Psycholinguistic And Cognitive Inquiries Into Translation And Interpreting — facebook.comLinguaLIB


Чтобы посмотреть этот PDF файл с форматированием и разметкой, скачайте его и откройте на своем компьютере.
Psycholinguistic and Cognitive Inquiries
Volume 115
Psycholinguistic and Cognitive
Inquiries into Translation
doi ./btl.
Cataloging-in-Publication Data available from Library of Congress:
  () /  (-\r)
      ()
      (-)
 John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microlm, or any
other means, without written permission from the publisher.
Table of contents
Acknowledgments
Part I. Psycholinguistic and cognitive intersections
Acknowledgments
In addition to the internal review process by the editors, each of the chapters pre
sented in this book was anonymously reviewed and evaluated by 29 international
scholars who were invited by the editors. As such, we would like to extend our
sincere gratitude to these researchers for helping in this important process. We
would also like to thank Prof. Yves Gambier, the series editor of the
Benjamins
Translation Library
, and two anonymous reviewers for their excellent suggestions
on the entire book manuscript. It is without a doubt that the expertise and guid
ance of all of these scholars has helped to diversify and strengthen the contents
of this book.
 \r
Psycholinguistic and cognitive intersections

e position of psycholinguistic and cognitive

Translation process research at the interface
Fabio Alves
Alves and Mees (eds.) 2010; Shreve and Angelone (eds.) 2010; Alvstad, Hild and
Tiselius (eds.) 2011) that provide novel insights into the intricacies and complexi
ties of the translation process. Nevertheless, as Alves and Hurtado Albir (2010)
argue, TPR still shows a tendency to borrow extensively from other disciplines
while striving to build its own tradition of empirical-experimental research. Most
of its instruments need to be validated and put to the test in exploratory and pilot
studies in order to guarantee the reliability of data and results. ey suggest that
more e\nort is also needed into re ning experimental designs, using larger and
more representative samples, and fostering the replication of studies. Jskelinen
(2011) considers that one of the most signi cant results of systematic TPR has
been to highlight the cognitive complexity of translating. She cautions, however,
that, although research questions and hypotheses in TPR have arisen within the
Chapter2. Translation process research at the interface
Cognitive Science and Translation Process Research
Similarly to TPR, cognitive science is a relatively young discipline. It fosters the
interdisciplinary scienti c study of the mind and its processes, drawing on, among
other disciplines, anthropology, arti cial intelligence, linguistics, neuroscience,
philosophy, and psychology (Mandler 2002). Cognitive science investigates intel
ligence and behavior, focusing on how information is repres
ented, processed, and
transformed through faculties such as perception, language, memory, and reason
ing in biological nervous systems. It also looks at how machines perform tasks that
emulate human cognitive activity which need some type of language, memory,
and reasoning in order to be accomplished. ere are three main epistemological
currents in cognitive science, namely cognitivism, connec
tionism, and embodied/
situated cognition. Each one of them, I will argue, has in uenced TPR on di\nerent
occasions and has somehow contributed to its development.
Cognitivism grew out of cognitive psychology in the late 1950s as a reaction
against behaviorism (Mandler 2002). It presupposes that human mental activ
ity can best be understood in terms of representational structures in the mind
and computational procedures that operate on those structures. For cognitiv
ists,
human cognition is viewed mainly as a modular activity that is heavily
special
ized and operationally encapsulated in order to enable information processing
Fabio Alves
certain node activations, regulates the con guration of these emergent processes.
In a way, connectionism presupposes some form of weak representations for the
Chapter2. Translation process research at the interface
of embodied cognition, the construction of meaning refers to a speci c identity
structurally coupled with the environment in its interactions. e core of the the
ory is the molecular system which constitutes a rst order system. Human beings
Fabio Alves
Chapter2. Translation process research at the interface
Fabio Alves
Expertise Studies and Translation Process Research
Like Translation Studies and cognitive science, expertise and expert performance
Chapter2. Translation process research at the interface
Fabio Alves
of expertise in translation (Shreve 2006; Alves and Gonalves 2007). In short,
expertise studies has demonstrated that expert knowledge and expert performance
are acquired skills.
Chapter2. Translation process research at the interface
segments, and few exceptionally long segments) and an integrated processing
mode
(with long average segment size, high production speed and short pauses, process
ing at clause/sentence level, few single-word segments, and many exceptionally
long segments). Dragsted (2005) shows that regardless of the level of text diculty,
novice translators mainly favor an analytic mode type of processing whereas profes
sional translators tend to work on an integrated processing mode when translating
familiar, easy texts but are likely to revert to an analytic processing mode when
faced with a text that they found dicult. Dragsted also suggests that this change
in behavior could be related to characteristics of long-term working memory, a
concept that she borrows from Ericsson and Kintsch (1995) and indirectly relates
Fabio Alves
Chapter2. Translation process research at the interface
Psycholinguistics and Translation Process Research
On a broader perspective, psycholinguistics can be understood as the st
udy of
human language processing concerning investigations of the psychological foun
dations of language (Garman 1990). It deals with written and spoken language,
their comprehension and production, and the nature of linguistic systems and
models of processing. In line with cognitive science and expertise studies, the
origins of psycholinguistics also relate to cognitive psychology. Psycholinguistics
is concerned with how people, children and adults alike, acquire, learn, under
stand and produce language in the context of rst, second, and multilingual
language use. It is also concerned with studies of reading and writing processes
among children and adults in di\nerent situations. erefore, as a eld of inquiry,
psycholinguistics shares with TPR a closely related object of study, namely how
understanding/reading relates to production/writing in cases of oral and/or writ
ten translation.
Fabio Alves
Groot and Christo\nels 2007). Models of bilingualism usually assume
the exis
tence of control processes that activate and/or inh
ibit language output so that
Chapter2. Translation process research at the interface
key logging. By providing information about saccadic movements and eye xa
tions, TPR studies can build on Just and Carpenters (1980) eye-mind assump
tion that eye xations point to stronger processing e\nort to investigate e\nort in
reading and correlate it to writing processes. Di\nerences in eye xations in source
Fabio Alves
Chapter2. Translation process research at the interface
Other TPR researchers have dealt with similar issues over the years and re ected
Fabio Alves
Muoz (2010a, 2010b), OBrien (2013) and Risku and Windhager (2013)
use the term cognitive translatology from a broader perspective than the one I
chose for the current chapter. While their thoughts can be extended to encom
pass cognitively related actions in anthropology, philosophy, sociology, and other
related disciplines, the focus of this chapter is more narrowly de ned on the links
Chapter2. Translation process research at the interface
Empirical and Experimental Research in Translation, TREC, (http://pagines.uab.
cat/trec/) nanced under the auspices of the Spanish Ministry for Science and
Technology (20112013) to strengthen links and actions among the TPR com
munity. Also, a series of international workshops on TPR organized by Susanne
Gpferich in Graz in 2009 and in Giessen in 2011, and by Ricardo Muoz in 2013
and in 2015 in the Canary Islands is also evidence of joint e\norts in that direction.
As stated in the very beginning of this chapter, TPR has now a nearly thirty-
year history within the discipline of Translation Studies and has de nitely come of
age. ere is con dence and hope that in the course of the next years, researchers
will harvest the fruit of hard work on TPR and see the eld bene t from adopting
Fabio Alves
Alves, Fabio, and Daniel Vale. 2011. On Draing and Revision in Translation: A Corpus Lin
guistics Oriented Analysis of Translation Process Data.
Translation: Computation, Corpora,
Cognition
Chapter2. Translation process research at the interface
Ericsson, K. Anders. 1996. e Acquisition of Expert Performance: An Introduction to some of
the Issues. In
e Road to Excellence: e Acquisition of Expert Performance in the Arts and
Sciences, Sports and Games
, ed. by K. A. Ericsson, 150. Mahwah, NJ: Lawrence Erlbaum
Associates.
Fabio Alves
Chapter2. Translation process research at the interface
Krings, Hans P. 1986.
Was in den Kpfen von bersetzern vorgeht. Eine empirische Untersuchung
zur Struktur des bersetzungsprozesses an fortgeschrittenen Franzsischler
nern
. Tbingen:
Narr.
Lajoie, Susanne P. 2003. Transitions and Trajectories for Studies of Expertise.
Educational
Researcher
32 (8): 2125. DOI: 10.3102/0013189X032008021
Lrscher, Wolfgang 1986. Linguistic Aspects of Translation Processes: Towards an Analysis of
Translation Performance. In
Interlingual and intercultural Communication: Discourse and
Cognition in Translation and Second Language Acquisition Stud
ies
, ed. by Juliane House and
Shoshana Blum-Kulka, 277292. Tbingen: Gunter Narr.
Lrscher, Wolfgang 1991.
Translation Performance, Translation Process, and Translation Strate
gies: A Psycholinguistic Investigation
. Tbingen: Gunter Narr.
Mandler, George. 2002. Origins of the Cognitive (R)evolution.
Journal of the History of the
Behavioral Sciences
38: 339353. DOI: 10.1002/jhbs.10066
Maturana, Humberto, and Francisco Varela. 1988.
e Tree of Knowledge
. Boston: Shamballa.
Mees, Inger, Fabio Alves, and Susanne Gpferich (eds). 2009.
Fabio Alves
Risku, Hanna. 2002. Situatedness in Translation Studies.
Cognitive Research Systems
533. DOI: 10.1016/S1389-0417(02)00055-4
Risku, Hanna, and Florian Windhager. 2013. Extended Translation: A Sociocognitive Research
Agenda.

