CBS in Language 37

Behind the Mind
Behind the Mind

Methods, models and results in
translation process research
Edited by
Susanne Göpferich
Arnt Lykke Jakobsen
Inger M. Mees
Copenhagen Studies in Language 37
Samfundslitteratur Press
Contents
INTRODUCTION: BEHIND THE MIND OF TRANSLATORS................................ 1
Susanne Göpferich
Towards a model of translation competence and its acquisition:
the longitudinal study TransComp .......................................................... 11
Gerrit Bayer-Hohenwarter
Translational creativity: how to measure the unmeasurable ................... 39
Kristian T.H. Jensen
Indicators of text complexity .................................................................. 61
Nataša Pavlović
More ways to explore the translating mind: collaborative
translation protocols ................................................................................ 81
Dorrit Faber and Mette Hjort-Pedersen
Manifestations of inference processes in legal translation ................... 107
Louise Denver
Unique items in translations .................................................................. 125
Brenda Malkiel
From Ántonia to My Ántonia: tracking self-corrections with
Translog ................................................................................................. 149
Ricardo Muñoz Martín
Typos & Co. .......................................................................................... 167
Fabio Alves and Tânia Liparini Campos
Translation technology in time: investigating the impact of
translation memory systems and time pressure on types of internal
and external support ............................................................................. 191
Maxim I. Stamenov
Cognates in language, in the mind and in a prompting dictionary
for translation ........................................................................................ 219
Notes on contributors .............................................................................. 253

INTRODUCTION
“If the human mind was simple enough to understand, we‟d be too simple
to understand it.” True words spoken by the American physicist Emerson
Pugh (1896–1981), which remind us of the complexity of human cognition
(i.e. the process of being aware, knowing, thinking, learning and judging1),
but which also inspire confidence that gaining access to the mind – in our
case the translator‟s mind – is within the realms of possibility. But, boy, is
it difficult! In this issue of Copenhagen Studies in Language (CSL), which
complements CSL 36, we continue our efforts to come closer to what lies
behind the mind of the translator. The idea for the two volumes arose while
Susanne Göpferich (University of Graz) spent time at the Copenhagen
Business School in order to complete her book on translation process
research (Göpferich 2008). CBS scholars were at that point working on a
major project (Eye-to-IT) funded under the EU FP6 programme.2 The
shared interest in translation process behaviour resulted in many fruitful
discussions and a decision to join forces to publish two titles containing
results of a number of studies.
Since Hans Krings‟s pioneering and groundbreaking work Was in
den Köpfen von Übersetzern vorgeht (Krings 1986), there has been a non-
abating interest in the cognitive processes involved in translation, and
scholars are gradually piecing together the clues to the workings of what
they variously refer to as the translator‟s brain, head, mind and “black
box”. In pre-computer days in the 1980s, Krings contributed greatly to our
knowledge by transferring procedures used by cognitive psychologists
(Ericsson and Simon 1980, 1984) to translation, asking his subjects to think
aloud while translating and instructing them to use different colours when
making changes to a text. The units of translation were subsequently
analysed and categorised and, with the help of the think-aloud protocols, it
was for the first time possible to gain insight into the mental processes of
the translator (see Krings 1986, 1987; and Pavlović, this volume, on the use
1
Webster‟s New World Medical Dictionary.
2
See http://cogs.nbu.bg/eye-to-it/ for a description of the project.
2 Susanne Göpferich, Arnt Lykke Jakobsen & Inger M. Mees
of introspective data in translation). The 1990s saw the development of

keystroke logging programs such as Translog (Jakobsen and Schou 1999),
which enabled researchers to trace keyboard activity and pauses. (See
Hansen 1999, 2002; Tirkkonen-Condit and Jääskeläinen 2000; Tirkkonen-
Condit 2002; Alves 2003 for studies using these research tools.) Round
about the beginning of the new millennium, the use of eye-tracking
technology was introduced into translation. The papers in CSL 36
(Göpferich et al.) are all examples of studies employing eye-tracking
devices (Tobii 1750),3 combined with key-logging, in order to observe and
model typical reading and translation behaviour.
As time goes on, we are learning more and more about what
happens during pauses in translation. This is a noble pursuit since there is
good reason to assume that pauses are not merely periods of silence where
nothing takes place. On the contrary, quite the reverse would appear to be
the case: pauses often represent intense cognitive activity, e.g. problem
solving. Progress has been made in this area, and we have moved from
having virtually no knowledge of the interruptions in the translator‟s flow
of text to being able to ascertain whether a pause occurred before or after a
particular unit of translation, how long it lasted and which words were
fixated. Thus from having no inkling as to what went on during pauses
lasting several seconds, the new technologies have provided us with
detailed information about gaze activity at millisecond intervals. The
challenge that remains ahead of us lies in interpreting this wealth of
additional information, since we still have to guess at what the translator
was actually contemplating while gazing at a particular word, but there can
be no doubt that the methodologies that have emerged in the electronic age
have enabled scholars to come up with increasingly nuanced interpretations
of the translator‟s cognitive processes.
Whilst CSL 36 incorporated studies employing eye-tracking
procedures to come to grips with gaze behaviour, this volume – Behind the
Mind – assembles a number of studies using methodologies other than eye-
tracking, i.e. concurrent verbalisation, key-logging, screen recording, etc.
The studies in CSL 37 view the translation process from different
perspectives, but most contributions have in common that, in addition to
3
See http://www.tobii.se.
Introduction 3
wishing to shed more light on translation processes per se, they also have
applications of such research in mind. For instance, many of the authors are
interested in learning more about the acquisition of translation competence
in order to improve the training of translators or in developing better
support applications. Consequently, they focus on the process behaviour of
professional vs. student translators, or advanced students vs. beginning
students. Once we know what characterises experts as opposed to
beginners, it will be possible to develop more reliable models of translation
competence and translation competence acquisition, and devise external
support applications for translators.
Since Susanne Göpferich‟s visit to CBS, she and her colleagues
have received funding for a major research project, TransComp, a
longitudinal study of the development of translation competence.4 In the
first paper of this volume, the design, aims and methods of the project are
described. Göpferich starts with an overview of how translation
competence and translation competence acquisition have been modelled in
the past, and reports on findings about the cognitive processes involved in
expert performance that have been obtained by cognitive psychologists
who have investigated the development of expertise in various domains
(e.g. playing chess and taxi driving). She then develops her own model of
translation competence as a framework of reference to be verified in
TransComp. In this longitudinal study, the development of translation
competence in 12 students of translation is investigated over a period of
three years and compared to that of ten professional translators. Their
translation processes will be analysed using a wide range of methodologies:
think aloud, key-logging, screen recording, webcam recording,
retrospective interviews, and questionnaires.
The second paper in this collection presents some preliminary
results of a PhD study conducted under the auspices of TransComp. Gerrit
Bayer-Hohenwarter‟s project aims at finding a new approach to measuring
creativity in translation and at establishing how this elusive ability develops
in students of translation as compared to professional translators. She bases
her method on the criteria novelty, fluency and flexibility, focussing
primarily on the cognitive procedures attributable to these dimensions. In
4
Funded by the Austrian Science Fund (FWF), project No. P20908-G03 (2008–11).
http://gams.uni-graz.at/fedora/get/container:tc/bdef:Container/get.
the present article she discusses how flexibility can be identified in

translations. Following a brief review of the literature, she suggests that
rather than using form-oriented shifts between ST and TT or a typology
based on scenes and frames, translational creativity can effectively be
analysed by adopting the creative procedures abstraction, modification and
concretisation. These represent cognitive shifts between ST and TT as
opposed to mere reproduction. The suggested procedures are tested on a
sample of 13 translations (nine students and four professionals) of one ST
item. In addition, a detailed analysis is provided of one set of intermediate
translations by one professional translator, which sheds light on the range
of procedures that translators are sometimes able to activate. Bayer-
Hohenwarter‟s study finds modest confirmation for the hypothesised lower
creativity in first-year students as opposed to that of professional
translators.
Like Bayer-Hohenwarter, Kristian T.H. Jensen‟s paper also deals
with measuring the immeasurable. Jensen‟s concern is to find a means of
determining the complexity of texts. The aim of his PhD project is to
investigate how texts of varying levels of complexity affect the gaze
behaviour of professional translators and student translators. Consequently,
it is essential to discover how relative differences in text complexity can be
measured. This paper examines how objective indicators such as readability
indices, word frequency and non-literalness can be employed to measure
the complexity of three different texts. Although aware that the objective
notion of complexity cannot be equated with the subjective notion of
difficulty, Jensen argues that these objective measures can contribute
towards predicting the degree of difficulty of some types of text.
Readability indices have normally been applied with respect to text
comprehension, but it is suggested that they can also be used to estimate
the production effort during a translation process. The assumption behind
the second criterion, word frequency, is that the less common a word is, the
more effort is needed to translate it. The final indicator, non-literalness, is
taken to be the presence of idioms, metaphors and metonyms in the texts.
Non-literal expressions are expected to involve a greater processing effort
than literal expressions. The results of Jensen‟s objective indicators of
complexity, together with the difficulties that emerge when applying them,
are discussed in this contribution.
Introduction 5
Another paper discussing methodological concerns in our attempts

at getting closer to the process of translation is that authored by Nataša
Pavlović. It deals with the use of a special type of verbal reporting, namely
collaborative translation protocols (CTP). This is a procedure based on
concurrent verbalisation of a pair or group of people translating the same
source text together, basing their decisions on mutual consensus. Like
Göpferich, whose TransComp project aims at exploring the development of
translation competence, Pavlović too wishes to pinpoint the difficulties
encountered by translators and the ways in which they solve problems in
order to deepen our insight into the acquisition of translation competence
and thereby improve translator education. As part of her PhD project,
translation processes of translators working into and out of their first
language (Croatian) were examined with respect to problems, solutions,
resources and decision making. CTP is compared with (a) think aloud
involving single subjects; (b) alternative tools such as integrated problem
and decision reporting; and (c) choice network analysis. It is concluded that
CTP enables us to understand both the social and cognitive aspects
involved in translation.
Another study using verbal reports (both individual and in pairs) is
that described in the contribution by Dorrit Faber and Mette Hjort-
Pedersen. The aim of their study, which in addition to using concurrent
introspection also employs Translog to log the translation process, is to
examine the cognitive processes in legal translation. Specifically, they are
interested in discovering whether the inference processes of students will
be manifested to a higher degree in their target texts than those of
professional translators. Legal texts are notoriously difficult, and in order to
process the content, it is often necessary during the comprehension process
to make explicit information that is only implicit in the source text (e.g. as
who performs an act and what, where and when the act is performed).
Conversely, legal translators also sometimes choose to omit elements in
their translation, thereby making implicit text that is explicit in the source.
For their analysis, two concepts from relevance theory are invoked:
reference assignment and enrichment (i.e. identifying the referent and
filling in missing information). In their paper they describe some tentative
results on the correlation between mental explicitation processes and
resulting instances of linguistic explicitation or implicitation. In addition,
they detail the problems involved in the set-up of experiments designed to

explore these correlations.
Like the paper by Faber and Hjort-Pedersen, the contribution by
Louise Denver also addresses the issue of inferencing using the same
process tools (a combination of think aloud and Translog). Denver‟s study
examines the translation of logical-semantic relations across sentence
boundaries. For this purpose she draws on both the “explicitation
hypothesis” and the “unique items hypothesis”. One specific aim of the
study is to examine the Danish connector ellers („else‟) when used uniquely
in translations from Spanish (L2) into Danish (L1). The data included both
product and process data from two groups of students (BA and MA) and a
group of professional translators. It was hypothesised that the use of
explicitations by means of the unique ellers would be markedly lower than
when the same connector was used at the propositional level with
alternative, disjunctive or conditional meaning.
Another experiment using Translog – this time, for the sake of
ecological validity, without the use of think aloud – is described in the
paper by Brenda Malkiel, who investigates the revision process of 16
beginning translation students. They translated two Hebrew texts into
English, with half working into their L1 and half into L2. The self-
corrections recorded in the logs were categorised in terms of, for instance,
self-corrections to grammar, self-corrections of meaning, and instances in
which a word or phrase was deleted and retyped verbatim. Since there was
no significant effect for source text or mother tongue, the 1257 self-
corrections in the 32 logs were treated as a single group. This contribution
is particularly interesting for its useful categorisation scheme, which was
constructed on the basis of the data rather than a pre-defined list. The
results indicate that actual corrections to the target text account for only
about 13 % of total self-corrections. The remaining 87 % relate to the
students‟ efforts to refine the target text. Malkiel also examines self-
corrections to false cognates, lexicalisable strings and culture-bound
expressions, and the extent to which self-corrections were predictable from
the language pair in question.
While Malkiel‟s study did not include changes to the spelling of a
word in the calculation of total self-corrections, since typos were taken to
reflect merely the translator‟s competence at the keyboard, the potential
Introduction 7
significance of typos is the very issue explored by Ricardo Muñoz. He puts

forward the interesting hypothesis that typographical errors may not be
caused only by imperfect typing skills, keyboard size, and the like, but also
by shifts or lapses in attention. It is thought that attention drops may
sometimes represent situations where cognitive resources have been
reallocated to support other mental activities such as problem solving. Four
advanced translation students were instructed to revise three drafts which
had been translated from English into Spanish by a different person. In
addition, they were asked to translate four English texts into Spanish
themselves and revise them afterwards. The data were obtained using
Translog and were classified according to an ingenious categorisation
scheme in order to discover whether such a quantitative approach could be
used to characterise the four subjects and the three tasks (i.e. the third-party
draft revisions, the translations, and the self-revisions of these translations).
Another process study using Translog is that by Fabio Alves and
Tania Liparini Campos. The purpose of the battery of experiments
conducted by these two authors was to investigate the way in which a
translation memory system (TMS) and time pressure affected the types of
support used by professional translators. The translations of 12 professional
translators were analysed in terms of internal and external support used for
orientation, drafting, and revision of a number of texts which were
translated into Brazilian Portuguese either from English or from German.
The text samples, language direction, subjects‟ experience as professional
translators, and their familiarisation with the TMS were controlled
variables. The findings illustrate the importance of internal support in all
tasks. Time pressure appears to reduce the number of revision pauses both
in drafting and revision phases and increases the need to rely on solutions
offered by the TMS; it does not have an impact on the types of support
used. All translators consulted the web, dictionaries and the spell checker,
but the experiments showed that they predominantly relied on their own
knowledge to solve translation problems.
The final contribution in this volume also deals with translation
support, specifically the prompting of translators. Maxim Stamenov
discusses the use of cognates (including false and partial cognates) and
suggests ways in which a prompting dictionary for such items can be
implemented in a human-computer interaction system. One concern results
from the spatial limitations imposed by a screen, and therefore the

challenge consists not in providing as much information as possible but in
selecting that most needed. Stamenov‟s contention is that much of the
information related to a word and its potential translations is known to the
users, who therefore merely need to be reminded of possible translation
matches. Eye-tracking technology is one means of detecting when and
where help is needed. It is argued that the conventional ways of structuring
entries in dictionaries of cognates need to be adapted if they are to work
optimally in a prompting dictionary. By means of a word with a rich
asymmetric polysemic structure it is shown how this can be effectuated.
The contributions in CSL 37 cover a wide range of aspects of the
translation process, all with the purpose of finding out what lies behind the
mind. The volume includes studies on the following: modelling translation
competence acquisition; measuring creativity and text complexity; using
collaborative translation protocols; examining the inferencing of logical-
semantic relations across sentence boundaries and also in legal translation;
tracking self-corrections, including typos; investigating the impact of
translation memory systems and time pressure on types of support used;
and developing a prompting dictionary for the translation of cognates. The
subjects in the studies comprised both professional translators and
advanced and beginning students. Different language pairs and
directionalities were examined, the following languages being represented:
German, English, Croatian, Hebrew, Bulgarian, Spanish and Brazilian
Portuguese. All studies have as their main objective to contribute towards
our knowledge of the development of translation competence, improving
translator training, and developing translation aids. The authors have
clearly demonstrated that the human mind is by no means simple, but –
even though there is yet some way to go – we hope that Behind the Mind
has also shown that we are not too simple to understand it.
We wish to dedicate this volume to our recently retired colleague and
friend Kirsten Haastrup in recognition of her achievements as a researcher
and her ongoing interest in process research.
Copenhagen and Graz, April 2009

The Editors
Introduction 9
References
Alves, F. (ed.) 2003. Triangulating translation. Perspectives in Process Oriented

Research. Amsterdam and Philadelphia: John Benjamins.
Ericsson, K. A. & Simon, H. A. 1980. Verbal reports as data. Psychological
Review 87: 215-251.
Ericsson, K. A. & Simon, H. 1984 (rev. edn 1993). Protocol Analysis.
Cambridge, Mass.: MIT Press.
Göpferich, S. 2008. Translationsprozessforschung: Stand – Methoden –
Perspektiven. (Translationswissenschaft 4): Tübingen: Narr.
Göpferich, S., Jakobsen, A. L. & Mees, I. M. (eds). Looking at Eyes: Eye-
tracking Studies of Reading and Translation Processing (Copenhagen
Studies in Language 36). Copenhagen: Samfundslitteratur.
Hansen, G. (ed.) 1999. Probing the Process in Translation: Methods and Results
(Copenhagen Studies in Language 24). Copenhagen: Samfundslitteratur.
Hansen, G. (ed.) 2002. Empirical Translation Studies: Process and Product
Jakobsen, A. L. and Schou, L. Translog documentation. In: G. Hansen, ed.
Probing the Process in Translation. Methods and results. (Copenhagen
Studies in Language 24) Copenhagen: Samfundslitteratur. 151-186.
Krings, H. P. 1986. Was in den Köpfen von Übersetzern vorgeht. Tübingen: Narr.
Krings, H. P. 1987. The use of introspective data in translation. In C. Færch and
G. Kasper (eds). Introspection in Second Language Learning. Clevedon and
Philadelphia: Multilingual Matters. 159-176.
Tirkkonen-Condit, S. 2002. Process research: state of the art and where to go
next? Across Languages and Cultures 3 (1): 5-19.
Tirkkonen-Condit, S. and Jääskeläinen, R. (eds). 2000. Tapping and Mapping the
Process of Translation: Outlooks on Empirical Research. Amsterdam and
Philadelphia: John Benjamins.
Towards a model of translation competence and its
1
acquisition: the longitudinal study TransComp
Susanne Göpferich
Abstract
The first part of this article gives a short survey of how translation
competence and its acquisition have been modelled so far and of what we
know from expertise research about the cognitive processes involved in
expert performance. Drawing on this, a model of translation competence is
presented as a framework of reference for the research project
TransComp, a longitudinal study which explores the development of
translation competence in 12 students of translation over a period of three
years and compares it to that of 10 professional translators. The model
will be used to generate hypotheses to be verified in TransComp. In the
second part of the article, the design of TransComp, the research
questions asked, and the methods of measuring those features which are
assumed to be indicators of central sub-competences of translation
competence will be presented. The article concludes with information on
the availability of the materials used for the study and the data collected in
TransComp.
1. Translation competence acquisition – providing empirical evidence

for our assumptions
The development of models of translation competence and, even more so,

the development of models of translation competence acquisition is still in
its infancy. One of the reasons for this is that investigating how translation
competence develops is only possible by means of longitudinal studies,
1 TransComp is funded by the Austrian Science Fund (FWF) as project No. P20908-
G03 (September 2008–August 2011).
12 Susanne Göpferich
and longitudinal studies are extremely cumbersome and time-consuming.

This also explains why longitudinal studies in the strictest sense of the
term, involving the analysis of translation products and processes of the
same persons at regular intervals during their training and later
professional career, do not yet exist. When modelling translation
competence and translation competence acquisition, we can so far draw on:
a) the results of studies comparing the translation processes of
translation students with those of professional translators (for an
overview, see Göpferich 2008: 168 ff.),
b) theoretical reflections on the components which make up translation
competence, and
c) results of investigations into the development of expertise in various
domains, such as playing chess, conducted by cognitive
psychologists (cf. Ericsson/Smith 1991).
Below a short survey will be given of how translation competence and its
2
acquisition have been modelled so far and of what we know from
expertise research about the cognitive processes involved in expert
performance. Drawing on this, I will develop my own model of translation
competence as a framework of reference for a longitudinal study of the
development of this competence: TransComp. The model will also be used
to generate hypotheses to be verified in TransComp. The design of this
longitudinal study, the research questions asked, and the methods of
measuring those features which are assumed to be indicators of central
sub-competences of translation competence will be presented in the second
part of this article. It concludes with information on the availability of the
materials used and the data collected in TransComp.
2. Central components of a translation competence model
There is consent among translation scholars that translation competence is

composed of several sub-competences. What sub-competences have to be
taken into account, and how they can be defined, is still a matter of debate
(see, for example, PACTE 2000; 2002; 2003; 2005; Shreve 1997; Wilss
2 For a more detailed description of the existing translation competence and translation
competence acquisition models, see Göpferich (2008: Ch. 6).
Towards a model of translation competence (acquisition) 13
1992). There is no doubt, however, t-hat at least the following three play a
decisive role: communicative competence in the source language and the
target language, domain competence, and tools and research competence
(see Section 2.3). Furthermore, there is general agreement that translation
competence involves more than the sum total of these three – and perhaps
other – sub-competences.
For both Hönig (1991; 1995) in his model of an ideal translation
process (Fig. 1) and Pym (2003) in his ―minimalist approach‖ to defining
translation competence, translation competence is composed of two main
sub-competences: (1) associative competence and (2) the competence to
develop a ―macro-strategy‖ (Hönig 1991; 1995) and to employ it
consistently. Pym describes these two competences as follows:
[T]he training of translators involves the creation of the following two-fold

functional competence (cf. Pym 1991):
The ability to generate a series of more than one viable target text (TT1, TT2 …
TTn) for a pertinent source text (ST) [This corresponds to what Hönig calls
associative competence.];
The ability to select only one viable TT from this series quickly and with justified
confidence. [This corresponds to Hönig‘s macro-strategy and the ability to employ
it consistently.]
We propose that, together, these two skills form a specifically translational
competence; their union concerns translation and nothing but translation. There
can be no doubt that translators need to know a fair amount of grammar, rhetoric,
terminology, computer skills, Internet savvy, world knowledge, teamwork
cooperation, strategies for getting paid correctly, and the rest, but the specifically
translational part of their practice is strictly neither linguistic nor solely
commercial. It is a process of generation and selection, a problem-solving process
that often occurs with apparent automatism. (Pym 2003: 489)
What Hönig calls ―macro-strategy‖ also appears at the heart of the PACTE
group‘s translation competence model, to which I will return in Section
2.2, and in Risku‘s (1998) ―cognition models of translation competence‖.
To understand what Hönig means by ―macro-strategy‖, we have to take a
closer look at his model.
2.1 Hönig’s model of an ideal translation process
According to Hönig‘s model (Fig. 1), translators first read the ST (upper
right corner of the model). Their source-text reception, however, differs
from that of ordinary readers in a non-translation-specific situation, since
their text reception is influenced by the translation task they have in mind.
ST
projected
uncontrolled
workspace
Schemes
Frames
SL-signs
MACROSTRATEGY
R
E
A
L
C
O
M
M
U
N
I
C
A
T
I
O
N
Figure 1. Hönig‘s model of an ideal translation process (Hönig 1991: 79;

