the 5- and the 6-grams.
This gradual change in connotation and the emergence of a semantic
prosody with the longer phrases is due to the length of the phrases and the
resulting increase in semantic, pragmatic and possibly syntactic restrictions in
their use: the longer a phrase, the more specific its function. The shorter a phrase,
the greater the possible variation of language co-occurring with the phrase. A
wider range of semantic, pragmatic and possibly syntactic variation may occur
with short phrases which can be used in a larger number of contexts. Conversely,
also the language co-occurring with the phrases becomes increasingly restricted
the longer the phrase is. Restrictions on the semantic context of the phrases
increase with their lengths. This results in pragmatic restrictions which are
mirrored in the semantic prosody of the phrases. This indicates that phrases do
not become increasingly text specific the longer they are, but rather situation- or
Further evidence: the case of
the end of
A second analysis of phrases provides further evidence for the claims made
above. The data for this analysis are the following phrases:
the end of
the end of the
at the end of the
at the end of the day.
The phrases all range among the top 15 phrases of the BNC of their respective
lengths disregarding the POS markup of the corpus.
The nucleus
the end of
occurs in the following realisations among the top
20 n-grams of the BNC:
Table 2: phrasal realisations of
the end of
Position on the respective n-gram list
the end of
the end of the
at the end of the
at the end of the day
by the end of
by the end of the
by the end of the year
In the following section of this paper, I am going to discuss the four phrases
containing the nucleus
the end of
either preceded by
or with
as part of the
extension, but excluding those phrases in which
precedes the nucleus. The
selected phrases are all among the three most frequent phrases of their respective
lengths in the BNC.
the end of
The dominant use of the phrase is mostly to express its literal meaning –
indicating a final state –, with the temporal dimension being dominant. This can
be seen for example in the following sample of concordance lines:
Members often meet up to carry out homework tasks together, and
contact each other between sessions. At
the end of
course, members are invited to exchange telephone numbers and
addresses, and small self-help groups often develop.
On one such incident I was in command of Venturous patrolling in the
Straits of Dover
at the end of
a very busy Bank Holiday,
during which we had followed a suspect yacht from just outside
Calais, and handed her over to our special unit in Dover.
And there's the target for go to go for. It's not necessarily erm a
target if you don't get it hard luck er that's
the end of
s your time with us, it's just a target that we would like we
know that we're gonna clear all our costs out of that first
Swayed by the prospects for economic revitalisation, governor Weld
gave MASSMoCA another$688,000 and until
the end of
1992 to
raise $12 million as proof of private-sector support. The
campaign fell $8 million short.
Non-literal usage is infrequent in realisations of the phrase that do not continue
with the extensions discussed in the following.
the end of the
Also the 4-gram is mostly used to express its literal meaning, again with a
predominantly temporal dimension. Evidence for this are the following
concordance lines:
Differences in patterns of collocation and semantic prosody
The EC Commission said that much the same in its submission to the
Energy Committee: "At the present time the FBR is the only
reactor type which could, if introduced early enough, extend
the lifetime of our uranium resources to
the end of the
century -- and beyond.”
My preconceived ideas about this course, which was held at
Manchester University, were completely erased by
the end of
first evening.
In fact, during the Hundred Years War (1337 to 1453), the people of
Bordeaux took the English side, and many vineyards were
destroyed in revenge. It was not until
the end of the
century that the first bottle of claret as we now know it was
put down for ageing at the famous Chateau Lafite in 1797.
So the patterns of moral respectability, far from being a simple
assimilation of the middle-class norm, were effects of
specific class experiences and a growing sense of class
identity. There were even signs, by
the end of the
century, of increased intermarriage between the skilled worker
and other strata of the working population, a sure indication
of a diminishing sense of social distance.
While the temporal bias in the usage of the phrase is visible, it is so to a lesser
degree than the bias of the 3-gram. A tendency toward a more metaphorical, that
is delexicalized, usage of the phrase becomes visible.
at the end of the
The 5-gram has got three main tendencies of usage. It firstly and primarily
indicates temporal finality and therefore carries its literal meaning. This tendency
is less dominant than with the 3- and 4-grams though. It secondly collocates with
spatial expressions. This is a move away from the mostly temporal dimension
dominant with the shorter phrases.
The third tendency is a more metaphorical usage of the phrase as part of a
fixed phrase. While most of the phrases in which this metaphorical usage is
apparent are part of the 6-gram
at the end of the day
, the metaphorical usage also
occurs when the phrase continues differently.
The following concordance lines give evidence for the three types of
However, they have one disadvantage.
At the end of the
season their
leaves are frequently dulled and disfigured by powdery mildew.
At the end of the
Mass there is a touching prayer as men go out into
the world linking the sacrament of the one bread which binds
all men in God with the bread and ale of human meeting:
While HRP can be claimed by both sexes, it predictably applies more
frequently to women. For more information, see "Pensions for
at the end of the
chapter or obtain leaflet NP 27
Looking After Someone at Home? How to Protect Your Pension
from your local Social Security office.
Gary Moore uses an idea similar to this
at the end of the
verse of the track Story Of The Blues, from his "After Hours”
Another common argument is to point out that everything in the world
must have a cause, but that
at the end of the
line there must
be an uncaused or "first” cause.
The collocations between the 5-gram and temporal and spatial expressions are the
most frequent in the sample of concordance lines from the BNC with the
temporal expressions dominating.
at the end of the day
The dominant usage of the 6-gram differs considerably from that of the 3- to 5-
grams. The phrase is predominantly used in a metaphorical context indicating the
end of something. While the temporal dimension observable with the other
phrases is still present, the main usage has shifted away from a concrete point in
time toward an undefined period or occasion. Evidence for this are the following
concordance lines:
When a bank creates a loan in a multi-bank system, the customers may
write cheques in favour of customers of other banks and
at the
end of the day
, all the banks have claims against each other.
If we are expecting a good level of practice from proprietors,
whether private proprietors or statutory, then we have not to
expect them to be out of pocket
at the end of the day
as a
It's not maybe as rare as you ought to have it but it really tastes
nice. With disasters and all it's come out reasonably well
the end of the day
The old prejudices still remain though, and
at the end of the day
far too few people really applauded this season's winner.
The phrase
at the end of the day
is non-compositional and its meaning cannot be
deduced from the single words. Its idiomatic character is clearly reflected in the
concordance lines above which show how the temporal core meaning of the
nucleus phrase is still part of the 6-gram and its idiomatic meaning. The
concordance lines also show a negative semantic prosody with negatively
connotated words frequently co-occurring with the phrase.
Systematisation of the findings on
the end of

and connections to
i do n't
The analysis of the four phrases above has shown a similar pattern to that of the
phrases containing the nucleus
i do n't
. In both cases, the analysis of concordance
lines has shown a pattern of progression of collocational meaning. Both sets of
phrases progress from a literal meaning of the shorter phrases toward non-
literalness with the longer phrases. The core meanings of the phrases remain part
of the longer phrases.
The semantic prosody of the phrases containing
the end of
develops from neutral – stating a literal end of something – toward a negative
Differences in patterns of collocation and semantic prosody
prosody. The 6-gram collocates with words and expressions such as
impossible, too few
. The phrase is used when describing future
consequences of actions or processes, and it is frequently used to discuss
insecurity, for example whether pupils will be able to find a job after leaving
school or whether certain security measures are enough to protect a building. In
these cases, the phrases express insecurity and hint at possible negative events.
The semantic prosody is predominantly negative.
This negative semantic prosody stands in contrast to the neutral prosody of
the shorter phrases. Both the 3- and the 4-grams mostly occur in their literal
senses and no prosody, either positive or negative, is perceptible. Concordance
lines of the 5-gram reveal instances of both neutral and negative semantic
prosody. The 3- to 5-grams do not have distinct prosodies.
This finding shows that the phrases containing
the end of
are divided into
two in terms of their semantic prosodies. The 3-and the 4-grams both have neutral
prosodies, the 6-gram has a negative prosody. The 5-gram takes an intermediate
position with no distinct prosody.
The distinction of the phrases into two groups in terms of their semantic
prosodies resembles that of the phrases containing
i do n't
. Both groups of phrases
have two groups of prosodies, namely a neutral one with the 3- and 4-grams and a
negative one with the 5- and 6-grams in the one case and the 6-grams in the other.
While the 5-grams do not have similar prosodies, they resemble each other in so
far as they both differ from the shorter phrases and resemble the 6-grams. We can
therefore talk about two groups of semantic prosodies.
Implications and conclusion
The findings from these analyses have several implications for corpus linguistic
First, frequent phrases of different lengths might derive from the same
shorter phrases. This is further evidence for the dominance of patterns in
Second, phrases function as units of meaning in language. The analyses
have revealed that phrases frequently have distinct semantic prosodies or fulfil
distinct functions in discourse. A phrase might, for instance, function as a
hedging expression.
Third, phrases of different lengths have different semantic prosodies or
fulfil different functions even though they are realisations of the same shorter
phrase. This means that a phrase does not have one meaning but that its meaning
depends on its realisation, that is its length. This in turn has implications for
phraseology in general since it attaches greater importance to the choice of
phrase-length for an analysis than has been done so far. The choice of phrase
length might predetermine the results obtained in an analysis and a different
choice might have generated different results.
Fourth, the analyses have shown that not all phrases have semantic
prosodies. In this paper, only the longer phrases, that is 5- and 6-grams, have
distinct semantic prosodies. This is independent of whether the phrase is
compositional as
i do n't know what you
or non-compositional as
at the end of the
. This shows that non-compositional phrases pattern in the same way as
compositional phrases.
Fifth, there appears to be a structural similarity between words and phrases
since both pattern differently with different realisations of the original unit (word
or phrase). While different realisations of a lemma have different collocations,
phrases of different lengths have different semantic prosodies and fulfil different
discourse functions. Similarities of prosody and discourse function seem to
depend on the length of the phrase.
I am grateful to William Fletcher who provided me with lists of the most frequent
1- to 8-grams in the
disregarding the POS markup.
Biber, D. & S. Conrad 1999. 'Lexical bundles in conversation and academic
prose', in: Hilde Hasselgard (ed.)
Out of corpora. Studies in honour of Stig
Amsterdam: Rodopi. 181-90.
Cortes, V. 2002. 'Lexical bundles in Freshman composition', in: R. Reppen, S. M.
Fitzmaurice, D. Biber (eds.)
Using corpora to explore linguistic variation.
Amsterdam, Philadelphia: John Benjamins. 131-145.
Firth, J.R. 1958. 'A synopsis of linguistic theory 1930-1955',
Studies in linguistic
. 1-32
Firth, J.R. 1951. 'Modes of meaning'
in: J. R. Firth (ed.)
Papers in linguistics
London: OUP. 1957. 190-215.
Fletcher, W. H. 2003/04.
Phrases in English
. database. 5
January 2007
Hoey, M. 2004. 'Lexical Priming and the Properties of Text', in: A. Partington, J.
Morley, L. Haarman (eds.)
Corpora and discourse
. Bern: Peter Lang. 385-
Leech, G. 1974.
. Harmondsworth: Penguin.
Louw, B. 1993. 'Irony in the text or insincerity in the writer? The diagnostic
potential of semantic prosodies', in: M. Baker, G. Francis, E. Tognini-
Bonelli (eds.)
Text and technology. In honour of John Sinclair
Philadelphia, Amsterdam: John Benjamins. 157-176.
Moon, R. 1987. 'The analysis of meaning', in: J. Sinclair (ed.)
Looking up. An
account of the COBUILD project
. London, Glasgow: Collins ELT. 86-
Differences in patterns of collocation and semantic prosody
Partington, A. 2004. ' "Utterly content in each other's company". Semantic
prosody and semantic preference',
International journal of corpus
. 9 (1): 131-156.
Sinclair, J. 1996. 'The search for units of meaning', in: G. Corpas Pastor (ed.)
lenguas de Europa: Estudios de fraseología, fraseografía y traducción
Granada: Comares. 2000. 7-37.
Sinclair, J. 1991.
Corpus, concordance, collocation
. Oxford: OUP.
Stubbs, M. 2001.
Words and phrases
. Oxford: Blackwell.
Stubbs, M. 2005. 'The most natural thing in the world: quantitative data on multi-
word sequences in English', Conference presentation at
Phraseology 2005,
Stubbs, M. & I. Barth 2003. 'Using recurrent phrases as text-type discriminators:
a quantitative method and some findings',
Functions of language.
10 (1):
Watt, I. 1960. 'The first paragraph of
The Ambassadors
: an explanation',
in criticism.
10. 250-74.
Top 20 3-grams in the BNC
Top 20 4-grams in the BNC
i do n't
one of the
the end of
part of the
do n't know
some of the
a number of
there is a
a lot of
# and #
there was
it's a
be able to
it was a
the fact that
you do n't
to be a
it's not
i do n't know
the end of the
at the end of
for the first time
on the other hand
between # and #
as a result of
the rest of the
in the case of
one of the most
# per_cent of the
the secretary of state
by the end of
from # to #
do n't want to
is one of the
to be able to
i do n't want
Top 20 5-grams in the BNC Top 20 6-grams in the BNC
at the end of the
by the end of the
i do n't want to
i do n't know what
as a result of the
in the middle of the
the secretary of state for
the other side of the
at the time of the
you do n't have to
for the first time in
at the top of the
i do n't think i
at the beginning of the
the end of the year
in the case of the
there are a number of
on the other side of
the end of the day
i do n't know how
at the end of the day
on the other side of the
ask the secretary of state for
to ask the secretary of state
from the point of view of
my hon. Friend the member of
by the end of the year
at the other end of the
i do n't think it 's
in such a way as to
the department of trade and industry
from # per_cent to # per_cent
in the second half of the
in the middle of the night
our next bulletin is at # p.m.
secretary of state for the environment
the secretary of state for the
at the end of the year
i do n't know what you
Stubbing your toe against a hard mass of facts:
corpus data and the phraseology of
Hans Lindquist
University of Växjö
In this paper Fletcher's database Phrases in English is used to extract frequently recurring
n-grams containing the verb 'stub' and the noun 'toe' from The British National Corpus.
After analysing some of these n-grams, the paper focuses on the formulaic sequence 'stub
one's toe' and investigates this in the BNC, The New York Times and The Independent.
Additional searches are also made on the World Wide Web by means of WebCorp. It is
found that the phrase is used with equal frequency in American and British English, but
that the American use differs in that approximately half of the tokens are non-literal, while
such use is relatively rare in British English. It is hypothesized that the non-literal use
originated in American English and that it may be spreading to other varieties.
Somewhere down there he stubbed himself against an ill-defined but
hard mass of fact, and brought it up to the surface to examine it.
(Michael Frayn: Towards the end of the morning, 1967; BNC G12
Stubbs (1996, 2001) pioneered the use of corpora in the study of semantic and
pragmatic meaning. In a number of more recent articles (Stubbs 2002, in press,
forthcoming a, forthcoming b), he has suggested methods for retrieving n-grams
from corpora in order to study frequent collocations and collocations with
frequent words. These methods constitute innovative ways of bringing data, hard
masses of fact, from the depths of large corpora to the surface where they can be
examined and analysed.
The present study will use some of the methodology suggested in these
papers to investigate the phraseological patterns, or, with Wray's (2002)
terminology, the formulaic sequences which form around two lemmas, the fairly
frequent body-part noun
and the rather infrequent verb
. My purpose is
first to describe and analyse the semantic and pragmatic features of recurring
formulaic sequences with these words, and, second, to discuss some theoretical
and methodological consequences of this type of study.
The paper will have the following structure. After a brief section on the
method and material there will be a few paragraphs on the history of the
individual words
. Then the phraseological tendencies of these two
words will be illustrated by a study of their occurrences in formulaic sequences in
the British National Corpus, which leads up to an investigation of the specific
to stub one
s toe
in the BNC,
The New York Times
The Independent
on the World Wide Web (by means of WebCorp). Finally there is a conclusion
and a Coda.
Method and material
The method used has been called "from lexis to n-grams" by Stubbs (forthcoming
a) and is described in some detail in Lindquist and Levin (forthcoming a and b).
Basically it means starting with a particular word or lemma, or a set of words or
lemmas, and investigating which recurring n-grams they occur in. Lists of n-
grams in the British National Corpus with the search word in all possible
positions can easily be extracted by means of William Fletcher's (2003/2004)
Phrases in English
, which includes all n-grams between 2 and 8 words
occurring 3 times or more in the BNC.
At the next stage, these recurring n-grams have to be manually analysed to
judge which may be considered to be formulaic sequences and which may be just
chance occurrences without interior structure and integrity. For instance, in the
BNC, the most frequent 3-gram beginning with
toe of his
(21 occurrences),
but this is not a likely candidate for a formulaic sequence to be stored holistically
in the brain. The second most frequent 3-gram beginning with
, however, is
toe the line
(16) which is clearly a formulaic sequence.
In the present paper, the phraseologies of
were also studied in
two sets of newspapers on CD-ROM: the
New York Times
and the
from 1990, 1995 and 2000. Here, searches for the words were made by means of
the program Wordsmith to create concordances which were then sorted and
analysed manually. Finally, searches were made on the Web through the
mediation of WebCorp (2007). However, as has been shown by e.g. Mair (2006)
and Lüdeling et al. (2007), mining the rich resources of the Web is complicated
due to a number of technical obstacles caused by the conflict between the
information the linguist wants to extract through for instance WebCorp and the
information that search engines like Google provide. Furthermore, due to various
technical limitations, WebCorp at present returns many fewer hits than direct
searches through Google. It is not the aim of the present paper to specifically
evaluate WebCorp or Google data in comparison with data retrieved from
traditional, tidy corpora or text archives like newspaper CD-ROMs. The Web
data has however been added as a complement to the findings based on the other
corpora. As has been pointed out by Mair (2006: 370), "corpus linguists of the
future will […] [be] working in a vast and expanding corpus-linguistic working
environment in which one of the chief skills required will be to identify the
resources which are relevant to the problem studied from a vast range of
possibilities". Comparing data from different corpora will often be a necessity (cf.
Lindquist and Levin 2000).
Stubbing your toe against a hard mass of facts
The verb
and the noun
The first record of
in the
is from 967, meaning "A stump of a tree or,
more rarely, of a shrub or smaller plant; the portion left fixed in the ground when
a tree has been felled […]" (
, n. 1. a.). As a transitive verb it occurs
c1400 with the meaning "To dig up by the roots; to grub up (roots)" (
, v
, n. 1. a.) and by 1577 it could mean "To reduce to a stub or stump" (
, v
, n. 6. a.). The first citation in the OED for
stub one
s toe
is from John
R. Bartlett's
Dictionary of Americanisms
1848: "'To stub one's toe', is to strike it
against anything in walking or running; an expression often used by boys and
others who go barefoot" (
, v
, n. 9. a.), and by 1927 another
frequent present-day meaning is recorded: "To extinguish (a cigarette) by
pressing the lighted end of the stub against a hard object. Freq. with
. Also
, v
, n. 12). In terms of semantic prosody (cf. e.g. Sinclair
1991, 2004; Louw 1993; Partington 1998, 2004; Hunston and Thompson 1999;
Levin and Lindquist 2007) it seems that in general
has negative prosody,
probably based on the original nominal meaning of something that is left when
valuable material has been extracted, and strengthened by the associations to
violent actions resulting in this extraction and later to various verbal metaphorical
uses leading to similar results. For some positive examples, however, see the
Coda below.
The earliest citation for the noun
is from c725, with the meaning
"Each of the five digits of the human foot" (
n. 1. a.) Through history
it has occurred in a number of phrases with figurative meaning like
stepping on
s toes
on one
s toes
toe to toe
(for a treatment of this particular
phrase, see Lindquist and Levin forthcoming b),
a toe in the door
to dig in one
and several others. Behind many of these figurative meanings seems to lie
either the balancing and gripping function of the toes or their exposed position at
the outer points of the longest extremities of the human body.
in the British National Corpus
The verb
occurs 100 times in the BNC, making it a fairly rare word with
1.03 occurrences per one million running words (compared for instance with
which occurs 106 times per one million words). Of the 100 verbal
were in the phrasal verb
stub out
, 62 of which referred to the physical putting out
of cigars or cigarettes as in (1), 2 referred metonymically to giving up smoking as
in (2), and one metaphorically to the crushing of an object as in (3). In one
example, from a poem where pigs are uncharitably likened to women at a jumble
stub out
seems to mean 'stick out' (4).
(1) She rolls off the bed and
the fag
. (A74)
(2) Now, though the office air is clean, the butt-crammed ashtray outside
testifies that smoking is far from
. (A4K)
(3) The car was so low it looked as if a giant had tried to
and it was
clear that getting out of the bucket seat gave the Greek momentary altitude
sickness. (FR3)
(4) Floppy hats, high-heeled trotters, massive hams, a double row of buttons
done up neatly, salmon pink on beige – they squeal and
their noses
, flushed and burning with the change of life ... (HRL)
In the remaining 31 examples,
is used about a number of different entities.
The most common is
, with 14 tokens. Most of the instances have concrete
meaning, as in (5), but there were also two where the meaning is figurative, as in
(5) Distracted, Luce
against a piece of raised planking and
tripped. (JY2)
(6) As a prominent figure in Rottweiler rescue, she's
on more
unfair bullying and downright idiocy than most. (C8U)
Other things that were stubbed and somehow damaged or terminated included
its shock absorbers
a blue trace
[on a screen],
miling exploits
another English prejudice
their ego
. In (7), however, the verb
seems to refer to a thrusting manner in which a goal was scored.
(7) After 70 minutes, Paul Robinson, a tall, long-necked forward,
Scarborough's second from close range. (A2S)
The "cigarette meaning" also occurs with
(three examples with
abstract meaning as in (8) and one with figurative meaning, as in (9)).
(8) A ringed hand held a thin cigar which – as if in impatient expectation of
her arrival – he
in a silver tray. (H82)
(9) Putting out the cigarette again, as though
Alice out of existence,
he said […] (EV1)
occurs 1616 times in the BNC, which gives it a frequency of 16.55 per
million words. In Adam Kilgarriff's frequency list based on the BNC ([1995]
1998) it has rank 4253, which can be compared with other body nouns like
(frequency 21,339, rank 484) or
(frequency 53,265, rank 176). Since
more than 15 times more frequent than
, it is not practical to study all the
concordance lines where it occurs with equal attention to detail. Instead, we will
Stubbing your toe against a hard mass of facts
look at the most frequently recurring n-grams with the forms
. As
mentioned in the method section, although the PIE program supplies n-grams up
to 8-grams, very few linguistically significant n-grams of that length recur. At the
5-gram level, however,
toe of his boot
occurs 10 times,
the toe of his shoe
times and
the toe of her shoe
3 times. This means that the most frequent long
formulaic sequences with
are ones where
does not refer to a body part at
all but rather to a part of a piece of footwear. In all, with various possessive
pronouns, there were 15
the toe of X
s boot
and 9
the toe of X
s shoe
. In addition,
there were occasional references to the toes of Doc Martens, wellingtons and flip-
flops. Typical examples are (10) and (11).
(10) Benjamin tapped
the toe of his shoe
on the soft carpet. (HH5)
(11) She raced for the protection of mast and water butt, caught
the toe of her
on a raised nail and went sprawling. (C85)
Another 5-gram with a frequency of 3 is
the toe of the club
, where
refers to a
protruding part of a golf club. Such technical meanings of
are not uncommon.
A toe in the water
also occurs three times, but is better treated as the 4-gram
toe in
the water
, which occurs 8 times. None of those examples refers to human toes
dipped into real water; they are all used figuratively as in (12) – (14).
(12) It's always best to dip a
toe in the water
first, rather than plunging in with
a programme of hopefully helpful ideas for the improvement of her life
and comfort. (C8Y)
(13) […] Cognos is still at the "
toe in the water
" stage with the AS/400
market. (CSH)
(14) But Nordstrom's catalogue is merely a
toe in the water
. (CR8)
Similarly, the 5-grams
covered from head to toe
(4 instances),
dressed from head
to toe
(4) and
clothed from head to toe
(3) are better treated under the 4-gram
from head to toe
(74 instances) and its variant
from top to toe
(19). While all
instances of
head to toe
refer to the human body being covered in clothes or
other material or being injured or treated or scrutinized in its entirety, as in (15) –
from top to toe
is also occasionally used about other concrete objects, as in
(18), or about abstract entities, as in (19).
(15) The likes of Naomi Campbell and Linda Evangelista were clad
from head
to toe
in leather, rubber, latex and PVC. (A7N)
(16) The King's legs were broken, there were injuries
from head to toe
(17) Her glance raked Polly
from head to toe
. (H7W)
(18) But John and Veronica Saunders still make time to decorate their home
from top to toe
. (ED4)
(19) The overwhelming impression left after the debate is of a Tory Party split
from top to toe
over Europe, and a Prime Minister unable to heal the rift.
Moving down to 3-grams, there are two noteworthy phrases:
toe to toe
with 6
tokens and
stubbed his toe
with 5. Counting all variants of the latter like
her toe
etc. the figure goes up to 14; these were treated above under
. Five of
the instances of
toe to toe
are used adverbially to refer to people standing opposite
each other at close range, often literally with touching toes, and exchanging blows
as in (20), while one token refers to the measuring of the distance from the toes of
one foot to the toes of the other. Lindquist and Levin (forthcoming b) also found
a number of cases where
toe to toe
was used non-literally, meaning 'in direct
confrontation' in an abstract sense.
(20) […] we stood
toe to toe
and swapped blow for blow. (H0A)
As has been pointed out by Sinclair (2003: 167–172), singular and plural forms of
nouns often occur in quite different contexts, and indeed the searches for
yielded totally different n-grams. One 6-gram with
occurred 5 times: the
line from the children's rhyme
head and shoulders, knees and toes
, which is
repeated over and over again in one single text. On the 6-gram level we also find
to keep you on your toes
(3), but this is better treated under
on X
s toes
, which has
a frequency of 169 with five main meanings as shown in Table 1.
Table 1:
On X
s toes
in the BNC
On his toes 12 21 1 2 2 38
On their toes 31 4 2
On your toes 9 9 2 1 21
On the toes
1 7 4 3 1 16
On her toes
3 11 1 1 16
On our toes 14 1 15
On its toes
9 3 1
On my toes
6 5 1 12
On yer toes
83 55 12 11 5 3 169
Stubbing your toe against a hard mass of facts
The five meanings 'attention', 'posture', 'embellishment', 'encroaching' and
'perceiving' are illustrated in (21) – (25).
(21) […] just as black people keep changing the name you are allowed to call
them in order to keep whitey
on his toes
. (ECU)
(22) Culley swung at him, coming up
on his toes
for the blow. ((FS8)
(23) On his legs were hose striped with red and gold, while his feet were
hidden in crimson velvet slippers with silver roses
on the toes
. (H9C)
(24) It claims it will stick to commercially-led Unix information not treading
on the toes
of existing non-commercial Unix networks […] (CTJ)
(25) […] come on look
on your toes
now, get up come on […] (KB8)
As Table 1 shows, the 'attention' meaning as in
keep sb on their toes
be on
s toes
is clearly the most common, followed by the 'posture' meaning as in
standing/running/dancing etc. on one
s toes
. The 'embellishment' meaning, where
something is applied or placed on the toes (usually the toes of a particular pair of
shoes, but sometimes on someone's bare feet), comes next, and then the
'encroachment' meaning as in 'treading on someone's toes'. Finally there are a few
cases referring to the fixing of one's gaze etc. on one's toes, and some unclear
A few more things can be said about these findings. First of all, the
distribution of meanings over forms is far from even. The singular
on his toes
her toes
and, to some extent,
on my toes
, are mainly about posture. With this
meaning there is also a gender difference: about men the reference is especially
fighting contexts like in (22) above, while about women it is to romantic contexts
like in (26).
(26) She once again burst into tears and, crossing rapidly to George, threw her
arm around his neck and stretching up
on her toes
, began to kiss him with
a fervour which shocked him. (C98)
The 'attention' meaning predominates clearly with the plural
on their toes
our toes
, and also with
on its toes
. (27) and (28) are typical examples.
On its toes
often refers to collectives that need to be on the alert, as in (29).
(27) "I also carry out random spot checks with a probe at various points during
the week, just to keep everyone
on their toes
," he adds. (HC3)
(28) "A lead of 2–1 guarantees nothing, except the fact that we need to be
our toes
against a side who will relish the big occasion." (CH3)
(29) The competitive situation keeps the governing party
on its toes
sensitive to the public's view of policy.
It is easy to see how the non-literal, metonymic 'attention' and 'encroaching'
meanings have developed from the physical acts of standing on one's toes, ready
to move quickly, on the one hand, and accidentally or intentionally stepping on
someone else's toes, on the other. The 'embellishment' and 'perception' instances
are just cases of fully transparent compositional constructions.
Stubbing one's toe(s) in the BNC, in The New York Times and The
Independent, and on the World Wide Web
In this section we will take a closer look at the construction
stub one
s toe(s)
in a
number of corpora. Table 2 shows the overall distribution. It is hard to make a
comparison between the corpora since the exact number of words is only known
for the BNC. The newspaper figures were arrived at partially by counting,
partially by extrapolation (for the procedure, cf. Lindquist 2007) and as regards
the World Wide Web, its constantly growing size can only be roughly estimated
(Keller and Lepata 2003: 467 and Mair 2006: 365–366 independently arrive at the
ballpark figure 100 billion words for the English part of the Web). The
unexpectedly low figures for WebCorp (accessed 4 March 2007) are due to the
limitation of the present prototype version of 200 accessed webpages (WebCorp
2001–2007: Guide). In addition, the reliability of the search engine figures is not
to be trusted (cf. Lüdeling et al. 2007) and the WebCorp figures are only included
for internal comparison.
Table 2: Frequency of the phrase
to stub one
s toe(s)
against something
Total number
of words
Stub one
s toe Stub one
s toes
BNC 100 M
1 14
NYT 180 M
5 23
IND 115 M
0 20
WebCorp ?
21 74
27 131
One conclusion that can be drawn from Table 2, in spite of the reservations
above, is that phrases with
in the singular are more frequent than phrases with
the plural
. This is largely a function of the fact that a majority of the subjects
are singular. As pointed out by Moon (1998:95), "nothing systematic accounts for
the way in which words denoting parts of the body inflect in some [fixed
expressions and idioms], in accordance with the number of the grammatical
subject or referend, but not in others." In the case of
stub one
s toe
, however, the
is normally used with singular subjects and the plural
with plural
subjects, even if there are exceptions, as in (30) and (31):
Stubbing your toe against a hard mass of facts
(30) Plus for someone like me who is always hiking through streams climbing
boulders and tripping over everything in my path it's a real asset to
discover I can't
stub my toes
in these sandals.
(31) I grew up in a small town in Maryland, and Dr. Roth was our family
doctor, he said. I remember calling him in the middle of the night. We
called him if we
stubbed our toe
. (NYT2000)
Furthermore it seems that the plural form is less current in British English (BNC
and IND) than in American (NYT) and mixed (WebCorp) varieties.
One way in which formulaic sequences increase their frequency is through
achieving non-literal, extended meanings which can be used in more contexts.
This has happened with
stub one
s toe
. Even if one can argue that the borderline
between literal and non-literal is not always crystal clear, and that some vestiges
of the literal meaning often remain in the non-literal uses and can be reactivated
in the mind of the language users, it is usually possible to ascertain from context
with reasonable certainty if a particular token is used (mainly) literally or non-
literally. In Table 3, the distribution between literal and non-literal meaning is
Table 3: Literal and non-literal meaning of the phrase
stub one
s toe(s)
Literal Non-literal Unclear Total
0 14
0 23
0 20
3 74
3 131
The American corpus (NYT
stands out from the two British corpora, BNC and
The Independent
, and the mixed web material. In American English, the meaning
stub one
s toe(s)
is evenly distributed between literal and non-literal, while in
the other corpora there is a clear predominance of literal examples (ratios between
6:1 and 10:1). The non-literal examples can for instance refer to unsuccessful
business encounters, as in (32), political debacles, as in (33), literary fiascos, as in
(34), and, very frequently, defeats in sports, as in (35).
(32) P. & G.
stubbed its toe
pretty badly with Carrefour, the international
retailer, said Mr. Flickinger, managing director of Reach Marketing in
Westport, Conn. (NYT 2000)
(33) Among the democrats who regularly crop up on lists of potential
Democratic candidates, a pair of northeasterners, Gov. Mario M. Cuomo
of New York and Senator Bill Bradley of New Jersey,
stubbed their toes
with less than impressive victories. (NYT1990)
(34) Lanford Wilson, who won the Pulitzer Prize for the romantic "Talley's
Folly" and who has written some of the most poetic and graceful plays in
the contemporary theatre,
stubs his toe
over "Burn This." (NYT1990)
(35) "This is not last year's team," he said. "We're a young team. We
our toes
. We're going through some growing pains but it doesn't excuse
our performance tonight." (NYT 1990)
This sports meaning can be seen as a possible link between the literal and the
non-literal use, since the mishaps of football teams and athletes are to some extent
physical in their nature, or strongly related to the physical, as when players
actually hit their toes against the turf, the goalposts or their opponents, as in (36)
(cf. the phrase
bite the dust
with similar meaning).
(36) Even worse to have someone
stub their toe
on your head – as happens to
Louis Saha, who feels the force of Andre Bikey's boot.
The study of lexical phenomena like collocation and phraseology requires very
large corpora. By means of
Phrases in English (PIE)
(Fletcher 2003/2004) and
the accompanying interface it is possible not only to look for already well-known
phrases found in dictionaries or retrieved from memory, but also for frequently
recurring n-grams of which we are not always consciously aware but which may
nevertheless be stored and retrieved holistically. This paper started out with such
n-grams retrieved from the British National Corpus by means of
, discussed
some of the most frequent types, and then zoomed in on one particular phrase,
stub one
s toe(s)
, which was found to be used both literally and figuratively. This
phrase was then further investigated in other sources of data (
The New York
Times, The Independent
) and on the Web (through WebCorp).
Beginning with the individual verb
in the BNC, it was found that it
was used about cigarettes, cigars et cetera in 69% of the cases and about toes in
14% of the cases. Studying the much more frequent noun
, it turned out that
the frequent 3-gram
occurred with a number of specific meanings,
some of which were literal and some figurative, and furthermore that different
meanings co-varied with different pronouns and the definite article, so that e.g.
his toes
was most likely to be about 'posture', while
on their toes
was most often
Stubbing your toe against a hard mass of facts
used about 'attention' and
on the toes
was used referring to 'embellishment'. It is
thus not enough to describe the meaning of the phrase or frame
on X
s toes
such, since its meaning is influenced by the word which is put into the X slot.
The phrase
stub one
s toe(s)
occurred most frequently in the singular,
which was seen mainly to be a trivial function of the number of the subject. Plural
or collective subjects were rarer in the British corpora, and consequently there
were fewer plural forms of
As for the meaning of the phrase, it was literal in 84% of the tokens in the
total material. In the purely American corpus, however, the meaning was non-
literal in approximately half of the cases (11/23). The cases of non-literal or
figurative meaning in the corpus referred to unlucky incidents in various areas of
human public endeavour: business, politics, literature and especially sports.
The picture provided by these findings is that the phrase
to stub one
, which was first attested in American English, now occurs with equal
frequency in British English, but that that the extended, figurative meaning of
encountering an abstract obstacle of some sort is primarily American and not
frequent in British English at all. This is an illustration of a phenomenon
discussed by Buchstaller (2006) in relation to attitudes towards certain linguistic
features when they are borrowed from one variety to another: the connotations are
not always taken over wholesale. Similarly, the figurative meaning of this phrase
does not seem to have been taken over by many British speakers (as yet).
The motto of this paper was taken from Michael Frayn's comic novel
Towards the
end of the evening
. (The title of that novel is by the way an instance of the string
of the
", which Stubbs (2002:232–235) has shown to be a very
frequent pattern in English; in Stubbs's material,
towards the end of the
is the
11th-most frequent exponent of this structure). At this stage, one may of course
ask oneself if
stub one
s toe
stub oneself
) is really a suitable metaphor for the
pursuit of corpus linguists. After all, as was mentioned in the
above, the semantic prosody of the phrase is almost always negative. However,
there are also some more positive contexts, like in (37) and (38):
(37) The young French explorers literally
stubbed their toes
on treasures
buried in the sand: obelisks and sphinxes; ruined temples at Karnak,
Dendera, and Luxor; broken colossal funeral statues in the Valley of
Kings. (Dora B. Wiener, 'With Bonaparte in Egypt'.
2000, 91:755)
(38) Otters quickly became scarce, but then three guys standing in a Yukon
stubbed their toes
on a thimble's worth of gold.
It is therefore possible to end this paper on a happier note: Wading through sand
dunes of words, standing waist-deep in swirling data, the corpus linguist will ever
so often stub his or her toe(s) on unexpected treasures.
Buchstaller, I. 2006. 'Social stereotypes, personality and regional perception
displaced: Attitudes towards the "new" quotatives in the U.K. ',
Journal of
, 10 (3): 362–381.
Fletcher, W. 2003/4.
PIE: Phrases in English
Frayn, M. [1967] 2005.
Towards the end of the morning
. London: Faber and
Hunston, S. and G. Thompson (eds.) 1999.
Evaluation in text. Authorial stance
and the construction of discourse
. Oxford: Oxford University Press.
Keller, F. and M. Lapata 2003. 'Using the Web to obtain frequencies for unseen
Computational Linguistics
, 29 (3): 459–484.
Kilgarriff, A.
[1995] 1998.
BNC database and word frequency lists
Levin, M. and H. Lindquist 2007. 'Sticking one's nose in the data. Evaluation in
phraseological sequences with
ICAME Journal
31: 63–86.
Lindquist, H. 2007. 'Viewpoint
: The spread and development of a new type
of adverb in American and British English', to appear in
Journal of
English Linguistics
35 (2): 132–156.
Lindquist, H. and M. Levin. 2000. 'Apples and oranges: On comparing data from
different corpora', in C. Mair and M. Hundt (eds.)
linguistics and
linguistic theory
. Amsterdam: Rodopi. 201–213.
Lindquist, H. and M. Levin forthcoming a. 'F
. The phrasal
patterns of two frequent nouns', to appear in S. Granger and F. Meunier
Phraseology: An interdisciplinary perspective
. Amsterdam:
Lindquist, H. and M. Levin forthcoming b. 'The syntactic properties of recurrent
phrases with body part nouns: The N
pattern'. To appear in U.
Römer and R. Schulze (eds.)
Exploring the Lexis-Grammar Interface
Amsterdam: Benjamins.
Louw, B. 1993. 'Irony in the text or insincerity in the writer? The diagnostic
potential of semantic prosodies.', in: M. Baker, G. Francis and E. Tognini-
Bonelli (eds.)
Text and technology: In honour of John Sinclair
Amsterdam: Benjamins. 157–176.
Lüdeling, A., S. Evert and M. Baroni 2007. 'Using web data for linguistic
purposes' in: M. Hundt, N. Nesselhauf and C. Biewer (eds.)
linguistics and the web
. Amsterdam: Rodopi. 7–24.
Mair, C. 2006. 'Tracking ongoing grammatical change and recent diversification
in present-day standard English: The complementary role of small and
large corpora' in: A. Renouf and A. Kehoe (eds.)
The changing face of
corpus linguistics
. Amsterdam: Rodopi. 355–376.
Moon, R. 1998.
Fixed expressions and idioms in English. A corpus-based
. Oxford: Clarendon Press.
Partington, A. 1998.
Patterns and meanings. Using corpora for English language
research and teaching
. Amsterdam: Benjamins.
Stubbing your toe against a hard mass of facts
Partington, A. 2004. '"Utterly content in each other's company". Semantic
prosody and semantic preference.'
International Journal of Corpus
, 9 (1): 131–156.
Sinclair, J. 1991.
Corpus Concordance Collocation
. Oxford: Oxford University
Sinclair, J. 2003.
Reading concordances. An Introduction
. London: Longman.
Sinclair, J. 2004.
Trust the text: language, corpus and discourse
. London:
Stubbs, M. 1996.
Text and corpus analysis
. London: Blackwell.
Stubbs, M. 2001.
Words and phrases
Corpus studies in lexical semantics
London: Blackwell.
Stubbs, M. 2002. 'Two quantitative methods of studying phraseology in English',
International Journal of Corpus Linguistics
. 7 (2): 215–44.
Stubbs, M. 2007. 'An example of frequent English phraseology: Distributions
structures and functions', in: R. Facchinetti (ed.)
Corpus linguistics 25
years on
. Amsterdam: Rodopi. 89-105.
Stubbs, M. forthcoming a. 'Quantitative data on multi-word sequences in English:
The case of the word "world"', to appear in M. Hoey, M. Mahlberg, M.
Stubbs and W. Teubert (eds.)
Text, discourse and corpora
. London:
Stubbs, M. forthcoming b. 'Quantitative data on multi-word sequences in English:
The case of prepositional phrases'. Lecture given at the Berlin-
Brandenburgische Akademie der Wissenschaften, 3 November 2006.
Wray, A. 2002.
Formulaic language and the lexicon
. Cambridge: Cambridge
University Press.
Stringing together a sentence:
linearity and the lexis-syntax interface
Oliver Mason
University of Birmingham
Following existing approaches to linear grammar we explore the application of
automatically identified multi-word units to the analysis of sentence structure. After
looking at several sample sentences we then move on to a discussion of routine use vs.
creativity in language.
The proposed new phraseological grammar does away with both syntactic and
functional categories and reduces syntax to a by-product of a linearising thought in the
form of phraseological units of meaning.
Phraseology is concerned with the study of units above the level of the single
word, which seem to become increasingly important with the widening
application of empirical principles to the field of linguistics. The single word,
while a convenient starting point, is not a suitable entity when describing either
sentence structure or aspects of meaning. A word has no meaning in isolation,
and even its syntactic environment is usually idiosyncratic when we consider
actual use rather than theoretical possibilities.
So far, multi-word units (MWUs) have been identified as units of meaning
(eg Danielsson 2001), as it is only in conjunction with other words that we are
able to decide which aspect of its meaning potential has been realised in a
particular instance. The phrasal environment of a lexical item thus serves as a
shorthand description of its use, if we consider the definition of meaning as use.
Stubbs (2001) gives the examples of
, which despite having
several distinct meanings in isolation cannot ever be confused when used in
authentic sentences. It is, of course, possible to deliberately invent sentences
where their use is ambiguous, but the important issue here is that this is not what
speakers do in real life.
However, MWUs are not only important in semantics, where they displace
the single lexical item as the central element, shifting the focus from lexical
meaning to phrasal meaning, but also in syntax, where they compete with abstract
descriptions ultimately based on phrase structure grammar (Chomsky 1957). As
Stubbs (1993) observes, grammarians are traditionally interested in structures
only, and view the lexical items as mere instantiations of the grammatical
categories which they belong to. More recent approaches (eg Sinclair 1991,
Francis 1991, Brazil 1995, Hunston and Francis 2000, Sinclair and Mauranen
2006), on the other hand, have demonstrated that lexical items are more than that,
and that instead there is a correlation between grammatical structures and the
words which occur in them. As is the case with everything in the description of
language, this correlation is not an absolute, but rather expresses strong
tendencies reinforced by everyday usage.
Some of these alternative approaches furthermore view sentence structure
not as hierarchical (as in analyses derived from phrase structure grammars) but
instead as linear. Such a linear sequence of units (
in the terminology of
Brazil (1995),
in Hunston and Francis (2000), and
in Sinclair and
Mauranen (2006)) is constructed mainly according to the principle of prospection,
where one unit places constraints upon the range of possible successor units.
Open Choice
Sinclair (1991) discusses two principles of grammatical descriptions, connected
to the Saussurian notions of
: the open-choice principle
treats each position in an utterance as a (complex) choice, basically like a slot that
is filled by an appropriate item (hence it is also referred to as 'slot-and-filler
model'). The idiom principle, on the other hand, states that the user has at their
disposal a set of larger units, so that they do not select individual lexical items (as
they would do following the open-choice principle) but instead larger chunks. He
argues that neither principle is sufficient to describe language, but that the idiom
principle is the more important of the two, which should be used by default for
describing texts. Only when a phenomenon cannot be accounted for by the idiom
principle should we fall back on the open-choice principle.
This view fits in well with a model that uses MWUs as their basic units, as
these would represent the larger chunks that make up utterances instead of single
lexical items. As we will see below, Sinclair was right in stating that both
principles are required for a more comprehensive description.
Multi word units and Phraseology
We now look at multi word units, which go beyond the single lexical item. While
they can of course be described intuitively, there are two principal ways of
automatically identifying them through computer algorithms. In this section we
will explain those algorithms, as they form the basis for extracting MWUs that
we later apply to our grammatical description.
When looking at computer-identified phrases in the past, most work has been
concerned with
-grams, where word sequences of a particular length are
extracted from a text. Values for
are typically in the range of 2 to about 8 (as on
Phrases In English
site). This is a great step forward from early studies
which were mainly limited to bigrams and trigrams; this step has been facilitated
Stringing together a sentence: linearity and the lexis-syntax interface
by advances in computing power and storage. The problem with
-grams is that
as their number quickly increases for larger values of
, their respective
frequencies quickly diminish, leading to large sparse matrices that are difficult to
Despite not being grammatically well-formed, these
-grams are usually
referred to as phrases, which is less awkward than the more general term 'multi
word units'. Stubbs and Barth (2003) use the term
to describe an
-gram, a
usage we will adopt here as well.
In our own work we collect all
-grams with
ranging from 2 to 7. In
order to filter out those that are not 'interesting' we assign to these chains a
weighting, based on its frequency of occurrence and length: since short chains are
usually more frequent while longer ones occur less often, using frequency alone
would favour short chains. But longer chains are more specific (and thus
interesting), provided they are also used often enough, so we take length into
account as well. The use of a weighting function has been inspired by Kita et al.
(1994) who applied it to a similar problem.
Another way to derive multi word units (MWUs) is described by Danielsson
(2001), who aims to identify units of meaning, starting with the premise that
single words do not carry meaning unless embedded in context. Danielsson uses
collocation to extract larger units. Renouf and Sinclair (1991) start off with a
gapped sequence of high frequency words, eg
as — as
, and investigate typical
fillers for the gaps in their templates. Mason (2006) reverses this procedure by
attaching words to a central node word which have a higher frequency than the
node word itself; this is based on the idea that context words (low frequency)
alternate with function words (high frequency). This procedure can also be
combined with
-gram extraction to produce a good range of MWUs for a
particular node word.
There are two different uses of MWUs apparent: first, one can extract all
MWUs from a text or corpus to look at the properties or distribution of MWUs
within the text (eg Stubbs and Barth, 2003), and second, one can look at the
MWUs involving a particular target word or phrase (eg Starcke, this volume, and
Mason 2006). In this paper we will be pursuing the second kind of analysis.
We combine the frames with the chains mentioned in the previous section,
and thus can collect for any particular word a good range of multi-word units. As
a starting point we take MWUs extracted from the written part of the British
National Corpus (BNC). These lists of MWUs are calculated for individual
words, and are filtered so that all items with a frequency of less than 1% of the
highest frequency MWU are discarded.
Before we apply those MWUs to the description of sentence structures we
will further investigate the transition from grammar to phraseology.
Linear Grammar
Linguists generally describe the structure of language in terms of sentences, and
those sentences are commonly assigned a hierarchical structure, represented as an
upside-down tree. Lexical items are of minor importance only; they get replaced
by syntactic categories in the first step, and from then on there are only nouns,
verbs, adjectives and a few other elements which make up the sentence. It is
obvious that the actual words do not play any significant role in this model, and
thus phraseology is completely irrelevant to it.
However, the study of collocations and work in phraseology has shown
that word class labels cannot adequately replace lexical items, as they are too
general and cannot describe the combinatorial idiosyncrasies of words. For an
example analysis of
see Sinclair (1991). Sinclair shows that even the most
frequent member of the class 'preposition' does not at all behave as it would be
expected to. Other studies have shown that even the various inflected forms of a
lemma have little in common when looked at in more detail (eg Stubbs (2001) on
CONSUME and SEEK). Francis (1993) describes the importance of phraseology
for the co-selection between lexical items and syntactic structures.
There are two related issues here: first, the use of hierarchical structures
for describing sentences, which necessitates abstract labels such as NP and VP for
phrasal units, and second, the use of abstract labels itself which leads to
overgeneralisation, as elements belonging to the same syntactic category usually
show divergent behaviour in authentic language.
In this section we will look at three alternative approaches, which
postulate a linear structure for utterances. The grammar devised by Brazil (1995)
focuses on spoken utterances, but there is in principle no reason why it should not
also be applied to written data. Brazil uses category labels, but is prepared to
relinquish them should they turn out to be superfluous, ie he is not fixated on
traditional grammatical terminology. Hunston and Francis (2000) in their pattern
grammar use a mixture of category labels (for both phrases and individual items)
and lexical items, in order to achieve a higher level of precision. Sinclair and
Mauranen (2006) in their linear unit grammar, on the other hand, do without word
class labels, and classify elements of utterances according to their role in
A Grammar of Speech
Brazil (1995) sets out to describe (spoken) utterances in a linear way, deliberately
focusing on the process rather than the product of language. In the absence of any
alternatives he tentatively adopts traditional categories such as nominal and
verbal elements, but he restricts himself to a purely linear structure. As an
underlying formalism he chooses a finite-state model, which seems to be
appropriate for his purposes.
His central concern is communicative function rather than supposed
grammaticality, and so he identifies target states in the speech chain which fulfill
Stringing together a sentence: linearity and the lexis-syntax interface
a communicative purpose. Various paths through the chain can either reach a
target state (which leads to a completed utterance) or an intermediate state (which
requires further elements for completion). Although it has not been adopted into
mainstream linguistics, this incremental approach works well without the need for
a constituent structure.
One reason why a finite state model works, even though Chomsky (1957)
states that it could not, is that authentic language is more restricted when it comes
to those features which a finite state model would have difficulties with. While in
theory there can be an infinite number of central embeddings in a sentence, this
simply does not occur in practice, especially not in spoken discourse. Taking this
into account thus allows us to use a simpler and less powerful apparatus for
language description.
Pattern Grammar
Hunston and Francis (2000) describe a grammar based on syntactic patterns
centred around lexical items, predominantly verbs, but also nouns and adjectives.
The syntactic behaviour of these words can be captured in a finite set of such
patterns, which are specific to the word in question (but does not have to be
unique across the vocabulary). In traditional terminology the closest equivalent
would be that subcategorisation information is contained in those patterns, eg
'V n' describes a verb that takes one object while 'V
n' describes a verb that
typically takes a prepositional phrase with
as its complement.
Pattern grammar, however, goes well beyond merely describing
subcategorisation: it can be used to model the structure of sentences as well.
Similar to Brazil's approach the description is a linear one, where a sentence is
seen as a sequence of patterns of its lexical items which are realised. These
patterns can either be end-on-end, or they can overlap. The latter phenomenon is
called 'pattern flow', as one pattern 'flows' into the next, as illustrated in table 1:
here the pattern
V n
of the verb
flows into the pattern
N that
of the
, which fulfills the role of the 'n' in the first pattern.
Table 1: Example of 'pattern flow'