e contributions of cognitive psychology
Daniel Gile
Chapter3. e contributions of cognitive psychology andpsycholinguistics
Daniel Gile
that much of it was primarily about description rather than experimental manipu
Chapter3. e contributions of cognitive psychology andpsycholinguistics
the linguistic form in which the message was expressed in the source language)
Daniel Gile
rst, and ndings from cognitive psychology were gradually integrated into them
as he discovered relevant work in that discipline. For more than a decade, in the
Chapter3. e contributions of cognitive psychology andpsycholinguistics
e tide turns
Daniel Gile
limited. e situation has now changed. ough the involvement of cognitive sci
Chapter3. e contributions of cognitive psychology andpsycholinguistics
memory, bilingual perception, bilingual memory, the acquisition of expertise, the
role of attention and resources and allocation of processing resources to di\nerent
tasks. (p. 6)
Daniel Gile
particular pragmatics, into his model as opposed to what he considers the domi
nant and almost exclusive paradigm of generalized information-processing mod
Chapter3. e contributions of cognitive psychology andpsycholinguistics
is performed in actual linguistically mediated communication. ey have a poten
Daniel Gile
Another obstacle lies in the choice of dependent variables. In cognitive psy
chology and psycholinguistics, the most frequently measured dependent variables
are reaction times and proportions of correct responses. Indeed, early studies
Chapter3. e contributions of cognitive psychology andpsycholinguistics
Technological progress has made the device far simpler and less invasive. Seeber
Daniel Gile
Chapter3. e contributions of cognitive psychology andpsycholinguistics
e practisearchers studies provide reasonably reliable evidence with respect
Daniel Gile
Chapter3. e contributions of cognitive psychology andpsycholinguistics
Daniel Gile
Chapter3. e contributions of cognitive psychology andpsycholinguistics
Daniel Gile
2007); so is Kahnemans System 1 vs. System 2 theory (Kahneman 2011) which
incorporates motivation as a determinant of performance, a factor which has been
neglected so far.
It would therefore make sense to introduce some topics in cognitive psychol
Chapter3. e contributions of cognitive psychology andpsycholinguistics
Danks, Joseph H., Gregory M. Shreve, Stephen B. Fountain, and Michael K. McBeath (eds).
Cognitive Processes in Translation and Interpreting
. ousand Oaks, London, New
Delhi: Sage Publications.
Daniel Gile
Chapter3. e contributions of cognitive psychology andpsycholinguistics
Moazedi, Laura. 2006.
Persnlichkeitsunterschiede von angehenden bersetzerinnen und Dol
metscherinnen. Klischee oder Wirklichkeit?
Diploma thesis, Karl-Franzens Universitt Graz.
Monacelli, Claudia. 2005.
Surviving the Role: A Corpus-based Study of Self-regulation in Simul
taneous Interpreting as Perceived through Participation Framework and Int
eractional Polite
ness
. Doctoral dissertation, Herriot Watt University.
Moser, Barbara. 1976.
Simultaneous Translation: Linguistic, Psycholinguistic and Human Inf
or
mation Processing Aspects
. Unpublished doctoral dissertation, University of Innsbruck.
Daniel Gile
Seleskovitch, Danica. 1978. Language and Cognition. In
Language Interpretation and Com
munication
, ed. by G. David and S. H. Wallace, 333342. New York: Plenum Press.
DOI: 10.1007/978-1-4615-9077-4_29
Seleskovitch, Danica. 1975.
Langage, langues et mmoire
 \r\r
Studies from psycholinguistic
andcognitiveperspectives

Discourse comprehension
Adelina Hild
Adelina Hild
Laying a foundation of the structure. Comprehenders use lower level repre
sentations (lexical and syntactic) to lay a foundation for their mental repre
sentation of larger units (sentences, episodes). is process demands cognitive
e\nort and is therefore time-consuming;
Establishing coherent links and mapping the incoming information on the
developing structure of the text. is process is accountable for the docum
ented
rapid access to referentially, temporally and spatially coherent segments; and
Shiing to initiate a new structure or sub-structure when the compre
henders cannot establish coherence with the existing representation of text.
Comprehenders have diculties accessing information that
occurs before the
Adelina Hild
Individual di\nerences in working memory capacity can account for qualitative
and quantitative di\nerences among adults in several aspects of language com
prehension. (Just and Carptenter 1992: 122)
Adelina Hild
Finally, in the abovementioned study, Dillinger (1989) also studied the e\nects
of experience on semantic and syntactic processing in SI. Brie y, he found quan
Prole features
GROUP
ExpertsNovices
Number of subjects88
Number of female subjects44
Mean age
3923
Adelina Hild
the Appendix). Finally, both texts scored 12 on the Flesch-Kincaid Grade index of
x12X13X14x15X16
Mister (x13)
some
the stages (X14)
e6 =: go through (x15, X16)
relationship (x15)
already