terminology adapted to Hönig 1995: 51)
The source text projected into the translator‘s mental reality becomes the
object of mental processing or, to be more precise, further mental
processing, because the first reception also involves mental processing.
This occurs in two different workspaces: the uncontrolled workspace and
the controlled workspace.
Processing in the uncontrolled workspace involves the activation of
frames and schemes, which are structured domains of long-term memory,
in associative processes (Hönig 1991: 79 f.; 1995: 55). These associative
processes give rise to expectations with regard to the prospective target
text. Expectations with regard to structure, style, and content of a text form
part of any comprehension process; in translation, however, they are
target-text-oriented (Hönig 1995: 55).
Using the projected source text, the prospective target text, and data
from their uncontrolled workspaces, competent translators develop a
translation macro-strategy. What goes into this macro-strategy are not only
the characteristics that are decisive for the target text, such as its function,
its audience, and the medium in which it will appear, but also the options
that translators have for searching information and verifying their
subjective associations, as well as for improving their subject domain
knowledge (Hönig 1995: 56 f.). Developing such a macro-strategy may
happen more or less automatically on the basis of the translator‘s
professional experience, or ―very deliberately, possibly with the aid of
translation-relevant textual analysis‖ (Hönig 1991: 80). Ideally, the
development of a macro-strategy precedes the actual translation phase, in
which both the uncontrolled workspace and the controlled workspace are
involved. In the controlled workspace rules and strategies are employed,
for which Hönig (1995: 50; my translation) provides the following
examples:
– Do not translate proper names.
– The English continuous form translates into German by adding gerade.
– Government means Regierung.
– Avoid repeating the same words in German texts.
These rules may lead to appropriate results in some cases, but not in all. To
be able to decide whether a rule is applicable in a specific situation,
translators again need a macro-strategy which controls the use of the
micro-strategies to be employed. Without a macro-strategy, translators run
the risk of getting lost in the maze of micro-strategies in their controlled

workspaces.
During the translation phase, the processes in the uncontrolled
workspace are complemented by an associative competence (or ―transfer
competence‖; Hönig 1991: 80), which comprises what some scholars, such
as Harris/Sherwood (1978), have termed ―innate translation ability‖.
Potential translation equivalents may become part of the target text
(―Product TT‖ in Fig. 1) in four different ways (Hönig 1991: 80; see also
1995: 56):
(1) As a linguistic reflex stimulated by the first contact between the projected st
[sc. source text] and semantic associations in the uncontrolled workspace.
(2) As an automatic transfer from the uncontrolled workspace after a
macrostrategy has been worked out.
(3) As a product of a microstrategy applied in the controlled workspace which
has been approved by monitoring.
(4) As a product of interdependent processes taking place in the controlled and
uncontrolled workspaces, whereby the final approval can be either by
uncontrolled (―automatic‖) or controlled (―cognitive‖) processes.
Decisions based on a macro-strategy thus, step by step, lead to a target
text, each portion of which is evaluated by deciding whether it fulfils the
requirements the translator has in mind. The target text leaves the
translator‘s mental reality, can be handed over to the commissioner and
become part of a real communication process (Hönig 1995: 56).
Hönig (1995: 56) does not claim that his model covers all relations
and interdependencies, but emphasizes that most didactic approaches do
not even take into account this minimal complexity. For Hönig the most
important insight into translation processes provided by his model for
translation pedagogy is the following:
The main reason why many students and teachers of translation are frustrated is
that they experience the complexity of their mental processes while translating,
but try to relieve themselves of this complexity because they do not really
understand the processes.
What is symptomatic of such relief efforts is clinging to rules which are supposed
to prove the absolute correctness of the way a certain word or phrase has been
rendered.
Typical signs of such relief efforts are striving for symmetrical matches between
source-text and target-text units and using the term equivalence. (Hönig 1995: 57;
my translation)
The relationship between the two central components of translation

competence, associative competence (―Assoziationskompetenz‖; Hönig
1995: 62) or ―transfer competence‖ (Hönig 1991: 84) on the one hand and
the ability to develop a macro-strategy and to employ it consistently on the
other, can be described as follows. Transfer or associative competence is
indispensable for translation; its use must be encouraged so that it can
become part of the translator‘s self-confidence. According to Hönig (1995:
62), it is the driving force in translation processes, the ‗engine‘ in the
uncontrolled workspace. Translations which fulfil their functions,
however, are achieved only when the results produced by this associative
competence are subjected to a macro-strategy. It is only the consistent
employment of such a macro-strategy that leads to real translation
competence. Hönig (1995: 62; my translation) points out:
It is only translation competence which gives translators the self-confidence
needed to make use of their associative competence and to avoid subjecting their
products to monitoring processes in the controlled workspace again and again.
In other words: Translation competence enables translators to translate as well as
they can on the basis of their associative competence. Without translation
competence, they will definitely translate worse than they could.
This also means, however, that translators who do not follow a macro-strategy but
possess an extensive associative competence may translate better than translators
who employ a macro-strategy (but lack associative competence).
2.2 The PACTE group’s revised translation competence model
The PACTE group‘s first translation competence model dates back to

1998, but was revised soon afterwards (see PACTE 2007: 330). In the
following, only the improved revised version shown in Fig. 2 will be
presented. According to this model, translation competence, which the
PACTE group defines as ―the underlying system of knowledge, abilities
and attitudes required to be able to translate‖, is composed of five sub-
competences and psycho-physiological components.
These form ―a system of competencies that interact, are hierarchical,
and subject to variation‖ (PACTE 2002: 43; see also PACTE 2000: 100).
The centre of the model is formed by the strategic competence, which the
PACTE group (PACTE 2005: 610) defines as follows:
The strategic sub-competence is the most important, solving problems and
guaranteeing the efficiency of the process. It intervenes by planning the process in
relation to the translation project, evaluating the process and partial results
obtained, activating the different sub-competencies and compensating for
deficiencies, identifying translation problems and applying procedures to solve
them.
Figure 2. The PACTE group‘s revised translation competence model

(PACTE 2002; 2005: 610; 2007: 331)
Comparing this concept to those in Hönig‘s model (Fig. 1) leads to the
conclusion that strategic competence is nothing but the competence to
develop an adequate macro-strategy and to employ it consistently. Like the
strategic competence, the macro-strategy decides what specific sub-
strategies need to be employed in a certain situation.
The fact that the macro-strategies and their sub-strategies are
situation- and problem-dependent seems to be enough of a justification for
the assumption that their development and selection do not involve purely
―declarative knowledge‖, but mainly ―operative knowledge‖ in the sense of
Anderson (1983). Translation competence involves both declarative and
3
operative (procedural) knowledge, the latter being the more relevant of the
two for the distinction between translation experts and translation novices
(PACTE 2003: 42).
3 What has to be taken into account in think-aloud studies is that according to

Anderson (1983) declarative knowledge can easily be verbalized, clearly be defined,
and is processed consciously, whereas operative knowledge is difficult to put into
words, can only be possessed partly, is acquired only gradually by means of practical
By bilingual sub-competence the PACTE group means ―pragmatic,

socio-linguistic, textual and lexical-grammatical knowledge in each
language‖ (PACTE 2005: 610) including ―interference control‖, i.e., the
ability to avoid interferences between the languages. They consider the
relevant knowledge to be mainly procedural (PACTE 2003: 91 f.).
By extra-linguistic sub-competence the PACTE group means
[p]redominantly declarative knowledge, both implicit and explicit, about the
world in general and special areas. It includes: (1) bicultural knowledge (about the
source and target cultures); (2) encyclopedic knowledge (about the world in
general); (3) subject knowledge (in special areas). (PACTE 2003: 92)
Instrumental sub-competence is made up predominantly of
procedural knowledge related to the use of documentation sources and
information and communication technologies applied to translation:
dictionaries of all kinds, encyclopaedias, grammars, style books, parallel
texts, electronic corpora, etc. (PACTE 2003: 93; 2005: 619).
Knowledge about translation sub-competence comprises
[p]redominantly declarative knowledge, both implicit and explicit, about what
translation is and aspects of the profession. It includes: (1) knowledge about how
translation functions: types of translation units, processes required, methods and
procedures used (strategies and techniques), and types of problems; (2) knowledge
related to professional translation practice: knowledge of the work market
(different types of briefs, clients and audiences, etc.). (PACTE 2003: 92)
To my mind, this category comprises rather inhomogeneous components
and therefore should be split up into two sub-competencies. As will
become evident in my model (Fig. 3), I propose subdividing it into
translation routine activation competence, which roughly includes the
aspects listed under (1) in the quotation above, and the translator‘s self-
concept, which differs in status from the other sub-competences. I will
come back to its different status in Section 2.3.
Psycho-physiological components comprise cognitive and attitudinal
components as well as psycho-motor mechanisms.
They include: (1) cognitive components such as memory, perception, and attention
and emotion; (2) attitudinal aspects such as intellectual curiosity, perseverance,
rigor, critical spirit, knowledge of and confidence in one‘s own abilities, the
ability to measure one‘s own abilities, motivation, etc.; (3) abilities such as
creativity, logical reasoning, analysis and synthesis, etc. (PACTE 2003: 93)
exercises, is processed automatically, and thus remains in the subconscious mind and
is not available for verbalization (cf. PACTE 2000: 102).
They have a special status in the PACTE model because they are not
translation-specific but form ―an integral part of all expert knowledge‖
(PACTE 2003: 91).
2.3 My translation competence model

4
On the basis of the models mentioned above, I have developed my own
translation competence model, which forms the framework of reference for
the longitudinal study TransComp (Göpferich 2007).
external sources of
information
and tools avaiblable
working conditions
communicative (e.g. time pressure)
competence domain
in at least competence
2 languages
strategic
tools and competence
psycho-
research motor
competence motivation
competence
translation routine
activation
competence
translator’s self concept/

translation professional ethos
norms
psycho-physical
topics covered and disposition
translation methods employed
assignment in theoretical & practical
translation training
Figure 3. Göpferich‘s translation competence model
As illustrated in Fig. 3, I differentiate between the following sub-

competences.
1. Communicative competence in at least two languages
This sub-competence corresponds to the PACTE group‘s ―bilingual
sub-competence‖. It comprises lexical, grammatical and pragmatic
knowledge in both languages. Pragmatic knowledge also includes
knowledge about genre and situation-specific conventions in the
4 For a ―psycholinguistic model of the translation process‖, see Kiraly (1995); for a
critical review of Kiraly‘s model, see Göpferich (2008: 137 ff.).
respective cultures. Communicative competence in the source language

is relevant primarily for source-text reception, whereas target-language
competence determines the quality of the target text produced. Target-
language receptive competence must not be neglected, however,
because it is needed for monitoring processes in which source-language
units and target-language units are compared for semantic equivalence,
for example.
2. Domain competence
This corresponds approximately to the PACTE group‘s ―extra-linguistic
sub-competence‖ and comprises the general and domain-specific
knowledge that, in addition to the knowledge mentioned above, is
necessary to understand the source text and formulate the target text, or
at least the sensitivity to recognize what additional knowledge is needed
from external sources of information to fill one‘s knowledge gaps.
3. Tools and research competence
This corresponds to the PACTE group‘s ―instrumental sub-competence‖
and comprises the ability to use translation-specific conventional and
electronic tools, from reference works such as dictionaries and
encyclopaedias (either printed or electronic), term banks and other
databases, parallel texts, the use of search engines and corpora to the
use of word processors, terminology and translation management
systems as well as machine translation systems.
4. Translation routine activation competence
This competence comprises the knowledge and the abilities to recall
and apply certain – mostly language-pair-specific – (standard) transfer
operations (or shifts) which frequently lead to acceptable target-
language equivalents. In Hönig‘s terminology, this competence could
be described as the ability to activate productive micro-strategies (see
also Section 5.2, especially footnote 10).
5. Psychomotor competence
These are the psychomotor abilities required for reading and writing
(with electronic tools). The more developed these competences are, the
less cognitive capacity is required, leaving more capacity for other
cognitive tasks. Psychomotor skills needed for typing may have an
impact on the cognitive capacity that will be available for solving
translation problems in a narrower sense, because from this the memory

capacity needed for performing psychomotor tasks has to be subtracted.
The poorer the psychomotor skills are, the larger the cognitive capacity
required by psychomotor activities is assumed to be.
6. Strategic competence
This corresponds to the PACTE group‘s ―strategic competence‖ and
controls the employment of the sub-competences mentioned above. As
a meta-cognitive competence it sets priorities and defines hierarchies
between the individual sub-competences, leads to the development of a
macro-strategy in the sense of Hönig (1995), and ideally subjects all
decisions to this macro-strategy. How strictly translators adhere to
employing this macro-strategy depends on their strategic competence
and their situation-specific motivation, which may be both intrinsic
(enjoying translating) or extrinsic (payment, fear of compensatory
damages, etc.).
The employment of the sub-competences mentioned above and their
central control are determined by three factors, which form the basis of my
model: (1) the translation brief and translation norms; (2) the translator‘s
self-concept/professional ethos, on which the contents conveyed and the
methods employed in theoretical and practical translation training courses
have an impact and which form the component of my model where aspects
of social responsibility and roles come in (cf. Risku 1998: 90; 2004: 76),
and (3) the translator‘s psycho-physical disposition (intelligence, ambition,
perseverance, self-confidence, etc.). Translators‘ psycho-physical
disposition may have an influence on how quickly their translation
competence develops: a critical spirit and perseverance in solving
translation problems may accelerate the development of translation
competence.
3. The PACTE group’s translation competence acquisition model
Since, as we have seen, the translation competence models developed so

far are still rather vague, it is obviously even more complicated to develop
a translation competence acquisition model. The only existing model of
translation competence acquisition is that of the PACTE group (Fig. 4).
Figure 4. The PACTE group‘s translation competence acquisition model

(PACTE 2000: 104)
The model is based on the PACTE group‘s revised translation competence

model, which I, however, propose to replace by my own model (Fig. 3).
According to the PACTE model, the acquisition of translation competence
involves the development of the individual sub-competences and, in
addition to this, the development of the integrative competence to fall back
on the individual competences and to prioritize them depending on the
respective assignment and communicative situation (―integration of the
sub-competencies‖). The development of these competences and their
integration do not only involve the accumulation of declarative knowledge,
but, above all, the restructuring of existing knowledge (PACTE 2000). The
PACTE group describes this as follows:
Thus, the novice stage in the development of translation competence could be
defined as the stage when the sub-competencies have been acquired, at least
partially, but they do not interact with each other. Therefore, the development
from novice to expert is not only a question of acquiring the missing sub-
competencies, but also of re-structuring the existing sub-competencies to put them
at the service of the transfer competence [i.e., the strategic competence in my
terminology]. (PACTE 2000)
According to the PACTE group, this integration and restructuring is only
made possible by a learning competence with specific learning strategies
(Fig. 4).
4. Findings in expertise research
The PACTE group‘s description of the development of translation

competence is supported by findings from expertise research. Some of the
results from this field of cognitive psychology that are relevant for process
research into the development of translation competence are the following:
1. Experts do not only possess a large amount of knowledge in their
specialized domain; this knowledge has also been restructured and
interconnected to a higher degree in the process of its acquisition; they
possess superior analytical and creative as well as practical skills; their
mental processes have been automatised to a higher degree (Sternberg
1997).
2. The high degree of interconnection of knowledge in their long-
term memories allows experts to retrieve it more quickly and with more
precision and to overcome limitations of their working memories
(Ericsson/Charness 1997: 15 f.; Neubauer/Stern 2007: 165 f.). They are
able to plan taking many factors into account (Ericsson/Smith 1991: 25 f.).
3. Experts have transformed declarative knowledge in their domain
of specialization into procedural knowledge (―proceduralisation‖); they
learn tactically (i.e., they store and automatise sequences of actions and
strategies they need for problem solving in their domain) as well as
strategically (i.e., they know how problem-solving processes in their
domain can be tackled most efficiently). Complex mental problem
representations help them in doing so (Anderson 31990: 267 ff.).
These specific features of expert performance are also reflected in
Risku‘s ―cognition model‖ of expert translation competence. Comparing
this model to her ―cognition model‖ of novice translation competence
reveals what must happen on the way from novice to expert (Risku 1998:
241 ff.).
5. TransComp – a longitudinal study into the development of

translation competence
As mentioned in Section 1, the above-mentioned assumptions and findings

have not yet been verified in longitudinal studies of the development of
translation competence if longitudinal study is understood in the strictest

sense of the term. There are findings from numerous contrastive studies,
though, comparing the translation processes of students of a certain
translation competence level with those of more advanced translators or
even translation professionals, which support the above-mentioned
assumptions (for a survey of these findings, see Göpferich 2008: 168 ff.).
To gain insight into the development of translation competence in its
continuity, to provide stronger empirical evidence of the assumptions
above, and to refine the models presented so far, the longitudinal study
TransComp was launched at the University of Graz in October 2007
(Göpferich 2007).
5.1 The design of TransComp
In TransComp twelve undergraduate students of translation at the

beginning of their first semester and ten professional translators have been
selected as subjects. The number of professional translators is smaller
because the project focuses on the students‘ development of translation
competence and the data from professional translators are needed for
purposes of comparison only.
The most crucial selection criterion for the student subjects was that
they had obtained ―very good‖ or ―good‖ grades for their A-level exams in
German (their mother tongue) and English (their first foreign language).
English also had to be the first foreign language they chose for their
Translation Studies program. Furthermore, the potential student subjects
5
had to take a test measuring their abilities for semantic differentiation.
The twelve student subjects have been divided into two groups of
six subjects each. If students drop out, they will be replaced by other
students with a similar profile.
Each student will have to translate ten English texts (eight extracts
from popular-science texts and two extracts from operating instructions for
household appliances) into German according to the scheme in Table 1.
5 By means of this test, we wanted to select the twelve best students from those who
had volunteered to take part in TransComp and met the other criteria mentioned
above. Since there were very few volunteers who fulfilled these other criteria, all were
accepted.
6
Table 1. Translation scheme
Group A (6 students) Group B (6 students)
st
Beginning of 1 semester Text A1, Text A2, Text A3 Text B1, Text B2, Text B3
Beginning of 2nd semester Text A4, Text A5 Text B4, Text B5
Text B1 (1 semester‘s lag) Text A1 (1 semester‘s lag)
Beginning of 3rd semester Text B2 (2 semesters‘ lag) Text A2 (2 semesters‘ lag)
Beginning of 4th semester Text B3 (3 semesters‘ lag) Text A3 (3 semesters‘ lag)
Beginning of 5th semester Text B4 (3 semesters‘ lag) Text A4 (3 semesters‘ lag)
Beginning of 6th semester
7
Text B5 (4 semesters‘ lag) Text A5 (4 semesters‘ lag)
End of 6th semester Text A1 (6 semesters‘ lag) Text B1 (6 semesters‘ lag)
The scheme takes into account that competence improvements may not
occur to a sufficient extent to be detected after only one, two or three
semesters, but may only become detectable after two or three years. It
allows us to check for progress over longer periods. It also takes into
account that progress may proceed in steps, with varying improvement
speeds over the whole period.
All texts will be translated only once by each student except for Text
A1 and Text B1, which will be re-translated after three years when the
learning effect can be assumed to have become highly attenuated, i.e.,
when the students can be assumed to have forgotten how they had
translated it three years before. Five of the professional translators will
translate Texts A1 to A5, the other five, Texts B1 to B5.
The source texts selected offer a repertoire of different translation
problems (lexical, syntactic, pragmatic, text-linguistic, culture-specific,
creativity-demanding and comprehensibility-related problems). Their
comprehension, however, does not require any specialized knowledge.
They were primarily chosen because they are relatively easy to understand,
but difficult to transfer into the target language. These texts have to be
translated in Translog 2006, which registers all keystrokes, mouse clicks,
and the time intervals between them. To guarantee ecological validity, the
6 ‗Lag‘ indications show the time elapsed from the moment the relevant text was
translated first to the moment it is re-translated for the purpose of comparison.
7 Unfortunately, we will not have data for a time lag of five semesters because this
would have involved handing out two more texts for translation to the subjects at the
subjects are allowed to use the Internet as well as any other electronic and
conventional resources they wish. Use of electronic resources is registered
by the screen-recording software Camtasia Studio (or ClearView
8
respectively ); use of conventional resources is documented by observers.
Originally we had planned that 50 % of the student subjects and
50 % of the professional translators would have to think aloud during the
experiments (level 1 and 2 verbalizations according to Ericsson/Simon
1999: 79); the remaining 50 % in both groups would be asked to comment
on their translation processes in immediate retrospective interviews for
which we wanted to use the screen records of their translation processes as
prompts. This would have allowed us to compare the data elicited by
means of the two methods with regard to their comprehensiveness and to
establish the degree to which the method interfered with the actual
translation process. If one of the two methods turned out to be superior for
answering our research questions, the plan was to proceed with this
method only. However, after the first experiment, we asked our subjects
how they felt about the think-aloud method, for which they had been
trained before the experiments in a trial run, or whether they would prefer
the retrospective method instead. All subjects answered unanimously that
they strongly preferred the think-aloud method. One of the reasons for this
can be assumed to be the fact that cued retrospection is very time-
consuming and the subjects did not want to spend even more time on the
experiments. The student subjects felt exhausted after each experiment and
wanted to rest, whereas the professional translators had their busy
schedules in mind. Therefore, we decided to use think-aloud in all
beginning of their first semester, which was not feasible due to time and staff
constraints.
8 For three of our subjects in the first wave, a Tobii eye-tracker (with the screen-
recording software ClearView) was also used. Due to the long duration of our
experiments (approx. 1 hour or more), in which we could not prevent the subjects
from moving out of the area in which eye-tracking registration is possible, and other
reasons, poor eye-tracking results were obtained, which do not allow any detailed
analysis. Since the added value we obtained from the eye-tracking results was
minimal compared with the effort involved in using this additional method, it was
decided to abandon it. Eye-tracking seems to be more useful when focusing on
specific aspects of the translation process, such as fixations in certain areas of the
screen, than when analysing translation processes as a whole. Cf. Göpferich (in press)
for methodological aspects of using eye-tracking.
experiments in order not to risk obtaining incomplete accounts in the

retrospective interviews due to the subjects‘ lack of motivation.
The think-aloud will be transcribed using XML mark-up according
to the Guidelines for Electronic Text Encoding and Interchange (version
P5) of the Text Encoding Initiative (TEI 2008). These guidelines have been
modified slightly to meet the specific requirements of translation process
studies (for a detailed description of the transcription rules used, see
Göpferich in press). In addition to the subjects‘ think-aloud, the transcripts
also include other activities such as search processes.
Immediately after each translation, the subjects have to fill in
questionnaires on how they felt during the translation process, on the
problems they encountered, the strategies they employed to solve them and
9
the extent to which they were satisfied with the results. After this, short
retrospective interviews will be conducted with the subjects. The goals
pursued in these short retrospective interviews are (1) to collect data on
phases in the translation process where the subjects stopped verbalizing
while thinking aloud (due to cognitive overload or other reasons), (2) to
find out whether the subjects are aware of certain problems they may have
encountered during the translation process, and (3) to make sure that the
terminology they use in their think-aloud and their retrospective interviews
is interpreted in the way they intended.
Although retrospective verbalization, in contrast to think-aloud
(concurrent verbalization), has the advantage of not interfering with the
translation process, think-aloud is used as a primary source of information,
at least for the student subjects, because ―[f]or tasks of longer duration, the
validity of think-aloud reports appears to be higher than that of
retrospective reports‖ (Ericsson/Simon 1999: xxii). Since we assume,
however, that in professional translators more processes will be
automatised and that they will take into account more potentially
translation-relevant factors in their translation processes leading to higher
cognitive load and thus to less concurrent verbalization, more emphasis
will be placed on the retrospective interviews when analysing their
translation processes (see also Hansen 2006; and Section 5.2). Since our
subjects translate into their mother tongue, i.e., the same language in which
9 For the availability of the materials used in the experiments, see Section 6.
they are thinking aloud, interferences should be smaller than in studies