V n
N that
Linear Unit Grammar
While both Brazil and Hunston/Francis make use of traditional category labels,
Sinclair and Mauranen (2006) abandon them completely. They also employ
different units of analysis, neither words nor phrases, but chunks, groups of
lexical items which by intuition belong together. They deliberately do not offer a
definition of 'chunk', but instead prefer it to remain as a pre-theoretical term. Each
chunk in a text gets assigned a functional label, classifying it either as message
oriented or (text-)organisation oriented. These two classes also have a number of
sub-classes, which are used eg to mark incomplete elements which require
completion (similar to Brazil's 'suspension'), or fragments (false starts and
repetitions in spoken utterances).
After the chunks have been classified, they suggest several steps for
further processing, which involves removing the organisational elements and
incomplete fragments, and results in a 'cleaned-up' version of the original
utterance. This revised utterance is more suitable for analysis with a traditional
grammar, as it will more closely resemble the kind of well-formed utterance that
such grammars have been designed to handle.
Their method of segmentation (using intuition) remains unsatisfactory,
though they suggest that there is sufficient overlap between different analysts to
assume its validity. It would also fit in with an individualistic view of language,
where we would be dealing with the internalised knowledge of language of an
individual: here we can easily accept that different speakers have different
internalised grammars, and different segmentations could be part of that.
However, if we use the MWU algorithm described above, then we would
have an objective means of deriving chunks, which is also capable of modeling
the linguistic experience of an individual, as it will be based on a particular
corpus. Different corpora will yield different sets of MWUs, which is consistent
with the differing language experience of different users.
Phraseology and Grammar
As we have seen in the above section, it is perfectly possible to do away with
hierarchical structure when describing utterances. Of the three approaches
mentioned here, two explicitly commend themselves for the description of spoken
language, namely Brazil's Grammar of Speech and the linear unit grammar of
Sinclair and Mauranen. However, there is no reason why written language should
not also be describable by a linear approach, as it essentially is a linear process by
which it is created. The only difference seems to be the possibility of additional
editing once a sentence has been written down, and the influence of the
conventions of writing systems and written genres.
In the remainder of this article we will present the outline of a grammar
based on phraseology. It will adhere to similar principles as the three linear
grammars, and it will be based on corpus data rather than intuitive judgments.
The basic premise goes back to Stubbs (1996:41) who states as the seventh
principle of neo-Firthian linguistics: "Much language use is routine." Essentially,
re-using bits of language has several benefits: it reduces the strain on the speaker
to create ever new and previously unheard constructions; it makes it easier for the
listener to recognise patterns in the language stream; it can help establish larger
Stringing together a sentence: linearity and the lexis-syntax interface
chunks and contexts for disambiguating words used with multiple senses and
Essentially we assume that pretty much everything has been said before,
though that is of course an over-simplification. There are indeed new and creative
constructions, but they are the exception rather than the rule. Most of language
will consist of chunks that have occurred before, just as we tend to re-use words
and only occasionally introduce new coinages. But it is not only the words
themselves that we re-use, it is also their contexts, as they are inseparable. And
their contexts are effectively multi word units.
Stubbs (1996) further notes that the study of language is basically part of
the social sciences, as language transmits (and creates) culture. But society is not
an independent entity, but a collection of individuals, each of them acting
independently (though according to behavioural rules which can be observed).
The same applies to any language, which exists only through the internalised
grammars of their speakers, which of course are all independent and therefor
distinct. This provides support for the use of corpus data, and it also highlights
the fact that any grammatical description of a language will only ever be an
approximation, as there cannot exist a single grammar describing a (natural)
Combining these points, we will try to identify fragments of a sentence to
be analysed in our corpus, but we will also not worry about gaps in the
description: these can either be caused by a lack of evidence (ie the fragment in
question has not been perceived before and is thus not part of the speaker's
linguistic experience), or it could be a genuine instance of creative usage. While
these two potential causes look pretty much identical, there is an important
difference between them: a creative use will also not be found in any other
corpus, whereas a case of lacking evidence might have occurred there.
The identification of the fragments is as follows: for each word token in
the sentence we look up all the MWUs found for that item in the reference
corpus. We then tabulate all the MWUs that can be matched with the syntactic
context. Eventually it is only necessary to keep track of the longest match, but for
the time being it will be interesting to also find shorter (overlapping) fragments,
as these might point us to issues of interest in the sentence.
Examples and Explanations
We will now look at several examples of the MWU analysis procedure applied to
sentences. All the sentences have been chosen at random from a variety of
(written) sources. To keep processing issues simple, all words have been
converted to lower case and punctuation has been removed. Where a MWU is
given, words in capitals are the node words of that particular MWU, ie the word
that has been used as the starting point in the extraction procedure.
The long dark tea-time of the soul
The first paragraph we will be looking at is from Douglas Adams' "The long dark
tea-time of the soul":
(1) at the top of the stairs was a minute landing which opened on one side
into a bathroom so small that it would best be used by standing outside
and sticking into it whichever limb you wanted to wash . (2) the door to it
was kept ajar by a length of green hosepipe which trailed from the cold tap
of the wash-basin out of the bathroom across the landing and into the only
other room here at the top of the house . (3) it was an attic room with a
severely pitched roof which offered only a few spots where a person of
anything approaching average height could stand up .
The beginning of the first sentence is very well covered by MWUs, as shown in
table 2.
Table 2: Fragment of sentence 1 of the Adams text