s', s'o t5,
e5

s', so' t14,
e6

t', t'n,
t',
t4=n
e above translation of a sentence from
Text1
into DRT parlance is expressed in
the graphical language of DRT. e box contains information on the discourse
referents (denoted by
) and event types (
Adelina Hild
Frequency data on the response categories in the form of percentage was then
analyzed by means of a series of independent z-tests for signi cance, using the for
Variable nameBrief description
Values
TEMPORAL COHERENCETime referencePast/Present Tense
Adelina Hild
Adelina Hild
for both texts and for both experimental groups. Figure1 below suggests that the
e\nect of the variable is only weakly modi ed by di\nerences in experience.
Present TensePast Tense
Figure1.
S1. MrX has reminded us of some of the stages through which
this relationship has gone
BUT
SDRS K
Adelina Hild
boundaries, it was expected that the extra processing e\nort required to generate
Figure2.
0%
IP
AP
Figure3.
Adelina Hild
segments (Vandepitte 1988), a ne-grained analysis which might serve to distin
guish experts from novices.
VARIABLE: DISCOURSE STRUCTURE
Values: [Episode1/Episode2 /Episode3]
Accuracy varied signi cantly across episodes as Table3 below suggests. e z-tests
also established that Episode 1 was processed signi cantly more accurately than
the subsequent episode (
Text1
z= 5.43, p 0.01;
Text2
z= 8.06, p 0.01), while
accuracy for the middle episode was the lowest. e data clearly demonstrate that
Episode 2 signi cantly di\ners from the other two discourse structures for the nov
Episode1Episode2Episode3z[E1/E2]z[E2/E3]
Text1
Experts81%75%70%2.89*2.19
Novices59%48%53%4.48*1.92
All
70%61%64%5.43*1.68
Text2
Experts90%74%87%6.79*7.00*
Novices67%51%60%5.37*3.78*
All
78%62%73%8.06*7.00*
e variations of accuracy across episodes suggest that episode boundaries were
computed during on-line processing and that experts and novices organize their
ST representation in macrostructures re ecting the event structure of the text.
ese ndings replicate a similar, but somewhat weaker e\nect of episode s
tructure
Episode 1Episode 2Episode 3
Experts
Experts
Figure4.
Adelina Hild
protocols (see Section5) they commented that they found one particular high PN
density segment particularly challenging, because, in their words, it was
convo
luted, entangled, could not make sense of it. ey responded by employing a
number of strategies that assisted the translation of the segment, such as impose
new structure, summarize, anticipate and select information based on its function
in the underlying script (the what-when-where script of the narrative text).
ese ndings are consistent with an earlier study by Mackintosh (1985) who
Low PN DensityMiddle PNHigh PN Density
Novices, Text1
Figure5.
TEXT1TEXT2
Figure6.
Adelina Hild
e z-tests calculated on the four basic response categories showed signi cant
accuracy variations for two of them. e highly statistically signi cant results (z=
6.86, p 0.0001) for response category
Paraphrase
demonstrated that subjects
were paraphrasing more for
Text2
. e di\nerence was very signi cant for the nov
ices (z= 2.3, p 0.01), but not so for the experts. e second highly signi cant
result was obtained on response category
Omissions
, indicating that signi cantly
more propositions were omitted from
Text1
(z= 4.61, p 0.0001). e result was
replicated for both groups although for the novices the di\nerence was bigger.
ese results are consistent with the experimental ndings reported by
Category
E(n= 8)N(n= 8)
Perception
Lexical access in SL
Syntactic processing
Text comprehension20
Translation16
Simultaneity
Unidenti ed 4 (8%) 11 (8%)
Total
52 (100%)140 (100%)
Table5.
Reports on sub-categories of problems for text comprehension (TC) for experts
and novices (frequencies relative to the total for a category are given in italics).
Category
E(n= 8)N(n= 8)
TC/
2039
Integration19
Background knowledge
Adelina Hild
e majority of verbalizations in the study referred to discourse processes both
low (perception, lexical search, syntax) and higher-level (text integration and use
of background knowledge) 58% for each group. Text comprehension appears to
be the most e\nortful among the processes reported by the participants Tables4
and 5 indicate that 38% (experts) and 28% (novices) of all problem-related ver
balizations deal with the challenges of constructing a coherent mental model of
the ST as the participants attempted to gain a deeper understanding of the novel
content. Even though background information was mentioned as a problem, it was
by far not the main problem in understanding the text.
Next, the verbalizations concerning strategic processing ware analysed. Eight
strategy types (for a description see the Appendix) were identi ed all of which
were applied to text comprehension problems. ey represent a mix of discourse-
speci c strategies (Kintsch and van Dijk 1983), general prob
lem-solving strate
Sum
Rest
OvergenD
Expl
Acc
Total
TC-Experts212316318
TC-Novices121151121
In conclusion, the protocol analysis outlined important ways in which the s
ubjects
cognitive processes di\ner at a strategic level. While for both groups higher-level
processes were reported as the most signi cant sources of processing problems,
Adelina Hild
With regard to the e\nect of skill on SI processing, the study demonstrated
Adelina Hild
Adelina Hild
Sanford, Anthony J., and Simon C. Garrod. 1981.
Understanding Written Language: Explorations
in Comprehension beyond the Sentence
. Chichester, UK: John Wiley and Sons.
Schank, Roger C. 1976. e Role of Memory in Language Processing. In
e Nature of Human
Memory
, ed. by Charles N. Cofer, 162189. San Francisco: Freeman.
Sinclair, John McHardy. 1985. On the Integration of Linguistic Description. In
Handbook of
Discourse Analysis
, ed. by Teun A. van Dijk, 1328. London: Academic Press.
Singer, Murray. 1990.
Psychology of Language
. Hillsdale, NJ: Lawrence Erlbaum Associates.
Tanenhaus, Michael K, and John C. Trueswell. 2006. Eye Movements and Spoken Language
Comprehension. In
Handbook of Psycholinguistics
(2nd ed.), ed. by Mattew J. Traxler and
Morton Ann Gernsbacher, 863901. London: Academic Press.
DOI: 10.1016/B978-012369374-7/50023-7
Teddlie, Charles, and Abbas Tashakkori. 2009.
Foundations of Mixed Methods Research
. Los
Angeles: Sage Publications.
Treisman, Anne M. 1965. e E\nects of Redundancy and Familiarity on Translating and
Repeating back a Foreign and a Native Language.
British Journal of Pshychology
56(4):
369379. DOI: 10.1111/j.2044-8295.1965.tb00979.x
Underwood, Geo\nrey, and Vivienne Batt. 1996.
Reading and Understanding
. Oxford: Blackwell.
Van den Broek, Paul. 1994. Comprehension and Memory of Narrative Texts: Inferences and
Coherence. In
Handbook Of Psycholinguistics
, ed. by Morton Ann Gernsbacher, 539588.
London: Academic Press.
Van Dijk, Teun A., and Walter Kintsch. 1983.
Strategies Of Discourse Comprehension
. New York:
Academic Press.
Processing problems (PP/)
Comprehension
Perception (P)Problems with hearing
Lexical access in SL (L)Failure to access meaning of a SL
chunk, which has been identi ed as
familiar
Syntactic processing (Syn)Failure to recognize syntax patterns
Text integration (TC/integ/)
Diculties in constructing a coherent
representation for SL chunks
Text comprehension (TC/bgkn)Comprehension diculties due to
lack of background knowledge
Adelina Hild
Processing problems (PP/)
Translation
(Tr/)
Strategy code (SC/)Brief description
Selection (S)Selection of one SL chunk for further processing because it is more
informationally or pragmatically salient
Here it was not so important that the relationship has a background
it is more important that it is a central point in the programme, so this
I decided to convey (E6)
Summarization (SUM)Rendering the gist of a SL segment
e translation took so much time, so I had to give the gist of this bit
here (N14)
Restructuring (Rest)Changing the original syntactic structure of a SL segment (usually
by transposing clauses or segments within the clause), in order to
improve the expression in TL, or anticipation of problems.
is part as the EU is more generally is was not well structured,
so I tried to re-structure it (E3)


UntrainedUntrained +
Students
Students
Students +
Professionals
Study


e EU environment is highly multilingual; up to 22 di\nerent languages are spoken in
http://business-humanrights.org
Actual stimuli are not listed to allow for use of the same materials in future studies. For a

Some list items were repeated in the text.
.
e procedure refers to tests relevant for the present report. e study was larger in scope

MeasureN
Range
Reliability
Age
Experience
years
Experience
days
Cattell
Letter
Corsi
Complex
Experience years
Experience days
Cattell
.57**.58**.47**
Corsi
.38**.30.25
Complex span.34*.37*.38**
Syntax
.37*.23
.19.12.19
Figures
.24.14
.34*.18
Negatives
.62**.20
.03.11.11
Vocabulary: Ratio
.26.33.30.10.37
Vocabulary: Unique
.27.29.18.37.28
Companies: Di\nerence.38*.32.39*
.11.07
Companies: Average
.51**.05
.37*.22
Median EVS.22.36.43**
.18.01.03
**
p
 .05, *.05
p
 .10
Experience yearsExperience days
Cattell
.22.12
.04.06
Corsi
Complex span.18.22
Syntax
Figures
Negatives
Vocabulary: Ratio
.25.02
Vocabulary: Unique
.30.02
Companies: Di\nerence.07.21
Companies: Average
Median EVS.33*.40**
Discussion