where the language in which the subjects think aloud differs from the
target language.
In addition to their translation products and translation processes,
the subjects‘ self-concepts will be analysed by means of questionnaires
(see Göpferich 2008: 264 ff.).
The results will be triangulated, set in relation to the quality of the
subjects‘ translation products, and used to correct, optimize and refine our
provisional translation competence model (Fig. 3) and the translation
competence acquisition model of the PACTE group (2003: 60; Fig. 4).
The quality of the translations, for which precise assignments have
been formulated and handed out to the participants, will be assessed
according to functional principles by three raters experienced in translation
and/or translation didactics in a discursive process.
5.2 The focus of TransComp and the ‘measurement’ of (the development

of) translation competence
TransComp will concentrate on the following components of translation

competence: (1) strategic competence, (2) translation routine activation
competence, and (3) tools and research competence (see the objectives in
PACTE 2005: 611). The reason for this selection is that we assume that
these competences are the main translation-specific competences in which
translation competence differs from the competence of bilingual persons
with no specific training in translation. These competences will form the
dependent variables in our study.
By requiring our students to have good or very good grades for their
A-levels in German and English and to follow the same curriculum during
our longitudinal study, we make sure that their communicative competence
in these languages can be considered a more or less controlled variable
(more or less because the impact of individual activities, such as stays
abroad, and personal factors, such as their intelligence, are beyond our
control). The same applies to the subjects‘ psychomotor competence. By
selecting source texts whose comprehension does not require any domain-
specific knowledge, we make sure that the subjects‘ domain competence
can also be regarded as a more or less controlled variable.
Since we will work closely together with our subjects for three
years, we expect to be able to characterize their psycho-physical
disposition, which may also have an impact on their development. During
the whole study, the subjects‘ development will be analysed against the
background of the controlled theoretical and practical input of their
translation training, which is assumed to shape their translator‘s self-
concepts and professional ethos. Here we start from the assumption that
the individual subjects go through stages of mental development which
also mark major stages in the development of the discipline of Translation
Studies from the equivalence-oriented paradigm to the functionalist
paradigm and beyond.
For analysing the development of the three sub-competences
mentioned above, numerous criteria, even criteria which we cannot think
of yet because they are not covered by the theoretical model used as a
starting-point, may be relevant. In such cases, working with think-aloud
protocols turns out to have the advantage that it offers the possibility of
collecting data in an unstructured way, i.e., in a way that is not biased by
our theoretical model. Krings (2001: 218) describes this as follows:
In this case [i.e., when working with think-aloud data], the researcher‘s model
predetermines data collection far less than in other models with a more rigid
structure […] in which the range of results is considerably restricted by the data
collection tools in direct proportion to the extent of their structuredness. Verbal-
report data, especially Thinking Aloud, is thus more ‗sensitive‘ to the structure of
the object area than methods with a more rigid predetermined structure. The
structures can only be developed in a reflexive process following data collection,
gradually approaching the object structure. Verbal-report data are thus particularly
suited to the investigation of objects whose structure is as yet little known.
We will follow such a reflexive process. As a first step, we will analyse our
corpus using the criteria described below, which can be derived from
findings from expertise research and/or findings from contrastive studies
of translation competence. In the course of our analyses we expect further
criteria to emerge which will then be added.
Starting from the assumption that strategic competence becomes
salient when problems occur and need to be solved, we will first analyse
the transcripts for problems that occurred during the translation process.
For this analysis we will use the primary and secondary problem indicators
suggested by Krings (1986). Having identified problematic items in the
translation process we will then analyse the transcripts for the strategies
employed to solve these problems. Special attention will be paid to the
mental processes involved in the translation of those items which are
thought to represent the repertoire of different potential translation
problems (Section 5.1). Both the translation problems and the strategies
employed will be classified. We will determine how they develop
quantitatively and qualitatively over the three years.
As Krings (2001: 310 f.) discovered, the application of his problem
indicators may be problematic for the professional translators for the
following reason:
While nonprofessional translators typically process many translation problems,
but usually consider them in isolation, the professional translators‘ mental
activities spread like waves from the translation problem across the entire text.
This fact renders it difficult to differentiate problematic elements of the translation
process from nonproblematic ones and thus strategic elements from nonstrategic
ones.
For this reason, we will use several other criteria which will be applicable
to both the students‘ and the professional translators‘ TAPs. The TAPs will
be analysed for passages where the subjects describe, comment on or
employ a macro-strategy in the sense of Hönig (1995). Indicators for such
passages are verbalizations about the function of the target text, the
expectations of the target-text audience, and other requirements that the
target text has to fulfil with respect to the translation assignment. We will
also analyse what aspects and how many different aspects the subjects take
into account in their problem-solving processes. Furthermore, we will
analyse the linearity with which the subjects proceed (for an
operationalization of this analysis, see Krings 1988).
Assumptions underlying these analyses are the following. The
higher a translator‘s translation competence, the more advanced the sub-
competences are and the better their interaction and coordination by the
strategic competence. This interaction and coordination should become
obvious from the number of aspects that are taken into account during
problem-solving strategies, in the repertoire of strategies employed, the
ability to implement a non-linear approach, and the macro-strategies
verbalized as well as the consistency with which they are employed. As
Tilp‘s (2007) exploratory study suggests, professional translators develop a
macro-strategy and use it as a criterion whenever they have to take
decisions, whereas students only gradually learn to develop a macro-

strategy, and even if they have one, they only gradually develop the
competence to employ it consistently.
Our analysis of the subjects‘ tools and research competence will be
based on the following assumptions. Whereas professional translators are
aware of the specific type of problem they encounter while translating
(higher meta-cognitive competence), novices have only a vague idea of the
type of translation problem they are experiencing. Furthermore, we assume
that with increasing communicative competence the translators‘ problems
shift from target-text production problems to comprehension problems or
the potential desire to understand the source text in more depth. This desire
may become visible in the subjects‘ think-aloud data on search questions.
Therefore, we will analyse the TAPs according to the following criteria.
What types of problem cause the subjects to search? What types of
information do they search for (lexical items, relationships between units
of information, encyclopaedic information on aspects of the source text,
etc.)? In addition to this we will ask the following questions: where do the
subjects search for solutions (in the source text, in their long-term
memories or in external resources)? Do they use reference works to solve
comprehension problems or target-text production problems, or both, and
are they aware of the type of problem they have? What type of reference
works do they consult for the different kinds of problems (monolingual
dictionaries, bilingual dictionaries, encyclopaedias, parallel texts, etc.)? Do
the subjects only search until they have found an acceptable target-
language equivalent or do they go beyond that, for example, to gain more
comprehensive understanding?
The subjects‘ translation routine activation competence will be
analysed in connection with their translation creativity. Here we start from
the assumption that novices have a rather restricted translation routine
activation competence, which increases in the course of their training. The
more their translation routine activation competence increases, the more
cognitive capacity they have available for more creative solutions, which
are solutions which take into account more translation-relevant aspects.
Indicators of increasing translation activation routine competence could be
―spontaneous interlingual associations‖ for more complex source-text
items (Krings 1986: 311, 317). Therefore we will analyse the TAPs for
such associations and their complexity. As one of several measures of the
subjects‘ process creativity, the number of alternative potential translation
solutions they produce or verbalize will be determined. In an exploratory
study by Krings (1988), a professional translator produced far more
alternative solutions (measured by the variant factor VF) in his translation
processes than foreign-language learners (1.24 for the professional vs. an
average of 0.69 for the language learners). Furthermore, the subjects‘
translation products will be analysed systematically for creative solutions
(on the operationalization of measuring creativity, see Bayer-Hohenwarter
10
in this volume). Special emphasis will also be placed on omissions,
corrections of content and additions made by the subjects. These changes
will be documented in concept maps. The TAPs will be analysed
systematically for reasons for these modifications.
All instances of the phenomena for which the TAPs will be analysed
(such as problem indicators, strategies, creativity of solutions, etc.) will be
marked with special tags in the XML transcripts.
Furthermore, our analysis will concentrate on verifying the
following assumptions.
1. At the beginning of their training, the subjects are more surface-
oriented, i.e., they concentrate on small linguistic items (lexical, syntactic,
and text-linguistic problems) without realizing that a skopos-adequate
target text also requires creative solutions in the more complex problem
areas. In the course of their training, their focus of attention shifts from the
former category of problems, which will then be solved more
10 In another study, we also plan to investigate whether translation routine activation

competence can be made visible by EEG or fMRT. Such an investigation involves
having the subjects read a short source text which offers (1) items which have a
standard target-language equivalent and can be expected to be transferred in a routine
process, (2) items which may not yet be translatable in a routine process by novices
but only by experts, and (3) items which require a unique translation solution and are
beyond the use of routine competence even in the case of experts. After reading the
entire text, the subjects will be shown the first of these items in the text and asked to
think of a target-language equivalent for it. After having found a solution, the subject
is expected to press a button. After a fixed interval, the next item will be displayed,
etc. Activation patterns in the brain are assumed to vary both in their strength and
their distribution depending on the degree of routine which the subjects can fall back
on during the translation of each item.
automatically, to the latter category. Increased automaticity in a category is

expected to result in fewer verbalizations with regard to problems in this
category.
2. At the same time, the translation strategies employed become
more global (on the distinction between local and global translation
strategies, see Jääskeläinen 1993) and more diversified.
3. The same applies to search strategies: whereas searches for
individual lexical items dominate at the beginning of their training, the
subjects gradually adopt a more global approach for more complex
knowledge clusters. To the same extent, the search in bilingual dictionaries
is reduced and the number of searches in monolingual dictionaries and
other resources (parallel texts).
4. During their training, the subjects become more self-confident and
visualize themselves more as text designers than as text reproducers. This
will be manifested as more profound shifts carried out during the
translation process and a stronger tendency to take the scenes evoked by
the source text as a point of departure for producing the target text instead
of the linguistic representation of the source text. This development will
also lead to more creative solutions.
5. The development described under 4 may also lead to an increase
in the size of the translation units, which will emerge from the log files and
TAPs.
6. Reflections on the skopos (function and audience) of the target
text and the translation brief will play a more decisive role towards the end
of the subjects‘ training whereas at its beginning the subjects proceed in a
more source-text-oriented fashion. Thus their individual development as
translators reflects the development in the theory of Translation Studies
over the last 60 years.
7. In the course of their training, the subjects‘ meta-linguistic and
meta-communicative competence, as well as their meta-cognitive
competence (Risku 1998: 163), increases. On the one hand, this gives them
better control over the translation process and may facilitate the
verbalization of what goes on in their minds. On the other hand, it may
also lead to increased automaticity, which will mean that less will be
available for verbalization.
6. Availability of materials and results
All materials used in TransComp, such as the source texts, the translation
assignments, the model translations, and the questionnaires, and all data
obtained in the experiments, such as the TAPs, the log files, and the screen
records, will be made available to the scientific community in an asset
management system (AMS), an open-source-based storage, administration
and retrieval system for digital resources. This also applies to the XML
transcripts. In this way the problems pointed out by Englund Dimitrova
(2005: 82 f.) are addressed. She criticizes that so far ―no single, widely
accepted model for coding and analysis‖ has been developed and that
―there does not yet seem to be an established way of reporting protocol
data‖. The AMS will contribute to the solution of this problem and allow
future multi-centre studies, in which, for instance, the same source texts
and assignments can be downloaded from the system and used with
subjects from other translation-oriented programs and with other language
combinations, whose data can then also flow into the system and be
compared with the ones from our own and other studies.
In October 2008, the third experimental wave of TransComp started.
The materials used as well as the data collected in the experiments so far
have already been stored in our AMS (Göpferich 2007). At the moment,
these materials and data are password-protected because the source texts
will also be used in future test waves of TransComp, and we have to make
sure that our subjects do not have access to them until the last text wave
has been completed. After this, password protection will be removed and
the data can be accessed freely.
References
Anderson, J. R. 1983. The Architecture of Cognition. Cambridge, Mass.: Harvard

Univ. Press.
3
— 1990. Cognitive Psychology and its Implications. New York: Freeman.
Englund Dimitrova, B. 2005. Expertise and Explicitation in the Translation
Process. (Benjamins Translation Library 64). Amsterdam, Philadelphia:
John Benjamins.
Ericsson, K. A. & Simon, H. A. 31999. Protocol Analysis: Verbal Reports as
Data. Rev. ed. Cambridge (Mass.), London (England): MIT Press.
— & Charness, N. 1997. Cognitive and developmental factors in expert

performance. In Feltovich, P. J., Ford, K. M. & Hoffmann, Robert, R. (eds).
Expertise in Context. Human and Machine. Menlo Park, Cambridge (MA),
London: AAAI Press/MIT Press. 3–41.
— & Smith, J. 1991. Prospects and limits of the empirical study of expertise: an
introduction. In Ericsson, K. A & Smith, J. (eds). Towards a General
Theory of Expertise: Prospects and Limits. Cambridge etc.: Cambridge
Univ. Press. 1–38.
Göpferich, S. 2007: Research Project TransComp: The Development of
Translation Competence.
http://gams.uni-graz.at/fedora/get/container:tc/bdef:Container/get
(19.04.2009).
— 2008. Translationsprozessforschung: Stand – Methoden – Perspektiven.
(Translationswissenschaft 4): Tübingen: Narr.
— in press. Measuring comprehensibility in specialized communication. In
Proceedings of the 16th European Symposium on Language for Special
Purposes ‘Specialized Languages in Global Communication’. Hamburg,
27–31 August, 2007.
— in press. Data documentation and data accessibility in translation process
research. The Translator.
Hansen, G. 2006. Erfolgreich Übersetzen: Entdecken und Beheben von
Störquellen. (Translationswissenschaft 3). Tübingen: Narr.
Harris, B./Sherwood, B. 1978. Translating as an innate skill. In Gerver, D. &
Sinaiko, H. W. (eds). Language Interpretation and Communication. New
York. 155–170.
Hönig, H. G. 1991. Holmes‘ ‗Mapping Theory‘ and the landscape of mental
translation processes. In van Leuven-Zwart, K. & Naajkens, T. (eds).
Translation Studies: The State of the Art. Proceedings from the First James
S. Holmes Symposium on Translation Studies. Amsterdam: Rodopi. 77–89.
— 1995. Konstruktives Übersetzen. Tübingen: Stauffenburg.
Jääskeläinen, R. 1993. Investigating translation strategies. In Tirkkonen-Condit,
S. & Laffling, J. (eds). Recent Trends in Empirical Translation Research.
Joensuu: Univ. of Joensuu. 99–120.
Kiraly, D. C. 1995. Pathways to Translation: Pedagogy and Process. Kent
(Ohio), London (England): Kent State Univ. Press.
Krings, H. P. 1986. Was in den Köpfen von Übersetzern vorgeht. Eine empirische
Untersuchung zur Struktur des Übersetzungsprozesses bei fortgeschrittenen
Französischlernern. Tübingen: Narr.
— 1988. Blick in die ‗Black Box‘ – Eine Fallstudie zum Übersetzungsprozeß bei
Berufsübersetzern. In Arntz, R. (ed.). Textlinguistik und Fachsprache.
AILA-Symposium Hildesheim, 13.–16. April 1987. Hildesheim: Olms. 393–
411.
— 2001. Repairing Texts: Empirical Investigations of Machine Translation Post-
Editing Processes. Kent (Ohio), London: Kent State Univ. Press.
Neubauer, A. & Stern, E. 2007. Lernen macht intelligent: Warum Begabung

gefördert werden muss. München: Deutsche Verlags-Anstalt.
PACTE 2000. Acquiring translation competence: hypotheses and methodological
problems in a research project. In Beeby, A., Ensinger, D. & Presas, M.
(eds). Investigating Translation. Amsterdam, Philadelphia: John Benjamins.
99–106.
— 2002. Exploratory texts in a study of translation competence. Conference
Interpretation and Translation 4.4: 41–69.
— 2003. Building a translation competence model. In Alves, F. (ed.).
Triangulating Translation. Amsterdam, Philadelphia: John Benjamins. 43–
66.
— 2005. Investigating translation competence: conceptual and methodological
issues. Meta 50.2: 609–619.
— 2007. Zum Wesen der Übersetzungskompetenz – Grundlagen für die
experimentelle Validierung eines Ük-Modells. In Wotjak, G. (ed.). Quo
vadis Translatologie? Ein halbes Jahrhundert universitäre Ausbildung von
Dolmetschern und Übersetzern in Leipzig. Rückschau, Zwischenbilanz und
Perspektiven aus der Außensicht. Berlin: Frank & Timme. 327–342.
Pym, A. 1991. A definition of translational competence, applied to the teaching
of translation. In Jovanovic, M. (ed.). Translation: A Creative Profession.
Proceedings of the 12th World Congress of FIT. Belgrad: Prevodilac. 541–
546.
—. 2003. Redefining translation competence in an electronic age. In defence of a
minimalist approach. Meta XLVIII.4: 481–497.
Risku, H. 1998. Translatorische Kompetenz. Tübingen: Stauffenburg.
— 2004. Translationsmanagement. Interkulturelle Fachkommunikation im
Informationszeitalter. (Translationswissenschaft 1). Tübingen: Narr.
Shreve, G. M. 1997. Cognition and the evolution of translation competence. In
Danks, J. H., Shreve, G. M., Fountain, S. B. & McBeath, M. K. (eds).
Cognitive Processes in Translation and Interpreting. Thousand Oaks: Sage
Publications. 120–136.
Sternberg, R. J. 1997. Cognitive Psychology. Fort Worth etc.: Harcourt Brace.
TEI. 2008. TEI P5 Guidelines. In <http://www.tei-
c.org/Guidelines/P5/index.xml> (14.06.2008).
Tilp, S. 2007. Übersetzen ohne Auftragsspezifikation? – Eine empirische Studie
zum Umgang mit unterspezifizierten Übersetzungsaufträgen. (Unpublished
Diploma Thesis): Graz: Karl-Franzens-Universität, Institut für Theoretische
und Angewandte Translationswissenschaft.
Wilss, W. 1992. Übersetzungsfertigkeit. Annäherungen an einen komplexen
übersetzungspraktischen Begriff. Tübingen: Narr.
Translational creativity: how to measure the unmeasurable
Gerrit Bayer-Hohenwarter
Abstract
The present article describes a new approach to measuring translational
creativity and its development in students of translation as compared to
professional translators. It reports preliminary results of my PhD thesis
(Bayer-Hohenwarter, in progress), which forms part of the longitudinal
study TransComp (see Göpferich 2008, and this volume).
Creativity is a concept that is difficult to define and even more
difficult to measure. An appropriate measurement method is crucial,
however, in finding out how translational creativity develops. The method
proposed here is based on the consensual creativity criteria novelty and
acceptability, and the prototypical creativity dimensions flexibility and
fluency. More specifically, the analysis reported on in this paper focuses
on cognitive procedures attributable to these dimensions. After a brief
review of the literature, a case is made for the inclusion of the creative
procedures abstraction, modification and concretisation in analysing
translational creativity. These procedures represent cognitive shifts
between ST and TT as opposed to mere reproduction. The applicability of
these procedures is tested on a sample of 13 translations (nine students,
four professionals) of one ST item and one set of intermediate translations
by one professional translator. This analysis modestly confirms the
hypothesised low creativity in first-year students as opposed to that of
professional translators.
Defining creativity
Any measurement of creativity first requires an adequate definition of the
concept. In psychology, creativity has been assumed to be an elusive
concept that seems to defy precise definition and measurement because of
its multicomponential nature. According to Wittgenstein‟s idea of family
resemblances (Wittgenstein 1958/1977, cf. Lakoff 1987/1990: 16), there
40 Gerrit Bayer-Hohenwarter
are many concepts which cannot be defined by common properties with

clear boundaries. Translational creativity is such a concept. Creative
translation products and processes can be characterised by qualities such as
rareness, outstanding quality, high cognitive effort, fluency or non-
literalness, but none of these individual qualities are mandatory.
Consequently, it is impossible to set up an exhaustive list of criteria that
can reasonably be regarded as necessary and sufficient for a definition of
translational creativity (cf. e.g. Amelang et al. 2006b: 46). Two criteria,
however, that any creative process or product must meet are novelty and
adequacy (e.g. Torrance 1988, Amabile 1996, Csikszentmihalyi 1997,
Gruber/Wallace 1999, Sternberg and Lubart 1999). For the purposes of the
present study, translational novelty is defined as a manifestation of (1)
exceptional performance that considerably exceeds translational routine,
(2) uniqueness or rareness within the TransComp data corpus (=
originality), and (3) a non-obligatory translational shift (cf. Kußmaul
2000a: 311), whereby not all aspects must be present. Acceptability is
defined as skopos adequacy.
As a first step, a review was carried out of several approaches to
specifying and classifying parameters or indicators of creativity, both in
translation studies and in psychology. It appeared that the factorial
approach suggested by Guilford (1950), the “father of creativity research”,
provided the most promising and comprehensive framework. It comprises
nine dimensions, or basic abilities, which are a prerequisite for creativity:
novelty, fluency, flexibility, ability to synthesise, ability to analyse, ability
to reorganise/redefine, complexity/span of ideational structure, and
evaluation. It seems possible to attribute all manifestations of translational
creativity, e.g. non-literalness, generativity as measured by Krings‟ variant
factor (1988, 2001), or Kußmaul‟s types of creative translation (2000a,
2000c), to one of these dimensions. Ideally, Guilford‟s framework will
enable us to define translator profiles based on specific strengths and
1
Contrary to Kußmaul‟s view, non-obligatory shifts are considered to be more creative
than obligatory shifts. If a literal translation sufficiently reaches skopos adequacy, the
production of a non-obligatory shift indicates a particularly high awareness of quality,
willingness to take risks, motivation to pursue one‟s search, etc. In the case of
obligatory shifts, however, relatively little problem sensitivity is required in order to
deviate from the ST structure.
Translational creativity 41
weaknesses in different areas, e.g. high fluency but little evaluation

competence or high flexibility but little fluency.
My PhD thesis focuses on novelty, fluency and flexibility, which are
commonly perceived as the prototypical creativity dimensions. Flexibility
is defined as the ability to transgress fixedness (e.g. literalness in
translation) and fluency as the ability to produce a large number of
translation variants and/or adequate translation solutions spontaneously or
even automatically.
This article focuses on how flexibility can be pinpointed in
translations. The idea is to identify the nature of creative cognitive
procedures in translation that express flexibility at the process-level. The
advantage of research into creative cognitive procedures is that these seem
to provide the key to finding ways of fostering creativity in students of
translation. In the following sections a review will be given of some types
of “creative procedures” that have been suggested in the literature and in
addition an approach to analysing them will be presented.
Creative procedures reviewed