AT the top of
AT the top of the
AT the top
AT the top of the stairs
THE top of the
the top of THE
THE top of
the top of THE stairs
THE top of the stairs
TOP of the
TOP of the stairs
at the top of the STAIRS
Here we have full coverage of the fragment
at the top of the stairs was
in a single
MWU, but also several partial overlaps, which would allow for minor variations.
The overlap continues with a MWU
was a MINUTE
; the following
is not
covered, but then the gap is closed with
which OPENED on
, and coverage
continues with
on one side
, which is identified as a MWU for each of its three
elements independently. Then there is no overlap with
into a
which again does not overlap with the next part covered by the analysis, shown in
table 3 (duplicate MWUs with different node words have been omitted).
Stringing together a sentence: linearity and the lexis-syntax interface
Table 3: Fragment of sentence 1 of the Adams text
Here we have a phrase that is less obviously repeated than
at the top of the stairs
and we can easily see how it is composed of smaller units 'flowing' into one
another. The remaining parts of the sentence contain one further stretch covered
by MWUs, namely
YOU wanted to
wanted to WASH
. In total, ¾ of the
sentence can be described through sequences of automatically derived MWUs.
In the second sentence we have a similar pattern, multiple MWUs
matching and overlapping. In the following representation we have put in
brackets words which are not covered, and in places where there is no MWU
flow/overlap we have placed a vertical bar:
The door to | it was kept | (ajar) | by a length of | (green hosepipe which trailed) |
from the cold tap | (of the wash-basin) | out of the bathroom | across the landing
and into the only other | (room) | here at the top of the house
The coverage rate is even higher than in the first sentence, at 79%. We can
identify some kind of larger constituents, which might be an approximation to
Sinclair and Mauranen's chunks, though they certainly violate the elements
posited by a traditional analysis. One could furthermore argue that the second
it was kept
, is a mere accident, a mismatch from a usage where
is the
subject of a clause rather than part of a postmodifier, and human analysts would
separate it into
the door to it | was kept
instead. However, an automated analysis
is never going to reproduce exactly what human beings would do, and we would
need to look at more data before being able to evaluate the quality of the results.
One interesting aspect is the first gap, where
was not found as part of
an MWU. Here we can go back to pattern grammar, where a pattern for
n adj/prep
, which would need to be transformed here due to the passive voice
used. The adjective
thus seems to indicate a choice point, a case where the
idiom principle (Sinclair 1991) cannot account for the variation, as there is a point
of lexical choice more suited for a slot-and-filler model. But by combining the
two, linear MWU analysis and pattern grammar, we can achieve a complete
analysis of this stretch of text.
We can also see that the
at the top of the
part is even a direct repetition
from the previous sentence, so it is not surprising that we get a similar result.
The analysis of the final sentence in this small sample is not that different
from the previous one:
It was an attic room with a severly | (pitched roof which) | offered only a few spots
| where a person of | (anything approaching average height) | could stand up.
The rate of coverage of these sentences is between 74% and 79%, which is
remarkable in that we have retrieved our MWUs from a general reference corpus
only. In section 6 below we will discuss the issue of routine usage versus
creativity in more detail.
Words and Phrases
Next we will look at a sample of academic prose, taken from Stubbs (2001). To
ease the analysis we will not tabulate the complete set of MWUs identified in the
text, but instead use the compressed format introduced in the previous section.
Coverage ranges from 54% (final sentence) to 73% (second sentence), so it is less
than the fiction sample discussed above:
a brief summary of the argument so far is the slogan | (meaning is use) .
(words) | do not have | (fixed) | meanings which are recorded | once and for all | (in
dictionaries) .
(they acquire or change meaning) | according to the social and
linguistic | contexts in which they are used .
(understanding) | language in use |
depends on a balance between | (inference and convention) .
(here) | are more
detailed | (examples which use textual) | data to show that our | (communicative
competence relies on) | knowledge of what is expected | (or typical) .
Regarding the first sentence we can observe that the first part is relatively well
covered, whereas the second part (which is a quoted statement introduced by the
first part) is not. In terms of the informational structure of this sentence Stubbs
makes use of well-established prefabricated units to introduce what is new,
namely the brief summary. We thus have an introductory phrase as the 'given' or
'theme', followed by the 'creative' formulation of the 'new' or 'rheme'.
In the second sentence the first word,
, does not occur in this usage
as an MWU; it is more frequently used in a non-technical sense, such as
in other
. The adjective
can be viewed similar to
in the Adams example in
that it is a point of high variation and can best be described following the open-
choice principle. The final part,
in dictionaries
, is probably not covered as
is a relatively specialised word too unusual for a general reference
corpus such as the BNC. It only occurs 348 times in the written part used for this
study, compared with 929 instances of the singular form
The next sentence (3) again begins with a sequence which is not covered
by MWUs, but after the first five words coverage is rather unexpectedly good.
In the fourth sentence we can see that we again have good coverage of the
middle part which relies more on core words (see Carter 2004). In the BNC,
Stringing together a sentence: linearity and the lexis-syntax interface
is not typically used in this way, but mainly in MWUs such as
inference that
We also have two MWUs in an end-on-end position, which tends to
indicate a higher level boundary, and especially the second one (
depends on a
balance between
) seems to be a highly re-usable stock unit. Typically at such
boundaries we find more variability in lexical choices, which reflects in a break
point when it comes to finding re-current sequences. Conceptually this is related
to the procedure Harris (1955) used to derive morphemes from an unsegmented
stream of phonemes. His distributionalist approach looked at points of maximal
variation in a sequence, such as how many different phonemes can follow a given
sequence. A local maximum would indicate that there was a morpheme boundary.
In this sentence, then,
language in use
would be a (higher level) unit, as would
depends on a balance between
, since there is no 'bridge' that links
, possibly due to too much variability.
Stubbs' final sentence has the least coverage of the five sentences
investigated here. We can explain this again through the technical vocabulary (eg
communicative competence
). It seems obvious that we would require a more
academic corpus in order to achieve a higher degree of coverage, as only the
more general parts of the sentences can be found in the set of MWUs.
Interestingly, the sentence-initial
mainly occurs as an MWU in
here are
here are the
, and
here are a few
; the singular
here is more
also occurs in
the MWUs extracted from the BNC.
Conference Call for Papers
The next example is a single sentence taken from a call for papers of a
(linguistics) conference. This is a fairly standardised kind of text which does not
leave much room for creativity, and hence it is unsurprising that coverage of the
sentence is very good:
The papers presented at the conference will be available in | proceedings on the
first day.
As with the previous sentences we have looked at, the MWUs overlap and link up
to form a longer sequence, similar to what Hunston and Francis (2000) describe
as 'pattern flow', and Gledhill (2000) as 'collocational cascade'. Looking more
closely at some example MWUs, we could say that in
papers presented at the
prospects the following items
presented at the
, whereas in
at the
the initial
the conference
, and so forth. Prospection is an
important organising principle in language use, as it restricts the expectations of
possible elements in the remainder of the utterance. As Sinclair and Mauranen
(2006) state, prospection is not fixed and prescriptive, but flexible, based on the
frequency of phraseological patterns, and can be used to great effect when
violated (in a similar vein as discourse prosodies can, see Louw (1993)).
then flows into
will be
, which could be classed as
an instance of 'colligation' in the sense used by Firth and later Hoey (2005): the
tends to occur frequently with expressions of futurity, in this
will be
There is only one point in this sentence where the flow of MWUs is
interrupted, between
. Here we can hypothesise the existence
of a higher-level unit boundary, which is not crossed by the MWU chunks.
Even though we speak of
the conference
, we need to be
careful about the scope of such statements: they only apply to the analysis of an
utterance, not its creation. While we could undoubtedly generate natural-sounding
utterances by randomly stringing together overlapping MWUs, we would ignore
the semantic aspect and the utterances would not be comparable to authentic
ones. But in the analysis we presuppose that the utterance we are looking at is
meaningful, so that the semantic dimension is implicit. There are certainly many
more (in fact, 565 in total) MWUs that begin with
, but out of these that
particular one has been chosen.
Especially in fairly standard situations (such as giving information about
conferences), we do not need to be creative. On the contrary, going back to
routine usages we make it easier for the recipients to understand what we are
saying, as it involves less effort to process something that one has already
encountered before. New or creative sentences, on the other hand, require more
decoding effort, as they cannot be matched against previously experienced
We could thus assume that the degree of MWU coverage changes
according to the text type: texts which are easy to read ought to be described
better using MWU chunks than highly creative ones or those which are more
difficult to read. This obviously has to take into account other considerations,
such as topic: since we model the speaker's linguistic experience through a
general reference corpus (the BNC), texts which make use of specialised
vocabulary will clearly have less coverage. But from a theoretical point of view
this poses no problem, as the BNC is only an approximation in the first place. If
we were to analyse an academic article, then we would get a higher coverage if
we used a corpus of academic language for the retrieval of MWUs. This is
consistent with the notion of a separate speech community, that of academics,
which have separate shared linguistic experiences from other communities.
We have examined several (authentic) sentences to see how far we get in terms of
accounting for their composition using multi-word units extracted from a corpus.
It is clear that there is considerable variation in the degree of coverage, which can
usually be explained by the nature of specific constructions (eg adjectives/adverbs
which allow for slot-type variability) or the use of technical vocabulary that
would not occur frequently enough in a general reference corpus. We have also
noticed, that 'stock' sentences are covered to a larger extent than more 'creative'
ones, an issue we will look at in the following section.
Stringing together a sentence: linearity and the lexis-syntax interface
Routine and Creativity
In section 4 above we mentioned Stubbs' principle that language use often is a
matter of routine, and that re-use rather than outright creativity dominates our
language experience. This is more so in spoken dialogue, but should also be
found in written language, even though there is more scope for creativity through
the added time available for careful composition and editing. As our approach to
the description of sentences involves the identification of repeated sequences in
the form of MWUs, we tap right into the re-use aspect of language production.
One possible application of our procedure which is a by-product of our
attempt at a grammatical description is therefore to determine the degree of re-use
of a sentence: if a sentence make use of a lot of 'existing' language, then we
should find more MWUs that match, and our coverage of that sentence should be
greater than that of a sentence whose creation involved more creativity.
We have already seen in the previous section that a stock phrase like the
one taken from the call for papers is almost completely covered, whereas with the
Stubbs data we could find certain introductory phrases which are routine whereas
the statement introduced involves less re-use. Other instances that were notable
were places where there is a lot of slot-like variation, as with the adjectives in
both the Adams and the Stubbs sample.
In this section we will be looking at several other extracts that we have not
already described in great detail, as we are only interested in quantitative analysis
of their coverage ratio. Before we look at the results, here a brief description of
the samples, presented in the compressed format, with sentence numbers printed
in bold:
: (see above)
: (see above)
: (see above)
: a section from a children's book,
My grandmother's clock
McCaughrean and Lambert
(in my grandmother 's house) | there is a grandfather clock | but it does not
(the hands on its big face never move)
(once) | i opened the door in the
front of the clock to find out why | and there was nothing inside | (but one
umbrella) | a walking stick and a picture of | (king zog)
: a section from the Independent newspaper, also analysed by
Sinclair and Mauranen (2006)
(mr kennedy now) | declares that it must be bold in its thinking | and ready to
| (plan long-term)
(sounding nice) | is no longer enough | (he argued)
now on the liberal democrats have | to present themselves as a party that |
(wants power) | and knows what it wants to do if it gets it
with that in mind
he | (announced two reviews) | one to take a broad | (look at policy) | the other
to look at | (tax) | policy as well as a number of internal | (reviews) | into the
party | ('s structure and communications)
: an extract from
The world according to Clarkson
by Jeremy Clarkson
last week the queen of england | (very) | kindly agreed to break off from her |
(waving duties) | and lend a hand with a television programme | (i'm) | making
about the | (victoria cross)
and so on | (wednesday) | i slipped into a whistle
and went to | (buckingham palace) | to see some | (prototype medals she'd) |
found in a cupboard
(sadly) | i never met | (my new researcher) | but i did
have a | (snout around) | the state rooms | which provided a rare insight into
the life of the royals
: an extract from a BBC news story taken from the BBC website
(a bickering new york couple) | have had a dividing wall | (constructed
inside) | their home as part of an | (acrimonious divorce)
(chana and simon
taub both 57 have endured) | two years of | (divorce negotiations) | but neither
is prepared to give up their | (brooklyn home)
(now a white partition) | wall
has been built | through the heart of the house to keep the | (pair apart)
taub) | asked a judge to allow him to erect the | (partition when the couple 's
divorce stalled over financial details)
(the taubs') | divorce has been
rumbling | through the new york | (divorce courts) | for two years
despite owning another home - just) | two doors away | (- the unhappily
married couple) | have decided to carry on living under the same roof
: seven sentences from
by Malcolm Gladwell (non-fiction)
(the videotape of bill and sue 's discussion seems) | at least at first to be a
random sample of a very ordinary | kind of conversation that | (couples) | have
all the time
no one gets | (angry)
there are no | (scenes no breakdowns no
i'm just not a dog | (person is how bill starts things) | off in a
perfectly reasonable tone of voice
(he complains) | a little bit but about the
dog | (not about susan)
(she complains too) | but there are also | (moments
when they simply) | forget that they are supposed to be arguing
when the
subject of whether the | (dog smells comes up for example bill and sue banter) |
back and forth | (happily both) | with half a smile on their lips
: six sentences from the
New Scientist
, also analysed by Hunston and
Francis (2000)
as a rule | (books proclaiming) | the solution of a mystery | (deal) | with
something that | (isn't mysterious) | or fail to deliver
(the sunken kingdom
falls into both categories
(plato 's atlantis) | is a mystery | only to those who
and the solution | offered in a readable and | (well-argued fashion) | is
not conclusive
(peter james distinguishes between believers chasing the real
atlantis and sceptics) | who are hostile to the very idea
but i fear he | (is
omitting) | a far larger | (category -) | those who find | (this) | a waste of time
: a sentence by Bill Bryson, analysed in Hoey (2005)
(in winter hammerfest is a thirty-hour ride by bus from oslo though) |
why anyone would want to go there | (in winter) | is a question | (worth
Stringing together a sentence: linearity and the lexis-syntax interface
: Hoey's rephrasing of the 'bryson' sentence, deliberately avoiding typical
lexical relationships for illustrative purposes
(through winter rides between oslo and hammerfest use thirty hours) | up in
a | (bus though why travellers would select to ride there then might be
All these extracts have been analysed as described in the previous section. From a
genre/register point of view they are not very wide-ranging, but there is some
scope for variation; all are written, two are fiction ('adams', 'clock'), two non-
fiction ('blink' and 'bryson'), several are newspaper/magazine ('bbc', 'independent',
'sunken'), one is a reprint of a newspaper column ('clarkson'), one academic prose
('stubbs'), and one is a made-up paraphrase ('hoey').
If we express the rate of re-use as a percentage of words of a text covered
by MWUs, we get a wide range from 11% for Hoey's non-idiomatic
reformulation of the Bryson sentence up to 100% for the formulaic call for papers
Table 4: MWU coverage per text (in %)
Conference 100.0
Independent 78.8
Perhaps unexpected is that some of the non-fiction samples tend to score lower
than the fiction ones; this suggests that coverage cannot simply be equated with
routine/lack of creativity (in a literary sense). It also shows features of a
readability measure, which makes sense if we consider that routine usages are
easier to decode.
It is interesting to note that the lowest coverage is for a piece of made-up
text, which Hoey deliberately created to sound clumsy and un-natural for the
purpose of comparing it with the original. Our method of analysis seems to be
able to pick up the non-naturalness, as almost none of the sentence's lexical items
are used in their expected contexts.
Hoey argues that from a traditional point of view there is nothing wrong
with his invented sentence, as it is perfectly grammatical; yet it is obviously odd.
His point is that it is the collocational and colligational patterns that are broken
which make the sentence un-idiomatic. However, if we look at it from the point
of view of phraseology, then we could argue that it is the multi-word units (ie the
basic elements of the idiom principle) which are ignored, that untypical words are
combined individually and taken out of their context (according to the open-
choice principle). From that angle, collocation (and colligation) are just
epiphenomena of multi-word units: simply because lexical items are used in a
fixed set of contexts, all the words that are also contained in these contexts occur
near to them more often than would be expected. The collocation algorithms
simply pick this up and identify those context words as significant collocates.
Reversing this process, Danielsson (2001) does in fact use collocations to
construct her units of meaning.
To summarise, this brief exploration of the routine re-use of language
fragments and creativity indicates that there is a link between the degree of
creativity as identified through multi-word units and certain properties of the
sentence: a high degree of re-use points to stock phrases; a low degree reveals
non-natural language; and anything in-between is 'normal'. Obviously, a lot more
further analyses are required in this area to establish valid benchmarks. This
would involve longer texts from a wider range of genres, and also cross-
comparisons with corpora other than the BNC for MWU extraction. It would also
be interesting to explore where poetry fits into this range.
Summary and Discussion
In this paper we investigated a phraseological approach to grammar. Based on the
notion of a multi-word unit, a re-current combination of words, we noted that
parts of sentences can be viewed as (often overlapping) sequences of MWUs.
This is a first step towards a 'proper' grammatical description; so far we are not
using grammatical categories, but neither does similar work by Sinclair and
Mauranen (2006).
What is the point of this description, when it does not provide labels of
either syntactic or functional nature? The answer is that it is a lexicalised
approach to grammar, where the function realised by a particular segment is the
meaning instantiated by the MWU. The expression of meanings of some sort is
the fundamental reason for using language, whereas syntax is merely a by-
product of the linearisation of thought.
This, however, is not the full story. There are gaps in the descriptions,
some of which we have attempted to patch up in the discussion of the samples
above. While the idiom principle goes a long way when composing structures,
there are instances where there are slots or wildcards which allow a greater
degree of choice. This choice cannot be captured with MWUs, as they rely on
repetition, but for the purpose of accounting for the full structure we could
combine the MWU approach with a set of local grammars, which would result in
a hybrid model, combining both the idiom principle and the slot-and-filler one,
only with constraints put on the possible items than can occur in a slot.
Another related problem that we are facing is that our definition of MWUs
might be too restrictive. There can be a lot of variability in the actual ordering of
words, with certain elements inserted or omitted, but none of that is currently
Stringing together a sentence: linearity and the lexis-syntax interface
captured by our MWU identification algorithm. Eventually we will want to be
able to relate similar MWUs to each other, rather than treating them as separate
So, what we have presented here is only the first step towards establishing
phraseology as a legitimate way of describing sentence structure. There is still a
lot of work to be done, but at the same time we have already established the main
principles, combining work in grammar with that in phraseology.
Brazil, D. 1995.
A Grammar of Speech
. Oxford: OUP.
Carter, R. 2004.
Language and creativity
. London: Routledge.
Chomsky, N. 1957.
Syntactic Structures
. The Hague: Mouton.
Danielsson, P. 2001.
The automatic identification of meaningful units in
. PhD dissertation, Göteborg University.
Francis, G. 1991. Nominal group heads and clause structure.
, p. 144—
Francis, G. 1993. A corpus-driven approach to grammar. In M. Baker, G. Francis,
and E. Tognini-Bonelli (eds)
Text and technology: in honour of John
, p. 137—156. Amsterdam: Benjamins.
Gledhill, C. 2000.
Collocations in Science Writing
. Tübingen: Narr.
Harris, Z. S. 1955. From phonemes to morphemes.
(2), p. 190—
Hoey, M. 2005.
Lexical Priming: A new theory of words and language
. London:
Hunston, S. and G. Francis 2000.
Pattern Grammar: A corpus-driven approach
to the lexical grammar of English
. Amsterdam: Benjamins.
Kita, K., Y. Kato, T. Omoto and Y. Yano 1994. Automatically extracting
collocations from corpora for language teaching. In T. McEnery and A.
Wilson (eds)
Corpora in Language Education and Research: A Selection
of Papers from TALC94
. Lancaster: Department of Linguistics.
Mason, O. 2006.
The automatic extraction of linguistic information from text
. PhD Dissertation, University of Birmingham.
Louw, B. 1993. Irony in the text or insincerity in the writer?—the diagnostic
potential of semantic prosodies. In M. Baker, G. Francis, and E. Tognini-
Bonelli (eds)
Text and technology: in honour of John Sinclair
, p. 157—
176. Amsterdam: Benjamins.
Sinclair, J. M. 1991.
Corpus, Concordance, Collocation
. Oxford: OUP.
Sinclair, J. M. and A. Renouf 1991. Collocational Frameworks of English. In K.
Aijmer and B. Altenberg (eds)
English Corpus Linguistics: Studies in
honour of Jan Svartvik
, p. 128—144. London: Longman.
Sinclair, J. M. and A. Mauranen 2006.
Linear Unit Grammar
. Amsterdam:
Stubbs, M. 1993. British traditions in text analysis—from Firth to Sinclair. In M.
Baker, G. Francis, and E. Tognini-Bonelli (eds)
Text and technology: in
honour of John Sinclair
, p. 1—33. Amsterdam: Benjamins.
Stubbs, M. 1996.
Text and Corpus Analysis
. Oxford: Blackwell.
Stubbs, M. 2001.
Words and Phrases
. Oxford: Blackwell.
Stubbs, M. and I. Barth 2003. Using recurrent phrases as text-type discriminators.
Functions of Language
(1), p.61—104.
'Sailing the islands or watching from the dock':
the treacherous simplicity of a metaphor. How we handle
'new (electronic) hypertext'
'old (printed) text'
Wolfram Bublitz
University of Augsburg
This paper looks at the validity of two tightly interrelated linguistic dogmas. They state
that the dyadic nature of human communication is an indispensable precondition for
negotiating meaning, which is understood as a dyadic, transitive and reciprocal act
requiring two interactants. It will be shown that since the advent of the new electronic
media, both dogmas have been subject to a process of gradual erosion. Some forms of
computer mediated communication have altered our understanding of participation as a
dyadic and focussed concept. Furthermore, despite their amazing possibilities and
extraordinary interactive potential (which, however, is at least partly counteracted by the
extremely high degree of fragmentarization), interacting with new electronic media does
not per se guarantee easier understanding, i.e. an easier access to the world 'behind the
screen' than when interacting with 'old' printed media. It is argued that the user's situation
is not essentially different from the familiar situation of the reader who is trying to
understand printed text.
Introduction: even eternal truths are not what they used to be
Outside grammar, there are not very many 'eternal truths' in the science of
linguistics. Arguably, that duality (i.e. dyadic orientation) is a basic feature of
human communication, is one of them, and that meaning is always negotiated
meaning, is another. These two dogmas are tightly interrelated. They cohere
because negotiating meaning is dyadic by nature in that it is a transitive and
reciprocal act requiring two interactants. Hence, the dyadic character of human
communication is an indispensable precondition for semiosis, i.e. the act or
process of meaning-making (in the Peircean sense).
As is sometimes the case with everlasting truths, however, an
unforeseeable change of their conditional fundaments can lead to their erosion.
As I will argue in my paper, this appears to have happened with duality as a
dogmatic feature of human communication. Since the advent of the new
electronic media, it has been subject to a process of gradual erosion and is no
longer unrestrictedly valid for both 'old' (spoken and written) and 'new' electronic
media. Some forms of computer mediated communication (CMC) in particular
have altered our understanding of participation as a dyadic and focussed concept,
and have also made negotiating meaning and thus understanding more difficult.
The latter may come as a surprise because the possibilities of the electronically
administered new media with their literally infinite number of audio-visual data
are widely regarded as an asset rather than as an impediment to composition and
thought. But, as we will see, the interactive potential of CMC is (at least partly)
counteracted by the high degree of fragmentarization (with all its consequences).
Thus, despite its extraordinary possibilities, interacting with this new medium
does not
per se
guarantee easier understanding, i.e. an easier access to the world
'behind the screen' than when interacting with old media. In actual fact, the user's
situation is not
different from the familiar situation of the reader who,
when reading a book, a handbook or a newspaper, is trying to understand, i.e. to
create his or her own inner world.
Communication is not as dyadic as is generally assumed
Among the long-established dogmas that spring to mind when studying how
communication works is the following: Human communication is most obviously
characterised by its speaker/writer – hearer/reader symmetry, i.e. its dual or
dyadic orientation. To communicate means for someone to communicate
someone else; it is a reciprocal act.
As a fundamental principle, this time-
honoured dictum has seldom been queried in its entirety, though every now and
then in some of its aspects (as I will show presently). A succinct description was
provided by Wilhelm von Humboldt. In an article about
as a grammatical
number (besides singular and plural), he reflected in a more general way on
as a universal communicative principle:
Besonders entscheidend für die Sprache ist es, daß die Zweiheit in ihr
eine wichtigere Stelle, als irgendwo sonst, einnimmt. Alles Sprechen
ruht auf der Wechselrede, in der, auch unter Mehreren, der Redende
die Angeredeten immer sich als Einheit gegenüberstellt. [...] Es liegt
aber in dem ursprünglichen Wesen der Sprache ein unabänderlicher
Dualismus, und die Möglichkeit des Sprechens selbst wird durch
Anrede und Erwiderung bedingt. (1827/1969: 138)
Duality is the most obvious defining feature of two-party talk as the archetypical
kind of spoken face-to-face communication in a homogeneous and focussed
social setting. Human verbal communication is by nature dialogic. At closer
inspection, however, neither the prototypical speaker nor the prototypical hearer
are monolithic concepts but fusions of various conceptual roles. To take a simple
example from the production side of verbal interchange: The
, the
and the
of a piece of text can be
different persons (e.g., a
government spokesman reading out a secretary's account of a cabinet minister's
ideas to a journalist),
different persons (the secretary reading out her own
account of a cabinet minister's ideas to a journalist), or just
person (the
cabinet minister telling the journalist herself her ideas). Or, focussing on the
reception side, that the
of an utterance is not necessarily its
'Sailing the islands or watching from the dock'
the interlocutor who is expected and entitled to reply) is a blatant truth that has by
now been accepted even in speech act and inference theoretical circles.

To account for such diversity, various theoretical frameworks have been
developed, in which the monolithic, unitary concepts of the producing and the
receiving participant are being deconstructed.
Among such frameworks we find
Goffman's (1981) well-known distinction between "production format" and
"participation framework" or, to use Levinson's (1988: 169 ff) rather more
comprehensible pair of terms "production roles" and "reception roles"; the
following outline of categories is taken (and slightly adapted) from Levinson
(1988: 169):
Production roles
: 'the sounding box'
: 'the agent who scripts the lines'
'the party to whose position the words attest'
Reception roles
A: ratified
addressed recipient
: 'the one to whom the speaker addresses his
visual attention and to whom, incidentally, he expects to turn over
his speaking role'
unaddressed recipient
: 'the rest of the 'official hearers', who may or
may not be listening'
B: unratified
'inadvertent, non-official listeners' or '
: 'engineered, non-official followers of talk'
Even though the two juxtaposed composite poles have been deconstructed into
several components, which can be adopted by one interlocutor or distributed
among several interlocutors, each can still be regarded as an integrated unit or
"Einheit" (in Humboldt's account), which is not unanalyseable and monolithic but
flexible and multi-facetted, allowing for various degrees of internal variation.
Thus, without losing its dyadic character, Goffman's participation model can be
adapted to cover one-many-talk or many-one-talk, or other complex forms of
interaction such as chaired panel discussions open to the public. And it can also
easily be expanded by implementing an even more refined sub-categorization of
its constituent roles. The analysis of the following two examples, for instance,
calls for the introduction of the role of
recipient, who is neither
nor simply
in Goffman's account).
(Four US high school students, one middle school student and the
recording teacher, Joan, are working in the school's writing lab.)
(1) Sue: oh you're you're from the Middle School
Cyndy: yep
Sue: I was wondering
Kim: oh yeah I didn't introduce you Cyndy this is Sue Sue this is Cyndy
(2) Sue: well you guys
Don: see my my introduction's like (using fingers to indicate two inches)
just like that long and that's it
Sue: wait are we being recorded
Kim: keep it I don't know I mean I don't know if that's too long
Don: yeah that's really short
Mary: she's taking a class in linguistics and she's not really looking for
what we're saying but how we say it (…)
Sue: ha
Mary: how they speak
Sue: okay
Joan: I promise you no one here will hear about this except for me
(Joan Wallace,
Mixed Sex Discourse
, USA 1994; private data; names of
students changed; adapted)
In (1), the exchange is opened by Sue, who, as
addresses Cyndy (by asking her). Cyndy is
addressed recipient
and re-addresses
Sue by answering her question. Kim, on the other hand, is not
addressed recipient
intended recipient
because she reacts to Sue's implicit reproachful request by
addressing both her interlocutors. In (2), two exchanges overlap. Don and Kim
address each other, are
as well as
, respectively and subsequently. With both her initial utterances, Sue
addresses everyone present. But only Mary accepts the role of
and re-addresses Sue by answering her question "Wait are we being
recorded?". For the length of this and the ensuing exchange, the other three
persons present assume the role of
unaddressed recipient
. However, following
Sue's "Okay", Joan, somewhat belatedly, reacts to Sue's original question (which
she indirectly confirms) and Mary's utterances, thus switching from
intended recipient
. (She cannot be
addressed recipient
, though, because Mary
refers to her in the third person singular.)
Goffman's participation model has been criticized for a number of
deficiencies and limitations, which predominantly concern individual problems of
definition and terminology or aspects of the assumed social setting,
but not for
its general dyadic construction. Despite the intricate patterns of categories, co-
categories and sub-categories at both the producing and the receiving side, it is
still a dyadic model of communication, which upholds the underlying
presumption that human interaction is principally and distinctly dyadic.
But blurred, fuzzy and generally unclear participant structure can even be
found in 'old' printed texts, e.g. in handbooks or encyclopaedias with their wealth
of references, cross-references, inserts, self-contained texts and strong iconic
orientation, which is manifest in images, drawings, graphics, layout, etc; they
reflect some aspects of modern electronic media, such as fragmentarization
(constituting a kind of
reverse remediation
in Bolter's (2001) sense).
And it is
obviously even more problematic for forms of new electronic media, which are
typically and in varying degrees multimodal, fragmented and interactive (cf.
'Sailing the islands or watching from the dock'
Hence, with the advent of the new electronic media the erosion of the
presumption that communication is and must be dyadic has reached a new
dimension. This becomes immediately evident when looking at two forms of
electronic communication, computer mediated chats and Wiki-media
Computer mediated
differ from traditional conversation, as
discussed by Goffman, in several ways.
Crucial differences are that their
structure is not necessarily dyadic and that they are not "focused social
interactions" occurring in "particular physical spaces" and involving "easily
identifiable participants with clearly defined roles and relationships" (Jones 2004:
23), who monitor each other's actions, attitudes and presence. The clear
distinction between participant roles is blurred in such online chats based on
multi-functional technological gear supported by various kinds of instant-
messaging software (like ICQ) (cf. Jones 2004). In these hybrid forms of human
communication (which share features of both speaking and writing), ongoing
interaction is basically multilateral, a multilogue, rather than bilateral and a
dialogue, with no clearly discernable participation roles of the kind described by
Goffman and other analysts.
To illustrate this point, I borrow one of Herring's (1999) examples of
overlap in CMC
, because "overlap in CMC is […] problematic. On the one hand,
temporal overlap in display of turns is not an option in one-way CMC, since one-
way systems force messages into a strict linear order. On the other hand, overlap
is rampant in computer-mediated environments. In dyadic
communication, users are unable to tell whether their interlocutor is in the process
of responding or not. They may become impatient and send a second message
before a response to the first has been received, resulting in incomplete or
interleaved exchange sequences […]. In group communication, unrelated
messages from other participants often intervene between an initiating message
and its response […]." (1999: 4) I have slightly adapted her example (which is
taken from CM group communication "on a public IRC channel") and added the
Participant: Exchange
1 2 3
1. ashna&#x-3a-;s3h;n6a;&#x-200; hi jatt
2. *** Signoff: puja […]
3. Dave-G&#x-3D-;-2;&#xv6e-;-7G;&#x-100; kally i was only joking around
4. Jatt&#x-3J-;š-2;&#xt1t1; ashna: hello?
5. kally&#x-3k6; -2l;l-1;y18; dave-g it was funny
6. ashna&#x-3a-;s3h;n6a;&#x-200; how are u jatt

7. LUCKMAN&#x-3L8;&#xU-1C;K-1;&#xM-15; -1N;&#x-100; ssa all
8. Dave-G&#x-3D-;-2;&#xv6e-;-7G;&#x-100; kally you da woman!
9. Jatt&#x-3J-;š-2;&#xt1t1; ashna: do we know each other?.
I'm ok how are you
10. *** LUCKMAN has left channel […]
11. *** LUCKMAN has joined channel […]
12. kally&#x-3k6; -2l;l-1;y18; dave-g good stuff:)
13. Jatt J-9; -2t;t10; kally: so hows school life, life in
geneal, love life, family life?
14. ashna&#x-3a-;s3h;n6a;&#x-200; jatt no we don't know each
other, i fine
15. Jatt J-9; -2t;t10; ashna: where r ya from?