Process and text studies of a translation problem
Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
Universiteit Gent

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
consist of more than one primary component and are therefore supposed to
require a more elaborated and indirect processing. Although reaction times
Chapter6. Process and text studies of a translation problem
which there is no keyboard activity, when a translator starts reading a source lan
guage sentence and begins producing (i.e., typing in) a translation. In other words,
a translation pause is a mentally productive timespan in which no keystrokes can
be recorded and its duration represents the demand of cognitive processing, in
particular that of working memory (see also Immonen 2011, 236). However, such
cognitive processing is not necessarily restricted to the beginning of sentences,
and this is also clear in Immonen and Mkisalos broader view of pauses (2010),

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
of the translated sentence was hit and before the second key of the lexical verb
in the sentence was struck. However, in nine cases lexical verbs only occurred
at the very end of the sentences (e.g.
Met zijn pensioen kon hij alleen goedkope
dingen kopen [gloss: With his pension could he only cheap things buy]; Dankzij deze
opdracht werd een nieuwe strategie ontwikkeld [anks to this task was a new strat
egy developed])
. Such word order may imply that a writer delays the expression of
the verb that is already present in his or her mind, but it is not clear at which point
exactly that knowledge is present. e data from these sentences have therefore
been excluded from further consideration. Together with the tran
slation onset
time, it is this initial position that is expected to reveal most translation problems
in the metonymic sentences. Findings have been recorded of b
oth unidiomatic
Chapter6. Process and text studies of a translation problem

Table1.
Translation

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
fewest pauses, i.e., the end position, has pauses that are almost three times as long
as mid sentence pauses.
When we relate these quantitative process ndings to the quality of the prod
First
tertile
Second
tertile
ird
tertile
NAccept-
ability
Language
errors
NAccept-
ability
Language
errors
NAccept-
ability
Language
errors
Chapter6. Process and text studies of a translation problem

length (middle tertile), we could advise translation trainees not to spend too much
time on diculties. However, such a conclusion may be premat
ure: shorter pauses
in the third tertile may actually lead to even more infelicitous translations for the
problems that the student has come across.

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
Discussion
Immonen and Mkisalo (2010) found that the processing of larger syntactic units
requires longer pauses in both monolingual text production
and in translation.
Chapter6. Process and text studies of a translation problem

that they investigated. In the condition with the metonymic
construction, the
Mean (s)Standard deviation
Untrained secondary school group

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
Most importantly, this case study generalizes the results from our previous study
e abbreviations L2a and L2b have been used rather than the notion of a third language, so
Chapter6. Process and text studies of a translation problem

Mean word translation time
L2b Untrained university student (N= 1)
Non-prototypical agent noun2.22
Prototypical agent noun1.92
L1 Untrained university group (N= 27)
Non-prototypical agent noun1.47
Prototypical agent noun1.39
L1Trained university group (N= 27)
Non-prototypical agent noun1.22
Prototypical agent noun1.22
However, here, too, the L2a
L2b students latencies for these nouns were not
the longest in the group of untrained students, with two students requiring an

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
it is remarkable that the L2A
L2b student produced none of these errors, most
of which concerned spelling, among the non-prototypical agent translations (see
Appendix B). As for lexical choice and priming, only three cases were found in
which the L2a
L2b student was primed where most of the L2
L1 students
were not: instead of
broer, neef
and
winst,
she opted for
broeder, kozijn,
and
pro
jt
as translations of
brother, cousin
and
projt
respectively
On the contrary, the
L2b student took great care not to be primed, looking for alternatives. She
actually did so in cases where the majority of students turned out to produce cog
nate translations (Appendix C).
e data from this third study raise questions about the generalizability of
123456789101112131415161718192021222324252627
Figure1.
Chapter6. Process and text studies of a translation problem

Conclusion
Although the three studies described above are underpowered and require fur
ther investigation with larger samples, the ndings are largely consistent with the
data from Vandepitte and Hartsuiker (2011). e ndings consistently reveal that

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
Chapter6. Process and text studies of a translation problem

Appendix A
Duration (s)

Sonia Vandepitte, Robert J. Hartsuiker, and Eva Van Assche
Appendix B
L2b translations (less acceptable items or errors in italics)
Non-prototypical agents
Prototypical agents
ENNLENNL
action actie actor acteur
address adresactressactrice
adviceraad
angelengel
Chapter6. Process and text studies of a translation problem

Appendix C
Non-primed translations by a minority of students (incl. L2a
L2b student)
ENNLstd no (total N= 58)
adviceraad12
justice gerechtigheid28
missionopdracht
passion
hartstocht
plague
pest
revolt
opstand
section
afdeling12
couple
paar

Post-editing machine translation
Eciency, strategies, and revision processes

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
many more plays a vital role within the translational work ow (cf. Hofmann 2012;
Mertin 2006). Computer-aided translation (CAT) tools and translation memory
(TM) systems have become integral parts of a majority of translators environ
ments (cf. Folaron 2010; OBrien 2012) to facilitate and support translation tasks
as well as making them more cost-e\nective without reducing quality. Machine
translation (MT), which in the meantime has come of age, is now heading in the
same direction. e growing needs for information and communication, as an
e\nect of globalization, are producing vast amounts of machine-generated transla
tions oen for personal use but with signi cant increase in commercial usage as
well. Figures from an AMTRA/SDL survey on MT
among managers of global
enterprises underline the general growing interest. 28% of the respondents are
either already using or planning to use MT and 57% of the surveyed are even in
favor of post-editing machine translation (PEMT) regarding the combination of
best of both human quality and machine eciency as a win-win situation.
us, more and more translators are facing a work load increase in post-editing
tasks since quality levels of MT output can vary signi cantl
y depending on the
system and language pair and are still far from human translation (HT) standards.
Post-editing (PE) is not a relatively new phenomenon in the translational
world. As a matter of fact, it is as old as machine translation (MT) itself and the
two have been lifelong companions in the form of a human translator post-edit
ing machine translation output (Koehn 2010b, 537). e developments outlined
above show that human translators alone cannot satisfy the demand because data
volumes are too massive and costs too high, PE is becoming a serious and vital
part of the translational landscape. Post-editing is de ned as follows:
Post-editing is the correction of machine-generated translation output to ensure
Chapter7. Post-editing machine translation

PE e\nort and technology
PE impact on translation education and the role of the translator

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
Types of users/purpose
Text intention and quality expectations are relevant for deciding what type of post-
editing will be the best choice for the purpose (cf. Specia 2011a). Costs and time
Chapter7. Post-editing machine translation

Technology
As research has shown, the quality of the MT output is essential for productivity
and hence cost eciency of post-editing (cf. Specia 2011b). High MT quality may
increase productivity, low quality can be very tedious and m
ay lead to frustration.
Considering that PE is closely linked to machine translation development, research
is very eager to consistently improve the technical aspect. Wh
at essentially started
in the 1960s with the development of basic machine translation systems such as
SYSTRAN
(an early rule-based MT system) translation soware has fou
nd its
preliminary peak in recently developed online real time translation tools (RTTS)
such as Google Translate, a new generation of interactive hybrid machine trans
lation systems for example TransType
http://www.systran.de/
A User-Friendly Text Prediction for Translators, http://rali.iro.umontre
al.ca/rali/?q=en/
TransType
experimental Web-Based Interactive Computer Aided Translation Tool developed by
the Machine Translation Group at the University of Edinburgh. http://www.caitra.org/
.
PET: a Tool for Post-editing and Assessing Machine Translation (2012) http://pers-www.wlv.
ac.uk/~in1676/publications/2012/AZIZ+LREC2012.pdf
Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
Education
Papers or studies on teaching post-editing are sparse. OBriens paper on this sub
Chapter7. Post-editing machine translation

a new way of working on it, for a new aim. (Laurian 1984, 237) and therefore
acknowledging that PE is di\nerent from translating might be the rst step towards
convincing those translators that acquiring the necessary new skills is a chance to
keep up with the changing demands in order to bene t from this growing branch,
the bright future prospects of which cant be ignored.
A translation experiment
We carried out an English-German translation experiment which is part of a pilot
study for the development of an up-to-date browser-based MT post-editing tool