A review of process-oriented psychological approaches showed that they
could not be used in my study for two main reasons: (1) traditional process-
oriented models, based on or similar to the four-stage model of preparation,
incubation, illumination and evaluation (see Preiser 1976: 42 f., for an
overview), seem to be too vague and thus inadequate for tracing the
development of creative competence; (2) more recent process-oriented
approaches (see Amelang et al. 2006a: 236 f. for an overview) require the
use of neuroscientific methods, which are beyond the scope of this study.
In translation studies and linguistics, analyses of creativity often rely
on procedures such as addition, omission, and modification. Different terms
for these concepts have been used and different typologies have been
developed (e.g. Ballard 1997, Ivir 1998: 138, Pellatt 2006: 52). The major
drawback of these typologies seems to be that they refer to purely form-
oriented ST-TT differences whereas a classification of creative procedures
needs to rely on cognitive categories. It will be necessary to focus on
cognitive categories if the goal is to find out how creative translators think,
how creativity develops and how to improve creativity in students of
translation.
Recent theories of creativity (e.g. Kußmaul 2000a) draw heavily on

prototype semantics (e.g. Rosch 1977) and more specifically, on scenes-
and-frames semantics (e.g. Fillmore 1976, 1985). Such approaches promise
to overcome the drawbacks of purely form-oriented categorisations of
translational procedures and have been used by Kußmaul (2000a, 2000b,
2000c, 2005) for developing his “types of creative translation”.
Fillmore‟s (1976, 1985) scenes-and-frames theory is based on the
idea that the language system is the mould that we can use to express our
ideas. This language system, or the “system of linguistic choices” in
Fillmore‟s terms, is called “frame” (1976: 63). Frames can trigger
associations with “scenes”, that is “not only visual scenes but familiar
kinds of interpersonal transactions, standard scenarios, familiar layouts,
institutional structures, enactive experiences, body image; and, in general,
any kind of coherent segment, large or small, of human beliefs, actions,
experiences, or imaginings” (1976: 63). Scenes and frames can activate one
another, relate linguistic knowledge to extralinguistic knowledge, and their
activation is assumed to be an essential process for the building of
meaning.
Kußmaul’s types of creative translation

Kußmaul (2000a) classifies translation shifts on a cognitive basis into
seven types: (1) Change of frames, (2) Framing, (3) Picking of scene
elements from one frame, (4) Picking of scene elements from one scene,
(5) Change of scenes, (6) Enlarging of a scene and (7) Re-framing. These
types of creative translation without question represent a very interesting
approach because it was the first time an attempt was made to use scenes-
and-frames semantics to identify and classify the cognitive procedures at
work during translation. Kußmaul‟s work was an important step, moving
beyond the concept of mere form-oriented shifts that have traditionally
been the focus of attention. In Guilford‟s framework, six of Kußmaul‟s
seven types could be attributed to the flexibility dimension, and Kußmaul‟s
second type (framing) would be an example of his novelty dimension.
However, a weakness of Kußmaul‟s approach is that the different
types are not easily distinguished or easily remembered, the reasons for
which partly seem to reside in the somewhat fuzzy use of the word frame,
in overlaps between the different types and in that the types appear to
belong to different levels of abstraction. As explained before, frame refers

to a system of linguistic choices whereas scenes are more abstract
extralinguistic entities. Within our current research framework, however, it
is in many cases very difficult to objectively trace the nature of scenes
evoked in the heads of translators by inferring from the linguistic elements
contained in the think-aloud protocols.
Kußmaul‟s typology provides us with extremely helpful and
inspiring insights for identifying creative translation procedures, but some
ground remains to be covered.
Three types of creative translation procedures

An attempt to measure translational creativity must also consider the
process level, i.e. creative procedures, if one wants to find out how
creativity develops, why certain translators produce more creative
translations than others and perhaps also how to foster creativity in students
of translation.
Apart from the scenes-and-frames semantics and prototype semantics
mentioned earlier, there is a third concept rooted in cognitive theory, which
has so far not received any attention in translation theory, though it would
appear to be very useful. This is the theory of basic-level primacy (see
below).
According to Lakoff‟s (1987/1990: 31ff) and Rosch‟s (1977)
interpretation of Brown‟s (1958) findings, humans operate on various
levels of categorisation, for example when they reason about something or
describe the world around them. With reference to the notion of different
levels of categorisation, it can thus be argued that TT renderings that
belong to the same level of categorisation as the corresponding ST element
can generally be considered “natural” and less creative than TT renderings
that belong to a different level of categorisation. This explains why “literal”
translations that are on the same level of categorisation as the
corresponding ST element are commonly (and reasonably) regarded as less
creative than non-literal translations. The same principle applies to
translation briefs: a brief that requires a translation according to the ST
function and with practically the same target group, except that it is from
the target culture, is generally considered routine and non-creative. Briefs
that include a variation of the translation‟s function, e.g. a specialised text
that needs to be translated for the general public or vice versa, are
considered more creative.
Moreover, according to the theory of basic-level primacy (Brown
1958, cf. Lakoff 1987/1990: 13 f), the basic level of categorisation, e.g.
dogs as opposed to mammals or poodles, is, among others, used most
frequently in natural language and is connected with most ease of cognitive
processing and linguistic expression. Whereas the basic level can be
considered “natural”, higher or lower levels of categorisation are said to be
products of the imaginative and thus the creative mind. From this, it can be
deduced that abstractions from lower or higher ST levels up or down to the
TT basic level respectively (e.g. TT dogs instead of ST poodles or TT cars
instead of ST motor vehicles) can also be considered less creative than TT
renderings that stay on the same low or high level of categorisation as the
ST. In fact, the use of umbrella terms on the basic level of abstraction is a
frequent strategy in sight translation or interpreting, whereas the use of
higher-order abstractions (e.g. motor vehicles instead of cars) can be
assumed to be more effortful. The use of lower-order categorisations
requires more activation of knowledge and for this reason seems to take
more effort. It is as yet uncertain to what extent these findings can be
applied to translation, and in order to move beyond the stage of speculation
a more detailed analysis with a larger data corpus is required.
A descriptive framework
A critical analysis of the approaches and findings mentioned above allowed
me to draw the conclusion that, instead of using form-oriented shifts or a
typology based on scenes and frames, one could perhaps more aptly
analyse translations with a view to the following three basic creative
procedures:
Abstraction ↑
Modification ↔
Concretisation ↓
As the arrows indicate, this basic typology refers to “directions of thought”,
i.e. upward, sideways and downward with reference to the ST element as
opposed to mere reproduction. These three procedures appear frequently in
Kußmaul‟s explanations of his types of creative translation and can also be
said to correspond to the levels of categorisation suggested by Brown

(1958). Abstraction refers to situations where translators use more vague,
general or abstract TT solutions. Modification refers to strategies such as
re-metaphorisation or changes of perspective. If the TT evokes a more
explicit, more detailed and more precise idea than the ST, this procedure is
called concretisation. It is assumed that actions such as paraphrase,
addition and deletion cannot be directly attributed to any one of these three
procedures. A paraphrase, for instance, can lead to a more abstract or a
more precise idea in the TT than that contained in the ST. The basic
creative procedures suggested may therefore have very different
manifestations at the form level, but this form is not decisive when creative
procedures are assigned. It is assumed that all translation products can
reasonably be assigned to either abstraction, modification, concretisation or
reproduction.
The abstraction, modification or concretisation procedures can all be
considered creative because they deviate from the initial level of
categorisation, i.e. the ST level. This can be justified for several reasons:
Abstraction figures prominently among strategies associated with
creativity in psychology (e.g. Ward et al. 1999: 191).
Concretisation, modification and abstraction can all be considered to
require more cognitive effort than reproduction. Whereas reproduction is
mere routine translation at an identical level of categorisation, concreti-
sation, modification and abstraction can be regarded as non-routine.
Among the many researchers who see creativity as a type of problem-
solving behaviour, cognitive effort as opposed to routine is commonly
held to be one of the most essential creativity criteria (e.g. Weisberg
2006).
Modification can be associated with flexibility in the sense of Guilford
(1950) and other creativity researchers inasmuch as seeing things from a
different perspective, or finding new uses for available resources, are
commonly regarded as aspects of creativity.
Apart from being a shift and thus a sign of flexibility, concretisation can
be associated with “depth of analysis”, i.e. going beyond the mere surface
of the apparent and obvious and giving details of what is assumed to be
the core meaning. Such depth of analysis is considered creative by e.g.

Rietzschel et al. (2007).
All three procedures, abstraction, modification and concretisation, can be
considered shifts and thus phenomena that can be categorised in Guilford‟s
dimension of flexibility. All three, if they are to be successful and result in
an adequate product, require deep and true understanding which, as
opposed to processing the surface of language, is considered a creative
process in its own right (e.g. Holman & Boase-Beier 1998/1999b: 15;
Bastin 2003: 350, Dancette et al. 2007) and different from what machines,
for example, can accomplish. Non-creative processes essentially consist of
reproducing the ST element; a lack of creativity can thus mostly be
attributed to fixedness on the ST structure. Fixedness can, however, also be
extended to a presumed authoritative ST validity (“what is written in the ST
is true”), or fixedness on a certain type of problem-solving strategy (e.g. re-
metaphorisation but not concretisation; body metaphors but not other
metaphors).
Assumptions
It is assumed that first-year students will tend to use more same-type
creative procedures in their translations than third-year students or
professional translators, i.e. more experienced translators will presumably
cover a broader range of creative procedures. The more creative procedures
that can be applied while translating a creativity-demanding ST element,
the more likely it is that the result will be a creative solution.
Another assumption is that advanced students and professional
translators will apply more unique procedures and produce more unique
solutions.
Furthermore, it is assumed that more competent translators will
display higher fluency in cases where they can fall back on routine
processes, i.e., they will produce more instant solutions.
The following section describes the results of a pilot analysis of
creative procedures that occurred in the translation of one ST unit in my
corpus.
Sample analysis of creative translation procedures for one ST unit
The first ST segment is from an English popular science book on how to

win friends. It comes from a chapter where it is reasoned that the behaviour
of dogs can be taken as a model for success in finding friends:
ST:
If you stop and pat him [the dog], he will almost jump out of his skin to show
you how much he likes you.
My sample corpus comprises the translations of nine students and four

professionals, which are listed in Table 1 below. I have awarded excellent
solutions “pass+”. Though such a judgment admittedly increases the
subjectivity of the rating, it is considered necessary because outstanding
quality is a frequently mentioned creativity criterion. It is even considered
legitimate, as creativity is based by definition on subjective judgments. By
maximally objectifying all other indicators, including process indicators, it
is believed that overall subjectivity is kept to a minimum. In the
“Procedures” column, the creative procedures abstraction, modification and
concretisation as defined above are indicated; if none of these procedures
applies, the label reproduction for a non-creative reproductive procedure is
given.
As can be seen from the table, the translations analysed can be
classified into four main groups:
Group A (TT2, TT6) The existing metaphor is re-produced and results in an
inadequate solution because the English to jump out of one’s skin in this case
does not mean “to be badly frightened” or “to be very much surprised” but
that the dog is overwhelmed by emotions and/or shows very strong
affection.
Group B (TT1, TT7, TT11; TT10) The meaning “dog shows very strong
affection” is rendered non-metaphorically and represents an abstraction.
TT10 is similar in that it also represents a de-metaphorisation and abstracts
the meaning but also different insofar as it describes the dog‟s seemingly
irrational emotional behaviour.
Table 1. Translations of the ST unit “dog jumps out of his skin”2
TT Sub- Wave Target text Rating Procedures

ject
TT1 JTH t1 wird er alles tun [um Dir zu pass Abstraction; de-
zeigen wie sehr er dich mag.] metaphorisation
TT2 JZE t1 wird er beinahe aus seiner fail Reproduction

Haut springen
TT3 ERE t1 wird er an dir hoch springen pass Concretisation; de-
metaphorisation
TT4 LPE t1 wird er sich fast überschlagen pass Modification; re-
metaphorisation
TT5 MLE t1 wird er beinahe einen Luft- pass Modification;
sprung machen re-metaphorisation
TT6 STO t1 wird er fast aus seiner Haut fail Reproduction
herausspringen
TT7 THI t1 tut er alles um dir zu zeigen pass Abstraction;
de-metaphorisation
TT8 BKR t3 wird er voll Übermut und pass + Concretisation;
Freude um dich de-metaphorisation
herumspringen explicitation
TT9 SFR t3 wird er sich fast überschlagen pass Modification;
re-metaphorisation
TT10 GLS t8 wird er sich wie verrückt pass Abstraction;
gebärden de-metaphorisation
TT11 HEM t8 dann wird er alles tun pass Abstraction;
de-metaphorisation
TT12 HOB t8 zerreißt er sich fast pass Modification;
re-metaphorisation
TT13 SCH t8 wird er sich fast überschlagen pass Modification;
re-metaphorisation
TT14 MT bringt er sich beinahe um pass Modification;
re-metaphorisation
TT15 GBH zeigt er Ihnen mit einem pass + Modification;
Freudentanz [wie sehr er Sie re-metaphorisation
mag]
TT16 GBH springt er an Ihnen hoch und pass Concretisation;
demonstriert Ihnen seine de-metaphorisation
Zuneigung einfach enrichment
umwerfend
2
The first column of the table gives a running number for the TT (e.g. TT1), the second gives the
abbreviation for the anonymised subject (or, exceptionally, MT for model translation or GBH for my
own translation), and in the third column the test wave is specified (t1 means translation at the
beginning of the first semester, t3 at the beginning of the third semester, t8 translation of professional
translator). The fifth column „Rating‟ specifies the global ratings in the sense of adequate or inadequate
with a view to the given skopos.
Group C (TT4, TT5, TT9, TT12, TT14) The ST metaphor is rendered by a

different TT metaphor. Most of the Group C TT metaphors, including model
translation TT14, provide violent images for the meaning “dog shows very
strong affection”. According to these images, the dog‟s affection puts his life
at risk (er zerreißt sich – literally: “tears himself into pieces”; er bringt sich
um – literally: “he kills himself”; er überschlägt sich – literally: “he
overturns”). TT5 is the only metaphor that provides a less violent image for
“showing one‟s affection” (einen Luftsprung machen, literally: “to jump into
the air”; translation: “to cut a caper”, “jump for joy”, “be exceedingly
happy”).
Group D (TT3, TT8): The meaning is concretised and (one aspect of) the
behaviour of the dog described non-metaphorically. In Kußmaul‟s
terminology, this would be picking one scene element from a scene. TT3
uses the scene element of the dog jumping at the person; TT8 uses the scene
element of the dog foolishly running around the person. It is also possible to
use another scene element like the dog licking the person. One could argue
that some people would not be happy at all if a dog jumped on them or
licked them and that translations of this kind thus do not fulfil their purpose,
which is to illustrate how dogs make friends and not how they deter people.
However, this solution is considered a borderline pass because everybody is
assumed to know that dogs just mean to show their affection and that the
reader can draw the intended analogy between the dog‟s behaviour and
friend-winning human behaviour despite their personal feelings about
certain aspects of dog behaviour.
TT15 and TT16 are my own: TT15 shows how, departing from the
concretisation of the dog‟s actual behaviour, a different metaphorical image
can be found for the meaning aspect “dog is overwhelmed by his
emotions”: einen Freudentanz machen corresponds to “dance a jig of joy”.
TT16 picks one element from the scene of a dog‟s behaviour and
compensates for the de-metaphorisation by including a pun in the second
part of the sentence. The pun is based on the notion of knocking someone
over (umwerfen) as the dog would if it jumped at someone in great joy; the
adjective derived from umwerfen can also take the meaning of a positive
adjective (roughly: “dazzling”, “drop-dead gorgeous”). This pun can be
seen as a form of linguistic enrichment. Under “linguistic enrichment” I

subsume all instances of a “neutral” ST element rendered with a TT
stylistic device that increases the rhetorical effect of a message such as a
metaphor, a pun, alliteration or rhyme.
Interestingly, five of the 13 TTs from the corpus have the German
counterpart of jump (= springen) included in some way in the target
language. This lexical link occurs not only in reproductions, but also in
concretisations and modifications (Luftsprung, hoch springen,
herumspringen). We could subsume such links, be they lexical or syntactic,
under the label “fixedness”. Such translational fixedness occurs when the
ST lexical elements trigger TT solutions with at least one structurally
similar element, or, in other words, the ST surface structure strongly
activates other metaphors, scene elements, idioms or other TL expressions
that build on the same linguistic element (here: jump).
An overview of the results for the “dog-jumps-out-of-his-skin”
translations is given in Table 2 with the goal of assessing how creative the
various solutions are. All same or same-type translations are included in the
same table row; the translations are classified according to their creative
procedures with all creative procedures marked in italics. In the row
“Creativity indicators”, all indicators that were observed for a particular
translation are listed. Acceptability is a necessary prerequisite; in the case
of a fail, no more creativity indicators need to be specified because
creativity must be excluded from the outset. Comprehension refers to a
creative comprehension process; this is true for all adequate solutions
because the English to jump out of one’s skin usually refers to the meaning
“be badly frightened” or “be very much surprised” and was used in a
different meaning only in the given context. P-flexibility stands for process
flexibility and refers to an abstraction, modification or concretisation
procedure and corresponding secondary procedures. The number given in
brackets refers to the number of secondary procedures observed; for
instance, solutions that include re-metaphorisation and enrichment are
deemed more creative than solutions with re-metaphorisation only.
Table 2. Creativity assessment of the translations of “dog jumps out of his

skin”
Dog jumps out of Primary Secondary Target Creativity Crea-
his skin proce- proce- texts indicators tivity
dure(s) dure(s) rating
Group A: Re-produce TT2_t1 x
wird er beinahe/fast (x)
aus seiner Haut TT6_t1
heraus/springen (x)
fahrt er fast aus der
Haut
Group B: Abstract De-meta- TT1_t1 Acceptability 2
wird er alles tun phorise TT7_t1 Comprehension
TT12_t8 P-flexibility (1)
wird er sich wie Abstract De-meta- TT10_t8 Acceptability 3
verrückt gebärden phorise Comprehension
P-flexibility (1)
Uniqueness
Group C: Modify TT4_t1 Acceptability 2
wird er sich fast TT9_t3 Comprehension
überschlagen TT11_t8 P-flexibility (1)
TT14
Group C: Modify TT5_t1 Acceptability 3
wird er beinahe Comprehension
einen Luftsprung P-flexibility (1)
machen TT13_t8 Uniqueness
zerreißt er sich fast
Group D: Concretise De-meta- TT3_t1 Acceptability 3
wird er an dir hoch phorise Comprehension
springen P-flexibility (1)
Uniqueness
Group D: Concretise De-meta- TT8_t3 Acceptability 5
wird er voll phorise Comprehension
Übermut und Explicitate P-flexibility (2)
Freude um dich Uniqueness
herumspringen Outstanding
quality
springt er an Ihnen Concretise De-meta- TT16 Acceptability 4
hoch und phorise Comprehension
demonstriert Ihnen Enrich P-flexibility (2)
seine Zuneigung Uniqueness
einfach umwerfend
zeigt er Ihnen mit Concretise Explicitate TT15 Acceptability 5
einem Freudentanz Modify Comprehension
P-flexibility (2)
Uniqueness
Outstanding
quality
Uniqueness refers to a unique solution within the given corpus; this,

however, is only a preliminary indicator that will presumably require
modification and re-naming as rareness when all (intermediate) translation
solutions of all experimental waves have been analysed. Outstanding
quality refers to particularly adequate, elegant or linguistically economic
solutions and is meant to compensate for the fact that some of the solutions
that are considered a pass are in fact a very bare pass. Finally, a creativity
rating is given by adding all creativity indicators together except
acceptability (which is the necessary prerequisite).
Generally, the creativity ratings that result from this assessment
procedure correspond with my own intuitive judgment. However, the
creativity indicators from Table 2 are not exhaustive. For instance, analyses
of additional segments from the corpus (Bayer-Hohenwarter, in progress)
have shown that the indicator “comprehension” is not applicable to all
segments. Moreover, in order to refine the analysis, the intermediate
solutions of the individual translators must be included. By way of
example, an analysis of HOB‟s problem-solving process is given below.
ITT stands for “intermediate target text”:
Table 3. Overview of intermediate translations for “dog jumps out of his skin”
(HOB_t8)
HOB t8 Target text Rating Procedures

ITT1 fahrt er fast aus der Haut fail Reproduction
ITT2 fährt er fast aus der Haut fail Reproduction
ITT3 macht er fast ‟nen Kopfstand fail Modification
ITT4 reißt er sich ein Bein aus fail Modification
ITT5 zerreißt er sich fail Modification
TT zerreißt er sich fast pass Modification;
re-metaphorisation
This example is an interesting account of how a creative solution comes

into being. The translator starts off by producing a literal translation that
demands relatively little cognitive effort. As this primary equivalent
association (cf. Krings 1986: 317) proves unsatisfactory even if downtoned