Here is part of her analysis:
Two extended dyadic exchanges are interleaved in this sample of chat,
one between Ashna and Jatt, and the second between Dave-G and
Kally. To complicate matters further, in line 13, Jatt initiates a third
exchange by addressing a question to Kally. [...] The perspective [as
represented by the lines] is anaphoric – the participant lower in the
diagram is considered to be responding "backwards" (or in this case,
upwards) to a previous participant in each case. Dotted lines indicate
interactions in which the message either initiates a new exchange with
an already active participant (as in 1 and 13) or responds to a turn not
'Sailing the islands or watching from the dock'
included in the example (as in 3). [The lines show] clearly that
exchanges overlap, rather than taking place in sequence – turns from
one exchange regularly "interrupt" another. (Herring 1999: 4 f)
The basically multilateral and multilinear character shapes,
mutatis mutandis
, also
written communication in
Wiki- or Web 2.0-based media formats
, which are
generally believed to support online collaboration among large numbers of users.
Hence, the meaning-making and even text-building actions can no longer be
assigned to individual but only to 'multiple authors', who, in exceptional cases,
may still be identifiable as individuals but are usually unidentifiable members of
collective networks.
Furthermore and going one step further, in computer based semiosis, even
the recipient side is involved in that the roles of author and user regularly
coincide. This leads to the central question of who is
the semiosis in these
cases. If the distinctions between participant roles blur or even vanish, if users
cannot be distinguished from authors and users become their own authors, how is
the essentially
action of creating meaning, i.e. of understanding
achieved? And, as meaning is always
meaning, who is negotiating
with whom (on what evidence or data input)?
To answer these questions, let me first explain what I mean by the two key
Negotiating is best seen as a metaphor
In agreement with hermeneutic, interpretive and usage based approaches within
semantics and pragmatics, I adhere to a theory of comprehension which views
understanding as a cooperative activity, resting on speakers' and hearers'
immediate and writers' and readers' delayed collaboration, rather than on each
person's autonomous and strictly individual action.

And while it is a platitude to state that understanding as an act and
cognitive process is a private affair in the sense that it happens in a single person's
mind, it is likewise true that a person's mind is not autonomous and isolated in the
sense that it is totally cut off from its
, including (the output of) other
people's minds. There is a constant
exchange of information between
an interactant's 'inside world' and the 'outside world'; understanding could not
happen otherwise. Only from a neurocognitive (and, incidentally, also analytical)
point of view are one interactant's understanding and meaning private,
idiosyncratic, unique and fully distinct from another's. From a semantic-
pragmatic point of view, however, which takes 'inter-action' as its focal point,
they are compatible and concordant to a degree that we can talk of 'collective acts
of comprehension' and 'shared meanings'.
It has been argued that understanding is not done (as an act)
(by interactively gathering and accumulating and arranging information), but that
someone (in a quasi autopoietic manner).
While I reject the
underlying highly mechanistic and de-humanized view, I readily concede, of
course, that the amount of cognitive effort required when understanding may vary
considerably. After all, much understanding is routine, with little or no
hermeneutic distance to be bridged. Understanding means reducing distances and
overcoming differences! Among them, relevant for our topic are the
(hearers have to ‚translate' speakers' language into their own and vice
versa), the
historical difference
(which separates quite literally the time of writing
and the time of reading of a text; or, when related to the comprehension process
itself, the difference between each of the emergent and successive states of
understanding, which are always only provisional, reflecting the status quo, to be
adapted and modified later), the
between familiar formal means of presentation and unfamiliar audio-visual,
iconic and related signs and formats as used in CMC), and the
episodic and
'semantic' difference
(referring to interactants' memories and systems of
knowledge, as described by Tulving 1972 and 1983). In various forms of CMC,
some of these distances turn out to be difficult for users to bridge when
negotiating meaning, as I will argue presently.
In communication, meaning is always jointly acquired and shared
meaning, which is communicatively valid. We are not free agents when it comes
to cooperate in order to understand; cooperation in understanding is an
anthropological constant. And this is where negotiating comes in, because
comprehension rests on the negotiation of
, which I wish to believe is still a human individual (and not an
electronic 'mind'), both
can be taken literally or
, negotiating is a bilateral, reciprocal, dialectical action between
the understanding person, the
, and his or her interlocutor, the
. What is
to be understood is not given, a priori existing, static meaning but emergent,
dynamic meaning, which is manifest in text or discourse. When negotiating, the
interactants therefore refer to the piece of text or discourse in question, but they
also need to and do rely on other sources of directly relevant information such as
non-linguistic signs (pictures, sounds, kinetic signs) or the situational and socio-
cultural setting. Negotiating in the literal sense hinges on speech acts such as
suggesting (e.g. readings of a word or clause) and accepting or rejecting, querying
and explaining, doubting and affirming, which themselves involve acts of
supplementing or completing as in the following examples:

(3) C you didn't have capital gains but of course you did háve
: .
a death du*ties*
C *dèath* duties

(Svartvik & Quirk 1980: 347)
(4) C because there was some pecùliar -
: convention about hyphens
which just
B seemed quite *àrbitrary*
C *it was* absolutely illògical - - (Svartvik & Quirk 1980: 134)
'Sailing the islands or watching from the dock'
In (3) and (4), by completing their interlocutors' utterances,
offer their
understanding of what is meant as part of a negotiating process. In both cases, the
offer is successful, i.e. accepted (also as part of the negotiating give-and-take) by
the other. This is different in the following examples, where negotiating takes a
different direction because the offer is 'wrong' and not accepted.
(5) A this is the one I could most live wìth . *the cardinals*
*the stàtues*
A well the càrdinal áctually

(Svartvik & Quirk 1980: 203)
(6) A the amount you get from - wèll . firstly the […]
amount you get from the sun doesn't còunt
A well no it dòes it's quite impòrtant

(Svartvik & Quirk 1980: 598)
In (7) and (8), supplementing other's talk is a means of negotiating shared stance
as well as shared meaning.
(7) C course it was Pòrt réally that kept them wàrm . in the eighteenth
and enormous quantities of food
C yes

(Svartvik & Quirk 1980: 340)
(8) A very óirish with a màss of great -
sort of grey háir
and a Catholic of course
A and a Catholic presúmably - *
*lots of* chìldren
A three grown-up chíldren - all márried [...]

(Svartvik & Quirk 1980: 748)
In these transcripts of spoken face-to-face conversation,
the other
is the
interlocutor who is present. In written communication, the author or writer is
either known or unknown, inferable or not, available (contactable) or not, more or
less displaced, depending on the kind of written communication (e.g. a personal
letter or a novel versus an unsigned newspaper editorial or advertisement copy
writing). If the reader has no access to the writer, he or she has to resort to
assumptions about the writer, i.e. has to
create the other
. Which leads us away
from the literal and into the realm of the figurative.
So far, we have assumed that
the other
is the negotiating self's human
collaborator. But since meaning is frequently negotiated not with the human
collaborator directly (as in the examples above) but, in a figurative sense, with the
verbal, non-verbal, situational and other available data,
the other
can actually
stand for the totality of such evidence. Accordingly,
the other as the totality of
data on which the self draws when negotiating meaning
stands metonymically for
the other as human collaborator
. Getting back to our dogma of the dyadic nature
of communication, while the understanding pole, i.e.
the self
, is still a human
individual and
in new electronic media, the producing pole, i.e.
the other
typically dissolves into a heterogeneous array of data.
The true explanatory power of
unfolds itself only when
is used as a metaphorical expression with the underlying concept
. In other words, negotiating meaning is
making assumptions about
what is meant by the author of an utterance, a text, a
picture, a sound etc
on the basis of
the relevant and available data. Negotiating
being a distinctly empathetic act, this means that the self also draws on the
(known or assumed) collaborator's inferred and construed linguistic and world
knowledge, episodic and conceptual memory, cultural background and emotional
frames of mind, as part of the cognitive scaffold which supports the emerging
To demonstrate how negotiating of meaning and text actually works in
CMC and what problems the negotiating (i.e. meaning understanding and text
building) user is confronted with, we have to cast a brief look at the concept of
(hyper-)text first and recapitulate its characteristic features.
Hypertext and the privilege of the eye
, which are nodes that are simply electronic versions of
conventional written texts and thus of no interest for our topic, the notion of
can be used when studying CMC, even though spoken and written language in
CMC is obviously not the sole and, sometimes, not even the most important
medium of information.
In accordance with the hermeneutic orientation of this paper, a
is seen
not as a document that is 'there' for the reader or user 'to find', i.e. not as input to
understanding, but as the output of the reader's or user's interpretation. Each user
creates his or her own text by relating perceivable data to the cognitive
framework of his or her mind. The emerging text is understood as coherent and
meaningful, with a topic, a purpose, a function, embedded in and dependent on a
situation, a socio-cultural environment, a set of other texts.
A chief difference between traditional printed media and new electronic
media is that for the latter's user to decide what (i.e. which data) constitutes a text,
can often be a serious challenge. It can be difficult to decide, which of the
perceivable assortments of signs displayed on the site can be taken to count as
text. There may be no clear demarcation lines between sequences of words
(which may look like a text, i.e.
or continuous text) and surrounding bits
of audio-visual data appearing in a kaleidoscopic wealth of signs, which range
'Sailing the islands or watching from the dock'
from pictures, graphics, pop-ups, bars, frames, links, films, sounds to layout and
Figure 1: Kaleidoscopic wealth of signs in a website of a Japanese pop duo
One can actually
the difference between 'old' printed media and 'new'
electronic media as a result of the often cited
iconic turn
, first noted by Mitchell
(1994) and Boehm (1994), who called attention to the powerful (and still
growing) contribution of pictures and other iconic signs (including metaphor) to
semiosis, i.e. to our semiotic approach to and interpretation of reality.

Considering the wealth of electronic possibilities of creating information
in CMC, it is reasonable to argue for a broader reading of
which takes into
account both symbolic (mostly linguistic) and iconic signs. To this end, the
notion of
has been coined. It refers to a much wider concept than
indeed, it incorporates text as one of its components. I adopt Slatin's definition of
as "an assemblage of texts, images, and sounds – nodes – connected by
electronic links so as to form a system, whose existence is contingent upon the
computer. The user/reader moves from node to node either by following
established links or by creating new ones." (Slatin 1991: 56)
From this definition, we can deduce five constitutive features of hypertext:
computer mediation
, (b)
, (c)
, (d)
and (e)
(a) Computer-mediation: Hypertexts consist of digitalized, electronically
mediated bits of information. Anything that can be written, spoken, drawn,
filmed, etc can be turned into digital signs to be put on the Web.
This in itself is
not a feature of hypertext that impedes meaning- and text-making negotiation.
(b) Multilinearity: It is generally distinguished between
medial linearity
. Medial linearity depends on and is conditioned by the
in varying degrees. It is weaker in a newspaper, where the reader can
easily deviate from the order in which the items are printed; it is stronger in a
video or audio tape, where the medium restraints the viewer's or listener's choice
of order. Conceptual linearity is given by the
, who is merely suggesting in
which order the written material can be perceived; the binding factor is stronger
in a work of fiction (but cf. below on hyperfiction) than in a travel guide or a
dictionary. Both types of linearity, medial and conceptual, can also be transferred,
, to multilinearity. Of course, the internet and thus hypertexts are
particularly well suited for conceptual multilinearity, which in itself is an asset
rather than an impediment to meaning- and text-making negotiation.
(c) Multimediality, also often called multimodality: The term
It can either refer to the
, the material devices that carry
and transmit information, such as newspapers, books, radios, tv-sets, MP3-
players, computers as well as mobile devices such as pdas, cell phones and,
though less material and more virtual, the internet. Or it can refer to the
representation format
such as the spoken words (utterances, discourse), the
written texts, pictures, illustrations, drawings, graphics, layout, typography, film,
melody. The alternative term
as in
is usually applied to the
latter reading because it generally refers to the ways in which information
materializes. Multimodality (-mediality) in CMC is a scalar feature, reaching
from monomodality (-mediality), e.g. in text-only CMC, to multi-modality
(-mediality), which may comprise textual, visual and audio modes. The internet
and hypertexts are multimodal (-medial) because of their rich diversity of
representation formats. On occasion, it may be difficult for users to find their way
through the wealth of kaleidoscopic data in order to negotiate meaning and
(d) Fragmentation
is among the most noticeable features of internet sites.
Every node, picture, pop-up, hyperlink etc is a fragmentary informational and
thus communicative unit, and as such a challenge for the user to ascribe meaning
and textuality to. Like multilinearity and multimodality, interactivity is based on
the fragmentary assemblage of different text clusters. In hypertexts, e.g., texts are
interactively aligned across different nodes (internodal), while multimodality is
mirrored in the fragmentary combination of text units within one and the same
node (intranodal). Furthermore, it is generally distinguished between
fragmentation across modes (e.g. textual and pictorial node) and fragmentation
within one mode. The latter is a characteristic feature of chat communication.
'Sailing the islands or watching from the dock'
Here, comments are broken down into smaller bits of incomplete information (or
"messages", Beißwenger 2005), which have to be linked up by users with the help
of cohesive means such as types of address, backchannels, turn-taking signals,
cross-turn references.
(e) Interactivity is a scalar feature. At one end of a cline, we find self-
contained asynchronous e-documents lacking in interactive potential, since they
do not allow for online feedback or manipulation by the user. At the other end,
there are chat environments (instant messenger, ICQ, IRC, etc) and video-
conferences, which have a high degree of interactivity because they allow for
synchronous and near simultaneous feedback and manipulation. The user can
obviously choose from a much wider range of options to participate in the
production, alignment and negotiation of content. Between these two extremes,
there are intermediary types of interactivity, among them purely physical acts (in
which the user connects self-contained 'text' units by, e.g., simply clicking
hyperlinks) and purely cognitive acts (in which the user relates textual and audio-
visual nodes to each other across the screen).
There is one more constitutive and quite salient feature of hypertext, the
(hyper-) link
. Links relate structured pieces of information, i.e. nodes, in an
electronic and non-linear way.
They can be used to interfere with existing data
or to create new data, e.g. when submitting to weblogs or using search engines.
Links are mechanistic
and as such a significant and apparent
component of negotiation. They substitute for identifiable authors (giving
instructions). As such, they can even be described as a residual feature of a dyadic
orientation of CMC.
The loss of the 'other participant' or the erosion of the dyadic principle
These properties of hypertext explain why different users can (and regularly do)
create their own distinctive hypertexts which are different from other users'.
Using the link function, they can easily and freely (i.e. multilinearly) navigate
through a broad spectrum of modally different informational fragments that are
easy to handle, i.e. to shift around, to replace, to (re-)arrange and to manipulate.
In doing so, users create meaning by negotiating with 'the other' in the figurative
sense explained above. The decisive factor, which promotes and supports
negotiation in this way, is, of course, the interactive nature of the medium.
However, on closer inspection it is a very restricted kind of negotiation,
whose outcome somehow thwarts the promising possibilities of the new media.
Their insufficiencies when creating meaning and building own hypertexts are
readily addressed, let me pick out a few.
Users can, of course, re-arrange the nodes, i.e. the fragments of
information they are confronted with in CMC, even though they are not of their
own but (frequently) of some anonymous author's making. Given fragments and
newly created fragments, however, may not always be compatible as to their
topic, their evaluative load, their register and function. Other-authored given
fragments may, therefore, resist manipulation and thus impede the free creation of
a new hypertext.
By the same token, bringing together textual and pictorial nodes which
were not adjacent and related before can cause incoherence. The difficulty of
making them cohere is due to a lack of 'traditional' and familiar cohesive means.
In hypertext, the user is faced with a large number of nodes, which are frequently
self-contained. Accordingly, all the relevant and expected means of linkage are
often missing such as inter-node gambits and discourse markers, referring
expressions other than definite descriptions and generally known proper names,
tag questions etc. Users are familiar with the problems of scanning the screen for
signs and signals that help to relate textual and audio-visual nodes to each other.
While we are accustomed to the linguistic, non-linguistic and cognitive ways of
establishing cohesion in non-electronic spoken or written communication, users
have to learn what (other) means and strategies are used in CMC to relate current
items or nodes to preceding or prospective other items or nodes. This is at least
partly due to the interplay of various modes of presentation; the reiteration of
pictorial elements, e.g., can support semantic connectivity proposed by the verbal
structures in the text. Of course, refined and sophisticated types of hyperlinks
have been developed to serve as cohesive devices. They are of a different, i.e.
mostly non-linguistic nature and are thus parasitic on the multimodal form of
hypertext (cf. Hoffmann 2006).
Hoffmann (2007), e.g., points out that hypertext users often miss
"important evidence in hyperlink anchors needed for the appraisal of hyperlink
trails. In this respect, all too often hyperlinks are insufficiently marked by their
authors. Most forget that visualizing the content of target nodes is essential for
determining which hypertext path one will follow. Likewise, informing the user
about possible link trajectories is a central motive in maintaining the cohesive
foundation of hypertexts. For these reasons, more and more hypertext authors use
extensive, multimodal means (audio-visual signs) to provide users with
information about the "hidden content" of their hyperlinks." And he takes his
examples from the websites of several universities, which by now "have caught
on to the new iconic possibilities of web design", making ample use of "images or
photos in their online presentations":
'Sailing the islands or watching from the dock'
Figure 2: Multimodal link anchor in university website
Figure 3: Multimodal link anchor in university website
In general, it seems safe to claim that cohesion relies strongly on the particular
mode used: speech, writing or, indeed, CMC.

There is another reason why coherence building in hypertext is a
challenge. Unlike readers of 'old' printed text, CMC users cannot operate on a
default assumption of coherence. They cannot assume as a matter of course that
what they will read or see next (when activating a link)
coherent (cf. Bublitz
2005a, 2006). However, work to downsize the problem is in progress. As
Hoffmann (2007) points out, an interesting feature, recently introduced by the
company© and appearing on some weblogs "may have the potential to
bridge coherence breaks between websites. Once a freeware program is installed
successfully on the hard drive of a webpage owner, users can direct their mouse
cursor over a hyperlink, and an additional window will appear instantly delivering
an appropriate preview of the respective target area. The preview picture includes
a search engine which can be used for looking up concrete words or phrases
within the future website." Here is an example (courtesy of Hoffmann):
Figure. 4:© applet used in Jill Walker's weblog
As Hoffmann (2007) points out, "it is highly probable that these simple
applications could at least provide partial or preliminary solutions to the cognitive
overload which stems from forward-looking planning strategies of hyperreading"
(cf. also Bublitz 2005a: 321 f).
'Sailing the islands or watching from the dock'
A further difference between 'old' printed text and hypertext nodes is that
the latters' propositional meanings are often much more readily understood than
interpersonal meanings. This is not only due to a lack of relevant means of
emotive prosody, empathetic orientation, evaluative judgement, ideological
stance etc. As any piece of text is created by a particular person (or several
persons) and addressed to an intended reader, browser, user, it carries and
displays subjectivity to a greater or lesser extent. But unlike with printed text, say
a book, the probability is rather high that hypertext users who do not belong to
the circle of intended addressees, visit nodes whose interpersonal, subjective
impact they do not understand. The same holds for meaning 'between-the-lines',
insinuations and allusions.
Generally, there is a much higher demand on the empathetic, knowledge
inferring skills of the hypertext user than on those of the traditional reader of
printed text. Pronounced empathetic proficiency is necessary for users to establish
and develop
common ground
, which is an essential prerequisite for
comprehension (and as such aimed at in any negotiating process) (cf. Bublitz
2006). The insufficiencies related so far clearly indicate that the establishment
and maintenance of common ground is much more difficult and sometimes hardly
possible in CMC.
The user as sailor or as watcher
Where do we stand? I have argued that the defining properties of hypertext can
seriously interfere with the users' effort to ascribe meaning, coherence and
textuality to the kaleidoscopic flux of online textual and audio-visual fragments
of information, and to build and thus create own new hypertexts. Comprehension
can become a challenge, because it rests on empathetic acts of negotiating, which
require a real or assumed other interactant. The lack of a human other forces the
user to negotiate with, i.e. to make assumptions about the assumed author's
linguistic and world knowledge, episodic and conceptual memory, cultural
background and emotional frames of mind, in order to establish a common
ground as a prerequisite for understanding. In computer mediated
communication, the simple dyadic set-up of prototypical speaker-hearer or writer-
reader communication has been replaced by a set-up with the human self as user
as one collaborator and the internet with its wealth of data as the other
That semiosis in CMC can be more demanding than in printed texts or
spoken discourse, is, on the face of it, not obvious at all. But actually, the user has
to rely on his or her own interpretive skills, knowledge and experience to a much
greater extent than the overwhelming wealth of electronically based devices and
mediated data has us expect. Somewhat ironically, such a literally unlimited pool
of audio-visual data offered by the internet can be an impediment rather than an
asset for understanding.
Things may be different when we move from negotiating meaning to
negotiating text, i.e. building own hypertext. Here, both the coincidence of the
roles of user and author and the multitude and diversity of electronically provided
means are definitely an asset. It is certainly regarded as an asset in
where the user as reader is strongly invited to act as author by actively interfering
with the composition of a story. That hyperfiction gains by the user turning
author is nicely captured in a revealing metaphor (taken from the opening
directions to a piece of hypertext fiction):
The difference between reading hyperfiction and reading traditional
printed fiction may be the difference between sailing the islands and
standing on the dock watching the sea. (Guyer & Petry 1991)
There are obviously two sides to this metaphorical coin, which focusses on the
recipient's (i.e. the reader's or user-reader's) stance. Underlying the metaphorical
are the two presumptions that reading is passively
happening to a reader, i.e. is a state rather than an act, whereas sailing is a
dynamic activity by the sailor. By adding the specification "the islands" the verb
acquires even greater power because the valency of "sailing" changes from
transitivity ('sailing a boat') to complex-transitivity ('sailing a boat through the
islands'). By pairing the two concepts, reading is turned into an undertaking or
even a venture, which demands from the reader to constantly make decisions
about the course to be taken, i.e. about what to do next, and to actually do it. The
presupposed inactivity or passivity as a semantic feature of reading is, of course,
confirmed in the second metaphorical concept
and further
emphasized by the supplement "standing on the dock". The message of the two
creators of the metaphors is clear: Printed texts are just there, for the reader to
take them or leave them, to understand them or not, while hyperfiction is of the
reader's own building and thus read and understood in the act of creating. And
there is no doubt that in their eyes, the latter stance is preferable to the former.
The obvious next step is to transfer the metaphors of 'the user-reader as
'the reader as watcher' from building hyperfiction to the more
general realm of building non-fictional hypertext and also to the question of how
understanding is managed in hypertext and printed text. I believe that only the
'sailing'-side of the metaphor is apt. CMC users can indeed devise and control
hypertext in an author-like manner. They 'sail the islands' in the sense that they
actively handle the data provided by the internet in various ways. Briefly
recapitulating, when building their own hypertexts, users
'Sailing the islands or watching from the dock'
- employ prospective, forward-looking planning strategies, much like
do when composing printed text;
- interact with the data provided by assembling fragments of information,
i.e. nodes in an unforeseen, unforeseeable and ad hoc way (i.e. in a manner
reminiscent of making a collage);
- exploit the multilinear and multimodal nature of hypertext to create
unusual but nonetheless meaningful sequences of (textual and audio-
visual) bits of information, thus moving from exclusively monomodal to
multimodal semiosis.
I contend, though, that the other side of the metaphorical coin does not hold. Even
though the metaphor refers to 'reading', what is actually meant is the act of
'understanding while reading'. Neither reading nor understanding can be likened
to standing (a state of immobility) and passively watching. As I have argued
before, understanding is the act of ascribing meaning (coherence, function etc) to
(textual or other sign-based) informational input. And that applies regardless of
whether the understanding individual is looking at a book or a computer screen, at
monomodal and steady print or multimodal and moving texts and pictures, is
listening or reading, or, indeed, watching.
The authors of the metaphor insinuate that
is a state rather than
an act or process. In their view, watching
appears as a one-dimensional mode of
perception which regards the perceiver as passively receiving visual sensations
together with their meanings, which are somehow inseparably attached to them.
But such view does not hold. It is hard to imagine how 'watching the sea' (or 'the
islands') does not involve recognition and thus conceptualisation of the perceived
the sea. And conceptualising, i.e. understanding a linguistic object
involves more than merely matching it with its mental image listed in the lexicon.
Depending on the nature of the object, i.e. the linguistic expression,
conceptualising can be a highly complex act of meaning construal, and even more
so when strings of expressions, i.e. texts are involved as is normally the case
when reading.