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
Figure1.
Screenshot of a post-editing session in Translog-II. e circles and dots in the
upper source text window represent gaze xations and gaze samples respectively.
Experimental design
A pilot study was conducted with Translog-II as a preparatory investigation into
post-editing processes, and as a baseline for later studies with the CasMaCat work
bench. Six English source texts, two of which were sociology texts from an ency
clopedia and four adapted versions from British newspaper texts were translated
in three di\nerent ways: 1. from-scratch translation (T), 2. post-editing the MT
output of Google Translate (PE) and 3. monolingual post-editing, i.e., post-editing
the Google output without access to the source text (E). e 6 English texts were
Chapter7. Post-editing machine translation

experiments, and hence no gaze data could be collected. An evaluation of the
Task
Translation (T)Post-editing (PE)
Monolingual post-editing (E)
Total
Text123456123456123456
788888878878788788139
ES111112101281012101281210910101011188
HI7767668128101211000000100
ZH033333353332021422
222222222222222222

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
English-German sub-project
e English-German sub-project was conducted at the eye lab of the Faculty of
Translation Studies, Linguistics and Cultural Studies (FTSK) of the Johannes
Gutenberg-University of Mainz in Germersheim in July and August 2012. 12
professional translators and 12 non-professionals (students), all of them German
native speakers translating from English (L2) into German (L1), participated in
the experiment and processed 6 texts each according to the pre-de ned permutat
ing pattern described above making sure all 6 texts were translated, post-edited
and edited in equal numbers. Key logging (Translog-II), non-invasive eye track
ing (Tobii TX300) and questionnaires before and aer the experiment served as
instruments for the multi-method data recording. Addition
al gaze data was gained
by screen recording the entire processing task.
e experiment was structured as follows: Aer a short introduction outlining
the general study context and the embedding of the experiment into the project,
a questionnaire was given to gather some personal participant information about
gender, visual aids, level of translation training and work experience as transla
tors in general, experience with machine translated output as well as post-editing
Chapter7. Post-editing machine translation

participants. Due to technical problems participant 19 from the professional group
Highly
satised
Somewhat
satised
NeutralSomewhat
dissatised
Highly
dissatised
Professionals9%45.5%9%27.5%9%
Students8%67%8%17%0
Overall9%56%9%22%4%
Concerning the satisfaction with the monolingual post-editing task, Figure2
shows that post-editing is obviously more satisfactory than editing. is early
positive impression is put into perspective by looking at the outcome of the ques
tionnaire, where the overwhelming majority, 83% of the participants, answered the
question if they had rather translated from scratch than post-edited MT output
with yes. e clear vote against post-editing is additionally underlined by the fact
that 78% of the participants explained that they would have preferred to trans
late from scratch. Furthermore, more than two thirds (69.5%) of the participants
stated that they actually had to post-edit 75100%
of the machine translated
.
Work experience here is de ned as working as a fully paid professional translator.
.
Evaluation results are presented either in percentage (%) or rating value (decimals).
.
See
InE

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
output. A tendency can be seen in so far that professionals went for always
and
100% whereas students rather opted for 75% for both scenarios. Nevertheless
editing seems to be the least satisfactory task.
As shown in Table3, the conscious and subjective rating of all four of the
given criteria of machine translated output is categorized as below average and
with an overall rating value of 1.24 (highlighted in bold) hence extremely nega
tive. e rating value is calculated based on a rating scale from 2 to 2 in integer
steps (i.e. 2, 1, 0, 1, 2) whereas 2 always refers to the most nega
tive answer.
us, rst the counts per answer are multiplied with the correlating integer, then
all results are summarized and divided by the total number of participants. For
the criteria style this means:
(15x2) + (4x1) + (30)+(11) + (02) = 33(sum) : 23(participants)= 1.43
neutralsomewhat
Figure2.
Comparison PE vs. E participant satisfaction.
So style was given the lowest rating with 1.43; accuracy was at the opposite end
with 1. e professionals evaluated signi cantly stricter
with an average rating of
1.46 for all four criteria compared to the students rating of 1.03. All in all, rat
ing here is rather homogenous among all participants. Only accuracy and gram
maticality were perceived di\nerently by professionals and students.
Table3.
Overview general rating of MT output.
Rating scale21012Rating
Well below
average
Below
average
Average
Above
average
Well above
average
Criteria
counts
(professionals/students)
Grammaticality
Style
Accuracy
Overall evaluation
(2/2)1.26
Overall rating value
Chapter7. Post-editing machine translation

e overall rating of the machine translated output pres
ented here gave clear
negative feedback with regard to all 4 criteria as hardly any rating was f
ound at
average or above. Both groups agreed on the necessity to change the majority of
the MT output herewith indicating that the MT quality was rather unacceptable.
Generally, the professional translators were more critical than the students. e
following sections will show how these subjective ratings are related to the empiri
cal key-logging and eye-tracking data.
Evaluation of translators unconscious reading and writing data
is section describes some ndings of the eye-tracking and key-logging experi
ments for the English to German translations. We initially give an overvie
w over
the actual translation behavior for the entire experiment and compare the e
tasks were distributed for the 24 participants.
Table4.
Task overview for all 24 participants: T: translation, P: post-editing,
E: monolingual editing.
Figure3 shows the translation/post-editing/editing times (respectively TT, PT and
ET) for each of the 23 translators (Translator 19 was taken out as only 2 translation

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
2131142418816112026152210913412231775
Figure3.
Average time in ms per word for 23 translators: Translation (TT), Post-editing
(PT) and Editing (ET).
Chapter7. Post-editing machine translation

It is a new way of considering a text, a new way of working on it, for a new aim,
we observe similar phases in both types of translation: an (optional) orientation
phase, a draing (or post-editing) phase in which the actual translation is pro
duced (or post-edited) and an optional nal revision.
Figure4 plots a post-editing session with a draing and a revision phase. e
horizontal axis represents the post-editing time; the vertical axis enumerates the
ST words (words are numbered from 1 to 160). e symbols in the graph indicate
the post-editing activities in time (keystrokes and xations) and map them on
the ST words to which they relate. Horizontal lines separate di\nerent source text
sentences (segments), numbered from seg1 to seg 9.
e post-editor starts from the beginning of the text working more or less
linearly until the end of segment 5. At time stamp 120.000 (aer two minutes),

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
02000060000100000140000180000220000260000300000340000380000420000460000
Seg:10
Seg:9
Seg:8
Seg:7
Seg:6
Seg:5
Seg:4
Seg:3
Seg:2
Seg:1
Figure4.
e progression graph plots post-editing session participant P03, text P1 (P03_P1) with a draing phase and a revision phas
e.
Green diamonds represent gaze activity on TT tokens; blue circles, gaze activity on a ST token. e size of the diamond
s and circles correlates
Chapter7. Post-editing machine translation

050000100000150000200000250000300000350000400000450000500000550000600000650000700000
Figure5.
Post-editing session P06_P2 shows almost no source text consultation.