(using the word fast – “almost”), the translator continues searching for
similar TL metaphors in his mind. Just as in the ST, his second intermediate
TT solution (ITT3) is a body metaphor (einen Kopfstand machen literally
means “to do a headstand”, but figuratively it means something like “to
work extremely hard”) and also produces an additional humorous effect if
one visualises a dog doing a headstand. The translator is obviously aware
of this and recognises the need for continuing his search. This time, his
bilingual associative competence allows him to fall back on an English
synonym of the ST element which, of course, is not an adequate TT
solution but serves as another point of reference. This strategy can be
considered a change of perspective and an unconventional method of
activating potential TT solutions that can be hypothesised to be unique or at
least rare within the TransComp data corpus. For this reason, this strategy
is an instance of (hypothesised) originality and (proven) flexibility; it helps
the translator in finding a successful definitive TT solution. ITT4 reißt er
sich ein Bein aus constitutes yet another re-metaphorisation with the focus
on the dog‟s body. However, it is again an inadequate solution because the
idiom is usually used in the negated form er reißt sich kein Bein aus,
meaning “he won‟t strain himself”. With ITT5 the translator takes up the
element reißen and produces yet another body metaphor which, however, at
least partly activates the same meaning as ITT3. By downtoning this
solution with German fast (“almost”), the comic effect produced by
visualising the literal meaning of a dog torn into bits is weakened and the
solution improves. This step-by-step procedure adopted by the translator
shows that he must invest high cognitive effort and that he can approach an
acceptable solution only gradually – a phenomenon that could be analysed
more profoundly within Guilford‟s dimension of “complexity/span of
ideational structure”. A more creative solution, however, would have
required even more determination and “creative strength”.
ITT3, ITT4 and ITT5 show how solutions can be found by building
on previous suggested solutions, by changing the voice and by specifying
the emotional state. It is felt that, in this example, visualisation (cf. e.g.
Kußmaul 2005) is a particularly useful strategy that helps in judging the
adequacy of intermediate solutions. As regards an overall creativity rating,
it is first of all argued that, at the process level, the unacceptability of an
ITT must not be counted as an immediate elimination criterion for

creativity. Translators often produce inadequate ITTs while fully aware of
their defect; these ITTs are just small steps in associative chains or other
creative production processes and it would be unfair to judge the quality of
the translation process according to the acceptability of the intermediate
result. Reproduction, however, is believed to be a valid elimination
criterion for creativity, also at the process level of ITTs. Instead, it is
believed necessary to judge the acceptability and quality of the translation
process according to the following criteria:
the number of intermediate translations that are considered to be a valid
indicator of fluency;
the creative procedures abstraction, modification and concretisation
inherent to ITTs as indicators of flexibility;
automaticity and spontaneity as indicators of fluency;
own idea vs. dictionary result as an indicator of novelty (cognitive
effort);
other interesting procedures such as changes of perspective or
visualisation.
The indicators “automaticity” and “spontaneity” result from the analysis of
the time interval between the reading of a particular ST segment and the
production of the corresponding TT element. If a TT element is generated
at once, i.e. within three seconds of the first encounter with the ST element
(excluding the pre-phase, where the ST is usually read without deep
analysis) and without any obvious signs of considerable cognitive effort
involved, I speak of an automatic translation; if a TT element is generated
within three seconds, but signs of considerable cognitive effort exist (e.g.
previous or subsequent comments or the production of translation
alternatives), I speak of a spontaneous translation. The creativity
assessment can thus be refined as follows:
Table 4. Creativity assessment for intermediate translations of “dog jumps out of

his skin” (HOB_t8)
HOB Target text Conditions of production Creativity

t8
ITT1 fahrt er fast main phase: written down after a long x (Reproduction)
aus der Haut pause
ITT2 fährt er fast main phase: after 3 minutes of producing x (Reproduction)
aus der Haut other TT and re-reading ITT1 but
without delay
ITT3 macht er fast main phase: generated spontaneously Modification;
‟nen after ITT2; dismissed without further re-metaphorisation
Kopfstand comment Spontaneity
ITT4 reißt er sich main phase: HOB produces English x (Modification)
ein Bein aus primary association fall over backward Change of
and uses it as a search term in online perspective
dictionary. ITT4 is one search result.
ITT5 zerreißt er main phase: solution taken from online x (Modification)
sich dictionary
TT zerreißt er main phase: generated and self-dictated Modification;
sich fast immediately after comment re-metaphorisation
Spontaneity
Generativity (4)
Procedures (1)
Spontaneity (2)
Others (1)
The total rating given in the last cell gives the overall creativity rating for
HOB‟s problem-solving process. Generativity (4) refers to the 5 inter-
mediate solutions whereby the difference between ITT1 and ITT2 is solely
grammatical and thus considered negligible. Procedures (1) refers to one
modification evident in ITT3 after the exclusion of the dictionary results
(ITT4, ITT5) and after the exclusion of the procedure assigned to the final
TT that had been counted before. Spontaneity (2) refers to the two instances
of rapid TT production, where in both cases the ST element had been
previously dealt with. Others (1) refers to HOB‟s generation of a ST
synonym (“fall over backwards”) that serves as a new starting point for
associations and dictionary research, and can be considered a change of
perspective. The sum of eight creativity points on this process level
together with HOB‟s three creativity points on the product level are
assumed to be a sound quantified basis for further comparisons with
translations by the same translator and translations of the same ST segment

by other translators with the same and with different competence levels. It
still remains to be decided if abstraction will receive a score identical to
that of concretisation and modification. Only an analysis of a larger sample
and comparison with intuitive creativity judgments will allow this decision
to be made.
A first comparison of these intermediate results by HOB with those
available from the other subjects seems to indicate that first-year students
more often produce fewer or no intermediate solutions, which can be
interpreted as a lack of problem sensitivity (Guilford‟s first creative
dimension). Interestingly, all three of the subjects for which the TAP
transcripts are available to date and who produced at least two ITTs stuck
to the procedure they chose at first: SFR produced three modifications,
HOB produced four modifications, HEM produced two abstractions. This
phenomenon can be interpreted as a type of successful procedural
association or, if the results were inadequate, one could speak of unwanted
procedural fixedness. A similar pattern of associations that can be
successful or unsuccessful was referred to earlier in this paper (e.g. reißt er
sich ein Bein aus – zerreißt er sich; or jump out of one’s skin – hoch
springen).
The analysis of 15 translations (13 from the TransComp corpus and
two produced by myself) modestly confirms the following predictions.
First-year students seem to have a stronger tendency towards reproduction,
produce fewer acceptable solutions, fewer unique solutions and fewer ITTs.
This is an indication of low creativity mostly in the sense of little problem
sensitivity, little originality, little flexibility and, at least partly, lower
fluency. However, these conclusions are based on an extremely small
sample and need to be tested on a larger corpus, before more reliable
conclusions can be drawn.
Conclusion
By way of conclusion I hope to have shown that measuring creativity is

worthwhile within Guilford‟s framework. The cognitive procedures
abstraction, modification and concretisation vs. reproduction can be
combined with other creativity indicators and assigned to
novelty/originality, fluency and flexibility. Abstraction, modification and

concretisation appear to be sound categories that bridge the gap between
traditional shifts at the form level and cognitive categories. Quantifying the
types of procedures involved and several other creativity indicators can
provide interesting results and allow us trace the development of
translational creativity. Not only can we find out which procedures are used
by beginners as opposed to experienced translators, but we can also find
out about the range of procedures that one translator is able to activate. We
can also trace how translators proceed from the primary equivalent
association via several intermediate solutions to the definitive target text by
means of procedures which have gradually been refined or which might
even be diametrically opposed. A qualitative analysis can show what types
of procedures at a micro level (e.g. fixedness on body metaphors) can be
successful or unsuccessful, how different creativity profiles of different
translators can have different or similar effects on the overall creativity of
their performance and how these profiles evolve over time.
References
Amabile, T. M. 1996. Creativity in Context. Update to the Social Psychology of

Creativity. Boulder/Oxford: Westview Press.
Amelang, M., Bartussek, D., Stemmler, G. & Hagemann, D. 1981/2006a. Zur
Theorie der Kreativität. Prozessmodelle. In M. Amelang, D. Bartussek, G.
Stemmler & D. Hagemann (eds). Differentielle Psychologie und
Persönlichkeitsforschung 6. Completely revised 6th ed. Stuttgart:
Kohlhammer. 235-237.
Amelang, M., Bartussek, D., Stemmler, G. & Hagemann, D. 1981/2006b.
Zentrale Begriffe. Konstrukte und Persönlichkeit. In M. Amelang, D.
Bartussek, G. Stemmler & D. Hagemann (eds). Differentielle Psychologie
und Persönlichkeitsforschung 6. Completely revised 6th edn. Stuttgart:
Kohlhammer. 45-48.
Ballard, M. 1997. Créativité et traduction. Target 9 (1): 85-110.
Bastin, G. 2003. Aventures et mésaventures de la créativité chez les débutants.
Meta 48 (3): 347-360.
Bayer-Hohenwarter, G. (in progress). Translatorische Denkflüssigkeit und
Flexibilität. Eine empirische Studie zur Entwicklung translatorischer
Kreativität als Komponente translatorischer Kompetenz. Graz: University
of Graz.
Brown, R. 1958. How shall a thing be called? Psychological Review 65: 14-21.
Csikszentmihalyi, M. 1997. Creativity. Flow and the Psychology of Discovery
and Invention. New York: Harper.
Dancette, J., Audet, L., & Jay-Rayon, L. 2007. Axes et critères de la créativité en
traduction. Meta 52 (1): 108-122.
Fillmore, C. J. 1976. Frames semantics and the nature of language. In S. R.
Harnad, H. D. Steklis & J. Lancaster (eds). Origins and Evolution of
Language and Speech (Annals of the New York Academy of Sciences 280).
New York: The New York Academy of Sciences. 20-32.
Fillmore, C. J. 1985. Frames and the semantics of understanding. Quaderni di
Semantica 6(2): 222-254.
Göpferich, S. 2008. Research Project TransComp: The Development of
Translation Competence. <http://gams.uni-graz.at/container:tc> [December
7, 2008]
Gruber, H. E. & Wallace, D. B. 1999. The case study method and evolving
systems approach for understanding unique creative people at work. In R. J.
Sternberg (ed.). Handbook of Creativity. Cambridge: Cambridge University
Press. 93-115.
Guilford, J. P. 1950. Creativity. American Psychologist 5: 444-454.
Holman, M. & Boase-Beier, J. 1999. Introduction: writing, rewriting and
translation through constraint to creativity. In J. Boase-Beier & M. Holman
(eds). The Practices of Literary Translation. Constraints and Creativity.
First published 1998. Manchester: St. Jerome. 1-17.
Ivir, V. 1998. Linguistic and communicative constraints on borrowing and literal
translation. In A. Beylard-Ozeroff, J. Králová & B. Moser-Mercer (eds).
Translator's Strategies and Creativity. Selected papers from the 9th
international conference on translation and interpreting, Prague,
September 1995. Amsterdam/Philadelphia: John Benjamins. 137-144.
Krings, H. P. 1986. Was in den Köpfen von Übersetzern vorgeht. Eine empirische
Untersuchung zur Struktur des Übersetzungsprozesses an fortgeschrittenen
Französischlernern. Tübingen: Narr.
Krings, H. P. 1988. Blick in die „Black Box‟ – Eine Fallstudie zum
Übersetzungsprozeß bei Berufsübersetzern. In R. Arntz (ed.). Textlinguistik
und Fachsprache. Akten des Internationalen übersetzungs-
wissenschaftlichen AILA-Symposions Hildesheim, 13.-16. April 1987.
Hildesheim: Olms. 393-411.
Krings, H. P. 2001. Repairing Texts: Empirical Investigations of Machine
Translation Post-editing Processes. In G. S. Koby (ed.); translated by G. S.
Koby, G. Shreve, K. Mischerikow & S. Litzer. Translation of the author‟s
Habilitationsschrift (1994). Kent: Kent State University Press.
Kußmaul, P. 2000a. Kreatives Übersetzen. (Studien zur Translation 10.)
Tübingen: Stauffenburg.
Kußmaul, P. 2000b. A cognitive framework for looking at creative mental
processes. In M. Olohan (ed.). Intercultural Faultlines: Textual and
Cognitive Aspects. Research Models in Translation Studies 1. Manchester:
St. Jerome. 57-71.
Kußmaul, P. 2000c. Types of creative translating. In A. Chesterman, N. G. San
Salvador & Y. Gambier (eds). Translation in Context. Selected Papers from
the EST Congress, Granada 1998. (Benjamins translation library 39)

Amsterdam: Benjamins. 117-126.
Kußmaul, P. 2005. Translation through visualization. Meta 50 (2): 378-391.
Lakoff, G. 1987/1990. Women, Fire, and Dangerous Things. What Categories
Reveal about the Mind. Chicago/London: University of Chicago Press.
Pellatt, V. 2006. The ASPRO model of creativity: assessing the creative handling
of the translation of „fat‟ and „old‟. In I. Kemble & C. O‟Sullivan (eds).
Proceedings of the Conference held on 12th November 2005 in Portsmouth.
Portsmouth: University of Portsmouth. 52-62.
Preiser, S. 1976. Kreativitätsforschung. Darmstadt: Wissenschaftliche Buchge-
sellschaft.
Rietzschel, E. F., Nijstad, B. A. & Stroebe, W. 2007. Relative accessibility of
domain knowledge and creativity: The effects of knowledge activation on
the quantity and originality of generated ideas. Journal of Experimental
Social Psychology 43(6): 933-946.
Rosch, E. 1977. Human categorization. In N. Warren (ed.). Advances in Cross-
cultural Psychology 1. London: Academic Press. 1-49.
Sternberg, R. J. & Lubart, T. I. 1999. The concept of creativity: prospects and
paradigms. In R. J. Sternberg (ed.). Handbook of Creativity. Cambridge:
Cambridge University Press. 3-15.
Torrance, E. P. 1988. The nature of creativity as manifest in its testing. In R. J.
Sternberg (ed.). The Nature of Creativity. Contemporary Psychological
Perspectives. Cambridge: Cambridge University Press. 43-75.
Ward, T. B., Smith, S. M. & Finke, R. A. 1999. Creative cognition. In R. J.
Sternberg (ed.). Handbook of Creativity. Cambridge: Cambridge University
Press. 189-212.
Weisberg, R. W. 2006. Modes of expertise in creative thinking: evidence from
case studies. In K. A. Ericsson, N. Charness, P. J. Feltovich & R. R:
Hoffman (eds). The Cambridge Handbook of Expertise and Expert
Performance. Cambridge: Cambridge University Press. 761-787.
Wittgenstein, L. 1958/1977. Philosophische Untersuchungen. Frankfurt:
Suhrkamp.
Indicators of text complexity
Kristian T.H. Jensen
Abstract
This paper reports on the preparatory steps that were taken prior to
conducting a number of translation experiments in which three texts of
varying levels of complexity will serve as source text stimuli to a group of
subjects. It explains how the relative differences in complexity can be
measured using a number of objective criteria such as readability indices,
word frequency and non-literalness. It is thought that a set of objective
indicators can to some extent account for the degree of difficulty
experienced by translators when translating a text. While we do not wish to
postulate that complexity (the objective notion) and difficulty (the
subjective notion) can be equated, it is thought that these objective
measures can help us gauge the degree of difficulty of some types of text
and thus help us in finding texts for experimental purposes. The aim of this
paper is thus to describe and discuss the potential of these indicators for
predicting text difficulty on the basis of illustrative examples from the texts
chosen.
1. Introduction
The present project aims at examining how variation in complexity in texts

for translation affects the gaze behaviour of professional translators and
novice translators. For this purpose, three texts (texts A, B, C; see
Appendix) were selected and manipulated to serve as translation stimuli.
In order to find out how the texts can be measured and compared in
terms of text complexity, we shall examine to what extent (a) readability
indices, (b) word frequency, and (c) non-literalness reflect complexity.
While readability indices have traditionally focused on difficulties related
to certain aspects of text comprehension only (Nation 2001: 161), this
paper suggests how readability indices can be used to assess the relative
62 Kristian T.H. Jensen
amount of both production effort and comprehension effort needed during a

translation process. The second indicator examined is word frequency. The
frequency with which a word appears in the real world as reflected through
corpora such as the British National Corpus is assumed to mirror the
amount of effort that a translator will have to put into the processing of it
(Read 2000: 160). Broadly, the less frequently a word appears in a corpus,
the more effort is generally needed to translate it. The final indicator is
non-literalness. In this paper, non-literalness is defined as the presence of
idioms, metaphors and metonyms in the experimental texts. Non-literal
expressions are expected to involve increased processing effort since
default-mode literal interpretation of them would be erroneous (Jakobsen et
al. 2007: 219-222). Consequently, a higher number of non-literal
expressions in a text is claimed to increase the level of complexity in a text.
2. Complexity and difficulty
The notion of a text‟s level of complexity should not be mistaken for a

text‟s level of difficulty. Though a linguistically complex text is often
experienced as a difficult text, the notions of complexity and difficulty do
not overlap completely. They express, respectively, an objective and a
subjective approach to text complexity. A translator with much experience
in technical translation may have little difficulty in translating technical
documentation for a particle accelerator, while a translator who mainly
works in the field of legal translation may consider such a text difficult.
Similarly, a translator who specialises in legal translation may experience
few problems when translating an affidavit, while the translator who is well
versed in translating annual reports may find this sort of translation
particularly challenging. The ease with which technical documentation,
affidavits and annual reports are translated may therefore differ hugely,
depending on the translator‟s area of specialisation.
The above examples illustrate that difficulty and complexity are two
different concepts which should not be confused with each other. The
notion of difficulty is subjective and elusive, and the perception of a text‟s
level of difficulty can therefore vary from one translator to another. By
contrast, the notion of text complexity is a more objective approximation to
relative text difficulty which can be based on one or more factual criteria.
Indicators of text complexity 63
While it is likely that text complexity and text difficulty do not coincide in
domain-specific language, a case can perhaps be made for saying that the
two will be relatively similar in general purpose texts. Here only general
knowledge is required on the part of the translator, and thus one can be
more certain that the problems encountered by the translator are related to
text complexity factors such as readability, word frequency and non-
literalness. For the purposes of this paper, we shall assume that complexity
in general-purpose texts is a strong indicator of difficulty even though this
will have to be put to the test at a later stage.
3. Indicators of text complexity
As stated above, three quantitative indicators were selected to compare the

levels of complexity in three experimental texts: readability indices,
calculations of word frequency and calculations of the number of non-
literal expressions, i.e. idioms, metaphors and metonyms. In addition to
selecting texts which were all general-purpose texts, all belonging to the
same genre (in this case, online newspaper articles), care was taken to
ensure that the texts were also comparable with respect to total length (837,
846, 856 characters, respectively) and length of the headlines (41, 44, 44
characters, respectively) so as to provide a uniform framework for
comparison of the selected indicators.
3.1 Readability indices
Assessing a text‟s level of readability, i.e. the ease with which the text is
likely to be read and comprehended, has attracted the attention of many
scholars. As early as 1935, Gray and Leary studied the effects of text
complexity on comprehension (Gray and Leary 1935), and in 1963 Klare
suggested that words and word frequencies are the most important criteria
in measuring the readability of a given text (1963: 18-19). More recently,
Nation (2001: 161-162) has pointed out that readability formulas mostly
focus on what is easily measurable, i.e. word length and sentence length.
Quoting Carrell (1987), he notes that other factors such as prior knowledge,
motivation, rhetorical structure etc. are valuable for assessing text
comprehensibility, i.e. readability, and could therefore successfully be
included as readability assessment criteria (Nation 2001: 161). These
factors cannot easily be included into an algorithm as they are difficult to

quantify, so most readability formulas are based on counts of syllables,
words and sentences as illustrated below.
Seven indices were used to assess the readability of the texts to be
used in the translation experiment: the Automated Readability Index (ARI),
the Flesch-Kincaid index, the Coleman-Liau index, the Gunning Fog index,
the SMOG index, the Flesch Reading Ease Score index and LIX. The first
five indices return the U.S. grade level that the reader must have completed
in order to fully understand the text. The last two indices return numerical
scores. Table 1 lists all seven indices and their individual formulas:
Table 1. Reading index formulas (obtained from Editcentral.com1 and

Bedreword.dk2)
Index Formula
ARI 4.71*characters/words+0.5*words/sentences-21.43
Flesch-Kincaid 11.8*syllables/words+0.39*words/sentences-15.59
Coleman-Liau 5.89*characters/words-0.3*sentences/(100*words)-15.8
Gunning Fog 0.4*(words/sentences+100*((words >= 3 syllables)/words))
SMOG square root of (((words >= 3 syllables)/sentence)*30) + 3
Flesch Reading Ease 206.835-84.6*syllables/words-1.015*words/sentences
LIX words/sentences+100*(words >= 6 characters)/words
Each of the seven formulas calculates the proposed complexity of text

differently. In general, countable properties such as characters, syllables,
words and/or sentences combined with one or more constants are used as a
basis for calculating a text‟s specific level of complexity.
All five U.S. grade level indices mentioned above reveal a
progression from text A through text B to text C. Text A is by far the least
complex text: to successfully comprehend the text a reader would have to
have completed 7.8 years of schooling. Text B is more complex than text
A, requiring 12.5 years of schooling for successful comprehension. Finally,
1
Editcentral is a website that returns the complexity scores of a text which is entered
into an online query box by the user. The website returns complexity scores for all
indices listed under Table 1 except for LIX.
2
Bedreword is a website from which add-in programs for Microsoft Word can be
downloaded. The BedreWord/Lixberegning add-in program calculates complexity
scores based on the LIX formula.
text C necessitates no less than 17.3 years of schooling in order to be

comprehended successfully.3
Figure 1. U.S. grade level indices scores
It is interesting to observe the strikingly linear progression in the number of

years of schooling required for successful comprehension of the texts by all
scoring methods: on average, text B would require 60 percent more
schooling than text A, and text C would require 122 percent more schooling
than text A. However, the relative uniformity of results by the different
scoring methods can be attributed to the fact that the calculations of all
indices are based on many of the same countable properties.
The Flesch Reading Ease Score index (FRES) (Editcentral.com) and
LIX4 (Bedreword.dk) both return numerical scores. In the FRES index,
higher scores (up to 100) indicate that the text is easier to read, while lower
scores (as low as 0) indicate that the material is more difficult to read. As
appears from Figure 2, here too there is a progressive increase in the
relative complexity of the texts. Text A scores 79.8, text B 59.4, while the
score for the most difficult text is 37.7.
3
7.8, 12.5, 17.3 years of schooling are average values for texts A, B and C,
respectively, based on all five U.S. grade level indices.
4
LIX is a Swedish abbreviation for läsbarhetsindex (i.e. readability index).
Figure 2. Flesch Reading Ease score
Björnsson divides the LIX-scale into five categories of difficulty which he

calls lix standards (Björnsson 1983): very easy texts (<25), easy texts (25-
35), average texts (35-45), difficult texts (45-55), very difficult texts (>55).
In line with the six previously mentioned indices, the LIX-scale also
suggests that there is a progressive increase in the relative complexity of
the texts, as illustrated below:
Figure 3. LIX score
Text A would thus be labelled easy, text B would be labelled average or

difficult while text C would be labelled very difficult.
These readability indices cannot give us conclusive evidence of how

difficult a translator perceives a text to be or how much translation effort is
needed to translate the text. The indices have been developed with text
comprehension in mind. However, tentatively, one could hypothesise that
the readability indices can also give us some idea of the amount of effort
that will be invested in the production of a target text, assuming that text
comprehension and text production (and memory) are inseparable parts of
the translation process (Gile 1995: 162-169). If it can be shown to be
possible to predict translation difficulty from source text readability and
complexity, we would have a tool that could give us an estimate of the
likely processing effort involved in the translation of a particular text. A
pilot study will test this hypothesis. In summary, the seven readability
indices described are all based on different methods of calculating the
complexity of a given text. All use counts of words, syllables or characters
to estimate the level of readability, but no two formulas are exactly alike.
Nevertheless, despite the different methods of calculation, they all indicate
that there is a progression in the levels of complexity of the three texts, and
therefore, it is assumed, also in the level of difficulty as experienced by the
reader/translator. Text A is expected to be perceived as the easiest text,
while text C will be perceived as the most difficult text, with text B
somewhere between texts A and C.
3.1.1 Limitations of readability measurements