What our metaphor does not cover is that watching can involve a higher or
a lower degree of cognitive effort, can be difficult or easy, depending on the
object watched. Applying this to understanding, it can be more difficult in some
forms of media, notably in CMC, and less difficult in others.
Conclusion: The loneliness of the user
This paper has started out with a recapitulation of two of those dogmas that have
widely been taken for granted in linguistics. Bringing them together, they state
that the dyadic nature of human communication is an indispensable precondition
for negotiating meaning, which is a dyadic, transitive and reciprocal act requiring
two interactants. The subsequent attempt to find proof in the new electronic
media for the universal validity of the two dogmas was not entirely successful.
They are no longer unrestrictedly valid for both 'old' (spoken and written) and
'new' electronic media. Some forms of CMC in particular have altered our
understanding of participation as a dyadic and focussed concept, and have also
made negotiating meaning and thus understanding more difficult.
It is the latter fact that may come as a bit of a surprise because the
development of the electronically administered new media with their infinitely
large quantity of audio-visual data was certainly not intended as an impediment
but as an asset for handling its possibilities and understanding its content. But the
extraordinary interactive potential of CMC is (at least partly) counteracted by the
extremely high degree of fragmentarization (with all its consequences as related
above). Wiki- and related CM media are thus to some extent deconstructive
However, this, I would like to argue, is no cause and certainly no need for
despair and lament. The multilinear, multimodal, fragmentary, intricate and
occasionally perplexing way of presenting information is by no means confined
to CMC, but is a totally familiar phenomenon to all of us. Even ordinary printed
text and discourse do not always transmit their messages in a linear, orderly,
explicit and straightforward way, which forces the hearer or reader to ascribe
order to disorder, to create his or her own linearity, to make fragments of
information cohere that are not cohesively connected, to infer the implicit from
the explicit, the additional from the given information, and all that in an
associative and occasionally roundabout way. Furthermore, the way users go
about ascribing meaning to a vast array of fragments, is suggestive of the
associative way human minds work when understanding. Contemplating can run
on different levels of modality simultaneously, can be extremely fragmentary
(like the outside-world) and quite unfocussed. Minds slip sideways. This
metaphor holds both in virtual reality, just think of Molly Bloom's mind slipping
sideways in her soliloquy (an often quoted long stream of consciousness passage
in James Joyce's
) and in the real world, where it can be applied to users
when moving on an unguided, self-constructed tour through the internet.
Despite its extraordinary possibilities, interacting with 'new' electronic
media does not
per se
guarantee easier understanding, i.e. an easier access to the
world 'behind the screen' than when interacting with 'old' printed media. The
user's situation is not
different from the familiar situation of the reader
who is trying to understand printed text. Like these readers, who have no one to
negotiate meaning with, we as users are on our own. When building fictional or
non-fictional text, we may be invited by the medium to sail or to surf with the
others, i.e. the community of authors, but when it comes to understanding, we are
on our own and as lonely as the 'old' readers sitting in their libraries surrounded
by thousands of books with no human interlocutors helping them to create their
own inner worlds.
'Sailing the islands or watching from the dock'
I wish to thank Jenny Arendholz, Volker Eisenlauer and especially Christian
Hoffmann for a number of valuable comments on an earlier version of this paper.
To reach a complete characterization of human communication, as, e.g.,
upheld by Bühler (1965) and going back to Plato, we need to add that
communication is '
something' and, of course, that it involves
as its
For a critique of the concept
in speech act theory and inference
theory cf. Clark and Carlson (1982) and Levinson (1988).
Cf. for a useful overview Levinson (1988); he presents and reviews a) the
traditional account (which is based on the grammatical distinction between
, 2
and 3
person, and relies heavily on the criterion of whether a
participant is present or absent), b) Shannon & Weaver's (1949)
communication model, and c) Goffman's (1981) theory (cf. below) before
introducing his own elaborate proposal (1988: 170 ff).
For an overview cf. Levinson (1988: 169 ff).
cf. Bolter (2001), and Eisenlauer and Hoffmann
Cf. Beißwenger (2005), Beißwenger (ed.) (2001), Hess-Lüttich and Wilde,
and Arendholz (2006).
Elsewhere (Bublitz 2006), I have juxtaposed and explained in more detail
the 'collaborative' or 'cooperative' and the 'autonomous' views of
comprehension, adopting Clark's (1992) terminology.
I.e., that creating meaning is autopoietic (literally 'self-creating'), to put it
differently. I take and adapt the term
from the system
theoretician Niklas Luhmann (1984), who defines a society as a social
system of communication.
In a wider sense, any
use of language can be taken to be a
means of negotiating meaning; for an overview of such use cf. Hübler and
Bublitz (2007). The following examples from the
London Lund Corpus
have been adapted; the remaining conventions refer to intonation, pauses (.
= brief pause, - = unit pause) and simultaneous talking (*… *).
Mitchell, who uses the term "pictorial turn" and Boehm, who talks of
"ikonische Wendung" (iconic turn) do, of course, take up Richard Rorty's
(1967) famous topos of the
(linguistic) turn
For an overview of hypertext definitions cf. Bublitz (2005a), Hoffmann
(2006), Huber (2002), Jucker (2002), Kuhlen (1991), Storrer (2000),
Some authors do not count this criterion among the defining criteria of
hypertext; as a consequence, they also apply the term hypertext to some
kinds of printed media such as encyclopediae and handbooks, cf. e.g.
Ansel Suter (1995), Bucher (1998), Kaplan (1995: 13).
Cf. Esser (2004), Kress and van Leeuwen (2001).
Also called "modularity", cf. Jucker (2003).
Nodes can be visual or aural, i.e. they can be read as text or seen as a
visual image and even heard.
This is why the term 'e-cohesion' is an apt description for electronically
mediated forms of cohesion in CMC.
That the user can literally (and not merely metaphorically) turn author is,
of course, primarily due to the interactive nature of CMC; after all,
hypertext has aptly been called "a medium for composition" and "not just
[…] a presentational device" (Slatin 1991a: 153).
Cf. for a similar account of
(to) see
both in a literal and a metaphorical
sense (as a metaphor of understanding) Bublitz (2005b).
'Sailing the islands or watching from the dock'
Ansel Suter, B. 1995.
Hyperlinguistics. Hypertext Lernumgebungen im
Akademischen Kontext: Eine Fallstudie.
Diss. Zürich: University of
Arendholz, J. 2006.
Kommunikative Unfälle in Chat-Gesprächen. Wie und warum
Online-Kommunikation misslingen kann.
MA-thesis. Augsburg:
University of Augsburg.
Beißwenger, M. 2005. 'Interaktionsmanagement in Chat und Diskurs', in: M.
Beißwenger and A. Storrer (eds.)
Chat-Kommunikation in Beruf, Bildung
und Medien: Konzepte - Werkzeuge - Anwendungsfelder
. Stuttgart: ibidem,
Beißwenger, M. (ed.) 2001.
Chat-Kommunikation. Sprache, Interaktion,
Sozialität & Identität in synchroner computervermittelter Kommunikation
Stuttgart: ibidem.
Berners-Lee, T.
Boehm, G. 1994. 'Die Wiederkehr der Bilder', in: G. Boehm (ed.),
Was ist ein
München: Fink. 11-38.
Bolter, J. D. 2001.
Writing space. Computers, hypertext, and the remediation of
. New Jersey: Mahwah.
Bublitz, W. 2005a. 'The user as 'cyberego': text, hypertext and coherence', in: L.
Moessner and C. M. Schmidt (eds.),
Anglistentag 2004 Aachen,
. Trier: WVT. 311-324.
Bublitz, W. 2005b. 'Seeing as a metaphor of understanding: the visible and the
invisible', in: A. Schuth, K. Horner and J. J. Weber (eds.),
Life in
language. Studies in honour of Wolfgang Kühlwein
. Trier, WVT. 135-149.
Bublitz, W. 2006. '
It utterly boggles the mind
: knowledge, common ground and
coherence', in: H. Pishwa (ed.),
Cognitive aspects of language and
. Berlin: Mouton de Gruyter.
Bucher, H.-J. 1998. 'Vom Textdesign zum Hypertext. Gedruckte und
elektronische Zeitungen als nicht-lineare Medien', in: W. Holly (ed.),
Medium im Wandel
. Opladen: Westdeutscher Verlag. 63-102.
Bühler, K. 1965.
Sprachtheorie. Die Darstellungsfunktion der Sprache
. Stuttgart:
Gustav Fischer. (First publ. in Jena, 1934).
Clark, H. H. and T. Carlson 1982. 'Hearers and speech acts',
58: 332-
Clark, H. H. and E. F. Schaefer 1992. 'Contributing to discourse', in: Clark: 144-
Clark, H. H. 1992.
Arenas of language use
. Chicago: UP.
Eisenlauer, V. and Chr. Hoffmann submitted, 'The metapragmatics of remediated
text design',
Information Design Journal
. Amsterdam: Benjamins.
Esser, J. 2005. 'Hypertext and taxonomies of text-types', in: L. Moessner and C.
M. Schmidt (eds.),
Anglistentag 2004 Aachen, Proceedings
. Trier: WVT.
Fritz, G. 1999. 'Coherence in hypertext', in: W. Bublitz, U. Lenk and E. Ventola
Coherence in spoken and written discourse: how to create it and
how to describe it.
Amsterdam: Benjamins. 221-232.
Goffman, E. 1981.
Forms of talk.
Oxford: Blackwell.
Graesser, A. C. and L. F. Clark 1985.
Structures and procedures of implicit
Norwood, NJ: Ablex.
Guyer, C. and M. Petry 1991.
Izme Pass
. Disk included in the magazine
on the Edge
. Reprinted in Coover, Robert
The End of Books
Herring, S. C. 1999. 'Interactional coherence in CMC',
Journal of Computer-
Mediated Communication
, 4(4).
Herring, S. C. (ed.) 1996.
Computer-mediated communication. Linguistic, social
and cross-cultural perspectives
. Amsterdam: Benjamins.
Hess-Lüttich, E. W. B. and E. Wilde. Undated. 'Der Chat als Textsorte und/oder
als Dialogsorte?'
Hoffmann, Chr. 2006.
Lost in hyperspace? Kohäsion und Kohärenz in
MA-thesis. Augsburg: University of Augsburg.
Hoffmann, Chr. 2007.
The conceptual origins of hypertext.
(Unpubl. ms.)
Augsburg: University of Augsburg, Dpt. of English Linguistics.
Huber, O. 2002.
Hyper-Text-Linguistik: TAH: ein textlinguistisches Analysemodell
für Hypertexte.
Diss. München: University of München.
Hübler, A. and W. Bublitz. 2007. 'Introducing metapragmatics in use', in: W.
Bublitz and A. Hübler (eds.).
Metapragmatics in use
. Amsterdam:
Benjamins. 1-26.
Humboldt, W. von. 1827/
1969. 'Ueber den Dualis', in:
Wilhelm von Humboldt.
Werke in 5 Bänden.
Bd. III:
Schriften zur Sprachphilosophie.
Wissenschaftliche Buchgesellschaft. 113-143. [First read in Berlin,
Akademie der Wissenschaften, 26 April 1827].
Jones, R. H. 2004. 'The problem of context in computer-mediated
communication', in: P. LeVine and R. Scollon (eds.),
Discourse and
Technology. Multimodal Discourse Analysis
. Washington, D.C.:
Georgetown UP. 20-33.
Jucker, A. H. 2002. 'Hypertextlinguistics: textuality and typology of hypertexts',
in: A. Fischer et al. (eds.),
Text types and corpora. Studies in honour of
Udo Fries
. Tübingen: Narr. 29-51.
Jucker, A. H. 2003. 'Mass media communication at the beginning of the twenty-
first century: dimensions of change',
Journal of Historical Pragmatics
Kaplan, N. 1995. 'E-literacies',
Computer-Mediated Communication Magazine
(3), March 1.
Kress, G. and T. van Leeuwen. 1996.
Reading images. The grammar of visual
. London: Routledge.
'Sailing the islands or watching from the dock'
Kress, G. and T. van Leeuwen. 2001.
Multimodal discourse: the modes and
media of contemporary communication
. London: Arnold.
Kuhlen, R. 1991.
Hypertext. Ein nicht-lineares Medium zwischen Buch und
. Berlin: Springer.
Lee, B. P. H. 2001. 'Mutual knowledge, background knowledge and shared
beliefs: their roles in establishing common ground',
Journal of Pragmatics
33: 21-44.
Levinson, S. C. 1988. 'Putting linguistics on a proper footing: explorations in
Goffman's concepts of participation', in: P. Drew and A. Wootton (eds.),
Erving Goffman. Exploring the interaction order
. Boston: Northeastern
UP. 161-227.
Luhmann, N. 1984.
Soziale Systeme. Grundriss einer allgemeinen Theorie
Frankfurt am Main: Suhrkamp.
Mitchell, W.J.T. 1994.
Picture Theory: essays on verbal and visual
. Chicago: UP.
Rorty, R. 1967.
The linguistic turn: recent essays in philosophical method.
Chicago: UP.
Shannon, C.E and W. Weaver. 1949.
A mathematical model of communication
Urbana, IL: University of Illinois Press.
Slatin, J. 1991a. 'Reading hypertext: order and coherence in a new medium', in: P.
Delany and G. Landow (eds.),
Hypermedia and literary studies
Cambridge, MA: MIT Press. 153-167.
Slatin, J. 1991a. 'Composing hypertext: a discussion for writing teachers', in: E.
Berk and J. Devlin (eds.),
Hypertext / hypermedia handbook
. New York:
Intertext Publication. 55-64.
Storrer, A. 2000. 'Was ist hyper am Hypertext?', in: W. Kallmeyer (ed.),
und neue Medien. Jahrbuch des Instituts für deutsche Sprache 1999.
Berlin: de Gruyter.
Storrer, A. 2002. 'Coherence in text and hypertext',
Document Design
3: 156-168.
Svartvik, J. and R. Quirk (eds.) 1980.
A corpus of English conversation
. Lund:
Tulving, E. 1972. 'Episodic and semantic memory', in: E. Tulving and W.
Donaldson (eds.),
Organization of memory.
New York: Academic Press.
Tulving, E. 1983.
Elements of episodic memory
. Oxford: OUP.
Linking the verbal and visual:
new directions for corpus linguistics
Ronald Carter and Svenja Adolphs
School of English Studies, University of Nottingham
This paper discusses an ongoing research project to investigate the compilation of a small
corpus and the development of appropriate software tools that enable a more multi-modal
approach to language data. The research draws on recent experience developed in the
development of spoken corpora to explore alignments of the verbal and the visual and, as
a starting point, does so with particular reference to gestures in communication and the
role of head nods in particular. Issues of appropriate data capture and description are
discussed alongside questions about the nature of language necessarily raised by language
research that goes beyond the textual.
Advances in the field of corpus linguistics over the past two decades have made it
possible to develop computerised multi-million word databases of spoken and
written language alongside powerful software tools to analyse this data
quantitatively and qualitatively, a development that has contributed to pioneering
research in many areas of communication studies and language description.
However, while the analysis of large-scale text corpora can provide insights into
language patterning and can help establish linguistic profiles of particular social
contexts, it is limited to the textual dimension of communication. Communication
processes are multi-modal in nature and there is now a distinct need for the
development of corpora that enable the user to carry out analyses of both the
speech and gestures of the participants in a conversation, and of how the verbal
and non-verbal complement one another. In other words, corpus linguistics and
discourse analysis might begin to be more closely aligned and descriptions made
of rich contexts of language use of the kind advocated and illustrated by Michael
Stubbs throughout his career.
Multi-Modal Communication
Recent work in multi-modal communication has seen advances in both theory and
practice. The theoretical starting point for much significant work has been
systemic-functional linguistics. Systemic linguistics is a theory that focuses on
meaning, choice and probability in language and on the significance of language
as a social phenomenon, underlining how particular choices of word, grammar
and structure encode different meanings in different contexts of language in use.
Carter & Adolphs
Foundational work in multi-modal communication such as Kress and van
Leeuwen (1996) has illustrated how choices of image can align with verbal
choices and this work has been extended in recent years to embrace the multi-
modal analyses of word, image and sound within different language varieties,
including cartoons, comics, film, information leaflets, maps, advertisements
(including TV advertisements), web pages and classroom textbooks (e.g. Baldry
and Thibault, 2004, 2006). The emphasis has been on how choices of one image
or camera angle or colour tone can cumulatively encode particular meanings. The
almost exclusive focus has been on written text.
A particular challenge for current research is therefore to integrate the
computer-enabled power of corpus linguistic methods, the theories and practices
of multi-modal linguistic research and, with particular reference to the analysis of
spoken discourse, the non-verbal signals of human gestures and bodily
communication. In other words, one key aim is to provide computerised analyses
of patterns of verbal and non-verbal meaning in ways that allow new
understandings of textuality to emerge.
What is a Gesture?
Human communication functions within a variety of direct and indirect 'semiotic
channels' (Brown, 1986: 409) which interact with, complement and 'counteract'
each other (Maynard, 1987: 590). The occurrence of such channels is affected by
modes of communication that differ widely according to their form, function and
context-of-use (see foundational work by Argyle, 1969 and Ekman and Friesen
1969, 1976) and more recent studies by Wilcox 2004 and Gu, 2006). However,
most studies have been undertaken within a research paradigm of psychology and
in experimental rather than naturalistic conditions.
To date, experimental studies of the multi-modal nature of discourse have
in general been designed to answer one or both of the following questions
(Kendon, 1994: 177):
1 If recipients are offered utterances which include gestures and if they are
permitted to see these gestures, do they interpret these utterances
differently than when they are not permitted to see them? (examples of
such studies include (Dobrogaev, 1929, reported in Kendon, 1980; Rogers,
1978; Riseborough, 1981).
2 If recipients are asked to make judgements about the gestures of others in
the absence of speech to which they were related, do they make such
judgements in a consistent way, and, if they do, do these judgements show
that they have some understanding of the utterance of which they were a
Studies of gesture and the multi-modal nature of communication have focused
upon gaze, (see Griffin, 2004 and Beattie & Shovelton, 1999, 2002) hand
movements (see Rimé & Schiaratura, 1991 and Thompson & Massaro, 1986),
head movements and other related gestures. In these studies the focus tends to be
on language use in experimental conditions and does not embrace spontaneous,
Linking the Verbal and Visual: New directions for corpus linguistics
natural conversation. In addition, such studies tend to be more concerned with the
gesture in relation to the basic content of talk, and do not explicitly explore the
links between specific forms of language and accompanying gestures.
Current gesture detection and recognition systems developed in computer
science within a tradition of automated vision recognition (see Nixon and
Aguado, 2002; Kapoor and Pickard, 2001)) often focus on precise, intentional
gestures. This is particularly true of hand gestures, where applications in sign-
language recognition and human-computer interaction mean that specific gestures
are made that are designed to be clearly distinguished by the observer. Gestures
made in authentic, face-to-face conversation, by contrast, are much fuzzier, their
form and meaning open to a greater degree of interpretation – a shake of the head
can, for example, indicate disagreement, disbelief, or confusion, creating
particular challenges for automated analysis of conversational gesture. Gestures
are unlikely to be uniquely identifiable and interpretation will need to take into
account other cues, such as the current role of the gesturer (speaker/listener) and
the co-text of the conversation (i.e., what occurs before and after a sequence of
gesture and talk). Furthermore, intentional gestures arise in a more constrained set
of situations than conversational gestures. As a result, image sequences are
usually acquired from a small, and known, set of viewpoints. Most intentional
gesture recognition systems assume that a lone participant is in clear view, facing
the camera from a short distance away. Many also assume the background to be
uniform and fixed. Real conversational gestures arise in a wide variety of
situations and involve dynamic activities from a variety of viewpoints and
distances and include multiple participants, cluttered backgrounds and other
moving people and objects.
However, for a corpus of gestures to be developed a record of the image is
required and current computer technology provides one of the best available
means of capturing such images digitally. The next sections report on a corpus-
based project to investigate such a phenomenon with a focus on naturally
occurring interactive two-party discourse.
Headtalk: an outline
is the first step in a project based at the University of Nottingham,
involving interdisciplinary research between applied linguists and computer
scientists, (in particular experts in vision recognition). The project aims to
combine both linguistic expertise and new computational techniques and
applications to provide the knowledge, research tools and procedures for
exploring the behaviour of some salient gestures in naturalistic conversation. An
initial focus on head nods was selected on account of their significance in
project team has collected to date (January, 2007) five hours
of video data, all based on face-to-face conversational episodes involving native
English speaking academics and students based at the University of Nottingham.
Carter & Adolphs
The participants were filmed face-on, in close proximity to the cameras in order
to create high quality, high resolution images, but were filmed in such a way as to
minimise the interference and invasiveness of the recording equipment, to make
the participants feel at ease and comfortable in the environment and to allow for
(relatively) natural, authentic communication. This data can be properly described
as 'multi-modal', as the transcribed recordings provide three different modes of
discourse, offering three separate streams of data for analysis: the audio, the
visual and the textual.
Utilising computer vision technology
The project utilises research in computer vision technology to allow the research
team to detect, recognise and extract descriptions of head nod movements. For the
detection and extraction of these movements, a variety of techniques were tested
on significant samples of the data. After numerous evaluations, a head tracker
was developed which can be placed upon the face of image data. Successions of
movements can then be monitored and matched to the basic up-down sequence of
a head nod in order to define where the movements occur, with the head tracker
tracking movement in the videos. The headtracker allows multiple targets to be
tracked in parallel, producing a description of the motion of each and showing
intermediate results as they are obtained. (For further description of the tracker
used (Cvision) see the Acknowledgements to this paper).
Head nods are vital for conversational maintenance and management (McClave,
2000) and often function as a form of 'back-channel' (Yngve, 1970), that is, a
'mechanism used for feedback' in discourse (Allwood et al, 1992), involving a
strategy which involves a form of 'minimal response', a way for the listener to
communicate that they have heard and perhaps understood a speaker's message,
while allowing the speaker to continue talking. Although there has been research
into and analysis of verbal back-channels, for example minimal responses or
'vocalisations' such as
, (Gardner (1998, 2002) integrated
explorations of the verbal and visual components of head nod behaviour of this
nature are limited. Preliminary linguistic analyses and classifications of each
stream of data, (i.e. the transcribed text of the talk, as well as the audio and the
video) was undertaken to determine patterns that may occur both within and
across each data stream. The findings were then compared with the computational
image analyses to define basic parameters for this particular gesture.
Coding back-channels
One of the key areas of concern of this project is how the head nods should be
encoded. In terms of verbal realisations of back-channels most existing schemes
focus upon grouping these in terms of their functions in discourse. This is a useful
point of categorisation as every back-channel has a function in discourse, even if
it may be unconscious to the interlocutor. Indeed, a wealth of research exists
which agrees that 'back-channels have more than one macro function' (O'Keeffe
and Adolphs forthcoming) as defined below (see also Schegloff, 1982; Maynard,
Linking the Verbal and Visual: New directions for corpus linguistics
1987, 1990). As a guide to the key functions, the framework provided by
O'Keeffe and Adolphs, has been adopted in the Headtalk project:

: Maintaining the flow of discourse (see Schegloff,
Convergence tokens
: Marking agreement and disagreement
Engaged response tokens
: High level of engagement, with the
participant responding on an affective level to the interlocutor.
Information receipt tokens
: Marking points of the conversation
where adequate information has been received.
While this basic categorisation can be a useful starting point in analysing verbal
realisations of back-channels, the question of how verbal and visual realisations
interact within and across such categories has remained largely under-explored.
For example, a back-channel such as
I see
can be
accompanied by a continuum of possibilities ranging from minimal head gesture
to an emphatic nod of significant duration. And duration can also comprise
several smaller nods within the same unit and still be linked to the same verbal
token. Much depends on how an interlocutor is responding, whether he/she is
simply maintaining contact or is signaling something altogether more engaged
and involved. It is not just verbal form or duration that are significant but such
factors as pitch and intensity govern how the form is interpreted and coded in
relation to its verbal counterpart. The relationship is a complex and elusive one
and a definitive coding scheme is still very much in development and will be
extended beyond this phase of the project.
For ease of transferability and consistency the data involved only native English
language speakers taking part in 45-60 minute PhD supervision sessions. This
meant factors such as intra and cross-cultural differences, which can potentially
influence the way in which individuals gesture or signal feedback, were
For the recording of the video participants were required to face each
other, with 4 cameras angled towards them and two microphones situated on the
floor between them. These images are displayed in a split screen and have been
positioned to ensure that they provide the highest quality images possible (see
figures 1 and 2).
Carter & Adolphs
Figure 1: Setup for video recording
In order to keep the images as large as possible, the following screen split for
capture was decided upon (figure 2):
Figure 2: Screen shot example of data collected with the modified recording set-
Linking the Verbal and Visual: New directions for corpus linguistics
Transcription of the data also created challenges as no universal, standardised
transcription conventions exist for this type of data. Therefore, for continuity, the
conventions used in the CANCODE (Cambridge and Nottingham Corpus of
Discourse in English) corpus based at Nottingham University (http://www.- were reapplied here for the purely
verbal component with Transana used to allow time stamping and annotation of
synchronised video and audio data streams (
Following the data collection and rendering of video to display both participants
as shown in figure 3, the next step is to develop a coding scheme for verbal and
visual signals of active listenership in the data. The main aims of the development
of the coding scheme are as follows:
- Defining and classifying verbal back-channel behaviour
- Defining and classifying non-verbal back-channel behaviour
- Combining verbal and non-verbal classifications and highlighting the
potential for exploring patterns and relationships between the two
In relation to the coding phase methodological approaches for each of these
processes needed to be closely explored. In order to create a new coding scheme,
preliminary linguistic and gestural analyses and classifications of the data were
undertaken. The findings were cross-compared in order to define basic
parameters for gesture-in-talk for use as a corpus coding scheme.
The basic linguistic functions that were used in the analysis of the
transcript are those outlined above (O'Keeffe and Adolphs, forthcoming). The
analysis of head-nods, on the other hand, is based on classifications established
with the use of computer-vision techniques. Five broad types of head-nods were
identified in our training data:
Type A: small (low amplitude) nods with short duration
Type B: small (low amplitude), multiple nods with a longer duration than type 1
Type C: intense (high amplitude) nods with a short duration
Type D: intense and multiple nods with a longer duration than type 3
Type E: multiple nods, comprising of a combination of types 1 and 3, with a
longer duration than types 1 and 3.
Using the functional categories, as well as the head nod classifications above, we
carried out a preliminary analysis of a 10 minute stretch of video extracted from a
longer MA supervision session. The participants in the session are a male
supervisor and a female student, both of whom are British. The extract was taken
from the middle section of the supervision, between 15.00 and 25.00 minutes.
The data was transcribed and annotated (see below). The overall word-count of
the transcript is 2156 words, of which 1401 words were uttered by the supervisor.
For the purpose of illustrating the coding scheme we will focus here only on the
description of back-channels used by the supervisor.
Carter & Adolphs
The supervisor uses 40 verbal back-channels in total, of which 18 are
accompanied by a nod and 22 are purely verbal. In addition the supervisor uses
24 nods which are not accompanied by a verbal signal. Thus, the supervisor uses
42 head nods, 18 of which are accompanied by a verbal signal.
So far, linguistic research has focused mainly on the classification of verbal back-
channels as outlined above. When we consider the discourse functions of the 40
verbalised back-channels used by speaker 1, the following breakdown emerges:
Continuer: 11
Convergence Token: 9
Information Receipt Token: 14
Engaged Response: 6
In order to analyses the interface between verbal and visual, we have, as a second
step classified the head-nods of the supervisor according to the different criteria
(amplitude and duration) that led to the five head-nod types outlined above. Our
analysis of the different types of head-nods used by the supervisor generates the
following results:
Type A: 13
Type B: 13
Type C: 12
Type D: 2
Type E: 2
We are particularly interested in this analysis to see whether any patterns emerge
in those instances where a verbal back-channel is accompanied by a nod. This is
the case in 18 of the back-channels used by the supervisor. In terms of linguistic
functions and head-nod types the 18 instance are categorised as shown below:
Continuer: 4
Convergence Token: 4
Information Receipt Token: 3
Engaged Response: 7
Type A: 6
Type B: 4
Type C: 6
Type D: 1
Type E: 1
An analysis of back-channel functions as coded with the use of the linguistic
coding scheme in relation to the type of nods that co-occur with the different
functions highlights a number of interesting trends. Half of the small nods of
short duration (type A) co-occurred with the information receipt function, while
half of the small nods of longer duration (type B) co-occurred with the function of
Linking the Verbal and Visual: New directions for corpus linguistics
a convergence token. All of the type C nods (i.e. short and intense nods) used by
the supervisor are accompanied by a verbal signal that has been classified as
carrying either the continuer or convergence function. Overall, it is important to
take a discourse level perspective to this kind of analysis, as preliminary
inspection of the data suggests that some of the functions of head-nods can be
aligned with the place at which they occur, i.e. where they are placed vis-à-vis the
main speaker's utterance.
These are preliminary results and more data needs to be analysed to see
whether there is any stable relationship between head-nods and linguistic signal.
However, this very brief illustration of the different coding schemes highlights
the need for an integrated analysis of verbal and visual, as the functions of back-
channels are modified through the use of head-movement, and it remains to be
established whether this modification is one of degree or of kind. One of the main
challenges of multi-modal corpus analysis and representation is that corpus
linguistics has traditionally focused on discrete items, such as individual words or
grammatical categories. The complexities of gesture and movement, on the other
hand, mean that they might not be able to be studied alongside traditional corpus
linguistic units of analysis in a straightforward manner. Baldry and Thibault
(2006: 181) point out that it is 'critically important [..] that corpus-based
approaches to text engage with the level of discourse analysis and discourse-level
meaning relations on various scalar levels of textual organisation'. While the
integration of scalar levels and discrete categories is likely to cause problems in
the development of an integrated framework, it also promises to lead to a much
richer description of patterns in social interactions.
Coding the Data: An emerging replay tool
As we have seen, the primary challenge for developing support for analysis of
multi-modal corpora is one of developing tools that provide an
approach to the representation of data. In general, there is a need to create tools
that support the 'marking up' or identification of multi-modal patterns and the
subsequent codification of recognizable patterns. Coding schemes for marking up
textual records and verbal aspects of talk already proliferate. However, there is a
paucity of such schemes for handling non-verbal elements: gestures, facial
expressions, gaze, head and body movement, posture etc.
Existing tools do not generally support the extraction of linguistic patterns
and thus fail, for example, to enable links between different types of listenership
and accompanying head movements to be established. There is a need to develop
new tools from the ground up to support linguistic analysis and, as an initial step
towards this, and by means of developing concrete requirements for technical
support, we have sought to exploit an emerging Digital Replay System (DRS)
that has been developed to support ethnographic inquiry (Crabtree et al. 2005;
French et al, 2006). The Digital Replay System provides some limited
mechanisms of representation and below we consider both their potential and
limits as a basis for articulating future requirements.
Carter & Adolphs
Figure 3: Digital Replay System
The Digital Replay System allows video data to be imported and a digital record
to be created that ties sequences of video to a transcribed text log, accompanied,
where appropriate, by samples of data that are also tracked by the adopted C-
vision recognition system (indicated in blue circles in Figure 3). The text log is
linked by time to the video from which the transcript is derived so that the text
log plays alongside the video. Further annotations can be added to the log to show
where gestures – head nods in this case – occur and these annotations are also tied
to the video. An index of annotations is produced and each can be used to go to
that part of the log and video at which they occur. The annotation mechanism
provides an initial means of marking up multi-modal data and of maintaining the
coherence between spoken language and accompanying gestural elements.
The data in the Digital Replay System is presented as a continuous, linear
sequence of communication. Yet within any sequence a substantial number of
utterances and gestures made by speaker and hearer overlap. This means that the
representation of gestural patterns can appear somewhat disjointed, as there is no
way at present to represent overlaps. The result of this is that it appears that head
nods last only for a specific time and only occur between two verbalisations,
which is inaccurate and misleading, as a nod may start after one verbalisation and
continue for a long period into the next. Head nods are, in short, variable. They
are not fixed in length and, given the limitations of the current incarnation of the
Digital Replay System, are difficult to code as events occurring over time and not
at particular moments in time. There is, then, a need to develop ways in which
their occurrence across utterances can be time stamped and marked out. One way
Linking the Verbal and Visual: New directions for corpus linguistics
in which this might be achieved is to represent the utterances and gestural
patterns made by parties to a conversation in individual transcripts, so to develop
a 'textured text' consisting of separate layers. However, this is not necessarily an
ideal solution since head nods need to be represented in relation to the behaviour
of both the speaker
the listener. Here it is important to identify the defining
parameters of the visual aspect of the particular gestural episode and align this
with the verbal realisation that may or may not coincide with it.
While there is necessarily a level of interpretation in transcribing spoken
discourse, the textual element of the head nod episode is relatively easy to
establish and patterns of common back-channels can be extracted from existing
multi-million word corpora. These types of patterns have been linked to particular
functions in the area of corpus linguistics. For example, minimal verbalisations
such as "mhm" have been linked to a continuer function while certain multi-word
units such as "that's right" have been linked to an agreement function. Yet, the
accompanying head movement, as well as the intonation pattern, can change the
function of the back-channel realisation, which in turn will affect the surrounding
discourse. It is therefore important to be able to establish some way of
recognising the visual elements in a
principled way
so that these can be studied in
relation to the verbal elements without adding a burdensome layer of
interpretative intervention to the initial representation of the data. There is, thus, a
need to marry vision recognition tools to machine learning techniques to reduce
the overhead of interpretive work and have these tools and techniques work
across utterances
to adequately represent the verbal-visual character of multi-
modal back-channels.
There is thus a need to
marry visual coding schemes to verbal coding
, which may then be exploited by machine learning techniques to codify
recognizable multi-modal patterns. In terms of using corpus linguistic techniques
to analyse patterns in language it becomes even more important that recognizable
patterns are consistently coded with reference to an agreed and replicable coding
scheme. If this is not applied throughout the corpus, any searches for patterns will
inevitably fail as they rely on the recurrence of consistent representations of
linguistic phenomena. Furthermore, coding schemes need to be developed in such
a way that that they can be shared across different research communities with
different community cultures and different representational and analytical needs.
In such circumstances, as analytical categories are re-classified in the light of new
audio-visual evidence and as new insights from different research communities
emerge, coding schemes need to be maintained as dynamically as possible.
The final concern of the corpus development is related to the way in which the
multiple streams of coded data are physically re-presented. Current corpora utilise
concordance tools. At the click of a button, appropriate citations of speaker
information, context of use and evidence of the specific conversation in which
each instance occurs, are easily available. With a multi-media corpus it is more
difficult to exhibit all features of the talk simultaneously. If all characteristics of
Carter & Adolphs
the specific instances where a word, phrase or coded gestures (in the video) occur
are displayed, the corpus would involve multiple windows of data with, for
example, 1000 instances of a head nod with an associated audio track of a
verbalised back-channel and the textual rendering of such (as seen in figure 4).
This would make the corpus confusing and impractical.
Figure 4: Representing concordances of multi-modal linguistic corpora
The basic solution to this problem is, as with current corpora, to present the data
in a 'textured' way and integrate relevant information, such as the codes and
further annotations, layering it behind main frames that display the key search
features in a similar way to current textual concordances. This would involve
marking up the transcript data with relevant information, such as codes for
different gestures or indeed with information on the function of each gestures as
well as corresponding speaker and time stamps, whilst linking it to other frames
of information.
When, for example, the code
is selected in the corpus, the user
will have access to the video and accompanying audio. Indeed such features may
be relatively straightforward when just marking up single gestures (this has been
the basic method used so far in our explorations of the supervision data), but, if
one were to mark up additional features, this would become even more complex,
especially when 'reading' concordances of multiple sources of data. So with
searches of the visual and audio information it is difficult to 'read' multiple tracks
of such data simultaneously, as is the case with current corpora and text. Our aim
is to create a balance between the amount of texture in the corpus, i.e. the
complexity and amount of information held in the corpus, and its ease of use.
This is still very much under development.
Linking the Verbal and Visual: New directions for corpus linguistics
Future Research Priorities
There are a number of lines of research arising from this project that require
investigation in the future. These include technical issues such as the
development of a recognition system to operate over the tracking data, and issues
of scope, such as the analysis of other gestures and the analysis of coupled
gestures and linguistic accompanying signals, such as hand movements
performed in parallel with head gestures.
has allowed us to gain a better
understanding of how we may describe and represent multi-modal language data
but has also generated a set of additional pertinent research questions in the
process. In addition to those outlined above, these also include theoretical
questions of how gesture and language integrate and whether they can be
described within a single framework. Major theoretical questions in this
connection include consideration of the extent to which gestures may be said to
be a language in the sense understood of language as a verbal medium. For
- Do gestures have rules and if so, how are the boundaries drawn?
- Do gestures have a syntax, that is, are they syntagmatically and/or
paradigmatically organised. Or do they not conform to such structuring?
- If the relationship between language and image can be modally connected,
as argued by theorists within a systemic linguistic tradition, and if images
can be interpreted according to paradigms of choice, is the same true for
gestures and for the relationship between human gestures and language?
- Is a system that is different to a linguistic system and are different
underlying theories needed to account for the sheer multiplicity of
different gestures?
- Many possible instantiations of headnods have been reported in this paper.
What happens when researchers begin to try to explain the many possible
meanings of hand gestures and their different cultural manifestations?
- What about 'body language' in the sense of movements encoded
interactionally by proxemics?
Another important priority for future research in this area is the development of
tools and methods to address ethical issues; for example, to anonymise video data
while still being able to extract the salient features that are the focus of the
analysis. Pixellating faces or using shadow representations of heads and bodies
can blur distinctions between gestures and language forms and, when taken to its
logical conclusion, anonymisation should also include replacing voices with
voice-overs and with other speakers. Ethical considerations of re-using and
sharing contextually-sensitive video data as part of a multi-modal corpus resource
need to be addressed further in consultation with end users, informants,
researchers and ethics advisors. The issues are especially acute when tools are
shared or are developed to be web-enabled.
project complements core strands of work to be carried out
by the e-Social Science Node at the University of Nottingham (see http://www.- As an extension to
, the Digital
Carter & Adolphs
Record project, hosted in the e-Social Science Node (see http://www.ncess.- allows for conversational gesture
recognition and mark-up of a wider range of different gestures, from hand
movements to gaze and facial expressions. This will enable researchers to start to
'complete the picture' of communication, to allow them to think about and explore
communication from a variety of different perspectives, something for which
has endeavoured to provide the ground.
Natural language is a focus of a diverse range of disciplines and the continued
explication of its real world, real time organization is of broad interest. The
impetus towards multi-modal corpora recognizes that natural language is an
embodied phenomenon and that a deeper understanding of the relationship
between talk and bodily actions, particular gestures, is required if we are to
develop more coherent understandings of the collaborative organization of
communication (see also Saferstein, 2004).
Core requirements towards meeting this goal include the development of
machine-based techniques that enable all visual and verbal patterns to be aligned
and enable common multi-modal patterns to be recognized. There is also a
pressing need to integrate visual and verbal coding schemes and to develop
techniques whereby these analytic schemes can be exploited in machine learning
environments to codify recognizable multi-modal patterns in large corpora of
data. In order to achieve these developments we need to gain a better
understanding of the particular requirements for recording, representing and
replaying each of the different modes, and the research presented in this paper
outlines some of the issues associated with this process.
The aim of this paper has been to begin to explore approaches that allow
researchers simultaneously to review and analyse video, audio and textual records
of naturally occurring communication. Such tools have the potential to provide a
major resource for researchers in the field of applied linguistics and
communication studies, film studies and drama in performance as well as in the
field of face-to-face and remote human interaction. The development of research
in this domain can also subsequently be extended to include pedagogic
applications in the analysis of cross-cultural communication for modern foreign
language learning as well as in professional discourse analysis, thus reinforcing
the essentially interdisciplinary potential of applied research of which Michael
Stubbs' work has been an exemplary instance.
The research on which this article is based is funded by the UK Economic and
Social Research Council (ESRC), e-Social Science Research Node
(, and the ESRC e-Social Science small
Linking the Verbal and Visual: New directions for corpus linguistics
grants project
(Grant N
RES-149-25-1016). Thanks are also due to
Dawn Knight for providing copy for research reports which have in part at least
formed the basis of this paper. All research reports so far are available via the
National Centre for e-Social Science
The HeadTalk demonstrator, Cvision, is an interactive program which
allows users to apply the visual tracking algorithm developed within the project
to selected targets in an input video clip. Cvision takes as its input an avi format
video file and produces a text file giving the estimated position of each target in
each frame of that video. This may be imported into the current version of the
DreSS 2 tool DRS. An output video may also be produced, if desired. This shows
the results of tracking overlaid on the input video images and is a useful
debugging and interpretation tool. Cvision allows multiple targets to be tracked in
parallel, producing a description of the motion of each and showing intermediate
results as they are obtained. Cvision is written in C++ and provided as a Windows
.exe file via User documentation
is also provided, and incorporates full details of the algorithm employed.
Allwood, J., J. Nivre & E. Ahlsen 1992. On the semantics and pragmatics of
linguistic feedback.
Journal of Semantics 9
: 1-26.
Argyle, M. 1969.
Social Interaction
. London: Methuen.
Baldry, A. and P. Thibault 2004.
Multimodal Transcription and Text Analysis
London: Equinox.
Baldry, A. and P. Thibault 2006. Multimodal corpus linguistics, in
System and
(eds Thompson and Hunston) pp. 164-183, London: Equinox.
Beattie, G. & H. Shovelton 2002. What properties of talk are associated with the
generation of spontaneous iconic hand gestures?
British Journal of Social
Psychology 41
, 403-417.
Beattie, G.W. & H. Shovelton 1999. Do iconic hand gestures really contribute
anything to the semantic information conveyed by speech? An
experimental investigation.
Semiotica 123
, 1/2: 1 – 30.
Berger, K.W. & G.K. Popelka 1971. Extra-facial gestures in relation to speech-
Journal of Communication Disorders 3
, 302 - 308.
Baldry, A. and P. Thibault 2004.
Multimodal Transcription and Text Analysis
London: Equinox.
Baldry, A. and P. Thibault 2006. "Multimodal corpus linguistics", in
System and
(eds Thompson and Hunston) pp.164-183, London: Equinox .
Brown, R. 1986.
Social Psychology
ed.). New York: Free Press.
Brundell, P. & D. Knight 2005.
Current Research and Tools to Support Data
Intensive Analysis for Digital Records in e-Social Science.
report. University of Nottingham.
Carter & Adolphs
Crabtree, A., A. French, C. Greenhalgh, S. Benford, K. Cheverst, D. Fitton, M.
Rouncefield and C. Graham 2005. "Developing digital records", DReSS
Working Paper 2, e-Social Science Research Node: University of
Ekman, P. & W. Friesen 1969. The repertoire of nonverbal behaviour: Categories,
origins, usage and coding.
Semiotica 1
: 49-98.
Ekman, P. & W. Friesen 1976. Measuring facial movement.
Journal of
Nonverbal Behaviour
1, 1.
French, A., C. Greenhalgh, A. Crabtree, M. Wright, P. Brundell, A. Hampshire
and T. Rodden 2006. Software Replay Tools for Time-based Social
Science Data. Paper delivered at the
annual international e-Social
Science conference
, June 2006, University of Manchester.
Gardner, R. 1998. Between speaking and listening: The vocalisation of
Applied Linguistics, 19,
Gardner, R. 2002.
When Listeners talk: Response tokens and listener stance
Amsterdam: John Benjamins.
Goldin-Meadow, S. 1999. The role of gesture in communication and thinking.
Trends in cognitive sciences 3
, 11: 419-429.
Griffin, Z.M. 2004. The eyes are right when the mouth is wrong.
Science 15
, 12: 814.
Gu, Y. 2006. Multimodal text analysis: A corpus linguistic approach to situated
Text and Talk 26
, 2: 127 – 167.
Kapoor, A. & R.W. Picard 2001. A real-time head nod and shake detector.
International Conference Proceedings Series.
Kendon, A. 1980. Gesticulation and speech: Two aspects of the process of
utterance. In M.R. Key (Ed.).
The Relationship of Verbal and Nonverbal
The Hague: Mouton. pp. 207-227.
Kendon, A. 1994. Do gestures communicate? A review.
Research on Language
and Social Interaction
, 3: 175-200.
Knight, D., S. Bayoumi, S. Mills, A. Crabtree, S. Adolphs, T. Pridmore & R.
Carter 2006. Beyond the Text: Construction and Analysis of Multi-Modal
Linguistic Corpora. Paper delivered at the
annual international e-
Social Science conference
, June 2006, University of Manchester.
Kress, G and T. van Leeuwen 1996.
Reading Images
Routledge, London.
Lock, A. (ed.)
Action, gesture and symbol: the emergence of language
, London:
Academic Press.
Maynard, S.K. 1987. International functions of a nonverbal sign head movement
in Japanese dyadic casual conversation.
of Pragmatics 11
, 589-
Maynard, S.K. 1990. Conversation management in contrast: listener response in
Japanese and American English.
Journal of Pragmatics
, 397-412.
McClave, E.Z. 2000. Linguistic functions of head movements in the context of
Journal of Pragmatics 37
, 7: 855-878.
Nixon, M. and A. Aguado 2002.
Feature Extraction and Image Processing
Oxford: Elsevier.
Linking the Verbal and Visual: New directions for corpus linguistics
O'Keeffe, A. & S. Adolphs forthcoming. Using a corpus to look at variational
pragmatics: Response tokens in British and Irish discourse. in K.P.
Schneider and A. Barron, ed.,
Variational Pragmatics
. Amsterdam,
Netherlands: John Benjamins.
Rimé, B. & L. Schiaratura 1991. Gesture and speech. In Feldman, R. & Rimé, B.
Fundamentals of nonverbal behaviour
. Cambridge: Cambridge
University Press. pp. 239-281.
Riseborough, M.G. 1981. Physiographic gestures as decoding facilitators: Three
experiments exploring a neglected facet of communication.
Journal of
Nonverbal behaviour 5
, 172 - 183.
Rogers, W.T. 1978. The contribution of kinesic illustrators toward the
comprehension of verbal behavior within utterances.
Communication Research, 5,
Saferstein, B. 2004
Digital technology and methodological adaptation: Text on
video as a resource for analytical reflexivity.
Journal of Applied
Linguistics 1
(2): 197–223.
Schegloff, E. 1982. Discourse as interactional achievement: some uses of "uh
huh" and other things that come between sentences. In D. Tannen (Ed.),
Analyzing discourse, text, and talk
. Washington, DC: Georgetown
University Press. 71-93.
Thompson, L.A. & D.W. Massaro 1986. Evaluation and integration of speech and
pointing gestures during referential understanding.
Journal of
Experimental Child Psychology, 42,
Wilcox, S. 2004. Language from gesture.
Behavioral and Brain Sciences 27
, 4:
Yngve, V. 1970. On getting a word in edgewise. In
Papers from the 6
Meeting, Chicago Linguistic Society.
Chicago: Chicago Linguistic
The novel features of text. Corpus analysis and stylistics
Henry G. Widdowson
'These are only hints and guesses,
Hints followed by guesses.'
The Dry Salvages
This paper takes up the problematic stylistic issue that Michael Stubbs addresses in his
study of Conrad's Heart of Darkness of the relationship between the analysis of a literary
work and its interpretation. Inspired by his example, and applying his 'quantitative stylistic
methods', I go in search of textual patterns and connections in the text of Conrad's novel
other than those he has noted, and consider what possible significance they might have.
The findings I come up with reveal features of the text that I would not otherwise have
consciously noticed. Whether these simply serve as explicit confirmation of a subliminal
literary awareness, or prompt new interpretative possibilities, is an open question. But
there is no direct correlation between textual findings and literary effects. The precision of
the analysis of the text does not lead to any greater precision in the interpretation of the
novel, but on the contrary leads to a heightened recognition of the necessary variability
and elusiveness of literary significance.
Of the many insightful enquiries that Michael Stubbs has undertaken into the
significance of corpus analysis for our understanding of language, that which, for
me at least, is the most intriguing concerns its contribution to literary stylistics.
Stylistics claims to provide linguistic substantiation for the interpretation of
literary texts. Since corpus analysis is
par excellence
a means of revealing textual
features in precise detail, it seems reasonable to suppose that it must be relevant
to the stylistic enterprise.
One advantage of corpus analysis by computer is that it can be so
comprehensive in its coverage of the textual facts: it can yield a quantitative
account of the recurrence and co-occurrence of all the words in a text. It is,
however, precisely because it provides such detailed information that it brings
into particular prominence the criticism that Stanley Fish levelled at stylistics in
general, long before corpora and computers came on the scene (Fish 1973/1996).
As Stubbs points out in a recent article (Stubbs 2005), Fish charges stylistics of
being circular and arbitrary in that it presupposes relevance in advance. The
analysis either selects literary features that are deemed to be significant and then
adduces linguistic features to substantiate their significance, or it selects linguistic
features and then claims that they are of literary significance. In the pre-corpus
period, stylistics is particularly vulnerable to the first charge: generally speaking,
what directed the selection of linguistic features was some impressionistic sense
of literary significance. It worked from the literature to the language. With corpus
analysis, however, we have the possibility of working in the other direction. Now
that we have the linguistic facts of texts available to us in such comprehensive
detail, we are in a position to make inferences from them about their literary
significance. We can at least be certain about the linguistic facts. The problem of
relevance, however, remains, and indeed becomes more difficult precisely
because we have so much linguistic information to deal with. How do we decide
what to select as significant?
This problem is both explicitly addressed and exemplified in the article
already referred to, in which Stubbs applies methods of quantitative stylistics to
Conrad's well-known short novel '
Heart of Darkness
This article is a fascinating
exercise in corpus analysis which reveals textual facts which are likely to be
unknown even to those readers who know the novel well. They certainly came as
a revelation to me.
But the article is of particular interest because the analysis raises more
general issues about text interpretation – about linguistic facts and literary
significance, about the limits of analysis, about the Fish dilemma. What I intend
in this contribution to this volume in honour of Michael Stubbs (henceforth MS)
is to take up some of the points that he makes in his article in order to reflect on
these wider issues.
The aim of his article is to apply a computer program to the text as corpus
data and to demonstrate how the software can 'identify textual features which are
of literary significance, including features which critics seem not to have noticed'
(Stubbs 2005: 6). As I have already observed, it is unlikely that the literary critic
will have noticed many textual features of the kind that computer software will
reveal, and one can acknowledge that the value of corpus analysis is that it can
provide textual substantiation to impressionistic interpretation. And indeed this
particular analysis provides convincing linguistic evidence to support what
literary critics have identified as the motif of dark indeterminacy that runs
thematically through the novel. Thus, for example, MS points out that the
computer reveals a high incidence of words denoting perceptual unclarity:
and so on, and of expressions of vagueness:
, and
. He notes that there is a repeated occurrence of
adjectives with a negative prefix like
and so on.
Inspired by this kind of analysis, one finds oneself scrutinising wordlist
and concordance for other findings which might be revealing. Use of the
Wordsmith Tools software (Scott 1997, Scott & Tribble 2006), enables me to
note, for example, that though these negatively prefixed adjectives occur
frequently, adjectives generally seem to be in short supply in the text. Only two
, and
) appear in the first 50 keywords (using BNC World as a
reference corpus), and the most frequently occurring are simple, descriptively
spare, monosyllabic (
long, great, black, white, old
). The description of river and
forest is almost colourless (of the colours that one might expect to figure in such a
occurs only 5 times,
only 4). One might conclude that it
is a rather featureless world that Marlow describes, a monochrome world in black
and white, a kind of abstraction.
The novel features of text. Corpus analysis and stylistics.
All of these textual features can be said to substantiate the general
impression that in Conrad's novel there is a pervasive presence of something
essentially negative and indeterminate. The very texture of the style is, we might
say, a representation of reality that can only be perceived, in the words of the
apostle Paul, 'through a glass darkly'. The textual facts, then, can be adduced as
evidence that 'This is a novel about the fallibility and distortions of human
knowledge' (Stubbs 2005:12).
But, of course, only some textual facts are adduced as evidence, and their
selection has been prompted by an impressionistic literary presumption that this is
indeed what the novel is about. We have yet to contend with Fish. For the
computer software will also reveal a whole host of textual features that the
literary critic, or anybody else for that matter, would also fail to notice but which
do not seem to be noteworthy. MS recognizes that textual features, however
selected, 'still require a literary interpretation'. But then it cannot be the case that
'the software can identify textual features which are of literary significance'. This
is because literary significance can only be assigned to
Heart of Darkness
as a
, not as a
In this article, MS almost always refers to
Heart of Darkness
as a book or
a text, hardly ever as a novel. But for the literary critic, of course, as for the
normal reader, that is what it is. It is not just a book. Even less is it a text: a text is
something you analyse, not something you read. MS, always admirably cautious
in making claims for his analysis, acknowledges that 'textual frequency is not the
same as salience, and does not necessarily correspond to what readers notice and
remember in a text.' (Stubbs 2005:11) But the point is that readers do not process
texts qua texts at all, and what they notice and remember are not textual features
as such but their discursive realization in newspaper articles, manuals, leaflets,
letters. And novels. The corpus analyst necessarily deals with
Heart of Darkness
as a text, a linguistic object. But the literary critic deals with it as a novel, a
discourse, a particular genre of verbal art. So they are naturally inclined to notice
different things: features of the text on the one hand, aspects of the novel on the
To return now to the point made earlier about the two possible directions
of enquiry in stylistics, it seems obvious that we need to identify literary features
first. In the present case, we need to consider not which linguistic features can be
analysed out of the text, but which features seem to be significant in realizing
different aspects of the novel. To take one simple example: the title. As part of
the text, it will be included in the data to be analysed. But as a title, an aspect of
the novel, it has a distinctive literary function which its textual features realize.
There are two things one might note about it. First it has no determiner (
Heart of Darkness
), and second it is ambiguous ('heart consisting of
darkness', cf
heart of gold
, vs 'at the heart, ie the centre of darkness'). What is the
significance, if any, of these linguistic features? One might suggest that the theme
of indeterminacy and uncertainty is already keyed in as a theme by the very title
of the novel.
A text consists of words which combine with each other in various ways to
form different kinds of linguistic pattern which the computer, of course, can
identify. A novel consists of characters and events which combine in various
ways to form narrative patterns which the computer cannot identify. But for the
textual patterns to have literary significance, it has to be shown how they
correspond or key in with the narrative patterns of the novel.
MS does talk about
Heart of Darkness
as a novel before proceeding to the
main business of analysing it as a text by providing an account of its narrative
structure. This, he says, is 'embedded in different frames' as follows:
1. The book starts with an unnamed narrator on a boat on the Thames
2. Marlow becomes the narrator, and talks about the Thames in Roman times.
3. Marlow tells of his visit to a European city
4. Marlow tells the story which takes up most of the book….. .
5. Marlow tells of his visit to Kurtz's fiancée back in the European city
6. [There is nothing corresponding to frame 2, but some vocabulary from
frame 2 is repeated in frame 7]
7. The book ends with a paragraph from the unnamed narrator back on the
(Stubbs 2005: 8)
Whatever criteria are used for identifying these frames, they are apparently not
textual. Indeed the only textual feature that is mentioned here, vocabulary, is
explicitly excluded. For if it were a factor then repetition of vocabulary would
presumably give some textual grounds for putting frame 7 in correspondence with
frame 2. Interestingly, later in his analysis, MS does, however, suggest that this
framework is marked by textual features, pointing out that the phrase 'waterway
leading to the uttermost ends of the earth' is repeated in the first and last frames.
He also mentions that the phrase 'the pose of a….. Buddha' that occurs in the last
frame is also a repetition, but in this case the first occurrence appears not in the
first frame but the second, after Marlow has already assumed the role of the
narrator. Such textual features, then, do not seem to be a reliable indication of the
narrative structure.
I shall return to these verbal repetitions presently. For the moment, the
point I want to make is that the framework that is proposed is an analysis of
narrative - an aspect of the novel, not of the text. It can only be based therefore on
a literary view on what is significant. For MS, who does the narrating is one
significant factor: when Marlow takes over the narration we shift into a different
frame. If only this factor were to be considered, then the novel would consist only
of frames 1 and 7 with Marlow's story in the middle. But a quite different factor is
introduced to distinguish the frames in between, namely the shift in the setting of
Marlow's story. This gives us frames 3 and 5 in the European city with Africa in
between. So in fact we have two separate kinds of narrative framework, and the
attempt to fuse them into one leads to the postulation of the non-existent frame 6,
which imposes a symmetry on the novel that appears to have no warrant in textual
The novel features of text. Corpus analysis and stylistics.
The question arises as to why these two aspects of the novel, narrator and
setting, should be taken as the only significant determinants of narrative structure.
One is led to wonder whether, if this is indeed the only structure that can be
discerned, it is, perhaps, not in itself of much significance anyway. Certainly the
analysis does not reveal what the significance might be: it concentrates on the
content of frame 4, which covers a good 75% of the text, which, on this account,
has no narrative structure worth mentioning at all.
In the case of the narrative framework, we have an aspect of the novel
described without substantiation from textual features that a computer analysis
might provide. But there are other kinds of narrative pattern that
shown to be
textually realized, and here we return to the repeated phrases cited earlier. In his
discussion of the way certain words are distributed in the text, MS indicates a
number of other instances of intra-textual repetition, where the words repeated are
not necessarily frequent in the text at all. Thus, for example, he points out the
description of the city that Marlow returns to is in several ways a lexical reprise
of its first description, and again that the words
are used in
reference to both Marlow and Kurtz. Such facts 'start to say something about the
structure of the whole text' (Stubbs 2005:12). Just what that something might be
is left to the reader to ponder on. And in my case, these observations provoked a
good deal of pondering. Though it is not clear to me what such facts tell us about
the structure of the text, they set in train all manner of speculative reflection about
the possible literary significance of intra-textual repetition of this kind.
So, with Wordsmith Tools at the ready, I set off in quest of other instances
of such repetition. MS provides us with the example of recurring words that
provide a textual link between the two descriptions of the city: 'high houses',
'narrow and deserted street', 'doors ponderously ajar' in the first description, 'a
ponderous door', 'between tall houses' in the second. Further enquiry about the
distribution of these words reveals that they also figure in the description of the
jungle. A stretch of the river is described as 'narrow, straight, with high sides',
there are 'high walls' of trees, the 'high stillness of primeval forest'. The word
only occurs three times in the text, and its other two occurrences are in
descriptions of the African scene:
And the village was deserted, the huts gaped black, rotting…
the waterway ran on, deserted, into the gloom…
Directing my computer in quest of other repetitions, I discovered that the word
occurs only twice in the text. Though its second occurrence in 'a
ponderous door' may echo 'doors standing ponderously ajar' and so serve to link
Marlow's two city visits, it also echoes its first occurrence where it appears in a
description of the progress of Marlow's boat upriver 'between the high walls of
our winding way, reverberating in hollow claps the ponderous beat of the stern-
What is one to make of these inter-textual lexical connections? What do
they hint at? What literary significance can we guess they might have?
If, as MS suggests, the use of
to describe both Marlow and
Kurtz indicates their similarity, then the use of these other examples of repetition
can presumably be said to have the same associative effect. If things that are
described in the same terms take on a similarity, then just as Marlow assumes the
likeness of Kurtz, so the city assumes the likeness of the river and takes on its
darkness. This, we may suggest, is supported by other distributional facts that the
computer reveals. As MS points out, 'the words
throughout the book, but increase in frequency at the very end'. This is the textual
hint. What possible literary significance might be assigned to it? On, then, to the
guesswork. And for this we need to set the computer aside for a moment, and
look at the text for ourselves and consider its novel effects.
The end of the book is where Marlow visits Kurtz's fiancée – the 'Intended'
- and he takes both the darkness and Kurtz's last whisper with him. At her door,
as 'the dusk was falling', he 'seemed to hear the whispered cry, 'The Horror! The
Horror!' She comes to meet him in the darkening room, 'dressed in black', with 'a
pale visage', 'dark eyes'. On her appearance, 'The room seemed to have grown
darker'. And as she speaks of the noble qualities of Kurtz, 'with every word
spoken the room was growing darker'. Marlow listened. 'The darkness deepened'.
As she talks, Marlow seems again to hear 'the whisper of a voice speaking from
beyond the threshold of an eternal darkness'. The repeated words are like a sound
track, a lexical leitmotif, that brings the room and the people in it into association
with the African river that has been described in the same terms. And this is then
confirmed by a quite explicit connection: a gesture of the Intended reminds
Marlow of another woman – the 'wild and gorgeous apparition of a woman' who
appears so dramatically and ominously out of the jungle earlier in the narrative:
'…I shall see her, too, a tragic and familiar Shade, resembling in this
gesture another one, tragic also, and bedecked with powerless charms,
stretching bare brown arms over the glitter of the infernal stream, the
stream of darkness.'
The two women become one and the realities of primitive savagery and apparent
civilisation are fused by the presence of a common darkness.
This is a darkness of deception and delusion as well. As MS points out, the
is, like
, also evenly distributed through the text,
often in negative form, but there is a cluster of positive instances at the end,
mainly spoken by Kurtz's Intended. She knows that Kurtz was good and noble,
and her belief is described as
'a great and saving illusion that shone with an unearthly glow in the
darkness, in the triumphant darkness.'
And here we come to what (to me at least) is the thematic climax of the novel.
Kurtz's last whispered words that have so haunted Marlow come back again:
The novel features of text. Corpus analysis and stylistics.
'The dusk was repeating them in a persistent whisper all around us, in a
whisper that seemed to swell menacingly like the first whisper of a rising
wind. "The horror! The horror".'
The threefold repetition of the word
, insistently evoking the dark reality
of Kurtz's world, is then immediately followed by another threefold repetition,
and one that is an emphatic assertion of the counter-reality of the Intended's
'Don't you understand I loved him – I loved him – I loved him'.
For a moment, this reality prevails, to such an extent that Marlow is drawn into it
himself and tells his lie to sustain it:
'The last word he pronounced was – your name'.
Her reaction is first to give a light sigh, and then in the tense and darkened room
she makes Marlow's heart stand still with 'an exulting and terrible cry, a cry of
inconceivable triumph and unspeakable pain'. This cry is a dramatic and climactic
moment in the novel. But it is also a textual echo. Where has the reader heard a
cry before? This, of course, is where the computer comes in. It reveals that the
word occurs 6 other times in the text. The first two occur in the phrase: 'a cry, a
very loud cry, as of infinite desolation', which bears some formal resemblance to
the Intended's cry, and breaks the stillness in a similarly sudden and startling way
– but this time in the heart of darkness itself. As indeed does the third occurrence:
'a cry arose whose shrillness pierced the still air.' The other occurrences of the
word relate to Kurtz – his 'cry that was no more than a breath', his 'whispered cry'.
But the cry of the Intended is not a whispered but 'exulting', not one 'of infinite
desolation' but 'of inconceivable triumph'. And here there is another echo, surely.
Back to the computer. And I find that the word
occurs only twice
in the book. Its only other occurrence appears just before Marlow arrives at the
Intended's door, when he describes the death of Kurtz in terms of 'a conquering
darkness,' and as 'a moment of triumph for the wilderness'. It is not now the
darkness that is triumphant, and in contrast to all the vague and menacing
indeterminacy that pervades the book, we have a straightforward assertion of
absolute certainty, which Marlow repeats.
'I knew it - I was sure!' She knew. She was sure.
Just how Marlow is supposed to actually say these words we cannot know – in a
tone of irony, incredulousness? But they serve to confirm this other reality which
he cannot deny. He cannot tell her the truth – to do so would have been to
condemn her to the other darker world – 'It would have been too dark – too dark
'..too dark altogether.' Here I am teased by another verbal echo, faint but
persistent. What does this phrase remind me of? I check the concordance for
, and there it is: 'too dark altogether', 'too beautiful altogether'. And this
latter phrase takes me to an earlier scene in the book: Marlow's visit to another
woman – his 'excellent aunt'. This aunt is a minor figure, and as far as her role in
the story is concerned, seemingly superfluous. Why then is she there at all?
Marlow describes the visit:
I found her triumphant. I had a cup of tea – the last decent cup of tea for
many days – and in a room that most soothingly looked just as you would
expect a lady's drawing-room to look, we had a quiet chat by the fireside.
Another lady's drawing room, but triumph here is associated with domestic
normality – the cup of tea, the chat by the fireside. But it turns out that the aunt
has the same kind of idealistic vision as does the Intended, and thinks of Marlow
in the same way as the Intended thinks of Kurtz – as a kind of 'emissary' or
'apostle' with a mission, 'weaning those ignorant millions from their horrid ways'.
Marlow comments:
'It's queer how out of touch with truth women are. They live in a world of
their own, and there has never been anything like it, and never can be.'
And he adds:
It is too beautiful altogether.
He expresses the same sentiment later, and in similar words, when, in his
narrative, he anticipates the lie he will tell to the Intended:
They – the women I mean – are out of it – should be out of it. We must
help them to stay in that beautiful world of their own, lest ours gets worse.
Oh, she had to be out of it.
The aunt and the Intended are thus brought into association, both inhabitants of
an illusory world of conventional ideals that has to be sustained by deception. A
reality 'too beautiful altogether' in contrast with the other reality which is 'too
dark altogether' – the darkness of some pervasive moral corruption that Marlowe
senses but cannot clearly discern, and which, as MS notes is reflected in the
vagueness of his language.
But there are times when his language is not vague at all, and here we
come to another aspect of the book as novel which the analysis of the text does
not itself reveal. The intra-textual patterns I have been tracing can be taken as
being indicative of the underlying theme of
Heart of Darkness
, and to lend
support to MS view that: 'This is a novel about the fallibility and distortions of
human knowledge'. What a novel is about is something, it would seem, that a
quantitative analysis of text is particularly well suited to identify: its theme is
reflected by the frequencies of linguistic features and their distribution, as MS
demonstrates so convincingly. But there is, of course, more to a novel than what
it is about. Its theme only becomes significant by the manner of its representation,
by the way it is activated by events and characters. What seems to me most
striking about the intra-textual patterns that I have noted is the way they give
dramatic expression to the theme in the representation of the women characters
and the events in which they figure.
The novel features of text. Corpus analysis and stylistics.
The aunt and the Intended are associated in that they have in common the
same idealistic view of the world, the same reality of conventional values. But the
contexts of their appearance are in striking contrast: one a cheery drawing room, a
quiet chat by the fireside, an atmosphere of relaxed normality, the other a sombre
and sepulchral room, the atmosphere charged with intense feeling, and dialogue
as different from a casual chat as it is possible to imagine. Neither woman is
described in any detail. In fact the aunt is not described at all. She is 'excellent'
and that's all. We have no indication about what she looks like. The Intended is
hardly less sparely described – just one or two simple monosyllabic adjectives:
'fair hair' 'pale visage,' 'dark eyes'. Why this absence of descriptive detail, one
might wonder. In the case of the aunt, one might suggest that since she is a minor
character no description is called for, but then other minor characters are
described in some detail. One of the women in the office where Marlow goes to
get his job, for example, is described in very particular terms – warts and all
Her flat cloth slippers were propped up on a foot-warmer, and a cat
reposed on her lap. She wore a starched white affair on her head, had a
wart on one cheek, and silver-rimmed spectacles hung on the tip of her
The aunt and the Intended are by comparison featureless. As such, they seem to
function more as thematic symbols than as individual characters: the aunt as
representing conventional normality, defined by the typicality of her drawing
room, and the Intended as representing too the darkness of delusion and deception
in which she is embroiled.
But there are other figures that do not blend in with the thematic
background, but are starkly foregrounded against it. The most striking instance of
this is 'the wild and gorgeous apparition of a woman' that, as pointed out earlier,
Marlow explicitly associates with the Intended. This is how the woman is
She walked with measured steps, draped in striped and fringed cloths,
treading the earth proudly, with a slight jingle and flash of barbarous
ornaments. She carried her head high; her hair was done in the shape of a
helmet; she had brass leggings to the knee, brass wire gauntlets to the
elbow, a crimson spot on her tawny cheek, innumerable necklaces of glass
beads on her neck; bizarre things, charms, gifts of witch-men, that hung
about her, glittered and trembled at every step.
There is nothing vague or indistinct about this description. The vivid impression
the woman's appearance makes on Marlow is etched on his mind in exact verbal
detail, and the description stands out as particular and precise because the words
in it are infrequent. It is this that makes the woman stand out against 'the gloomy
border of the forest' 'the immense wilderness'. But at the same time she is also
'like the wilderness itself, with an air of brooding over an inscrutable purpose'.
The clarity of the perception registered in this descriptive precision contrasts with
the vaguely expressed sense of strangeness and foreboding and accentuates it. As
MS notes, much of the language of the text, the recurrence of words to do with
darkness, uncertainty, negation and so on reflects the underlying theme of the
novel. In a way they serve as a backdrop, a mise-en-scene. But it is the events and
characters, figures against this ground, that activate the theme and give it
dramatic force, create the literary significance that make the text into a novel.
And these are often described in language that a quantitative analysis of the text
would not register as remarkable.
The observations made by MS in his article, particularly those that point
out the recurrence of certain words and phrases, often themselves infrequent, in
different parts of the text, have set me off looking for similar intra-textual links,
and using them as hints to possible literary significance. Hints followed by
guesses. And one bit of literary guesswork has set me in quest of other hints – in
frequency lists, in concordances – looking for possible bits of textual evidence to
support a particular literary interpretation. Hints and guesses. It is a fascinating
But not one that Stanley Fish would be likely to approve of. For it is, of
course, open to the charge of circularity. This does not, however, make it invalid
as a process of exploring significance which, prompted by the MS analysis, I
have been pursuing here. On the contrary, circularity of a kind, is an essential
feature of this process, for, to quote T.S. Eliot again:
…the end of all our exploring
Will be to arrive where we started
And know the place for the first time.
Little Gidding
The Fish objection only applies to the positivist claim that stylistics establishes a
correlation between linguistic features and literary effects so that one can be read
off from the other. But this is not the claim that stylisticians generally make.
What they are principally concerned with is not correlations but correspondences,
with ways in which textual features can be adduced to give warrant to different
literary interpretations, not to ratify one of them as definitive. Stylistic analysis
does not seek to foreclose on a particular interpretation but to open up alternative
possibilities. It does not claim to discover meanings which are inscribed in a text
and which may have eluded literary critics, but to provide the means for exploring
one's own reactions to the text. Herein lies its educational value – for it offers an
alternative to the traditional teaching of literature. Rather than being the passive
recipients of the second hand interpretations of literary critics, students can be
enabled (empowered even) to take the initiative and engage actively and directly
with literary texts themselves. (Widdowson 1975,1992)
The novel features of text. Corpus analysis and stylistics.
Stylistics, then, is all about hints and guesses. As Verdonk puts it:
' can serve not only to substantiate an impressionistic sense of
meaning, but also to suggest the possibilities of reading other
interpretations into a text'