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
110120130
90100
7080
5060
3040
1020
90100
7080
5060
3040
1020
050000100000
Figure6.
Four di\nerent types of orientation phases observed during post-editing: le: careful reading of ST (P05_P6),
middle le: reading of the ST (P11_P3), middle right: rst reading TT then ST (P04_P3), right: reading ST followed
by reading of TT (P07_P02).
Chapter7. Post-editing machine translation

46485052545658
292000296000300000304000308000312000316000320000324000328000332000336000340000
Figure7.
Post-editing pattern, bottom: read ST segment, edit, than consult/check ST; top: read TT segment, th
en amend MT
output and monitor changes. Both segments are taken from session P02_P5.

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
According to our observations, the revision phase is, in many cases, similar to
from-scratch translation where post-editors as well as translators mainly re-read
Chapter7. Post-editing machine translation

to in uence the cognitive processing e\nort during the translation task whereas
post-editing is not a\nected. is might be an indicator that PEMT improves the
eciency of working with complex texts. In order to understand why this is the
50
100
150
200
250
300
350
400
450
Text 1Text 2Text 3
Text 1Text 2Text 3
Figure8.
Processing of source text, Translation vs. Post-editing.
In the remainder of this section we will look the post-editing patterns for Text 3
more closely, as this is the most dicult one. Text 3 consists of 145 ST words while
the Google translation into German was 148 words in length. As indicated in
Table5, the Google translations of Text 3 were post-edited by 8 participants (P01,
P04, P10, P11, P15, P17, P21 and P23, see also Table4) each of whom performed
Participant
P01P04P10P11P15P17P21P23
Insertions
445234228305391448278164

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
It appears that some post-editors conduct a large number of editing activities (e.g.
P01) while others execute relatively few keystrokes (e.g., P10). Some post-editors
.
e InE\n value can be 1 in translation, since a blank is counted as part of a word, but not
always realized, e.g. when a full stop immediately follows the word without blank.
Chapter7. Post-editing machine translation

59
711
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
53
55
57
59
61
63
65
67
69
71
73
75
77
79
81
83
85
87
89
91
93
95
97
99
0
1
2
3
4
5
6
7
8
P01_P3
P04_P3
P10_P3
P11_P3
P15_P3
P17_P3
P21_P3
P23_P3
Figure9.
Overlaid InE\n measure (vertical) for translations of the 145 ST words of text 3 (horizontal) of 8 post-edito
rs, P01, P04, P10,
P11, P15, P15, P21 and P23.

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
Despite the fact that di\nerent post-editors di\ner by more than 100% in their
overall editing activity (cf. Table5), the graph in Figure9 reveals that post-editors
oen modify the same translations, or sequences of translations: instead of a uni
form distribution of the edit operations over the entire text, some passages of ST
seem more dicult to translate (or their MT output to adjust) for all post-editors
than others, resulting in common areas of
InE
peaks. For instance, the whole
phrase around ST words 37 to 47 His withdrawal come in the wake of ght ar
ing, for which Google produced a literal and unacceptable German translation,
Chapter7. Post-editing machine translation

processing the source text. Machine translated output can thus be assumed to be
an ecient preparatory step for producing a translation, which again contradicts
the personal judgments of the translators involved in our experiment.
In addition, the InE\n measures presented in Figure9 show that the post-edit
ing e\norts are concentrated on single, more dicult constructions in the text but
are not evenly distributed over the text. is contradicts the translators statement
that they had to post-edit 75100% of the machine translated output.
Overall, the subjective rating clearly shows a negative attitude towards
machine translation (see Table2). e ndings concerning the reading and writ
ing data, however, do not support this negative picture. Future work on the qual
ity of the machine translation output compared to the post-edited and translated

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
collaborative translation and post-editing projects. As a basis, empirical research
on PEMT is needed to identify processing patterns and recurrent strategies within
the process of post-editing. Such empirical results could lead to standardized rules
for PEMT, which in turn could revolutionize the training and teaching of transla
Chapter7. Post-editing machine translation

Carl, Michael. 2012b. e CRITT TPR-DB 1.0: A Database for Empirical Human Translation
Process Research. In
Proceedings of the AMTA 2012 Workshop on Post-Editing Technology
and Practice (WPTP 2012)
, ed. by Sharon OBrien, Michel Simard, and Lucia Specia, 918.
Stroudsburg, PA: Association for Machine Translation in the Americas (AMTA).
Carl, Michael, and Martin Kay. 2011. Gazing and Typing Activities during Translation: A Com
parative Study of Translation Units of Professional and Student Translators.
Meta
952975. DOI: 10.7202/1011262ar
Carl, Michael, Barbara Dragstedt, and Arnt Lykke Jakobsen.. 2011. A Taxonomy of Human
Translation Styles. Accessed from: http://translationdirectory.com/articl
es/article2321.
php, June 8, 2014.
Choudhury, Rahzeb, and Brian McConnell. 2013. TAUS Translation Technology Landscape
Report. Accessed from: http://www.translationautomation.com/rep
orts/taus-translation-
technology-landscape-report, October 29, 2013.
Doherty, Stephen, and Sharon OBrien. 2009. Can MT Output be Evaluated through Eye Track
ing? MT Summit XII, Ottawa, Canada.
Doherty, Stephen, Sharon OBrien, and Michael Carl. 2010. Eye Tracking as an MT Evaluation
Technique.
Machine Translation
24 (1):113. DOI: 10.1007/s10590-010-9070-9
Doherty, Stephen, and Joss Moorkens. 2013. Investigating the Experience of Translation Tech
nology Labs: Pedagogical Implications.
Journal of Specialised Translation
19: 122136.
Accessed from: http://www.jostrans.org/issue19/art_doherty.php, July 21, 2013.
Fiederer, Rebecca, and Sharon OBrien. 2009. Quality and Machine Translation: A Realistic
Objective?
Journal Of Specialised Translation
11: 5274. Accessed from: http://www.jos
trans.org/issue11/art_ ederer_obrien.pdf, February 3, 2013.
Folaron, Deborah A. 2010. Translation Tools. In
Handbook of Translation Studies: Volume 1
, ed. by Yves Gambier and Luc van Doorslaer, 429436. Amsterdam: John Benjamins.
DOI: 10.1075/hts.1.tra9
He, Yifan, Yanjun Ma, Johann Roturier, Andy Way, and Josef van Genabith. 2010. Improving
the Post-editing Experience using Translation Recommendation: A user Stud
y.
Proceedings
of the 9th Annual AMTA Conference
, 247256, Denver. Accessed from: http://doras.dcu.
ie/15803/, June 8, 2014.
Hofmann, Sascha. 2012.
Prozessgesttztes bersetzen: vom funktionsorientierten bersetzungsprozess
zum Geschsprozessmodell fr die Dienstleistung bersetzen
. Lichtenberg: Harland media.
Hvelplund, Kristian Tangsgaard. 2011.
Allocation of Cognitive Resources in Translation an Eye-
tracking and Key-logging Study
. PhD thesis, Department of International Language Studies
and Computational Linguistics, Copenhagen Business School.
Jakobsen, Arnt Lykke. 1999. Translog Documentation.
Copenhagen Studies in Language
24:
Jakobsen, Arnt Lykke. 2011. Tracking Translators Keystrokes and Eye Movemen
ts with Trans
log. In
Methods and Strategies of Process Research Kapitel 4
(Benjamins Translation Library,
Vol. 94), ed. by C. Alvstad, A. Hild, and E. Tiselius, 3755. Amsterdam: John Benjamins
Publishing Company. DOI: 10.1075/btl.94.06jak
Kirchho\n, Katrin, Anne M. Turner, Amittai Axelrod, and Francisco Saavedra. 2011. Applica
tion of Statistical Machine Translation to Public Health Information: A Feasibility Study
.
Journal of the American Medical Information Association
18 (4): 473478. Accessed from:
http://jamia.bmj.com/content/18/4/473.full.pdf+html, March 13, 2013.
DOI: 10.1136/amiajnl-2011-000176
Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
Koehn, Philipp, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico,
Nicola
Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej
Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open Source Toolkit for
Statistical Machine Translation.
ACL Companion Volume. Proceedings of the Demo and
Poster Sessions
, 177180. Prague, Czech Republic, June 2007. Association for Computa
tional Linguistics.
Koehn, Philipp. 2009. A Web-Based Interactive Computer Aided Translation Tool.
Proceedings
of the ACL-IJCNLP 2009 Soware Demonstrations
, 1720. Suntec, Singapore.
Koehn, Philipp. 2010a.
Statistical Machine Translation
. Cambridge: Cambridge University Press.
Koehn, Philipp. 2010b. Enabling Monolingual Translators: Post-Editing vs. Options.
Human
Language Technologies: e 2010 Annual Conference of the North American Chapter of the
Association for Computational Linguistics (HLT/NAACL)
, 537545. Los Angeles, Califor
nia: Association for Computational Linguistics,. Accessed from: http://www.aclweb.org/
anthology/N/N10/N10-1078.pdf, June 8, 2014.
Koponen, Maarit. 2012. Comparing Human Perceptions of Post-editing E\nort with Post-edit
ing Operations.
Proceedings of the 7th Workshop on Statistical Machine Translation 2012
181190. Montreal, Canada, Association for Computational Linguistics. Accessed from:
http://www.statmt.org/wmt12/pdf/WMT23.pdf, March 5, 2013.
Koponen, Maarit, Wilker Aziz, Luciana Ramos, and Lucia Specia. 2012. Post-editing Time as
a Measure of Cognitive E\nort.
Proceedings of the AMTA 2012. Workshop on Post-editing
Technology and Practice (WPTP 2012)
Krings, Hans. 2001.
Repairing Texts: Empirical Investigations of Machine Translation Post-edit
ing
Processes
. Translated (German into English) by G. Koby, G. Shreve, K. Mischerikow, and
S. Litzer; ed. by G. Koby. Kent, OH: Kent State University Press Translation Studies Series.
Chapter7. Post-editing machine translation