Not surprisingly, there are limitations to the applicability of readability
indices as assessment tools of complexity. Noam Chomsky‟s famous
nonsensical sentence Colorless green ideas sleep furiously (1957: 15),
which was used to illustrate the need for structured models of grammar,
would score lower by all seven readability indices (i.e. Chomsky‟s sentence
is less complex and therefore requires less cognitive effort in order to be
understood) than the following newspaper headline: Protesters storm
president’s compound; most readers would nonetheless find that
interpretation of Chomsky‟s sentence requires far more effort than the
newspaper headline, which is not at all surprising, given that Chomsky‟s
sentence is meaningless. Readability indices reflect complexity scores that
are based only on the countable properties discussed earlier, and both
sentences are therefore automatically assumed to be equally grammatically
acceptable and meaningful. The indices do not distinguish between sense

and nonsense, i.e. the semantic acceptability of a text is ignored. The
example above therefore clearly illustrates that readability scores serve
poorly as the sole indicator of text complexity in instances where language
norms are violated.
Readability indices are only sensitive to a text‟s level of complexity
based on its surface structure and they are therefore unable to interpret
meaning-related properties of an expression, referred to below as the
internal lexical properties of sentences or texts. To give an example, the
words sesquipedalian5 and supercomputers are identical with respect to
their numbers of syllables and characters, variables which are often applied
when calculating readability. Most people would nevertheless experience
far more difficulties with comprehending the first word than the second,
which is related to the fact that far fewer people are familiar with it. The
consideration of word frequency (word familiarity) and the density of non-
literal expressions is an attempt to compensate for the inability of
readability indices to capture text complexity caused by infrequent,
unfamiliar lexical items or by the occurrence of items requiring non-literal
interpretation. Below, these additional indicators of text complexity are
discussed in more detail.
Another weakness of using readability as an instrument for
measuring degrees of difficulty is the very nature of the tool. We saw that
Editcentral returned scores for the number of years of schooling needed to
understand the text without difficulty (Grades 1-12 corresponding to 6-18
years of schooling), but the population group we are dealing with here
consists of experienced translators who a have a university degree, and as
such they are well beyond Grade 12. While the readability scores will be a
good indicator of how successfully a schoolchild will cope with
understanding a text, it can be assumed that very few of the words and
sentences which pose problems to Grade 1-12 pupils will be at all difficult
for a translator. In other words, it is perhaps not all that crucial whether one
needs 7.8, 12.5 or 17.3 years of schooling to read a text with ease, since it
can be argued that once reading of longer words and sentences has been
mastered, it is no longer relevant to know how long it took to reach this
5
Merriam-Webster: sesquipedalian: (1) having many syllables; (2) given to or
characterised by the use of long words.
stage. Unless it can be shown that the texts requiring longer schooling also
result in more processing effort on the part of the translator, readability
indices may turn out to be less useful than they first might appear.
3.2 Word frequency
One characteristic that was considered a potential indicator of the internal

lexical levels of complexity of the three texts was word frequency. Word
frequency is a powerful measurement tool for estimating the relative
amount of effort that needs to be invested in the cognitive processing of a
particular word. Based on the general assumption that the more frequently
a word occurs in a language, the more likely it is to be known to the
recipient (Read 2000: 160), we can hazard the suggestion that the more
frequently a word occurs, the less cognitive effort is needed to process it,
and conversely, that the less frequently a word occurs, the more cognitive
effort is needed to process it. Figure 4 illustrates the distribution of the
words that appear in the three texts into K1 and K2-K10 frequency bands.
Words that belong to the K1 frequency band are most common (1-1000
most frequent words), while words that belong to the K2-K10 frequency
bands are less common (1001-10000 most frequent words) (Cobb 2008;
Heatley & Nation 1994). Table 2 lists the less frequent words in the three
texts, i.e. words that belong to the K2-K10 frequency bands.
Figure 4. Word frequency scores of texts A, B, C

Table 2. K2-K10 words in texts A, B and C

Text A Text B Text C
Guilty, imprisoned, Ahead, alarming, analysts, China, Chinese, criminal,
medicine, murder, nurse, below, climbed, cough, embarrass, emphasizing,
patients, sentences, customers, fuel, nurses, extensive, humanity,
strangely, trial, victims, professionals, racing, international, maintains,
weak, burden, motive salary, struggles, suffered, negative, oil, sought,
warned, supermarkets, suffering, wake,
inflation, sector, withdrawal, bulk, flaring,
escalating, insistence, soar, gesture, rattle, halt,
soaring, healthcare ongoing, protest,
Olympics, fallout, Sudan,
Sudanese, atrocities
Nearly 90 percent of all words that appear in text A belong to the K1

frequency band while the remainder are less frequent. In text B the
distribution is 82 percent K1 words and 18 percent K2-K10 words. In text
C, only 72 percent of all words belong to the K1 frequency band, while
more than a quarter of the words, viz. 28 per cent, belong to the K2-K10
frequency bands.6 Assuming that a relationship between word frequency
and processing effort exists, cf. Read (2000) above, text C would involve
relatively more processing effort than text B, and text A would require least
effort of all three texts.
3.2.1 Limitations of word-frequency measurements

Just as in the case of readability indices, word-frequency measurements
also have their limitations. Although frequency of occurrence is often
related to degree of familiarity with words, this is far from always the case.
Whether a word is known or unknown will obviously vary from translator
to translator. But just because a word is less frequent, it does not mean that
it is necessarily difficult to understand or translate. Among the K2-K10
words, we find below, China, Chinese, international, murder, to name but a
few, none of which can be assumed to be particularly difficult to translate.
Although the word murder in the sentence “Yesterday, he was found guilty
of four counts of murder following a long trial” (Text A) belongs to the K2-
K10 frequency band and the word counts does not, it is probable that a
6
The frequency bands are based on the British National Corpus.
translator translating into Danish would have more difficulty translating the
latter than the former.
3.3 Non-literalness
3.3.1 Idioms
Another indicator that could mirror the internal lexical properties of a text
is the occurrence of idioms. Idioms are conventionalised multi-word
expressions that are often non-literal and indivisible (Fernando 1996: 30),
which means that they rarely permit interpretation based on the idiom
constituents alone and that they do not permit constituent variation. The
meaning of this type of multi-word expression must be known to the hearer
or reader as a whole in order for them to interpret it correctly (Jakobsen et
al. 2007: 218-219). Research has shown that translating idioms is more
time-consuming than translating non-idiom containing text (Jensen 2007:
67; Jakobsen et al. 2007: 235). For example, the idiom kiss goodbye to
(sth) lacks a counterpart similar in form in Danish, and must therefore be
translated either into another idiom similar in meaning but lexically
dissimilar, e.g. vinke farvel til (ngt),7 or by paraphrase, a process which
requires increased time consumption. This increased time consumption is
interpreted as indicating that more cognitive effort is involved in this type
of translation. Further, the increase in the amount of time invested in idiom
translation relates to the fact that the „default‟ mode of interpretation is
compositionally motivated; evidence suggests that sentences are first
interpreted literally and then, when this strategy fails, non-literally (Gibbs
1994 and Jensen 2007: 69). The non-compositional and non-literal nature
of idioms may cause interpretation difficulties and increase the amount of
cognitive effort needed for its translation. Such additional effort is believed
to be one of the main contributors to the increase in time consumption
associated with idiom translation. Another important contributor is believed
to be translators‟ (and interpreters‟) difficulties in finding an appropriate
and satisfactory rendering of the original idiom. Many translators spend
vast amounts of time looking for a target-text idiom as a translation of the
source-text idiom – even if none exists (Jensen 2007: 64). Conversely,
translators sometimes resort to time-consuming cognate avoidance, fearing
7
Back-translation: wave goodbye to (something).
that direct transfer of the constituents that make up the idiom in language A
into language B is erroneous (Jakobsen et al. 2007: 234). In line with the
above observations, the presence of idioms in the experimental texts was
considered an important contributing factor to establishing the level of
complexity of a text. Like word frequency, idiom density, is taken to
indicate the ease (or difficulty) with which particular lexical items in a text
are expected to be processed.
Figure 5 below summarises how many idioms there are in each of
the three experimental texts, and Table 3 lists which idioms are present in
each text. As appears, idiom density is highest in text C, which contains
four idioms, while texts A and B contain one and two idioms, respectively.
Figure 5. Idiom density of texts A, B, C
Table 3. Idioms in texts A, B, C

Only the awareness of Families have to cough up an Spielberg shows Beijing red
other hospital staff put a extra £1,300 card
stop to him and to the
killings
Prices are racing ahead of Spielberg pulled out of the
salary increases Olympics
His withdrawal comes in
the wake of fighting
Khartoum bears the bulk of
the responsibility
Figure 5 illustrates that there is a progression in the number of idioms in

the texts indicating an increase in the levels of complexity. However,
caution should be exercised when using idiom density as a measurement of
text complexity because some idioms appear to be more challenging in
terms of processing effort, viz. idioms that can be translated only by means
of a paraphrase or by a target text idiom that is similar in meaning but
different in form. Target language idioms that are similar in both meaning
and form are in fact processed considerably faster than other idioms during
the translation process (Jensen 2007: 67). For instance, “Families have to
cough up an extra £1,300” (Text B) would be relatively easy to translate
into Danish as there is a similar (though not quite identical idiom in this
language (“hoste op med”). On the other hand, the sentence “Only the
awareness of other hospital staff put a stop to him and to the killings” (Text
A) would be a difficult task. Even though Danish has an idiom that is
similar in form (“sætte en stopper for”), it is normally only used in
connection with non-animate objects and the sentence would therefore
require that the idiom was translated by a non-idiom (“standse” = “stop”).
Since the translation of idioms is often by no means a straightforward
matter, using idiom density as an indicator of text complexity and for
predicting translation problems only provides a rough suggestion.
3.3.2 Metonyms and metaphors

Metonyms and metaphors can also give the experimenter an idea of the
potential effort that will have to be invested in processing these two types
of expressions. Unlike idioms, which are compositionally polymorphemic,
metonyms and metaphors can be made up of just a single free morpheme.
Similar to idioms, however, both types of expression are non-literal.
Assuming that a relationship exists between non-literalness and processing
effort, in line with the observations made in the idiom discussion above,
both types of expressions are subject to relative increased cognitive effort
spent on their comprehension and production.
Figure 6. Metonym and metaphor density of texts A, B, C
The figure above shows that text A is by far the least complex text in terms
of the metonym/metaphor density indicator. This is not surprising, since
there are no metonyms or metaphors in this text. Text B contains six
metaphors and one metonym, and text C contains five metaphors and six
metonyms, totalling eleven occurrences of these types of expression. A
progression supporting the patterns reflected by the other indicators
described in this paper is clearly identifiable. As with idioms, metonym and
metaphor density can only be said to be rough indicators of complexity.
Table 4 below lists which metonyms and metaphors are present in each of
the experimental texts.
Table 4. Metaphors and metonyms in texts A, B, C8

(None) Hit with increase (mp) Beijing (mt)
Prices/bills soar (mp) Rattle the Chinese government (mp)
Prices (…) have climbed (mp) China (mt)
Cut interest rates (mp) Sudan (mt)
Struggles to keep inflation … Fighting flaring up (mp)
under control (mp)
Escalating prices (mp) Has sought to halt (mp)
Government (mt) Negative fallout (mp)
Close ties
Close links
Government (mt)
Khartoum (mt)
8
Mp.: metaphor. Mt.: metonym.
Like our other indicators, metonym and metaphor density has to be treated
with caution as possible indicators of translation difficulty. None of the five
metonyms (i.e. Beijing, China, Sudan, Government, Khartoum) is difficult
to translate. As far as the metaphors are concerned, the same reservations
apply as in the case of idioms: some will be easy to transfer to the target
language, others will require more cognitive effort. The amount of
processing will vary from language to language. Again, further testing is
needed.
Summary of text complexity indicators
As Table 5 illustrates, all indicators used in this paper point in the same
direction: text A is the least complex text, text C is the most complex, and
text B lies somewhere in between. Pursuing the idea that a relationship
does in fact exist between text complexity and the perceived difficulty of a
text, text A is the easiest text, text B is more difficult than text A, but less
difficult than text C, which is the most difficult text of the three
experimental texts.
Table 5. Summary of text complexity indicators
Level of complexity Least In- Most

between
Indicator
ARI Text A Text B Text C
Flesch-Kincaid Text A Text B Text C
Coleman-Liau Text A Text B Text C
Gunning Fog Text A Text B Text C
SMOG Text A Text B Text C
Flesch-Reading Ease Score Text A Text B Text C
LIX Text A Text B Text C
Word frequency Text A Text B Text C
Idiom density Text A Text B Text C
Metonymy & metaphor density Text A Text B Text C
The list of measurement tools that have been discussed in this paper as
indicators of complexity in texts for translation and therefore as indicative
of subjective (i.e. perceived) text difficulty is by no means exhaustive.
Other measurement tools could be used. For example, the introduction of

false cognates into the texts could quite well prompt translators to resort to
other and more time-consuming translation strategies, which could be
interpreted intuitively as an indicator of increased complexity. Similarly,
the ratio of passive to active verbs may also be considered indicative of a
text‟s level of complexity, since verbs are employed differently in different
languages.
4. Applicability of the indicators
The question remains: how useful is this approach for indicating an

individual text‟s level of difficulty based on the text complexity indicators
discussed so far? It would be bold and naïve to claim that these indicators
can be used straightforwardly to predict the exact amount of cognitive
effort that is needed to translate one specific text by every translator in the
world. There are too many uncontrollable variables, too many different
texts, and too many minds to devise an instrument that will reveal text
complexity and text difficulty correlations within a single text reliably. But
the set of indicators prove to be useful when comparing the relative
difference in the levels of complexity of two or more texts. If one employs
several indicators that all point in the same direction, it is reasonable to
assume that they can be used as a reliable measure. In the present
experiment three different indicators are used, all leading to the same
result, namely that text A is less complex than text B and that text C is more
complex than text B. It would therefore appear that they constitute a
powerful instrument for assessing relative differences in levels of
complexity and, presumably, also differences in relative text difficulty. We
cannot, however, escape the fact that text complexity measurements at this
stage can only give us a rough indication of how difficult a text will be
perceived by the translator. The hypothetical relationship between the
objective notion of text complexity and the subjective notion of text
difficulty will require further research and more testing.
5. Conclusions
The potential indicators of complexity in texts for translation, which

consisted of seven readability indices, calculations of word frequency and
calculations of the number of idioms and the number of metaphors and
metonyms in the texts, were applied to the three texts, and every indicator
pointed in the same general direction. The three types of indicators all
conclude that there is a difference between the levels of complexity of the
three texts. What is more, the three general indicators show uniformly that
text A is the least complex text, text C is the most complex text while text B
is in between texts A and C in terms of relative complexity.
The set of tools selected to serve as a basis for comparing the relative
difficulty between the experimental texts could be expanded. Both
qualitative and other quantitative tools could have been added to the list of
potential indicators.
As we have repeatedly noted, a text‟s level of difficulty and level of
complexity cannot be assumed to be synonymous, but it is argued that we
can go some way towards predicting the probable degree of difficulty of a
text by employing a battery of objective measures. The extent to which the
two concepts overlap will obviously require further testing. Questions to be
examined include: (1) Can readability indices, which – it should be
remembered – have been developed for pupils in Grades 1-12, reveal
anything about the degree of difficulty a translator will experience in
translating a text? More specifically, does a text such as Text A, which
requires 7.8 years of schooling, involve less processing effort than Text C,
which requires 17.3 years of schooling? Or is the difference evened out by
the time one has obtained a university degree? (2) Do readability indices
merely predict ease of comprehension or do they, implicitly, also reveal
anything about ease of translation? (3) To what extent is word frequency
comparable to word familiarity? (4) Are idioms and metaphors/metonyms
on the whole more difficult to translate than non-literal expressions? There
are various ways of investigating the issues above. The following
procedures could be carried out. One could:
look at the amount of time needed to complete the translation. If, for
instance, it takes longer to complete Text C (which on the basis of
the objective measures above was found to be the most complex of

the three passages) than Texts A and B, this would show that it was
also experienced as being more difficult.
establish the number and length of writing pauses in connection with
the objectively complex segments as compared with less complex
segments;
conduct retrospective interviews in which translators are asked about
the segments they experienced as difficult;
assess the quality of the translation; if solutions are less adequate for
complex units, one can assume they were experienced as more
difficult;
investigate the degree of automaticity; if more different solutions are
considered in the case of units that are objectively complex before a
final decision is made, this would indicate that the translator finds
the segment more difficult.
If a large number of translators exhibit the same patterns with respect to the
above parameters, it should be possible to assess the value and validity of
the objective measures described in this paper, and there should
consequently be a sound basis for concluding that the difficulty of texts can
be assessed using objective indices.
References
Bedreword.dk [accessed 19 March 2009 from http://www.bedreword.dk].

Björnsson, C.H. 1983. Readability of newspapers in 11 languages. Reading
Research Quarterly 18 (4): 484.
Carrell, P. L. 1987. Readability in ESL. Reading in a Foreign Language 4 (1):
21-40.
Chomsky, N. 1957 Syntactic Structures. The Hague/Paris: Mouton.
Cobb,T. 2008. Web Vocabprofile [accessed 19 March 2009 from
http://www.lextutor.ca/vp/ ], an adaptation of Heatley & Nation‟s (1994)
Range.
Editcentral.com [accessed 19 March 2009 from http://www.editcentral.com].
Fernando, C. 1996. Idioms and Idiomaticity. Oxford: Oxford University Press.
Gibbs, R. 1994. Figurative thought and language. In M. A. Gernsbacher (ed.).
Handbook of Psycholinguistics. San Diego: Academic Press. 411-446.
Gile, D. 1995. Basic Concepts and Models for Interpreter and Translator
Training. Amsterdam/Philadelphia: John Benjamins.
Gray W.S., & Leary, B.E. (1935) What Makes a Book Readable. Chicago:
Chicago University Press.
Heatley, A. and Nation, P. 1994. Range. Victoria University of Wellington, NZ.
[Computer program, available at http://www.vuw.ac.nz/lals/.]
Jakobsen A.L., Jensen, K.T.H. & Mees, I.M. 2007. Comparing modalities:
idioms as a case in point. In F. Pöchhacker, A. L. Jakobsen & I. M. Mees.
(eds). Interpreting Studies and Beyond. Copenhagen: Samfundslitteratur
Press. 217-249.
Jensen, K.T.H. 2007. Idiom Translation from English into Danish. An Empirical
Study of Cognitive Effort and Translation Strategies in Idiom Translation.
MA thesis, Copenhagen Business School.
Klare, G. R. 1963. The Measurement of Readability. Iowa: Iowa State University
Press.
Nation, I. S. P. 2001. Learning Vocabulary in Another Language. Cambridge:
Cambridge University Press.
Read, J. 2000. Assessing Vocabulary. Cambridge: Cambridge University Press.
Appendix
(Text A) Killer nurse receives four life sentences

Source: The Independent (4 March 2008)
1 Hospital nurse Colin Norris was imprisoned for life today for the killing of
2 four of his patients. 32 year old Norris from Glasgow killed the four women
3 in 2002 by giving them large amounts of sleeping medicine. Yesterday, he
4 was found guilty of four counts of murder following a long trial. He was
5 given four life sentences, one for each of the killings. He will have to serve
6 at least 30 years. Police officer Chris Gregg said that Norris had been acting
7 strangely around the hospital. Only the awareness of other hospital staff put
8 a stop to him and to the killings. The police have learned that the motive for
9 the killings was that Norris disliked working with old people. All of his
10 victims were old weak women with heart problems. All of them could be
11 considered a burden to hospital staff.
Number of characters with spaces: 837

Length of headline in characters with spaces: 41
(Text B) Families hit with increase in cost of living

Source: The Times on 12 February 2008
1 British families have to cough up an extra £1,300 a year as food and fuel
2 prices soar at their fastest rate in 17 years. Prices in supermarkets have
3 climbed at an alarming rate over the past year. Analysts have warned that
4 prices will increase further still, making it hard for the Bank of England to
5 cut interest rates as it struggles to keep inflation and the economy under
6 control. To make matters worse, escalating prices are racing ahead of salary
7 increases, especially those of nurses and other healthcare professionals, who
8 have suffered from the government‟s insistence that those in the public
9 sector have to receive below-inflation salary increases. In addition to fuel
10 and food, electricity bills are also soaring. Five out of the six largest
11 suppliers have increased their customers' bills.

(Text C) Spielberg shows Beijing red card over Darfur

Source: The Daily Telegraph on 13 February 2008
1 In a gesture sure to rattle the Chinese Government, Steven Spielberg pulled

2 out of the Beijing Olympics to protest against China‟s backing for Sudan‟s
3 policy in Darfur. His withdrawal comes in the wake of fighting flaring up
4 again in Darfur and is set to embarrass China, which has sought to halt the
5 negative fallout from having close ties to the Sudanese government. China,
6 which has extensive investments in the Sudanese oil industry, maintains
7 close links with the Government, which includes one minister charged with
8 crimes against humanity by the International Criminal Court in The Hague.
9 Although emphasizing that Khartoum bears the bulk of the responsibility for
10 these ongoing atrocities, Spielberg maintains that the international
11 community, and particularly China, should do more to end the suffering.