(Verdonk 2002: 78)
In his review of that book, MS calls this a 'weak' defence of stylistics and takes its
author to task for not rising to the Fish challenge to 'defend a stronger position'
(Stubbs 2004: 129). In this article on
Heart of Darkness
MS does rise to the Fish
challenge, but interestingly does so by in effect arguing for the validity of the so-
called 'weak' position himself, justifying his stylistic analysis, very much along
Verdonk lines, by concluding that:
'..observational data can provide more systematic evidence for
unavoidable subjective interpretation'. (Stubbs 2005: 22)
As a text
Heart of Darkness
consists of observational data that can be analysed by
computer. As a novel, however, it can only be subjectively interpreted. This
means that what counts as evidence for interpretation can never be objectively
determined, and any claim that it can (the 'strong' position) is mistaken. Hints and
guesses are all we can reasonably expect. But the point about the computer is that
it can provide so many hints for us to guess the significance of. This is what
makes Michael Stubbs' article so stimulating – his own textual findings set the
reader off in quest of others.
What, for me at least, is revealing about its 'quantitative stylistic methods',
is that the results of the analysis are so different from its effects: the very
precision of the findings provoke very imprecise speculation about their
significance. The more you pin down and quantify features of the text, the more
aware you become that features of the novel cannot be pinned down and
quantified. They remain elusive, subjective, and variable, and cannot be reduced
to textual terms. But, as MS says, it is not the claim of stylistics that they can be,
or should be. His kind of analysis does not tell us what
Heart of Darkness
but what it might mean to different readers. And herein lies its value, and
particularly its educational value: it demonstrates ways in which textual features
can be explored, and how such exploration can open up possibilities of novel
Fish, S. 1973. 'What is stylistics and why are they saying such terrible things
about it?' In S. Chatman (ed.)
Approaches to Poetics
. Columbia University
Press. Reprinted in Weber, J.J. 1996.
The Stylistics Reader
. London:
Arnold. 94-116.
Scott, M. 1997.
WordSmith Tools
(Software). Oxford: Oxford University Press.
Scott, M. & C .Tribble. 2006.
Textual Patterns
. Amsterdam: John Benjamins.
Stubbs, M. 2004. Review of Verdonk, 2002
Applied Linguistics
Vol 25/1 126-
Stubbs, M. 2005. 'Conrad in the computer: examples of quantitative stylistic
methods'. In
Language & Literature
. Vol 14/1. 5-24.
Verdonk, P. 2002.
. Oxford Introductions to Language Study.
Oxford:Oxford University Press.
Widdowson, H.G. 1975.
Stylistics and the teaching of literature
. London:
Widdowson, H.G. 1992.
Practical Stylistics
. Oxford: Oxford University Press.
Hocus pocus or God's truth:
the dual identity of Michael Stubbs
Guy Cook
Open University
Text analysis is in an anomalous position: hovering on the borders between the sciences
on the one hand, and the arts on the other. As science it seeks to be descriptive rather than
prescriptive, replicable by other analysts, expounding objective facts about language use.
As an art it evaluates and prescribes, imposing the writer's views upon the external world,
saying as much about the analyst as the analysed. This chapter explores the position of
Michael Stubbs in relation to this dichotomy, suggesting that, while he advocates
objectivity, and has made an outstanding contribution to linguistic description, his
achievement - almost despite himself - is also to be an evaluator and interpreter. Like a
good literary critic, he is worth reading not only for what he tells us about the external
world (which is a great deal) but also for his own unique ideas.
"Pure induction will never get you from empirical observations to
interesting generalizations. You have to know where to look for
interesting things. As Grice (1958:173) puts it: 'you cannot ask [....]
what something is unless (in a sense) you already know what it is'.
However, this is true only 'in a sense', since the aim is to say
systematically and explicitly what something is: and that is where
empirical, observational analysis can contribute. It is not possible (or
desirable) to avoid subjectivity, but observational data can provide
more systematic evidence for unavoidable subjective interpretation."
(Stubbs 2005)
Hocus pocus or God's truth
In the early 1950s, when Michael Stubbs was just starting school in Glasgow, the
lexicographer and semanticist Fred Householder, reviewed a book called
'Methods in Structural Linguistics' by Zellig Harris and evoked a distinction
between two positions in linguistics.
"On the metaphysics of linguistics there are two extreme positions,
which may be termed (and have been) the 'God's truth' position and
the 'hocus pocus' position. The theory of the God's truth linguists [...]
is that language 'has' a structure and the job of the linguist is (a) to find
out what the structure is, and (b) to describe it [...]. The hocus pocus
linguist believes that a language (better, a corpus, since we describe
only the corpus we know) is a mass of incoherent formless data, and
the job of the linguist is somehow to arrange and organize this mass,
imposing on it some structure [...]." (Householder 1952)
Householder describes both positions as extremes. He assumed perhaps what we
would now call a negative discourse prosody for both phrases,
and that nobody
would want to be identified as either. He criticises Harris for being too much of a
"God's truth" linguist, but implies that a "hocus pocus" position would be just as
flawed. The good linguist should be somewhere between the two poles and not at
either end.
Householder was thinking of the division of his time between armchair
structuralist linguists using examples drawn by intuition from their own minds,
and empirical anthropological linguists going out and studying language
behaviour. Given how linguistics has developed since, there are quite a few
complications in applying Householder's distinction to linguists today. For one
thing, there has been a revolution in corpus linguistics, in which that Glasgow
schoolboy went on to play a leading part. The modern linguist is no longer
limited to "only the corpus [the linguist] know[s]", but has access to millions of
words beyond their own immediate experience. Nor are their corpora "a mass of
incoherent formless data"; linguists like Michael Stubbs seek and find in them
patterns and connections undreamed of in Householder's time. So it is now the
corpus linguist who believes "that language 'has' a structure and the job of the
linguist is (a) to find out what the structure is, and (b) to describe it".
Stubbs himself has paid attention to this dichotomy, though in different
terms. Considering the ideas of Saussure, he writes:
"In a famous and influential statement, Saussure (1916) argued that
'far from the object of study preceding from the point of view, it is
rather the point of view that creates the object'. Due to advances in
technology, new observational methods have made it possible to
collect new types of data and to study patterns which had previously
been invisible, but the point of view does not create the patterns. What
we see certainly varies according to point of view, and it follows that
any view is partial, but it does not follow that what we observe has
been created by the point of view or by the observational tools."
(Stubbs 2002a:220)
Saussure in other words was too hocus pocus, and has moreover been overtaken
by events. But Saussure's work, though contested and discussed, is far from
dismissed by Stubbs. It is a recurrent point of reference in his work. Saussure's
ideas, my intuition suggests, are ones with which he has a love/hate relationship.
In this chapter, I shall use this distinction between 'God's truth' and 'hocus
pocus' liberally and with poetic licence, interpreting it beyond the context of the
time it was written to mean simply that there are two opposite tendencies in
linguistics. In both, language is seen as ordered, but in the first case (God's truth)
the order is an objective one, out there to be discovered, while in the second case
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
(hocus pocus) it is something subjectively imposed upon the language by the
analyst. Like Householder though I shall treat this as an idealisation, and
charitably assume that no-one is so foolish as to be actually at one extreme or the
other. We are all somewhere on a line between the two poles, pulled now in one
direction, now in the other.
The distinction of God's truth and hocus pocus is a light-hearted echo of
the many "unavoidable dualisms" (Stubbs 2000) "which trouble linguistics" and
"other disciplines"
"dualisms of subject and object, internal and external, agency and
structure, process and product,
, language use and
language system, pragmatics and semantics, communication and
language, creativity and rules, intended action and unintended
consequences" (Stubbs 1996: 56-57).
These are all dichotomies with which he has resolutely engaged, even agonised
over. What I want to do in this chapter is assess where the work of Michael
Stubbs belongs in relation to some of these dualisms, aware that in some hands,
the weapon of corpus linguistics is indeed wielded as though it were God's truth,
while for some of its critics it comes close to being hocus pocus: distorting the
living actuality of language by freezing and dissecting it. And my conclusion will
be that Stubbs is not where he appears or claims to be on that line! "What we see"
as Stubbs writes of Saussure "certainly varies according to point of view, and it
follows that any view is partial". In the partial view to be presented here, I shall
suggest that on the continuum from hocus-pocus to God's truth, Stubbs is, like a
quantum particle, actually in two places at once, depending on the observer.
I shall acknowledge the influence of Stubbs' ideas, but seek to develop two
points. The first is the inevitable role of evaluation in language analysis. The
second is the issue of where exactly "patterns which had previously been
invisible" are. Are they in the mind or only in the corpus? If they are in the mind,
are they conscious or subconscious, and how might we access them? If they are
both in the mind and in the corpus, how might we relate the two?
Arts or Sciences
Let us begin with a related dichotomy, partly epistemological and partly
institutional, between the arts and humanities on the one hand and the sciences on
the other. In recent years, the tendency has been for linguistics, however
institutionally placed, to emphasise its scientific credentials. "Mere" scholarly
disquisition based upon reading and reflection has fallen out of favour, and for
many journals, research assessments and examinations all valid conclusions must
be based upon experimental or observational data. Rigorous analysis, quantitative
measures, testable hypotheses, replicability, reliability, validity, are the order of
the day.
This allegiance has large implications for how we write and what we write
about. If we are wholly scientific, then our aim should be only to discover
ive facts about language as though it were an
out there in the world,
untainted by our own intuitions, beliefs and values. Our aim is not to mould the
data to our own
ive vision of the world in the hocus pocus way, but like
good natural scientists to provide a description and explanation of what exists
independently - to try to approach God's truth. The facts we discover should
therefore be the same, whoever the investigator. They should be exactly
replicable by another researcher following the same rigorous procedures. The aim
is for the personality of the investigator to melt away. The good researcher is one
who becomes a clone of all other good researchers, emulating the proper scope of
science as description and explanation of the material world, but not evaluation.
Like the botanist describing a flower we should say everything possible about it,
but not whether we think it is a pretty flower or the right one for the garden.
Taken at face value, the writing of Michael Stubbs would seem to be very
much part of this movement, taking linguistics away from the arts and
humanities, and towards the social and even natural sciences. (He often, for
example, emphasises an aspiration to provide data which can be checked.) But the
matter is more complicated. Curiously, he seems to avoid the word "science" in
describing his methodology. A search of the eleven papers available
electronically on his website (from which I created a mini-corpus for the purposes
of this chapter), reveals only fifteen uses of "science"/ "sciences"/ "scientific" /
"scientist(s)". Of these, nine occur because he is discussing scientific vocabulary.
A further five occur in discussing trends in the "social sciences", which at times
he seems to see as not scientific enough!
The remaining two occurrences are
because he is drawing an analogy - of which he is fond - between scientific
observational instruments and corpus software. So he seems, quite studiously, to
avoid describing corpus linguistics as scientific.
And there seems to be good
reason for this. His own writing (e.g. 2002a: 232-238) is given over to discussion
of subtler distinctions, elaborating in particular upon the nature of brute, social
and subjective facts, as discussed by thinkers such as Popper and Searle. Thus
however alienated he may feel from the contradictions of post-modern relativism,
he is very much the social scientist, aware that linguistics is studying social rather
than the brute facts of concern to the natural sciences. Like many natural
scientists, however, he too is concerned with rationality, rigour and evidence - but
in a different way.
Nevertheless, despite this avoidance of the term "scientific" and an
informed take on the object of linguistic enquiry as different from that of the
natural sciences, we might reasonably say that the methods advocated by Stubbs
share more with those of science than of the arts. Verifiable replicable facts about
language are what he seeks. And the imposing personality of the analyst is
downplayed. His critique of Critical Discourse Analysis (CDA) (Stubbs 1997) for
example takes the movement to task on exactly this ground of a lack of rational
rigour, and for the imposition of the analyst's prior beliefs in analysis. At one
point for example, he criticises CDA for being
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
"reminiscent of the moral crusade against the vulgarising mass media
and increasingly mechanized and capitalist society which was carried
out by F.R. Leavis and his colleagues in
in the 1930s."
This is not to say though that he is out of sympathy with Leavis' literary critical
any more than he is out of sympathy with the aims and opinions of
CDA. (He expressly says that he is not.) It is rather that he finds such an approach
inappropriate for linguistics. What he seeks (notwithstanding the usual
philosophical caveats) is to be as objective as possible.
In the arts and humanities on the other hand, in literary criticism for
example, there is also a legitimate but unscientific imposition by the writer upon
their data, an assertion of ideas coming from inside as much as outside. (That is
not to say that the
of study does not impose constraints upon what is said:
the critic should not get their facts wrong.) Thus the literary critic (quite
appropriately for the discipline) interprets what they find in the light of their own
aesthetic or moral values, moulding it to their own system of thought. Thus we
read a literary critic like F.R. Leavis (to take Stubbs' own example) not only to
find out about the literature he critiques, but also to find out about his unique
personality and view of the world. We are learning not only about the object of
study, let us say for example the novels of D.H. Lawrence (Leavis 1955), but also
the subject of the analyst. We enjoy, or get irritated by, the writer's company, and
learn from his ideas, both matters of fact and matters of opinion. Rather as we
would in reading the novels of D.H.Lawrence himself.
And in terms of style, the good literary critic should, like a good creative
writer, be idiosyncratic and distinctive, with their own quirks of style - which
Leavis certainly had. The good scientist (and linguist pace Stubbs) though, will
aim to write about the facts s/he has discovered as lucidly and elegantly and
possible, but to do so by removing all personal touches, idiosyncrasies and
embellishments. Michael Stubbs' own lean and lucid style would be a very good
model to follow. In earlier times, he was drawn to the story Cat in the Rain by
Ernest Hemingway (Stubbs1983:194-219), a writer whose style though powerful,
is famously clear, terse and unpretentious. Stylistically, Stubbs is in many ways
the Ernest Hemingway of linguistics - though paradoxically, as with Hemingway,
Stubbs' apparently depersonalised style is instantly recognisable and highly
A further paradox with making any neat division between hocus pocus and
God's truth is that hocus-pocus ideas, once uttered - especially if eloquently
expressed - become part of the external world which the God's-truth scholar seeks
to describe. Perhaps the arch example is Freud, who persuades through his skill as
a storyteller (Fish 1986/1987), and whose concepts, however unscientifically
arrived at, became part of people's experience, even if they were not so before.
But I suspect that Stubbs, like Popper (who also included Marx in this category),
would see such creations as "pseudo science" (Popper 1963). A converse paradox
is that God's-truth thinkers, once they have found order, try to impose it upon the
"formless mass" of their hocus pocus opponents.
Literary Criticism or Stylistics
Let us pursue the dichotomy of literary critical and linguistic analysis of texts a
little further, as it is one with which Michael Stubbs is often concerned. Writing
of this distinction in 1975, H.G. Widdowson characterises these two activities as
"the ultimate purpose of literary criticism is to interpret and evaluate
literary writings as works of art and (...) the primary concern of the
critic is to explicate the individual message of the writer in terms
which make its significance clear to others" (Widdowson 1975:5)
"the linguist, on the other hand, is primarily concerned with the codes
themselves and particular messages are of interest in so far as they
exemplify how the codes are constructed. Given a piece of literature, a
poem for example, the linguist will be interested in finding out how it
exemplifies the language system, and if it contains curiosities of
usage, how these curiosities might be accounted for in grammatical
terms." (ibid.)
The literary critic in other words is concerned with doing something to the object
of study: interpreting and evaluating. That is to say, reading something into it.
The linguist on the other hand is concerned with "finding out" something from it.
It is the hocus pocus and God's truth distinction all over again - and formulated in
this way, Michael Stubbs is clearly for the latter.
Widdowson on the other hand sees "the purpose of stylistics" (a subject of
the book from which these quotations above are taken) as
"to link the two approaches by extending the linguist's literary
intuitions and the critic's linguistic observations and making their
relation explicit." (ibid.)
He refers presumably to a linking of the two perspectives in the service of
interpretation, rather than of the other literary critical activity he mentioned:
Stylistics or Quantitative Stylistics
Stubbs seems to accept the distinction between literary criticism and linguistics,
but, unlike Widdowson, to be far from happy with the position of stylistics
between the two. He refers to it as leading "an uneasy half life, never fully
accepted, for many related reasons, by either linguists or literary critics" (Stubbs
2005). His solution is to distance stylistics further than does Widdowson from
"the linguist's literary intuitions", to shift it more towards the linguistic than the
literary critical, to make it in others words more scientific, less - if you will -
hocus pocus. He has done this by positing a further dualism, opposing traditional
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
stylistics on the one hand to corpus or "quantitative stylistics" (Stubbs 2005) on
the other. He does this partly on the practical ground that conventional stylistics
struggles with longer works (it is "unworkable for novels") and partly on the
more theoretical ground that a linguistic analysis of an individual text or extract
cannot reveal facts of the same interest and reliability as an analysis setting those
findings "against a background of what is normal and expected in general
language use" - comparing in other words the language of (a) particular literary
text(s) with a reference corpus. Conventional stylistics is in his view at fault, like
CDA, for making arbitrary choices of which features to analyse, and then making
assertions about those features without objective comparative evidence.
Thus where Widdowson had a trinary distinction in which stylistics played
a mediating role between two extremes
literary criticism stylistics linguistic description
Stubbs sets up a new trinary distinction
(literary criticism) conventional stylistics quantitative stylistics
but appears to have dropped literary criticism as a candidate for the kind of
linguistic work he is interested in - hence my bracketing of it here. In doing this
he also introduces, by default (because the conventional stylistician can only
handle short texts, and the quantitative stylistician can handle large ones), a
further distinction, between
intensive analysis
extensive analysis.
But there is an important distinction to be made here. This concerns whether
Stubbs is advocating
the intensive analyses of conventional stylistics
with the extensive ones of quantitative stylistics, or whether he see the two as
complimentary. The evidence points to the latter. He writes:
"the most powerful interpretation emerges if comparisons of texts
across corpora are combined with the analysis of the organization of
individual texts" (Stubbs 1996:34)
Furthermore, in his many discussions of dualisms, he professes increasingly a
desire to move beyond or even reconcile them. He is not apparently for picking
one side and discarding the other. In
Words and Phrases
(Stubbs 2002a:228)
writing again of Saussure, he cites Hodge and Kress's (1988:17) critique of
Saussure for constantly dividing the subject matter of linguistics into binary
oppositions and then dropping one half of the resulting opposition. (
dropped in favour of
, then a diachronic study of
in favour of a
synchronic one, then a syntagmatic analysis of
langue in favour of a
paradigmatic analysis). And he approves of Hodge and Kress's disapproval.
Where his own dichotomy between conventional and quantitative stylistics is
concerned, he is apparently not following this pattern. He is adding not
subtracting, enriching not impoverishing.
Over recent years, Michael Stubbs has used this augmented stylistic
technique to carry out a number of analyses of relatively short novels and stories,
notably James Joyce's
(Stubbs 2002a: 123-144), Joseph Conrad's
Heart of
(Stubbs 2005) and Henry James's
Turn of the Screw
(Stubbs 2007). All
of these analyses have yielded significant literary insights: an achievement of
great note as these are among the most discussed authors and texts in literary
criticism and conventional stylistics. Indeed, that is his express reason for having
chosen these works.
, for example, is chosen on the grounds that it is:
"well known, and widely available to readers who want to check my
analysis [and] has been the subject of many literary critical and
stylistic analyses [and] we can therefore compare the computer's
results with the interpretations of trained critics" (Stubbs 2002a:125)
This is not a case of throwing up interesting insights out of nothing in other
words, but of adding interesting insights to a very well ploughed field. And it has
also inspired others to follow his lead: see for example O'Halloran (2007a) on
But what of literary criticism, whose business is, according to Widdowson,
"to interpret and evaluate literary writings as works of art"? Is it completely
beyond the pale? Something with which Stubbs is simply not concerned? The
answer appears to be both yes and no.
On the one hand, Stubbs has argued very effectively for a quantitative
dimension to support interpretation, which can be seen as augmenting literary
criticism. Thus just as conventional stylistics added to literary critical
interpretation by linking it to linguistic analysis, so quantitative stylistics has
added the insights of corpus analysis. It is a cumulative process, of a kind of
which Michael Stubbs seems to approve.
On the other hand, however, he is apparently not concerned in his own
work with the other aspect of literary critical activity: evaluation. As he writes in
a different context (discussing truth conditions):
"Truth and falsity are also problematic with respect to evaluative
utterances. If someone says
That's super!
, then that may tell us
something about the speaker, but little about the world." (Stubbs
But as to other logicians, one might ask the following question, especially
considering that in Stubbs' express view (2002a:232-235), linguistics is not
concerned only with the physical world. What conception of "world" excludes
speakers and their utterances from membership? There are social and
psychological as well as brute facts.
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
Part Two:
I feel somewhat out of place in this
for I am persistently and
unrepentantly guilty (as will be clear from this chapter itself) of all the crimes of
which Michael Stubbs is so determined a scourge: arbitrary selection of texts and
features within them, and the mixing of evaluation and description. I have also
been a critic of applications of corpus linguistics in language teaching (Cook
1998, 2001b), and the subject of some counter attack (Carter 1998 and Hunston
2002: 192-197).
Where Stubbs situates my own work on the hocus-pocus spectrum is clear
from his remarks on my book
The Discourse of Advertising
(Cook 1992
). It is
one of three books chosen in chapter one of his
Text and Corpus Analysis
1996: 14-21) to demonstrate shortcomings of three respective methods of
analysis. And of it he writes:
"The method is simply that of confident personal literary judgment
(...) he picks out his own favourites (which is what all literary critics
do), concentrating on memorable or famous examples, but does not
analyse the majority which merely provide useful information and/or
are just banal."
In these lines, a number of activities are singled out as inappropriate for stylistics:
- confident personal literary judgment
- picking out favourites
- concentrating on the memorable or famous
- ignoring the majority
- ignoring the banal
What I want to do in this section is to argue for the inevitability, and even
desirability, of all of these activities in certain types of analysis, particularly
literary analysis. I do not do this in a spirit of refuting Stubbs innovations in
literary analysis, but rather of suggesting, in his own spirit of reconciling rather
than reducing dualisms, a way of adding to and enriching both findings and
methods. (I shall return to the relation of Stubbs to those with whom he disagrees
in my conclusion.)
I should like to argue next that this second component of literary criticism,
evaluation, cannot be so completely sidelined. It is part and parcel of language
analysis, and - problematic though it is for linguists - it cannot be left to literary
criticism. I do not mean by this only that linguists should take stock of the
importance of evaluative judgments by language users and their effect on
language use and usage,
but rather that evaluation is an inevitable part of the
process of linguistic analysis itself, i.e. something linguists do as well as the other
language users they study.
Stubbs' response, I suspect, would be that while he, like everyone else, has
his own "favourites" among literary texts, such preferences are "simply" not part
of the process with which he is professionally concerned. In this respect one half
of the literary critical enterprise (evaluation) is indeed left behind, while the other
(interpretation) is embraced and improved. This is related to the issues of hocus-
pocus and God's truth continuum in the sense that evaluation, even more than
interpretation, is of its nature an imposition upon the data, rather than something
arising from it.
Evaluation or calculation: discourse prosodies
There are many examples of "confident personal .... judgment" in corpus
linguistic analysis. Take for example the identification of negative and positive
discourse prosodies (Sinclair 1991:112, Louw 1993): one of corpus linguistics
outstanding contributions to the understanding of word meaning. There can be no
doubt that the many studies of discourse prosody - including very significant ones
by Michael Stubbs himself (1995, 1996, 2001a, 2002a) - provide invaluable
insights into the relation between linguistic choices and their effects, of
tremendous usefulness to discourse analysis in general and literary stylistics in
particular. But the point I want to make is that only a part of this process of
establishing a discourse prosody is automatic and objective. Thus the first stage is
initiated by the software's statistical information about the collocates of a chosen
word. But the second stage - saying whether those collocates are 'negative' or
'positive' - is guided by the analyst's "confident personal .... judgment". That is
not to say they are wrong in their conclusions, but only that the established
method for reaching such conclusions is partly subjective and evaluative. Thus in
a recent cross-linguistic study, for example, Xiao and McEnery (2006) conducted
an extensive analysis of the degree to which English and Chinese conventional
translation word equivalents also have similar discourse prosodies. (The study is
itself an illustration of the power of corpus linguistics to contribute to a range of
sub fields, in this case translation and lexicography, and of the influence of
Michael Stubbs to whom Xiao and McEnery frequently refer.) They list the
following collocates of
, revolution, order, increase, death
, downfall
, war
government, situation, action, improvement, policy, reduction, result
and state.
The underlined words are described as "
negative, while the other
collocates are either positive or neutral" (my italics). This is a standard enough
procedure, of a kind followed routinely in corpus-linguistic analysis. My point is
not to question the value of Xiao and McEnery's study, or dispute that these terms
are indeed negative, but rather to make explicit the method by which corpus
linguists reach such conclusion. Though I agree with such judgments in practice,
I want to emphasise that there is no empirical basis for saying that a particular
word - "death" for example - is negative. It is something which the corpus linguist
feels intuitively to be "obviously" the case.
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
In addition, whether prosody is negative or positive will vary with to some
degree with individual or cultural context. Even "death" is not negative in all
contexts: for a suicide bomber, believer in human sacrifice, or Spartan warrior.
Less far-fetchedly, the designation of "revolution" as "positive or neutral" is
surely highly disputable, especially in a Chinese context, where it could be seen
as positive by some people and markedly negative by others. Discourse prosody
in other words is context and reader dependent. Indeed, the phrase "God's truth"
is a superb example of a phrase whose prosody will be negative in some contexts
and positive in others. A Google search reveals that it is used in two contradictory
ways: positively by the fundamentally religious, and negatively by others as a
synonym for bigotry.
Of course one could counter these arguments in two ways. Firstly one
could survey opinion, specifying if necessary for which populations a given word
is negative or positive. But this seems to involve the very circularity of which