OBrien, Sharon. 2012. Translation as Human-Computer Interaction.
Translation Spaces
(1): 101122. Accessed from: http://doras.dcu.ie/17541/1/Translation_as_HCI_OBrien.
pdf, January 31, 2013. DOI: 10.1075/ts.1.05obr
OBrien, Sharon, Johann Roturier, and Giselle de Almeida. 2009. Post-Editing MT Output
Views from the Researcher, Trainer, Publisher and Practitioner. Accessed fro
m: http://
www.mt-archive.info/MTS-2009-OBrien-ppt.pdf, July 12, 2013.
Prez, Celia Rico. 2012. A Flexible Decision Tool for Implementing Post-editing Guidelines.
Localisation Focus
11 (1): 5465. Accessed from: http://www.localisation.ie/resources/loc
focus/LocalisationFocusVol11_1Web.pdf, June 8, 2014.
Plitt, Mirko, and Franois Masselot. 2010. A Productivity Test of Statistical Machine Transl
tion Post-Editing in a Typical Localisation Context.
e Prague Bulletin of Mathematical
Linguistics NUMBER
Poulis, Alexandros, and David Kolovratnik. 2012. To Post-edit or not to Post-edit? Estimating
the Bene ts of MT Post-editing for a European Organization.
AMTA-2012: Workshop on
Post-editing Technology and Practice. Proceedings
, San Diego, October 28, 2012. Accessed
from: http://www.mt-archive.info/AMTA-2012-Poulis.pdf, June 8, 2014.
Przybocki, Mark, Greogory Sanders, and Audrey Le. 2006.
Edit Distance: A Metric for Machine
Translation Evaluation
. National Institute of Standards and Technology (NIST). Accessed
from: http://www.mt-archive.info/LREC-2006-Przybocki.pdf, July 22, 2013.
Schae\ner, Moritz, and Michael Carl. 2014. Measuring the Cognitive E\nort of Literal Transla
tion Processes. In
Workshop on Humans and Computer-assisted Translation
, ed. by Ulrich
Germann, 2937. Association for Computational Linguistics, 2014.
Specia, Lucia. 2011a.
Quality Estimation of Machine Translation
. Abstract to Video talk on the
Homepage of DCU, School of Computing, Dublin City University. Posted by gconway
07/07/2011. Accessed from: http://www.computing.dcu.ie/bl
ogs/dr-lucia-specia-quality-
estimation-machine-translation-4th-july-2011, June 8, 2014.
Specia, Lucia. 2011b. Exploiting Objective Annotations for Measuring Transl
ation Post-editing
E\nort.
15th Annual Conference of the European Association for Machine Translation
, 7380,
Leuven, Belgium.
Specia, Lucia. 2013. Statistical Machine Translation. In
Emerging Applications of Natural Lan
guage Processing: Concepts and New Research
, ed. by S. Bandyopadhyay, S. K. Naskar, and
A. Ekbal, 74109. Hershey, PA: IGI Global.
Tatsumi, Midori, Takako Aikawa, Kentaro Yamamoto, and Hitoshi Isahara. 2012. How Good
Is Crowd Post-Editing? Its Potential and Limitations.
Proceedings of the AMTA 2012
Workshop on Post-editing Technology and Practice (WPTP 2012)
. Accessed from: http://
amta2012.amtaweb.org/AMTA2012Files/html/8/8_paper.pdf, June 8, 2014.
Temnikova, Irina. 2010. A Cognitive Evaluation Approach fo
r a Controlled Language Post-
Editing Experiment.
7th International Conference on Language Resources and Evaluation

Michael Carl, Silke Gutermuth, and Silvia Hansen-Schirra
Van der Meer, Jaap. 2013.
Choose your Own Translation Future
. Accessed from: http://langtech
news.hive re.com/articles/share/27171, January 28, 2013.
Vashee, Kirti. 2011
. An Exploration on Post-Editing MT Part I
. Accessed Blog from: http://kv-
emptypages.blogspot.de/2011/02/exploration-of-post-editing-mt-part-i.html, June 8, 2014.
Wagner, Emma. 1985. Post-Editing Systran A Challenge for Commission Translator.
Termi
nologie et Traduction
Wagner, Emma. 1987. Post-editingPractical Considerations. In
e Business of Translation
and Interpreting
, ed. by Catriona Picken, 7178. London: Aslib.
Winther-Balling, Laura, and Michael Carl. 2014. Production time across languages a
nd tasks:
A large-scale analysis using the critt translation process database. In
e Development
of Translation Competence: eories and Methodologies from Psycholingu
istics and Cogni
tive Science

On a more robust approach to triangulating

Igor Antnio Loureno da Silva
is is a description of soware like Translog (Jakobsen and S
chou 1999), available at http://
translog.dk. ere are other alternatives, such as Inputlog, allowing participants to use regu
lar
text editors (for a comparison of some of these alternatives, see: http://www.writingpro.eu/log
Available at: http://www.qsrinternational.com/products_NVivo.aspx.

Igor Antnio Loureno da Silva
Igor Antnio Loureno da Silva
(see OBrien 2013 for a discussion on the interdisciplinary nature of the trans
lation studies), and why many researchers aliated to expertise studies in fact
focus on expert performance (for an account of expertise and expert performance,
see Ericsson, Charness, Feltovich, and Ho\nman 2006), that is, on the a
nalysis
of behaviours. Among several examples in this direction are Shreves (2006: 154)
questioning under what conditions and in what ways does translation compe
tence evolve to support expertise? and Jakobsens (2005) claim that instances o
peak performance are those in which over 60 keys are pressed sequentially without
being interrupted by any pauses shorter than 2.4 seconds.