More ways to explore the translating mind:
collaborative translation protocols
Nataša Pavlović
Abstract
Investigation of translation processes has intensified over the past two

decades largely due to the application of the introspective verbal reporting
method known as think aloud. In this paper, a less widely used method of
research into translation processes is examined: the collaborative
translation protocol (CTP). References are also made to integrated
problem and decision reporting (IPDR) and choice network analysis
(CNA). The paper reviews the main issues concerning CTP, reporting on
the author’s experiences in using the method to investigate translation
processes.
1 Introduction
There are several reasons why researchers might be interested in studying

translation processes. The basic desire to find out what goes on in the
translator‟s mind – to understand the phenomenon of translation, to be able
to describe, explain and perhaps even predict the processes of translators –
is certainly the main drive behind any research. Some of the more applied
goals of particular studies might include understanding the processes of
professional, or expert, translators with a view to identifying the features
that constitute expert competence in translation. The findings of such
studies might then be applied by practitioners – translators and translation
teachers – in order to improve the practice of translation. Other researchers
might decide to examine the opposite end of the spectrum and focus on the
processes of translation students or novice translators. By identifying the
problems students and novices encounter and the ways they try to deal with
82 Nataša Pavlović
those problems, researchers hope to shed some light on the acquisition of

translation competence and ultimately help improve translator education.
The experiences reported on in this paper are based on a project that
falls into this last category. I conducted a series of studies between 2005
and 2007 as part of my PhD project, involving novice translators – students
from the University of Zagreb, Croatia, who had recently passed their final-
year exam in translation. The aim of the study was to describe and compare
their translation processes in two directions – into and out of the subjects‟
first language (Croatian), as the latter direction is important for translators
working with a language of limited diffusion. The processes were
compared in terms of problems, solutions, resources, and decision-making
(actions/interactions and verbalizations). More details on the pilot study
can be found in Pavlović (2005), while the main study, including the
control experiments, is reported on in Pavlović (2007).
As in any research, the choice of methodology to be used in
examining translation processes depends on the aims of the project.
Translation processes are notoriously difficult to study because they are
highly complex cognitive (and social) phenomena, and unfortunately no
research method provides direct access into the “black box” of the
translator‟s mind. Nevertheless, translation process researchers have
adopted methods from other disciplines, and in some cases devised their
own methodology, which allows them to catch a glimpse, albeit indirect, of
the translating mind. While arousing curiosity for centuries, translation
processes have only become the subject of systematic empirical studies
over the past two decades. This development has largely been due to the
application of the introspective verbal reporting method known as think
aloud and the resulting protocols (TAPs). The history, achievements and
limitations of research employing TAPs have been discussed at length
elsewhere (see especially Jääskeläinen 2002, Bernardini 2001 and
Tirkkonen-Condit 2002; a summary of the pros and cons can be found in
Pavlović 2007: 39-45 and Göpferich 2008: 22-32). From the early studies
using TAPs in the 1980s (e.g. Krings 1986, 1987, 1988) to today‟s projects
that make use of keystroke logging programs such as Translog (2006), eye-
tracking (e.g. O‟Brien 2006, Eye-to-IT project 2007) and even EEG
monitoring, translation process researchers have used a variety of methods,
Collaborative translation protocols 83
discussed particularly in several dedicated volumes (e.g. Hansen 1999,

2002; Tirkkonen-Condit and Jääskeläinen 2000; Alves 2003).
In this paper, a less widely used method of research into translation
processes is discussed: collaborative translation and its resulting protocols
(CTPs). References are also made to the accompanying pre- and post-
process questionnaires, as well as to integrated problem and decision
reporting (IPDR) and choice network analysis (CNA). The paper reviews
the main issues concerning CTPs, reporting on the author‟s experiences
from several studies in which the protocols were used.
2 Collaborative translation protocols
The think-aloud method (and its resulting protocols, also known as

concurrent verbal reports) is a way of data elicitation in which individual
subjects are asked to think aloud, i.e. to verbalize their thoughts
concurrently with cognitive processing involved in the primary task
(Ericsson and Simon 1984/1993: xiii). As an off-shoot of this research
method, translation tasks involving more than one person have also been
used, producing joint translation protocols or, in the case of pairs working
together on a task, dialogue protocols. Recently Pavlović (2007) proposed
the term “collaborative translation protocols” for protocols obtained from
collaborative translation tasks. Collaborative translation tasks are tasks in
which a pair or group of people translate the same source text together,
basing their decisions on mutual consensus. In such a task, the construction
of the source text meaning and the emergence of the target text are a result
of individual cognitive processing, as well as the interaction among the
members of the group.
2.1 Some advantages of CTPs over other methods
2.1.1 Naturalness and stress reduction

Most advocates of CTPs point out that they are elicited in “a more natural
situation [than TAPs] since there is a real partner to work with and one
does not talk only to oneself” (Kussmaul 1991: 91-92). House (2000: 159)
argues in favor of what she calls “dialogic think-aloud tasks,” in which
subjects might engage in “more „natural‟, less strained and less pressured
introspective exercises that resemble „real life‟ activities much more than
the laboratory-type individual thinking-aloud practices.” The need to talk
arises not from the instructions given by the researcher (who need not be
present at all), but from the collaborative endeavor itself: “here is the need
to explain and justify one‟s translation, to make suggestions for
improvement, to ask for advice and criticism, all of which are features of
natural discourse” (Kussmaul 1991: 91-92). Or, as Séguinot (1996: 88)
says,
in a standard protocol analysis subjects are constrained to think, but not justify
their thinking. In the natural discourse situation where both subjects were
responsible for the task, the translation was negotiated, sometimes with overt
reasoning.
She believes that these rationalizations do not invalidate the approach as

they arise naturally from a real-life task, and are not “a construct of the
experimental situation.” According to Barbosa and Neiva (2003: 152), the
dialogue protocol, “owing to its very interactive nature,” compels the
subjects to “express, comment on and even justify their strategies in the
process of negotiating solutions for problems without the need for external
intervention or prior training in the think-aloud technique.”
While it may be true that for experiments involving collaborative
tasks the subjects need not be trained in any research-related technique,
some experience in collaborative translation might be desirable. The
success of a collaborative session is not a given, especially considering the
issues related to group dynamics, which are discussed in more detail below.
Having two or more people who have never before worked together take
part in an experimental situation involving collaboration may produce
wonderful results just as easily as end in a complete disaster (or anything
between the two extremes). A definite advantage of Séguinot‟s study is that
she was able to find people who regularly work together. In a similar vein,
the subjects who took part in the studies reported on in this paper had
worked together on collaborative tasks for some time before the
experiments, in their translation classes. As a consequence, they were used
to the situation and no further training was required. For them, the
experiments were a natural occurrence (insofar as any experiment can be
„natural‟). Most of them report a great degree of enjoyment in the tasks:
4.55 in the pilot and 4.7 in the main study, on a scale of 1 (“not at all”) to 5
(“very much”). The protocols showed all the groups to be very relaxed,
joking a lot and talking freely about the task at hand. The open-ended
questions in the post-process questionnaires reveal that most subjects found
the atmosphere in their groups friendly, relaxed and stress-free.
2.1.2 Richness of data

In addition to the fact that data tend to be more spontaneous in CTPs than
in TAPs, they also tend to be more plentiful, and perhaps richer: Séguinot‟s
study “makes use of the dialogic situation to increase the amount of
verbalization in the think-aloud protocol” (Séguinot 2000: 145). Likewise,
House (2000: 159) observes that the “introspective data produced by pairs
of subjects were generally less artificial, richer in translational strategies
and often much more interesting.”
In my main study, collaborative translation tasks involving a 250-
word source text yielded protocols that were between 7,000 and 9,000
words long. From these CTPs I was able to identify more than a hundred
problems (units or aspects of translation that the subjects focused on),
hundreds of tentative solutions that were considered, and so on. As I did
not use think aloud in my studies, I am unable to offer comparative figures
for that method. I can, however, compare CTPs with integrated problem
and decision reporting (IPDR; Gile 2004), a method used in my control
experiments. IPDR is a kind of diary that accompanies a translation and
consists of notes about the problems that the translator encountered in the
task, the tentative solutions considered, the resources consulted and the
reasons for adopting a particular solution in the end. This handy research
tool that requires little effort on the part of the researcher nevertheless has a
disadvantage in that data tend to be scant in comparison with either verbal
reports or retrospection with replay (cf. Hansen 2006: 10). Compared to the
7-9,000-word long CTPs, the IPDRs that accompanied the translation of
the same source text in my study were typically several hundred words
long, listing no more than a dozen problems (an exceptionally long one was
around 800 words long and listed 37 problems).
CTPs are a rich source of data also in the sense that they convey the
„messiness‟ of human translation processes in all its glory. IPDR is usually
a neat summary – frequently self-censored – of what was done during the
task. In contrast to this, CTPs show all of the many steps – in the case of
students and novices, often unnecessarily convoluted – taken to get there.

This feature of CTPs becomes all the more prominent if we compare it with
yet another research method that can be used to study translation processes:
choice network analysis (CNA; Campbell 2001). In this method, “clues in
translations by multiple subjects can be pooled in order to make inferences
about the processes that typically operate in particular types of subjects
translating particular texts under specific conditions” (Campbell 2001: 31).
All the variations encountered in multiple translations of the same source
text are used to construct “choice networks,” which in turn “reveal a range
of differences and similarities in the behavior of the subjects” (2001: 32).
The main disadvantage of the networks, as also stressed by Campbell
(2000: 31-32), is that their neat appearance evokes a model of “serial
processing,” i.e. a model of decision-making that involves sequential steps.
On the other hand, the verbal protocols clearly show that while the subjects
normally progress in a more or less linear fashion from the beginning of the
source text to its end, this progression is not nearly as ordered or organized
as the networks would suggest. The networks are rather a post festum
reconstruction of what might have gone on if the human brain was a
sophisticated computer program. It is highly doubtful that any human
translation process – group or individual – would ever follow the decision-
making steps shown in Figure 1 in such a disciplined, orderly way.
Secondly, choice networks show only the various possible results of
the decision-making processes, without the reasons why some solutions
made it to the final versions of certain translations and others did not. From
the networks alone, we cannot see whether the solutions arrived at were
produced spontaneously or found (confirmed, etc.) in external resources.
We cannot see how many other solutions each translator considered before
opting for the one we find in the final version of the translation. We do not
have an insight into the actions or interactions that might have been taken
as a response to the problematic points in the text.
ST segment in Fig. 1: Gospodarski razvoj i proces kristijanizacije (osobito

intenzivan u IX. st.) glavni su čimbenici u procesu stvaranja hrvatske države. [Gloss:
Economic development and the process of Christianization (particularly
intense/intensive in the 9th century) are the main factors in the process of the creation
of the Croatian state.]
Figure 1 – CNA based on 60 translations of the same ST sentence

In the context of research on translation processes, in particular

research that aims to be applied in the area of translator education, these
limitations seem to present a distinct disadvantage compared to verbal
reports such as CTPs.
2.1.3 A glimpse of how meaning is constructed

Séguinot (2000: 146) remarks that collaborative settings have a further
advantage of allowing us to see “the integration of world knowledge with
[the translators‟] understanding of the text as they argue for particular
versions.” In other words, they “show how meaning is gradually built
during a conversation” (Salmi 2002: 86). In Example 1, the subjects are
translating from English into Croatian, discussing the sentence about the
Book of Kells from a travel guide to Ireland. The problematic part refers to
the Vikings “being unable to read” and thus “ignoring” the Christian
manuscript.1
ST sentence: Also surviving are some of the monks' exquisitely illuminated

manuscripts, such as the Book of Kells, which the Vikings, being unable to read,
ignored.
Irena: Yes. koje su Vikinzi ignorirali jer ih nisu mogli pročitati [which the Vikings
ignored because they couldn't (=were unable to) read them]
Marina: jer nisu znali čitati [because they couldn't (=didn't know how to) read]
Sanja [typing]: koje su Vikinzi [which the Vikings]
I [to M]: Oh, right
S: So, koje su... as in plural?
M [nodding, to S]: koje su. [To I:] being unable to read... nisu mogli... [they couldn't]
How do you mean nisu mogli [they couldn't]? It... It sounds as if...
S [overlapping, unclear] As if they couldn't
M: As if they physically couldn't...
I: All right, nisu znali [didn't know how to]
M: As if they were blind and so they couldn't
1
The protocols have been translated into English for the sake of the reader, with
relevant phrases left in Croatian (in bold letters, with a gloss in square brackets). The
words printed in Arial were originally spoken in English by the subjects.
S: Or as if they couldn't read only that particular book, for some reason
I: No, they couldn't read it because they didn't know the Latin script
S: Ah, that
M: The thing is... Yes
I: They had the runes and all that. And they knew how to read those
S: Aha
I: But they didn't know the Latin script. Because they were not in contact...
M: In any case, nisu znali pročitati [didn't know how to read it]
I: nisu znali pročitati, yes
M: And this refers to those manuscripts. In fact, the manuscripts survived because the
Vikings didn't know how to read them and so
I: They were unimportant to them
M: Yes
S: So... Book of Kells, comma, koje su Vikinzi ignorirali jer ih nisu znali pročitati
M [nodding]: pročitati
I: pročitati
M: But are we going to [say] ignorirali [ignore] or...?
S: zanemarili [neglected]
M: ostavili [left (behind)]
S: zanemarili
I: Well, ostavili
S: It sounds as if they left them somewhere and went away
I: They did, more or less
M: They did, in fact
I: [waving her hand dismissively] 'Ah, a book!'
M [overlapping]: because they were plundering and taking that which was of value to
them, and books... 'What are we going to do with a book? And besides we can't read it'...
And then they left it
S: In that case, OK
Example 1
2.2 Some disadvantages of CTPs compared to other methods
2.2.1 Unwanted social interaction

According to Ericsson and Simon, explanations, descriptions, justifications
and rationalizations – all of which are seen as socially motivated
verbalizations – and in fact any kind of social interaction during the think-
aloud session is strongly discouraged, since “social verbalizations may be
quite different from the sequences of thoughts generated by subjects
themselves while solving problems, performing actions, and making
evaluations and decisions” (1984/1993: xiv).2 In collaborative translation
sessions, the social interaction is not only present; it is their prominent
feature. Consequently, CTPs generally involve a considerable degree of
rationalization (justification, explanation, etc.). Strictly speaking, this fact
invalidates the protocols in terms of Ericsson and Simon‟s criteria
mentioned above.
Nevertheless, I believe CTPs can still be considered a valid research
tool, as long as it is made clear that they are not a result of think aloud
proper, but an alternative method. Depending on what we are investigating,
we might even decide to make use of adversity: the very reasons why CTPs
are not TAPs might be the very things we would like to find out about
translation processes, for instance, why particular translation decisions are
made, and not others.
In Example 2, the subjects are discussing a sentence about St. Patrick
from the same guide to Ireland. The expression they focus on is trackless
forests. It is interesting to follow the decision-making process and the
arguments they use in assessing tentative solutions (that is, solutions
proposed for a particular translation problem).
ST sentence: Later, he travelled widely in France and Italy, returning to Ireland in 432
to spread the word of Christ through the trackless forests.
2
We might argue that the very circumstances of research constitute a social situation.
Even if the subjects „forget‟ the existence of the experimenter at the conscious level, we
cannot be sure that their verbalizations are not monitored, at least unconsciously.
I: [overlapping] I think that the point is that it was overgrown, not so much… that he
travelled around regardless of whether there was a path or not
S: Maybe kroz divlju šumu [through a wild forest] in the sense that it was not… I don‟t
know
M: kroz… [through]
I: bez obzira na postojanje puta [regardless of the existence of a path] [laughs]
S: Yeah
M: Yeah, right
S: Kill the artistic whatever
M: How did you say, nepreglednu [vast]?
I: Yes
S: Why not?
M: kroz šumovite šume [through foresty forests] [laughs]
S: nepregledna šumovita prostranstva [vast expanses of forest]
M: bespuća [wilderness]
I: nekultivirana područja [uncultured areas]
M: Nooo…
S: You can use that of people, not of areas
I: How do you say when the soil has not been tilled?
S: neobrađena zemlja [untilled land]
I: There is an expression…
M: šuma [forest]
S: krš [karst]
I: pustinja i prašuma [desert and virgin forest]
M: tundre i tajge [tundra and taiga] [laughter throughout this exchange]
I: OK, OK... Come on, let’s get back on track
M: trackless forests
S: There you have it
M: Why don‟t you read the sentence?
S [reading TT from the screen]: Kasnije je putovao po Francuskoj i Italiji te se
vratio u Irsku 432. godine kako bi širio riječ Božju kroz… nepregledne šume, I
don‟t know [later he travelled through France and Italy and returned to Ireland in 432 to
spread the word of God through... vast forests]
M: That sounds as if he was walking through the forest and [waving her hands
expressively] la-la-la… Get it?
I & S [nodding]
S: Maybe he was preaching to the birds like that guy, what was his name? [referring to
St. Francis of Assisi]
M: Exactly
I: Maybe… maybe we can [say], like, da je širio… [that he spread]
S [cutting in]: putujući kroz [travelling through]. Maybe we can [do it] like that, make
a transition like that. Širio je riječ Božju, a pritom je putovao kroz takva područja
[he was spreading the Word of God, and at the same time he was traveling through that
kind of areas], which doesn‟t mean that he would, as she said, stand in the middle of the
forest and [laughs, waving her hands]…
M: talk to himself
S: chapter so-and-so, verse so-and-so [overlapping]
M: because here it says: he found, in this land… a largely peaceable people. So there
were people in those forests…
I: [overlapping, unclear]
M: Ah, land refers to the country…
I: What I think he wanted to say by this „trackless forests’ is that this was an island full
of trees, there were no roads, there were no… I mean, this is 432
M: Yes, yes
S: That‟s why I suggest we put putujući kroz [travelling through]
M: [nods, overlapping] putujući [travelling]
S: because that implies that he didn‟t necessarily stay in the forest and spread the Word
of God, but on his way he would pass through such areas
I & M [smile, nodding]
I: Fine, all right
S: But what did we say, which adjective did we take [to use with] forest?
M: We didn‟t
I: [overlapping] Nepregledne, guste, neobrađene, neraskrčene… [vast, thick, untilled,
uncleared]
S: neraskrčene šume... [uncleared forests] But it‟s more like...
I: zarašćene [overgrown]
M [laughs]: zarašćene!
S: zarašćene… It‟s not a Croatian word at all. It‟s slang

I: zarasle [overgrown]
S: No… šuma can‟t be zarasla. I mean… It‟s, like, zarasla, in need of a shave
M: That was my first association, too. Um… [pause] nepregledna [vast], maybe. That‟s
kind of… It means that it‟s full of trees
I: neprohodna [impenetrable, trackless]
M & S: neprohodna!
S [applauds]
M [bows] kudos!
S: [indicating I] Person of the month
M: Employee of the month
Example 2
Based on the CTPs featuring verbalizations such as those illustrated by

Examples 1 and 2, I was able to identify nine types of arguments used in
assessing tentative solutions: personal preference, „sounds better,‟ free
associations, „sounds as if,‟ „it is (not) said that way,‟ „it is (against) the
rule,‟ pragmatic/textual reasons, TT reader, and „what the author wanted to
say‟ (Pavlović 2007: 95-105).3
I was then able to compare different translation processes (in my
case, processes in two directions of translation) in terms of types of
arguments that were predominant in each case (2007: 147-50). Even
without going into details, it is clear that a lot can be learned from this kind
of analysis. For instance, if types of arguments used by novices and by
experienced translators are compared, or if predominance of certain types
of arguments is correlated with translation quality, the findings could shed
some light on translation expertise and could be applied in translator
education.
2.2.2 Group dynamics

Another problem that is reported in the literature is related to the
psychodynamic interaction processes that take place between the subjects.
Séguinot points out that the subjects have an interpersonal relationship to
3
In vivo codes – words used by the subjects themselves – were used to label some of
these arguments in order to capture their salient property in an easily recognizable way
(Strauss and Corbin 1998: 105 and passim).
maintain, while Kussmaul (1995: 11-12) warns that one of the subjects may
become a leader “not because of his or her superior capabilities, but
because of personality features.” Likewise, a subject may “hold back his or
her ideas for reasons of politeness.” Barbosa and Neiva (2003: 151) make
similar observations. Indeed, according to psychologists, people‟s decisions
are sometimes guided by what they call a “feel-good” criterion. Depending
on the cultural norms, social situation, and/or personality traits, this may
take the shape of either exaggerating one‟s superiority over others, or else
exaggerating their commonalities with group members (Wilson 2002: 38).
It has to be stressed that in (individual) think-aloud studies, in spite
of Ericsson and Simon‟s admonition that the social component should be
excluded at all cost, there is indication (e.g. Jakobsen 2003; Tirkkonen-
Condit 1997) of subjects engaging in similar tactics in an attempt to save
face or manage uncertainty. This supports the view that social factors are
inevitably present in any kind of research, individual or collaborative,
although admittedly to varying degrees. These „tactics‟ often operate at a
non-conscious level, and post-process elicitation procedures such as
interviews or questionnaires may not always reveal everything.
In my experiments involving CTPs, all groups collaborated closely.
Of all the actions/interactions, joking was by far the most frequent type. It
seems that, for all the groups, joking was a way of maintaining a positive,
creative and cooperative atmosphere conducive to free associations and
brainstorming, and one in which differences of opinion were less likely to
be perceived as face-threatening (Kussmaul 1995: 48). In the protocols of
Groups A, B and D the jokes were more task-related and seemed to be less
distracting. Although in all the groups the subjects knew each other and had
worked together before (in class), this was especially true of Group C,
where all three subjects were close friends. As a result, their attention often
wandered from the task at hand. Thus in this group‟s L1 translation task, as
many as 16 prompts were found among the verbalizations. By prompts we
mean expressions such as “let‟s move on” or “come on, let‟s get back to
work.” It would be misleading to say that Group C did not take the task
seriously; rather, this was their style of working. It did, however, cause
them to make some mistakes, when a correctly translated element never
made it to the typed version because the typist was laughing and the other
two members of the group were also distracted and consequently failed to
notice that the element had gone missing.
For collaborative work to be a successful experience, certain
conditions have to be fulfilled, as can be deduced from the subjects‟
comments in the post-process questionnaires. All the team members being
“open to different suggestions” would be among those desirable
ingredients, as would “appreciating each other‟s opinions” or “listening to
each other‟s ideas.” Thus Vlatka [all names are fictional] says about her
group‟s collaboration: “No one forced their own opinion, but we all gave
our suggestions freely.” Or, as Marta explains: “Every idea somebody had
was discussed and either accepted or rejected but in such a way that
everyone was happy with the one decided on in the end.” Ivan adds to the
list the “relaxed atmosphere” and “contribution by all team members.”
Nevena also emphasizes the “friendly and cooperative atmosphere” and
Tina mentions the fact that “everyone did their share” of the work. Here are
some more comments:
I was completely satisfied, although usually I do not like group work. It all
depends on the people in the group. This time both of my colleagues were very
cooperative; and they are good students and have a very good knowledge of
English. I think each of us contributed equally. It is really important that people
agree to work together and that they are all, more or less, at the same level of
knowledge and willing to take each other‟s suggestions. Translating in groups is
terrible if one or more members is bad at grammar or orthography or refuses to
cooperate.
That things can go wrong in a collaborative working environment is

evident from this subject‟s description from the control experiments:
We just couldn‟t get along. When one of us suggested something, the others
dismissed her answer immediately and suggested theirs. Every one of us thought
that her own suggestion was the best. I felt like the entire work was about who
would [be right]. That‟s why it lasted so long. At the beginning we took an hour
to do the first couple of sentences.
Complementing CTPs with introspective data that provide an insight

into group dynamics therefore seems necessary for this kind of study (see
also 2.3).
2.2.3 Environmental validity of CTPs