Michael Stubbs has been so critical elsewhere, as it is using people's intuition to
access facts which are supposed to be unavailable to intuition! (Though the
standard social-science assumption is that inter-coder agreement is itself a kind of
Alternatively, staying within the heuristic framework of corpus generated
data, one could seek out the discourse prosody of the common collocates of the
word in question, for example "death". But this would then, if we are to be
rigorous, necessitate a further search for the collocates of these collocates, taking
us into a game of everlasting deferral, more suited to a Derridean
deconstructionist analysis that the new empiricism of corpus linguistics.
We should not be surprised though by this need in linguistics to combine
empirical observation of external objective factors with internal subjective
judgment for it nicely mirrors the ontological status of language, as discussed
extensively by Stubbs (e.g. 2002a:226-242). Language is simultaneously both
within us and without us, a mental and physical object. On the one hand it is an
internal subjective fact which can be perceived internally in our minds, and on the
other observable countable physical traces in the world (in the shape of marks on
paper and screens and sound-wave vibrations) whose reality as language is only
created by the perception of a human subject. Marks on paper are only words if
they are perceived to be so by someone who reads that language! It is this which
accounts for the ambiguous status of linguistics as a discipline and explains
perhaps Michael Stubbs' careful avoidance of the term 'science'.
Verse or poetry: what would be a corpus of poems?
A similar point about the combination of objective fact and subjective intuition
pertains to the assembly of the corpora from which the objective facts emerge.
This is not a problem for finite specialist corpora, but it is one for general
corpora. Thus one might have for example a corpus of
The Guardian
in 2003 which is complete - every word in every copy of the newspaper in that
period - and searches of such a corpus would indeed yield hard indisputable facts
about the language of
The Guardian
in this period. But when we come to the
notion of a general corpus of English (or any other language) then we encounter
serious problems, if we are to rule out the role of "confident personal ...
judgment". Thus the written component of the British National Corpus for
example - often taken as a standard - when broken down into components is
based upon quite arbitrary choices about proportions of different text types,
making one wonder how the conclusions drawn from it about British English in
general might be different if the proportion of say novels in it were higher or if
the proportion of academic prose were lower. Again as with my remarks on
discourse prosody it is important to be clear what exactly it is that I am
criticising. I am not saying the selection of texts for the BNC is a bad one, nor am
I offering an alternative. The point is that any selection in a general corpus must
be arbitrary. As Stubbs himself says, acknowledging exactly the problem I have
"the concept of a representative sample of the English language makes
little sense.... A sample can be representative only if the population to
be sampled is homogeneous, and this is possible only in special cases,
say with a specialized sub-genre corpus (such as editorials from
quality newspapers or research articles on biochemistry). Every time
we enlarge a corpus, we increase the heterogeneity of the data, and
there will always be text-types which we have not sampled, or which
are arguably underrepresented. Unfortunately, the same problem
arises with the concept of a balanced corpus: who is to say what
percentage of the corpus should consist of weather forecasts, lonely
hearts ads, business reports, the lyrics of pop songs, or whatever?"
(Stubbs 2002a:223)
So the problem is recognised, and has been much discussed since the early days
of computerised corpus linguistics (Francis 1979). Yet to quote Stubbs' own
criticism of a methodological weaknesses in CDA:
"the fact that this is noted from time to time by practitioners does not
get CDA out of this particular Catch 22" (Stubbs 1997).
So the same must presumably apply to corpus linguistics!
This however is not the point about corpus assembly I wish to argue, though it is
an introduction to it. Stubbs and others may well be right in seeing increasingly
large corpora, and the cross checking of findings from different corpora (Stubbs
2002a:223-224) as the best practical - if fallible - solution to the problem, and he
has certainly obtained significant results by pursuing it. For the sake of argument,
I am happy to take this particular objection (which I have raised elsewhere) as
. But I would like to pursue further a particular problem about the
construction of literary corpora, as I believe it illustrates my point about the
necessity of using evaluation as a component in analysis. It is an instance of the
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
inevitability of being at times hocus pocus, and abandoning the search for God's
If we are studying literary discourse, then an example of a finite corpus
(equivalent to
The Guardian
in 2003) would be the published works of a single
author, or of one particular work such as
Heart of Darkness
The Turn
of the Screw
- as in Stubbs' analyses alluded to above. This finite corpus can be
compared with a general corpus as Stubbs has done. But what if we want to
compare such a finite literary corpus with literature in general? Then I believe we
encounter a problem which only the literary critic can solve.
This is most easily illustrated with the case of poetry (a genre which
significantly does not figure prominently in any of the standard general corpora
such as the BNC or COBUILD).
Suppose that one had a finite corpus, let us say
the published poems of W.B.Yeats, and wanted to compare the language used in
it with the language used in poetry
in general
(or some more manageable but still
general category such as nineteenth and twentieth century poetry
in general
How could one construct the necessary reference corpus: that is to say, one of
in general
? There would really be two options, reflecting different types
of definition of the word 'poem' itself.
Let us consider how poetry is defined outside the academic linguistic or
literary critical world, by a dictionary which is not corpus informed. Here is
Collins Concise Dictionary Plus
definition of the word poem:
a composition in verse, usually characterised by words
chosen for their sound and suggestive power as well as for their sense,
and using such techniques as metre, rhyme and alliteration.
literary composition that is not in verse but exhibits the intensity of
imagination and language common to it:
a prose poem
resembling a poem in beauty, effect etc. [C16: from L.
, from
Gk, var. of
something created, from
to make]"
This is, to say the least, a bit of a muddle. The first sense is in part a mechanical,
technical and undiscriminating definition, suggesting that anything, however
dreadful, can qualify as a poem, provided it has metre, rhyme and alliteration.
Mixed in with this, however, is the notion that the words are chosen for their
"sound and suggestive power". This seems to beg a host of unanswerable
questions about authors' intentions, and whether other genres lack these qualities.
The second definition repeats and even complicates the vagueness of the first. But
the third definition, though tautological (a poem is "anything resembling a
poem"!) introduces a quite different notion from either the mechanical technical
criteria or the good intentions of the poet. It talks instead about the effect of
Nevertheless, despite its muddle and internal contradictions, this
dictionary definition does capture a dualism which is acute in the case of poem,
but true of other genres too, and poses a serious problem for corpus linguistics in
that it cannot be solved by its own methods alone, but must appeal to others for
According to one sense of the word 'poem' you can proceed without
evaluation of examples. Just collect everything that claims to be a poem and/or
fulfils certain formal criteria - it is set out in lines on the page, or whatever. For
certain purposes that is perfectly legitimate. Let me give you an example. In a
recent research project, Brigitte Nerlich and colleagues examined the social and
psychological effects of the Foot and Mouth epidemic in Britain in 2001, arguing
against the media and governmental characterisation of the problem as an
economic and health issue. They discovered that the traumatic experience of the
epidemic in which 5 million farm animals were slaughtered and disposed often in
sight of their owners and their families had occasioned an outpouring of poetry
(in this first sense) from most unlikely sources: vets, government health
inspectors, farmers, and schoolchildren. And she and her team collected and
analysed a corpus of these poems (Nerlich and Döring 2005). The poems in this
corpus are not necessarily good ones from a literary critical point of view, but it
does not matter for these purposes.
But for other purposes it does matter. Of course one could compare the
poems of W.B.Yeats with such a corpus, but the findings would be very different
from those if one compared his poetry with other "good poetry". A mechanical
definition of poetry (rhyme, rhythm, lineation and so on) is not enough to capture
what is fully meant by this term. For "poetry", as the clumsy dictionary definition
suggests, is an evaluative as well as merely descriptive term. To identify
something as "poetry" is to applaud it for having achieved a certain kind of effect.
So a corpus of poems in this second sense does need to concentrate on "the
famous and memorable" and ignore the "the majority or the banal".
How is corpus linguistics to deal with such subjective and evaluative
criteria? I can think of a number of answers, but they all seem to involve a degree
of circularity and internal contradiction. To say that one would choose only
published poetry simply dodges this issue, as it shifts the burden of evaluation
away from the analyst and on to the publisher, and to an extent the public -
insofar as what is published is a response to demand. To say that one would take
poems from the "literary canon" (as evidenced for example in university literature
syllabuses) is even less satisfactory, for that would ultimately rely upon the
subjective evaluations of literary critics, whose faulty diagnoses and "confident
personal literary judgment" we are supposed to be leaving behind.
One other solution might be to test findings on public. One of the effects
cited in the clumsy dictionary definition, for example, is "beauty". Now this
effect, as the truism observes, is a subjective one, "in the eye of the beholder".
(Though within a certain discourse community, there could be quite a degree of
consensus about what is beautiful language. Considering Shakespeare's language
to be beautiful is not a minority view!) So one could objectify (and thus quantify)
if not the beauty of a piece of language itself, at least the extent of its perception
as beautiful among a particular population.
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
Evaluation and saliency
"I cannot prove that a jury did interpret a summing-up in a particular
way: I cannot look inside their minds. But I can attempt to show that
patterns of language are likely to be interpreted in a certain way,
because that is how they are likely to be interpreted in everyday life.
The linguist has to try and show how a reasonable person, doing his or
her best to understand, is likely to interpret language." (Stubbs
Evaluation is to a degree a conscious and explicit process: something which the
reader actively does. There is then the possibility of correlating corpus findings
foregrounding some particular linguistic feature with the reaction of actual
readers, as a way - albeit partial and fraught with problems - of "looking inside
their minds". Let us turn to this possibility next.
Now one apparent way out of these difficulties might be, as we are dealing
with matters of effect and evaluation, to test out how readers do evaluate and
react to certain texts and to certain features within them. Which poems do they
find "memorable"? Which features within them do they find "beautiful" and so
forth? In fact, just such a procedure of reader research is advocated by Stubbs
(1997) in his critique of CDA, as a way out of its usual reliance upon the
intuitions and personal responses of the analyst. This is one of his main criticisms
of CDA when he laments the failure (since remedied in some CDA work
) to
actually make any enquiries of real readers. What he seems to be advocated then
are two alternatives to relying upon the analyst's intuition. The first is an appeal to
corpus findings to ascertain how patterns in a single text relate to those in the
language as a whole (or as near to a whole as one can model). The second is to try
to ascertain the effects of linguistic choices, not by guesswork with reference to
oneself , but by asking actual readers for their response. Stubbs himself, we might
note, had made pioneering use of reader responses in his stylistic analysis of the
Hemingway story
Cat in the Rain
(Stubbs 1983: 194-217) asking teachers, rather
than himself, to summarise the story as a way of accessing what features were
salient for them.
I myself now strongly favour such an approach, partly in response to
having taken on board (despite my points here) some of Stubbs' criticisms. In a
series of research projects examining the discourse of public debates over food
issues, I have combined corpus analysis with interviews with text writers
with focus groups in which readers discuss their responses to sample texts (Cook
2004; Cook, Robbins and Pieri 2006; Cook, Pieri and Robbins 2004). In this way,
I and my colleagues have some evidence of which frequently occurring linguistic
choices were made consciously by the text writer, and which were noticed by the
text reader. Using a corpus from four British newspapers
of all articles
published in the first six months of 2003 that mention genetic modification in any
We have been able to show, for example, that a frequent use of modals
such as "could", "would" and "may" in hard-news reports about developments in
genetic-modification technology is actually noticed without prompting by
Thus our corpus analysis showed such constructions as the following to
be typical - and there were many many more than these few (emphases added
could become
extinct because a vicious disease is wiping the
fruit out.
The fungus, Sigatoka, is devastating plants in Africa. (...............) Emile
Frison is top banana at the International Network for the Improvement of
Banana and Plantain. He told New Scientist magazine that the only hope
of saving edible bananas
may be
to create controversial GM versions -a
new Frankenstein food. That
would involve
taking a gene from a disease-
resistant, non-edible banana and injecting it into the threatened fruit.
16 January 2003)
Decaffeinated brews
could soon be
cheaper and tastier after Japanese
scientists grew GM coffee plants with 70 per cent less caffeine than
normal. (
19 June 2003)
GM variety
could end
danger to children
THE threat of peanut allergies
could soon be
wiped out by genetic
engineering, scientists have revealed. (
Daily Mail
17 February 2003)
GENE therapy
could help
men undergoing surgery for prostate cancer to
enjoy normal sex lives, scientists said yesterday. (
Daily Mail
29 April
(...............) Experts in Texas genetically modified ricin and found it killed
off tumour cells in mice.
Potentially lethal side effects were cut fivefold. Doctors who used ricin in
tests on lymphoma patients believe it
could help
develop new drugs
known as magic bullets.
A hoard of ricin, extracted from castor beans, was found in London in
(Sun, March 10 2003)
Such modals, we hypothesised, seek to confuse actual with hypothetical
developments, and, if they are missed by the casual reader, are likely to create a
more benign image of GM technology than it actually deserves.
However this
hypothesis that readers do not notice this detail was not borne out by our findings.
We showed the following text (chosen with the help of corpus analysis as typical)
to our focus groups (recruited to represent different types of interest in food
GENETICALLY modified crops
help endangered birds such as
lapwings and skylarks thrive again in Britain, says a study. It follows the
development of herbicide resistant GM crops and a herbicide which can
kill weeds much later in the year. (......) The study was funded by GM
giant Monsanto. (
Daily Mail
, 15 January 2003)
We received the following spontaneous commentary (and again these are only
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
help endangered birds. They haven't proved that it does. (Farmers)
It just says 'could' * Yes I just spotted that * Which is very ambiguous,
isn't it? (Charity workers)
Well another thing that's in there. It says it
help. It doesn't say it
So therefore it's not necessarily scientifically proven either (Parents)
It's not saying modified crops will help endangered birds i.e. lapwings, it
just says 'such as' and it says it
kill weeds it doesn't say it
weeds. So the terminology's very vague. (Undergraduates)
There are many problems in pursuing such a methodology. It risks for example
equating writers' and readers' own reports of their intentions and reactions with
their actual intentions and reactions. It can rely too much upon the intensive
reading and discussion generated in focus groups and interviews, when actual
reading of newspapers is likely to be much more casual and less attentive
(O'Halloran 2003). Nevertheless, for all its shortcomings, some such procedure is
perhaps the best we have available if we are to distinguish between aspects of
texts which are evident to the linguist (with or without the help of corpus
analysis) and those which are salient and effective for the reader.
But for corpus linguistics, in addition to all the standard reservations about
elicited data, there is a danger of a further contradiction. Stubbs (1996:21) makes
the claim that "the deep patterning" revealed by corpus analysis is "beyond
human memory and observation". Two questions come to mind. The first is to
wonder why, if the patterns of language evidenced by corpora are not consciously
available to one person, what the point is of asking anyone. The second is to
wonder why, if an analyst's intuitions are to be discarded, they are better replaced
with the intuitions of others.
What is needed is an integration of a corpus approach dealing with the
records of language behaviour, and a cognitive approach dealing with how the
mind produces or responds to that behaviour. The patterns of language revealed
by corpus analysis need to be matched up with its acquisition and representation
in the mind. There are certainly interesting attempts to do this (Wray 2000, Hoey
2005), though they tend to focus predominantly on subconscious rather than
conscious interpretation and evaluation. Despite such pioneering studies, the lack
of such integration is at present the main weakness of corpus linguistics, but also
perhaps its main way forward. Interestingly, and significantly, Stubbs too seems
to think so too:
"When Chomskyan linguistics took a decisive step away from
studying behaviour and its products, to studying the cognitive system
which underlies behaviour, this led to the discovery of many
interesting facts about language. Equally, when corpus linguistics took
a decisive step towards the study of patterns across large text
collections, this also led to the discovery of many new facts. The
approaches are often seen as being in opposition, and the dualisms are
perpetuated, but the long-term aim must be to integrate the insights
from different approaches." (Stubbs 2002a:242)
CODA The hocus pocus Stubbs
The discussion above has sought to pursue a number of issues arising from
Michael Stubbs' work on corpus linguistics. To some it may seem inappropriately
over-critical for a
, but I hope that, on the contrary, it follows Stubbs'
own lead. Though critical of my work and disapproving of my methods, he has
persistently engaged in dialogue with me: in print, by letter, by email, in
conversation. I claim no special status for this. The evidence is that he does this
with many people. He shuns the academic fashion for ignoring those with whom
he disagrees, and allowing the discipline of linguistics to be safely
compartmentalised. Indeed, he seems unable to leave the arguments of those with
whom he disagrees alone, but thrives on engagement. This is as much the case
with those who are narrow and ill-informed (Borsley and Ingham 2002, answered
by Stubbs 2002b) as with those whose criticisms are of substance and depth
(Widdowson 2000 answered in Stubbs 2001b) or powerful voices from the past,
such as Saussure. He sharpens his own ideas by these encounters, eschewing the
easier but more popular option of nailing his colours only to one methodological
mast, and battening down the academic hatches in order to sleep more easily at
night, untroubled by contradictory voices. And he is uncannily well informed,
with a disconcerting habit of predicting and countering objections to his method
and approach, even before they are uttered. It is difficult to catch him out. Many
of my criticisms of corpus linguistics above are supported by quotations from
Stubbs himself. Encounters with him have certainly made me think, and
considerably changed my opinions over the years - for despite my quibbling, I do
recognise the extraordinary richness of his way of analysing language.
So where is he on the God's-truth hocus-pocus continuum? He is by no
mean a linguist for whom the corpus is a sausage machine, mincing up the living
language and delivering it in manageable chunks. He has also exerted his own
influence upon it. He seeks, like Saussure, to structure and understand the world
he discovers, and to persuade others who do not agree. Thus although he may see
himself as approaching God's truth, he is actually also hocus pocus in the best
sense. He has exerted his own influence upon linguistics and shaped its
development. And like a good literary critic, he is worth reading not only for what
he tells us about the external world (which is a great deal) but also for his own
unique ideas. I have always read his work with great interest, as much as to find
out how he sees language, as to know what that language is "really" like.
My thanks to Kieran O'Halloran for his comments on earlier drafts of this
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
I use Stubbs' own current term 'discourse prosody' rather than 'semantic
prosody'. (Stubbs 2002a)
He makes some wry criticisms of the influence of post-modern relativism.
See for example the conclusion to Stubbs 2000: "That's logic...
nevertheless, some facts are based on publicly-accessible empirical
evidence. The post-modernists among you may argue that I don't realise
the implications of my own text. But I can reply that, in order to study
intertextuality, we need both historical and corpus methods. That's
He refers for example to "books such as Kuhn (1970) on paradigms of
thought" (Stubbs 1997) where most people, including Kuhn himself would
refer to Kuhn's ideas as referring to paradigms of science rather than
thought. In my wider less reliable reading of his work outside what is
available electronically, I know of only one exception, where he does
seems to equate the proper practice of linguistics with good science in a
critique of Chomsky (Stubbs 1996:29).
Though his choice of the loaded phrase "moral crusade" does suggest so!
Though still too long for the conventional stylistician to handle. Stubbs
reports that Eveline is "a little over 1800 words"; Heart of Darkness "less
than 40,000 words"; Turn of the Screw "a short text of only 42,880
Now in a second edition (Cook 2001a). The book's main aim is to show
(through conventional stylistics analysis of adverts in comparison with
literary works) that those linguistic features commonly identified with
'literariness' are as prevalent in ads as in literature.
As argued by Deborah Cameron (1995:1-33), and implicit in Stubbs' own
work on debates over standard English in schools (e.g. Stubbs 1976, 1980,
1986, 1995).
A word whose discourse prosody is also analysed by Stubbs (2002a:45-
For example, the written component of the BNC is roughly 60% books,
25% periodicals, 10% other published material, 10% unpublished written
material, and 5% material written to be spoken. Broken down in another
way, it is 75% informative writings, and 25% "imaginative" writing (BNC
website Although these proportions
are described as "enough to justify the claim that it characterises modern
British English" (Burnage and Baguley) the basis for them appears to be
intuition and subjective judgment.
Perhaps for the reason that it is too idiosyncratic to contribute to
statements about normality. As Hunston and Francis remark: "[O]ne of the
outcomes of using large quantities of data is that some of it may be
discarded, in the sense that instances of word-play or language that is
strange because it is being used in strange circumstances, are deliberately
ignored in terms of the general description of the language. (Hunston &
Francis 2000:17).
See for example Wodak et al. 1999
A similar supplementation of corpus analysis through interviews with
writers can be found in Harwood 2007.
The Sun
The Daily Mail
The Guardian
The Times
These were identified automatically by simply searching for every use of
terms such as "GM", "Genetic modification" "genetically modified" etc.
Thus the search was objective, even if our search times were inevitably
For a definition and corpus study of "hard news" see O'Halloran 2007b
See Cook 2004 for extensive elaboration and critique of the arguments.
Borsley, R. and R. Ingham 2002. 'Grow your own linguistics? On some applied
linguists' views of the subject.'
LINGUA International Review of General
112: 1-7.
Burnage, G. and G. Baguley. 'The British National Corpus' (accessed February 10 2007)
Cameron, D. 1995.
Verbal Hygiene
. London: Routledge.
Carter, R. 1998. 'Reply to Guy Cook.'
English Language Teaching Journal.
Carter, R. 1998a. 'Orders of reality: CANCODE, communication, and culture.'
English Language Teaching Journal.
52(1): 43-56
Collins Concise Dictionary Plus
1989. London and Glasgow: Collins.
Cook, G. 1992.
(first edition)
The Discourse of Advertising
. London: Routledge.
Cook, G. 1998. 'The Uses of Reality: A Reply to Ronald Carter.'
Language Teaching Journal.
52 (1): 57-64.
Cook, G. 2001a. (second edition)
The Discourse of Advertising
. London:
Cook, G. 2001b '"The philosopher pulled the lower jaw of the hen": ludicrous
invented sentences in language teaching.'
Applied Linguistics
. 22 (3): 366-
Cook, G. 2004.
Genetically Modified Language
. London: Routledge.
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
Cook, G., P.T. Robbins and E. Pieri 2006. '"Words of Mass Destruction": British
newspaper coverage of the GM food debate, and expert and non-expert
reactions.' Public Understanding of Science. 15 (1): 5-29
Cook, Guy, E. Pieri and P.T. Robbins 2004. ' "The scientists think and the public
feels": expert perceptions of the discourse of GM food'
Discourse and
15(4): 433-449
Fish, S. 1986. 'Withholding the Missing Portion: Power, Meaning and Persuasion
in Freud's "The Wolf Man".'
Times Literary Supplement
, August 29: 935-
938. (Longer version in F. Meltzer (ed.) (1987)
The Trial(s) of
. Chicago: University of Chicago Press. 183-209.)
Francis, W.N. 1979. 'Problems of assembling and computerising large corpora' in
H. Bergenholtz and B.Schaeder (eds.)
Empirische Textwissenschaft.
Berlin: Scriptor. 110-23.
Grice, H. P. 1958. 'Postwar Oxford philosophy', in
Studies in the Way of Words
1989. Cambridge, MA: Harvard University Press. 171-80.
Harwood, N. 2007. 'Political scientists on the functions of personal pronouns in
their writing: an interview-based study of "I" and "we".'
Text & Talk
27(1): 27-54.
Hodge, R. and G. Kress 1988.
Social Semiotics
. Oxford: Polity.
Hoey, M. 2005.
Lexical Priming: A New Theory of Words and Language.
London: Routledge.
Householder, F.W. Jr. 1952. 'Review of Harris, Zellig S. (1951),
Methods in
Structural Linguistics.
International Journal of American Linguistics
Hunston, S. 2002.
Corpora in Applied Linguistics.
Cambridge: Cambridge
University Press.
Hunston, S. and G. Francis 2000.
Patterns of Grammar: a corpus - driven
approach to the lexical grammar of English
. Amsterdam: John Benjamins.
Kuhn, Thomas 1970.
The Structure of Scientific Revolutions
. Chicago: Chicago
University Press.
Leavis, F.R. 1955.
D.H. Lawrence: Novelist
. London: Chatto & Windus; Toronto:
Clarke, Irwin.
Nerlich, B. and M. Döring 2005. 'Poetic Justice? Rural policy clashes with rural
poetry in the 2001 outbreak of foot and mouth disease in the UK.'
of Rural Studies
Popper, K.R. 1963.
Conjectures and Refutations.
London: Routledge & Kegan
Louw, B. 1993. 'Irony in the text or insincerity in the writer? The diagnostic
potential of semantic prosodies.' In M. Baker, G. Francis & E. Tognini-
Bonelli (eds.)
Text and Technology
Amsterdam: Benjamins. 157-176.
O'Halloran, K. 2003.
Critical Discourse Analysis and Language Cognition.
Edinburgh: Edinburgh University Press.
O'Halloran, K. in press 2007. 'The subconscious in James Joyce's
: a
corpus stylistic analysis which chews on the "Fish hook"',
Language and
O'Halloran, K. in press 2007a. 'Critical discourse analysis and the corpus-
informed interpretation of metaphor at the register level'
28 (1)
Saussure, F. de 1916. Cours de Linguistique Générale. Paris: Payot.
Sinclair, J. McH. 1991.
Corpus, Concordance, Collocation
. Oxford: Oxford
University Press.
Stubbs, M. 1976.
Language, Schools and Classrooms.
London: Methuen. (2nd
edition 1983)
Stubbs, M. 1980.
Language and Literacy.
London: Routledge.
Stubbs, M. 1983.
Discourse Analysis.
Oxford: Blackwell.
Stubbs, M. 1986.
Educational Linguistics
Oxford: Blackwell.
Stubbs, M. 1995. 'Educational language planning in England and Wales: multi-
cultural rhetoric and assimilationist assumptions.' In O. García & C.
Baker, eds.
Policy and Practice in Bilingual Education.
Multilingual Matters. 25-39.
Stubbs, M 1995. 'Collocations and cultural connotations of common words.'
Linguistics and Education
. 7(4): 379-90.
Stubbs, M. 1996.
Text and Corpus Analysis.
Oxford: Blackwell.
Stubbs, M. 1997. 'Whorf's children: critical comments on critical discourse
analysis.' In A. Wray and A. Ryan (eds.)
Evolving Models of Language
Clevedon: Multilingual Matters. 100-16.
Stubbs, M. 1998. 'German loanwords and cultural stereotypes.'
English Today
, 53
(14): 19-26.
Stubbs, M. 2000. 'Society, education and language: the last 2000 (and the next
20?) years of language teaching.' In H. Trappes-Lomax ed.
Continuity and
Change in Applied Linguistics
. Clevedon: Multilingual Matters.
Stubbs, M. 2001a. 'On inference theories and code theories: corpus evidence for
semantic schemas.'
21(3): 437-65.
Stubbs, M. 2001b. 'Texts, corpora and problems of interpretation.'
22 (2): 149-72.
Stubbs, M. 2002a.
Words and Phrases: Corpus Studies in Lexical Semantics
Oxford: Blackwell.
Stubbs, M. 2002b. 'On text and corpus analysis: A reply to Borsley and Ingham.'
LINGUA International Review of General Linguistics 112:7-13.
Stubbs, M. 2005. 'Conrad in the computer: examples of quantitative stylistic
methods.' Language and Literature, 14, 1: 5-24 .
Stubbs, M. 2007. 'The Turn of the linguists: text, narrative, analysis.' Unpublished
Widdowson, H.G. 1975.
Stylistics and the Teaching of Literature
. London:
Widdowson, H.G. 2000. 'On the limitations of linguistics applied.'
21(1): 3-25.
Wodak, R., R. de Cillia, M. Reisigl and K. Liebhart 1999.
The Discursive
Construction of National Identity
(Transl. A. Hirsch and R. Mitten)
Edinburgh: Edinburgh University Press.
Hocus Pocus or God's Truth: the dual identity of Michael Stubbs
Wray, A. 2000. Formulaic sequences in second language teaching: principle and
Applied Linguistics
Xiao, R. and T. McEnery 2006. Collocation, Semantic Prosody and Near
Synonymy: A Cross-Linguistic Perspective.
Applied Linguistics

Приложенные файлы

  • pdf 15860552
    Размер файла: 2 MB Загрузок: 0

Добавить комментарий