Igor Antnio Loureno da Silva
(a) accumulating signi cant episodic memory under the conditions of deliberate
practice, (b) applying goal-directed pattern recognition to domain-relevant events
represented in episodic memory, where the goal is the recognition and storag
e of
patterns that will identify task-relevant problems, e.g., patterns calling for action
to be taken, (c) attaching domain-relevant meaning to such patterns and link
For a comprehensive account of working memory and other types of memory, see Dragsted

Igor Antnio Loureno da Silva
text. Cross-boundary segments are those segments that do not follow the normal
production ow and cross the boundaries of the word, group, or clause. Cross-
sentence segments cross the boundaries of the sentences, but the two parts of the
involved sentences can be identi ed as a word, group, clause and/or sentence.
In sum, segmentation at higher ranks can be indicative of higher capacity of
operating the translation task and a higher development of ones working mem
ory
(Dragsted 2004), and thus, indicative of a higher level of expertise. Representation
at higher ranks are also indicative of a higher level of expertise (da Silva 2007)
given that expertise involves, as Chi (2006a, 2006b) points out, the ability to orga
nize and process knowledge in depth.
Procedures for identifying and classifying segments and representations will
Subjects expertise was preliminarily assessed on social grounds, that is, their peers recog
nition of them as experts in the eld, their work at a leading centre, and their publications in
relevant journals in the area. For a social notion of expertise, see Collins and Evans (2007).
See: http://admin-apps.webonowledge.com/JCR/static_html/notices/notices.htm.
Igor Antnio Loureno da Silva
e two experts in sickle cell disease translated a research article on sickle cell
disease as their domain-speci c task, and a research article on Chagas disease as
their non-domain task. e two experts in Chagas disease translated a research
article on Chagas disease as their domain-speci c task, and a research article
on sickle cell disease as their non-domain task. e two texts were the same for
both groups.
Data elicitation techniques were questionnaires, direct observation, key log
ging, screen recording, and free retrospective protocols. Fo
r segmentation and
.
is value is obviously arbitrary, and particularly motivated by group membership (see A
lves
2003) and therefore a need of ensuring comparability across studies. Dragsted (2004, 2005) sug
gests a exible approach using heuristics. Jakobsen (2005) suggests 2.4 seconds for studies con
Igor Antnio Loureno da Silva
di\nerent parental nodes, as subnodes cannot overlap in the same node. If this
happens, this is a sign that the categories (subnodes) were not adequately de ned.
Figure1 shows an example of some nodes. e source column displays how
many les contain a given node (each le refers to one subjects translation task),
and the reference column displays how many hits are found for that node. Note
that somehow the columns are rightly displayed only for the parenta
l notes, but
not for the subnodes. Anyway, the last numbers in each row correspond to the
references, and the penultimate correspond to the sources.
Figure1.
Nodes used in NVivo.
e present analysis focuses on the third node (Representao da tarefa, i.e.,
task representation) and its subnodes: Grupo (group), Orao (claus
e), Palavra
(word), Segmento no-sinttico (cross-boundary segment), Sentena (sentence),
and Transentencial (cross-sentence segment).
Aiming at reducing subjectivity in the codi cation process, such categories
were previously established and described in the soware. To improve internal
Subject
Task
Time (in seconds)
S1DS8,765
NDS7,255
S2DS3,070
NDS4,220
S3DS4,240
NDS5,101
S4DS3,560
NDS3,662
Note: DS = domain-speci c; NDS = non-domain-speci c.
As Table2 shows, the domain-speci c tasks are not necessarily performed faster
than the non-domain speci c tasks. S1 took longer to translate the domain-spe
ci c task than the non-domain speci c task, and the amount of time S4 spent
on both tasks is considerably close. Table2 also shows that S1 is the subject that
spent more time to perform both tasks, and S2 and S4 were the fastest subjects to
Igor Antnio Loureno da Silva
Subject
Task
Word
Group
Clause
Sentence
Cross-
sentence
Cross-
boundary
Un-
matched
Total
S1DS34.83%35.96%19.10%1.12%3.37%5.62%89
NDS44.68%40.43%
2.13%1.06%
S2DS34.33%35.82%19.40%1.49%7.47%1.49%67
NDS47.73%28.41%13.64%1.14%3.41%3.41%
S3DS34.57%16.05%
1.23%2.47%12.35%34.57%81
NDS33.33%38.89%12.22%1.11%3.34%11.11%90
S4DS18.75%34.38%21.88%15.61%9.38%32
NDS28.33%33.33%23.33%8.33%6.68%60
Note: DS= domain-speci c; NDS= non-domain-speci c.
Source text
sndromes falciformes(SF)constituemum
conjunto
demolstias qualitativas
sickle cell syndromes(SCS)
constitute
Figure3.
S3s linear representation.

Igor Antnio Loureno da Silva
As mentioned above, S3 and S4 stand out in the group for having higher per
centages of cross-boundary segments, particularly in their domain-speci c tasks.
Table3 also shows a comparatively high percentage of unmatched segments for
S3, which singles this subject out from the rest of the group. is particular nd
ing was rst brought up in Figure2 and will be further explored in Section5. e
following paragraphs will report on the representation data.
Subject
Task
Word
Group
Clause
Sentence
Cross-
sentence
Cross-
boundary
Un-
matched
words)
S1DS58.56%36.47%
2.32% 2.65%
1,188
NDS19.82%48.84%12.76%15.89%2.69%1,117
S2DS48.26%26.88%15.58%
777
NDS25.78%46.50%13.99%13.73%442
S3DS18.03%32.26%23.81%24.11%1.79%897
NDS10.25%62.00%25.90%1.85%619
S4DS57.15%18.42%10.70%13.73%387
NDS56.22%32.43%11.35%363
Note: DS= domain-speci c; NDS= non-domain-speci c.
Table4 shows that the number of words varies largely across subjects, with S1
and S4 being the subjects who speak the most and the least, respectivel
y. It also
displays a tendency in subjects explanations and comments to point to repre
sentation of the task at word rank for S1, S2 and S4 in their domain-speci c task,
this tendency being reversed in their non-domain task for S1 and S2, who have
a higher percentage of representation at group rank. S4 has a consistent repre
sentation at word rank for both tasks. S3 clearly stands out for representing at
As the subjects are experts in areas of the medicine domain, and not linguistics or related

Igor Antnio Loureno da Silva
Pagano and da Silva (2008) cross-examined process data and text produc
tion and also found S3 as clearly standing out in the sample. If one examines S3s
As mentioned in Section3, subjects usually verbalized in Portuguese. ese and o
ther quotes
were translated into English.
Igor Antnio Loureno da Silva
expected to take place at lower ranks for the non-domain task when compared to
Igor Antnio Loureno da Silva
References
Alves, Fabio. 1995.
Zwischen Schweigen und Sprechen: Wie Bildet sich eine Transkulturelle
Brcke?
Hamburg: Dr. Kovac.
Alves, Fabio. 2003. Traduo, Cognio e Contextualizao: Triangulando a Interface Processo-
Produto no Desempenho de Tradutores Novatos.
D.E.L.T.A.

Igor Antnio Loureno da Silva
Gile, Daniel. 2004. Integrated Problem and Decision Reporting as a Translator
Training Tool.
e Journal of Specialised Translation
Gpferich, Susanne, and Riitta Jskelinen. 2009. Process Research into the Development of
Translation Competence: Where Are We, and Where Do We Need to Go?
Across Lan
guages and Cultures
10 (2): 169191. DOI: 10.1556/Acr.10.2009.2.1
Halliday, Michael A. K., and Christian M. I. M. Matthiessen. 2004.
An Introduction to Functional
Grammar
(3rd ed.). London: Edward Arnold.
About the contributors
Fabio Alves
is a professor of translation studies at the Universidade Federal de
Minas Gerais in Brazil.
Michael Carl
is an associate professor of human and machine translation at the
Copenhagen Business School in Denmark.
Ivana e\rkov

Index
acquisition , , , , ,
, , , , , , , ,
, , , , , ,
, 
activated long-term memory
model 
analytical processing 
articulatory suppression
, , 
audience 


Приложенные файлы

  • pdf 16384140
    Размер файла: 5 MB Загрузок: 0

Добавить комментарий