Another criticism leveled against CTPs is that in professional translation
practice, people do not usually work collaboratively on the same text.
Teamwork, increasingly frequent among professionals, usually entails a

division of labor, whether by sectioning the text or by dividing the roles
(e.g. a terminologist, a reviser, a project manager, and so on). Therefore,
environmental validity of studies involving CTPs might be questioned.
Jääskeläinen (2000: 74) warns that to conclude that CTPs are a better
source of information about translating would be “premature, since the
studies in which the two types of data have been compared contain other
variables which may account for the differences between the two
experimental conditions.”
In the main experiments of my project CTPs were used as the
research method in both directions (translation into the subjects‟ first
language and out of it). But I also wanted to see if I could compare group
and individual translation processes of the same source texts, along the
same parameters (number and type of problems, tentative and selected
solutions, resources, quality of the final product and decision-making
process). To this end, control experiments were conducted involving a
larger number of comparable subjects working (at home) either
individually or in groups. They were asked to accompany their translations
with IPDR, and choice networks were also created on the basis of their
translations. Post-translation questionnaires similar to the ones used in the
main experiments added to the picture.
From these control experiments it was found, in brief, that groups
and individuals encountered similar problems and considered similar
solutions. They also used similar resources. However, groups on average
tended to produce better translations (translations with fewer elements that
needed revision, in particular „unpublishable‟ elements). A possible
explanation is that collaborative translation seems to be more „fluent‟ in the
sense of there being a much larger number of tentative solutions to choose
from than in individual translation; collaborative translation also tends to
entail more stringent output monitoring, as the subjects themselves suggest.
Thus Nevena says: “It is excellent when you can choose from a large
number of different ideas,” and Tanja reports she likes group work
“because many more ideas appear.” Mislav similarly states that in
collaborative translation “others will often have better ideas or they might
just help you think of something you would not think of at that moment.”
Another subject explains:
Translating in a group is very efficient because there are always more ideas and
solutions than one can come up with when translating alone. Discussion can
always result in good ideas and I believe the choices we made were often better
than they would have been had we translated on our own. Ideas were discussed,
analyzed, criticized and always led to good conclusions.
As far as output monitoring is concerned, Ivan says that he likes

collaborative work because it “gives [one] the opportunity to get one‟s
ideas double-checked before sending them out in print.” Marija also says
that it is “great to have someone with whom you can discuss things you‟re
not sure about.” Vlatka echoes their sentiments when she says: “When I
translate alone I sometimes lack the final checkup when I‟m trying to
decide which of several options is the best. This is not the case in group
translation.” Marta makes a similar point, saying (about her group‟s L2
translation task):
I‟m happy about our brainstorming for the best solutions, and [having] three
minds instead of one to check if the English sentence sounds “right”. Since there
were three of us, each noticed a different detail and I think we covered all of
them, some of which a single person might have overlooked much more easily.
Due to these differences between individual and collaborative translation, it

remains to be seen to what extent it is possible to make generalizations
about the former mode on the basis of the latter (or indeed about translation
processes as such on the basis of either TAPs or CTPs). For now, we have
reason to assume that the experiments with either TAPs or CTPs offer
valuable insight, albeit incomplete or indirect, into those processes. While
it is true that CTPs may not necessarily be a better source of information
than TAPs, they are certainly an additional source, and a useful one.
2.3 Using pre- and post-process questionnaires with CTPs
In my studies involving collaborative translation, the post-experiment

questionnaire asked the subjects to rate, on a scale from 1 to 5, the
collaborative session along several parameters judged to be relevant for
group dynamics. These included the relations in your team (ranging from
“very conflicting” to “very cooperative”), the atmosphere (ranging from
“very dull” to “very creative”), and the subjects‟ satisfaction with the way
your team worked (ranging from “very dissatisfied” to “very satisfied”).
They were furthermore asked about how much they felt they contributed
toward the final version of the translation, whether the other members of
the team did their share of the work, whether they had an opportunity to say
what they wanted, whether the other members of the team listened to what
they had to say, whether their suggestions were accepted for the final
version and, if not, whether they were happy with the solutions that their
colleagues decided on. In addition, there were some open-ended questions
in which the subjects could write further comments. Most subjects who
took part in the studies used this opportunity to comment on some aspect of
collaborative work. Diana thus admits: “I kind of had the feeling
sometimes that I was pushing too hard with the suggestions that I liked,”
but another subject in her group says (emphasis added): “We were open to
all suggestions and no one tried to put their solutions in front of everyone
else‟s.” In the same group, the third subject expresses her satisfaction with
“the way we „respected‟ each other‟s opinions and fully collaborated. I […]
didn‟t feel insecure.” In another group, a subject remarks that “the work in
groups depends on the participants‟ characters, and I had the impression
that Mirna was less frequently in the spotlight than Ana and I.” But Mirna
herself observes: “We worked together many times before so we function
well as a group.” The first subject, Jasna, says: “Sometimes one of us
preferred one translation (hers) but had to compromise,” and Ana remarks
in a similar vein: “Sometimes you simply have a different opinion and
don‟t agree with your colleagues, but not everything can always go
smoothly […]. It was fun working together like this.” In one of the pilot
groups, two subjects express their satisfaction with the way things went:
“Everyone‟s ideas were considered and discussed, everyone had a duty,
organization was good,” “the division of work was great.” But the third
subject expresses reservations about one of her colleagues‟ knowledge,
admitting also that they do not get on well outside class. In the actual
protocol, her irritation with the other is sometimes apparent; without her
comment it would have been difficult to explain. Alternatively, protocols
may display a great deal of what the researcher may perceive as
confrontation or argument, while in reality this may not be how the
participants see it. In some cultures, social situations, and perhaps age
groups, such confrontation need not necessarily spell conflict or animosity,
as this comment illustrates:
The creative, fun and cooperative atmosphere during our work was not
interrupted with positive conflicts about different versions and suggestions for
translation answers, we did enough brainstorming, research and consulting with
other resources. I find this type of working fun, creative and prefer it over doing
it solo.
The first-person perspective provided by the questionnaires can thus help

the researcher interpret the results obtained from the collaborative
protocols.
2.4 Group size in studies involving CTPs
Research has shown that effective group size is relative to the type of task
(Bruffee 1999: 26). For translation tasks involving short, non-domain-
specific texts, groups of more than three members seem to be too large, as
non-domain-specific texts are not likely to involve a clear division of labor
(e.g. terminology management). It could also be argued that in groups of
four, some subjects might not have the opportunity to speak their minds, or
that several subjects might speak all at once, which would make the
transcribing of (parts of) the sessions difficult or impossible.
CTPs from my pilot study (Pavlović 2005), which involved three
groups of three members each, showed that in each group two people
tended to talk more than the third. On the other hand, this third person did
have an important role to play. In one group, for example, she was the one
writing. In the other, the third person often asked questions, or made the
others return to the source text when she thought they had drifted too far
away from it. It was also easier for them to make decisions in groups of
three than it would have been in pairs or fours, as they couldn‟t get stuck in
a 50-50 stalemate. Having a third member make a decision when two had
conflicting opinions seemed to go a long way towards defusing potential
tension in the group. Groups of three are not without justification referred
to in the literature on collaborative tasks: “Working groups, especially
long-term working groups, seem to be most successful with three
members” (Bruffee 1999: 26).
2.5 Audio and video recording in studies involving CTPs
In addition to audio recordings, some experimenters have used a video

camera to register the subjects‟ behavior. This has been done to
complement the think-aloud protocols and provide a more detailed picture

“by the confrontation of introspective data with empirical observations of
the subject‟s nonverbal behavior registered on videotape or in detailed field
notes” (Barbosa and Neiva 2003: 143). But, as Kovačič (2000: 102) points
out, “videotaping the experiment […] leaves the researcher with the same
problem of identifying non-explicit messages and classifying e.g. facial
expressions, nods of approval and disapproval, etc.” Another problem with
video recordings is that “the translation process may become unnatural
when it is taped. If the subjects cannot ignore the video camera, they may
feel that they are being observed and consequently change their behaviour”
(Hansen et al. 1998: 63). Occasionally, a video camera has been used to
capture changes in the translated text, and was therefore pointed at the
computer screen, not at the subjects (e.g. Séguinot 2000: 145). Today this
can be done by screen recording programs, many of which are freely
available on the web.4 Bernardini (2001: 255-6) points out that if validity of
a TAP study is to be ensured, the least invasive environmental conditions
have to be set up, and this means “renouncing the wealth of information
provided by video-recordings so as to check the well-known tendency of
subjects to monitor their verbal performance more carefully in this
condition.” Instead, she suggests techniques such as eye-movement
tracking and sound recording, as well as use of computer programs that
record keyboard strokes performed by the subject.
Interestingly enough, a study (reported in MacIntyre and Gardner
1994) in second language acquisition, in which a video camera was used to
arouse anxiety in a group of students, found no significant differences in
self-reported anxiety between that group and the control group. Although
other studies have produced different results, this at least warns against
taking for granted the anxiety-inducing effect of video recordings. The
level of anxiety can be reduced by setting up the experiment in a relaxed
environment. Thus Hansen (2006) reports, based on her studies, that there
might be a greater stress factor involved in the experiments at the office
compared with those conducted at home.
In my pilot study (Pavlović 2005), collaborative translation sessions
were recorded on digital audio equipment. The recorder used was tiny, and
4
It is perhaps worth noting that screen recordings take up large amounts of disk space.
was therefore relatively easy to ignore and forget about. The recordings
show that the subjects were very relaxed; there is a lot of talking, laughing
and an occasional swear word that they would not have used in front of the
researcher. However, at the transcription stage two problems emerged. One
was that, although the speech is relatively clear and it was possible to tell
what the subjects were saying, it was not always possible to determine
exactly who said what. This was especially the case in one group in which
all three subjects were women, two of whom had very similar voices.
Another problem, which may be even more important, is that it was not
always possible to tell exactly what the subjects were doing, for instance,
which resource they were consulting. Occasionally they could be heard
reading a definition of a word or an example of usage, but it was not clear
whether they were reading it from a dictionary, and if so from which one,
or from an Internet page. For studies that set out to examine the use of
external resources in translation, this kind of information, or lack thereof,
might prove critical. Relying on the subjects to provide the missing
information after the experiment has been transcribed may be a risky
undertaking, as they may not remember all the details. Going through the
audio recording straight after the experiment may not always be feasible
either.
Other researchers seem to have encountered similar problems to the
ones mentioned above, and decided to use video recordings in order to
overcome them. Thus Dancette (1997: 88) says that her subjects were
videotaped in order to “record some behaviors that they would not
necessarily mention, such as looking up a word in a specific dictionary, and
to see what they were doing when they were silent.”
If a video camera is used to this end, it should be small and
unobtrusive, and it should be positioned out of the subjects‟ field of vision,
such as at an elevated position. Alternatively, researchers (e.g. Livbjerg and
Mees 1999: 136) have used a method of being seated in an adjacent room,
separated from the subjects by a glass panel through which they could
observe the process without being in their line of sight. Whenever a
reference work was used, it was noted down.
In my main study, I decided to use a video camera for the reasons
mentioned above. Comparing the pilot experiments, in which only an audio
recording was made, with the main experiments, which were video-taped, I
can see no difference in the subjects‟ behavior that could be attributed to

the two kinds of recording system. What I can say is that having a picture
to go with the sound has made transcribing and interpreting the protocols a
lot easier, and I would highly recommend the use of video if the aims of the
study justify it.
3 Conclusion
In this paper I have presented some of my experiences in using CTP to

investigate translation processes, comparing this research tool with TAP,
IPDR and CNA. Relevant issues such as post-process questionnaires, group
size and video recording have also been discussed.
As also observed by other researchers, CTPs have a number of
advantages over think-aloud protocols involving single subjects, as well as
over alternative tools such as IPDR and CNA. However, using
collaborative translation tasks also has its drawbacks. The most notable of
these is that the exact relationship between individual and collaborative
translation has not been fully investigated, and generalizations about one
condition on the basis of the other are difficult to make.
The use of CTPs seems to be particularly suited for studies aimed at
improving translator education, as collaborative translation tasks can be
natural and stress-free for the subjects, while at the same time providing the
researcher with a deep insight into the students‟ translation processes. For
example, a lot can be learned about how and why certain mistakes are
made, what kinds of problems novice translators encounter in particular
types of texts, what kind of solutions they opt for and for what reasons,
how they construct the meaning of the source text and gradually build the
target text on the basis of that construction, how they integrate their
knowledge of the world with the text at hand, and how groups of students
working collaboratively on a translation task can help each other master
particular translation skills.
While such studies involving collaborative translation that takes
place in the learning context may or may not tell us something about the
processes in the minds of individual professional translators, they will
certainly help us understand better the social, as well as cognitive aspects
of acquiring this complex competence we call translation.
References
Alves, F. (ed.) 2003. Triangulating translation. Perspectives in Process Oriented

Research. Amsterdam and Philadelphia: John Benjamins.
Barbosa, H. G. and Neiva, A. M. S. 2003. Using think-aloud protocols to
investigate the translation process of foreign language learners and
experienced translators. In F. Alves (ed.). Triangulating Translation.
Perspectives in Process Oriented Research. . Amsterdam and Philadelphia:
John Benjamins. 137-155.
Bernardini, S. 2001. Think-aloud protocols in translation research: achievements,
limits, future prospects. Target 13 (2): 241-263.
Bruffee, K. A. 1999. Collaborative Learning: Higher Education, Inter-
dependence and the Authority of Knowledge. Baltimore and London: Johns
Hopkins University Press.
Campbell, S. 2001. Choice network analysis in translation research. In M.
Olohan (ed). Intercultural Faultlines. Research Models in Translation
Studies I: Textual and Cognitive Aspects. Manchester: St. Jerome
Publishing. 29-42.
Dancette, J. 1997. Mapping meaning and comprehension in translation. In J. H.
Danks, G. M. Shreve, S. B. Fountain, M. K. McBeath (eds). Cognitive
Processes in Translation and Interpreting. Thousand Oaks: Sage. 77-103.
Ericsson, K. A. and Simon, H. A. 1984/1993. Protocol Analysis. Verbal Reports
as Data. Cambridge, MA and London: Massachusetts Institute of
Technology.
Eye-to-IT. Project homepage. http://cogs.nbu.bg/eye-to-it/?summary. Visited
February 2007.
Gile, D. 2004. Integrated problem and decision reporting as a translator training
tool. JoSTrans 2: 2-20. http://www.jostrans.org/issue02/art_gile.pdf. Visited
March 2005.
Göpferich, S. 2008. Translationsprozessforschung: Stand – Methoden –
Perspektiven. Tübingen: Narr.
Hansen, G. (ed.). 1998. LSP Texts and the Process of Translation (Copenhagen
Working Papers in LSP 1). Copenhagen: Samfundslitteratur.
Hansen, G. (ed.) 1999. Probing the Process in Translation: Methods and Results
Hansen, G. (ed.) 2002. Empirical Translation Studies: Process and Product
Hansen, G. 2006. Retrospection methods in translator training and translation
research. JoSTrans 5. http://www.jostrans.org/issue05/art_hansen.pdf.
Visited February 2007.
Hansen, G. and the members of the CBS translation project. 1998. The
translation process: from source text to target text. In G. Hansen (ed). LSP
Texts and the Process of Translation (Copenhagen Working Papers in LSP
1). Copenhagen: Samfundslitteratur.
House, J. 2000. Consciousness and the strategic use of aids in translation. In S.
Tirkkonen-Condit, and R. Jääskeläinen (eds). Tapping and Mapping the

Philadelphia: John Benjamins. 149-162.
Jääskeläinen, R. 2000. Focus on methodology in think-aloud studies. In S.
Tirkkonen-Condit and R. Jääskeläinen (eds). Tapping and Mapping the
Jääskeläinen, R. 2002. Think-aloud protocol studies into translation: an annotated
bibliography. Target 14(1): 107-136.
Jakobsen, A. L. 2003. Effects of think aloud on translation speed, revision and
segmentation. In F. Alves (ed.). Triangulating translation. Perspectives in
Process Oriented Research. Amsterdam and Philadelphia: John Benjamins.
69-95.
Kovačič, I. 2000. Thinking-aloud protocol – interview – text analysis. In S.
Krings, H. P. 1986. Was in den Köpfen von Übersetzen vorgeht. Tübingen:
Gunter Narr.
Krings, H. P. 1987. The use of introspective data in translation. In C. Færch and
G. Kasper (eds). Introspection in Second Language Learning. Clevedon and
Krings, H. P. 1988. Blick in die „Black Box‟ – Eine Fallstudie zum
Übersetzungsprozeß bei Berufsübersetzern. In R. Anrtz (ed.). Textlinguistik
und Fachsprache: Akten des Internationalen übersetzungswissenschaft-
lichen AILA-Symposions. Hildesheim: Olms. 393-412.
Kussmaul, P. 1991. Creativity in the translation process: empirical approaches. In
K. M. Leuven-Zwart and T. Naaijkens (eds). Translation Studies: the State
of the Art. Proceedings of the First James S Holmes Symposium on
Translation Studies. Amsterdam and Atlanta: Rodopi. 91-101.
Kussmaul, P. 1995. Training the Translator. Amsterdam and Philadelphia: John
Benjamins.
Livbjerg, I. and Mees, I. M. 1999. A study of the use of dictionaries in Danish-
English translation. In G. Hansen (ed.). Probing the Process in Translation:
Methods and Results (Copenhagen Studies in Language 24), Copenhagen:
Samfundslitteratur. 135-150.
MacIntyre, P. D. and Gardner, R. C. 1994. The effects of induced anxiety on
three stages of cognitive processing in computerized vocabulary learning.
Studies in Second Language Acquisition 16: 1-17.
O‟Brien, S. 2006. Eye-tracking and translation memory matches. Perspectives:
Studies in Translatology 14(3): 185-203.
Pavlović, N. 2005. Directionality Features in Collaborative Translation
Processes. Unpublished D.E.A. thesis. Tarragona: Universitat Rovira i
Virgili.
Pavlović, N. 2007. Directionality in Collaborative Translation Processes.
Unpublished Ph.D. thesis. Tarragona: Universitat Rovira i Virgili.
Available on: http://www.tesisenxarxa.net/TDX-1210107-172129/
Salmi, L. 2002. Computers, documentation and localization: a methodological

perspective. Across Languages and Cultures 3 (1): 83-90.
Séguinot, C. 1996. Some thoughts about think-aloud protocols. Target 8 (1): 75-
95.
Séguinot, C. 2000. Management issues in the translation process. In S.
Philadelphia: John Benjamins.143-148.
Strauss, A. and Corbin, J. 1998. Basics of Qualitative Research: Techniques and
Procedures for Developing Grounded Theory. 2nd edn. London: Sage.
Tirkkonen-Condit, S. 1997. Who verbalises what: a linguistic analysis of TAP
texts. Target 9 (1): 69-84.
Tirkkonen-Condit, S. 2002. Process research: state of the art and where to go
next? Across Languages and Cultures 3 (1): 5-19.
Tirkkonen-Condit, S. and Jääskeläinen, R. (eds). 2000. Tapping and Mapping the
Philadelphia: John Benjamins.
Translog. Tool for analyzing text production processes. www.translog.dk. Visited
November 2006.
Wilson, T. D. 2002. Strangers to Ourselves. Discovering the Adaptive
Unconscious. Cambridge, MA and London: The Belknap Press of Harvard
University Press.
Manifestations of inference processes in legal translation
Dorrit Faber and Mette Hjort-Pedersen
Abstract
Legal texts are linguistically complex and difficult to understand for lay
persons. From a cognitive point of view it may therefore be assumed that
linguistic explicitations and implicitations will be frequent phenomena in
legal TTs because translators will tend to leave traces of their hard-won
understanding in the TT. On the other hand, legal translations have legal
consequences in the real world. From a legal point of view it may therefore
conversely be assumed that explicitations and implicitations will be
relatively rare phenomena in legal TTs because adding or removing
information may change the legal scenario. This article describes and
discusses tentative results on the correlation between cognitive processing
of legal texts and linguistic explicitation and implicitation in legal
translation performed by student and professional translators, respectively.
In addition, the article describes and discusses various problems in setting
up experiments designed to reveal the nature of this correlation.
1. Introduction
The aims of this article are twofold: (a) to describe some tentative results of
an ongoing research project on the correlation between mental explicitation
processes and resulting instances of linguistic explicitation or implicitation
in legal translations performed by student translators and professional
translators; (b) to describe and discuss the problems involved in the set-up
of experiments designed to shed light on the nature of this correlation.
Our project, which is an attempt to find out more about the mental
processes involved in legal translation, consists of three studies using
Translog and either dialogue protocols („think aloud in pairs) or concurrent
think aloud combined with retrospective interviews. In the first, translations
by student translators have been analysed (Hjort-Pedersen & Faber, in
108 Dorrit Faber & Mette Hjort-Pedersen
press); in the second, which is now in progress, the translation processes of

the student translators will be compared to those of professional translators.
In the present paper, analyses of the two professionals for which data have
so far been collected will be compared with those of the student translators.
In the final study, the translations will be assessed by legal professionals.
A number of studies of explicitation have been conducted over the
past years, and explicitation has been claimed by some scholars to be a
translation universal (e.g. Blum-Kulka 1986; Klaudy 1998; Klaudy and
Károly 2005; and Pápai 2004). Explicitation may be triggered both by
linguistic factors and a desire to cater for the needs of the target language
audience. Implicitation is apparently a much less frequently used strategy
(see Klaudy and Károly 2005).
Looking at legal texts and legal translations in particular, one
characteristic feature of such texts is that they are linguistically complex
and difficult for lay people to understand, for instance because of their
frequent use of nominal constructions, passives, culture-bound terms and
elliptical phrases (e.g. Šarčević 1997; Kjær 2000; and Chromá 2005).
When a translator embarks on a translation of a legal text, it is therefore
often necessary in the comprehension process to mentally explicitate
information that is only implicit in the source text, such as who performs a
certain act, what is the act in question, and where and/or when is the act
performed. From a cognitive point of view, it may therefore be assumed
that the inference processes or mental explicitation a legal translator will
have to go through when understanding a legal text will leave traces in a
corresponding target text in the form of linguistic explicitation. This
assumption is supported in Pym (2005), who hypothesises that the harder
the source text, the harder translators work, and the more likely they are to
make their renditions explicit.
Conversely, depending on the purpose of the translation, legal
translators may choose to leave out linguistic elements in their translation,
thereby making implicit textual elements that are explicit in the source.
Examples 1/1a and 2/2a below serve to illustrate a possible correlation
between mental and linguistic explicitation in the translation of a Danish
statutory text into English. Examples 1 and 2 are excerpts of the Danish
Inheritance Act. The translations produced by student translators are shown
in 1a and 2a.
Inference processes in legal translation 109
Example 1
Er der ingen arvinger efter § 1, stk 2, arver arveladers forældre
Where there are no beneficiaries under s. 1, subs. 2, the parents of the deceased
inherit (our close translation)
In the cognitive processing of the Danish ST, a translator might mentally

explicitate what the nature of the rules contained in s. 1, subs. 2 is, and this
mental explicitation might then materialise in the translation:
Example 1a
Where there are no beneficiaries under the intestacy rules in s. 1, subs. 2, the
parents of the deceased inherit.
The causes underlying this explicitation could either be the translators‟ own
need to know what section 1 applies to, in which case they allow the
information to remain explicit in the TT. Alternatively, translators might
feel that TT readers should have this information made explicitly available
to them to facilitate comprehension of the legal system.
However, legal texts are also characterised by the fact that they may
have legal consequences or establish rules that are intended to apply over
time. Seen from a legal perspective, it may therefore conversely be
assumed that explicitation and implicitation will be relatively rare in legal
translation because of the risk of an unintended change of legal meaning or
an unintended specification of an intended source-text vagueness, e.g.
Endicott (2005: 46):
the use of vagueness in normative texts is a technique of central importance.
While it always brings with it the form of arbitrariness that precision could
avoid, that form of arbitrariness is often insubstantial. The value of vagueness
means that lawmakers need it for their purposes.
Example 2 and its translation in 2a serve to illustrate the point of linguistic

explicitation in the form of specification resulting in an arguably
problematic TT solution because the vagueness in the source text was
intended.1 This is in contrast to 1 and 1a, where the choice of linguistic
explicitation is unproblematic:
1
See 3.3 for categories of explicitation and implicitation.
109
Example 2
Et adoptivbarn og dets livsarvinger arver […], med mindre andet følger af

reglerne i adoptionslovgivningen.
An adopted child and his or her issue inherit […] unless otherwise provided by
the provisions of the adoption legislation. (Our close translation.)
In the cognitive processing of 2, the translator might mentally explicitate

the exact type of legislative instrument that constitutes „adoptions-
lovgivningen‟ at the time of translation, and this mental explicitation might
subsequently be rendered in the translation:
Example 2a
An adopted child and his or her issue inherit […] unless otherwise provided by
the provisions of the Adoption Act.
The term „adoptionslovgivningen‟ in the inheritance act is vague in that it is

underspecified as to the nature of the legislative instrument or instruments
that contain provisions on the right of inheritance of adopted children. This
is probably intended on the part of the drafters of this provision. The
inheritance act will apply until it is repealed, and it is important that the
present provision on rights of inheritance of adopted children is broad
enough to cover any additional rules that might be laid down in any other
and different types of legal instruments over time. In 2a this presumably
intended vagueness would be eliminated in the target text by the
specification of the nature of the legislative instrument.
Finally, examples 3 and 3a serve to illustrate a case of implicitation
in legal translation. When leaving property to somebody by will, an English
testator might choose to insert the following provision in his or her will:
Example 3
I devise and bequeath my Mercedes to Cindarella.
The verbs „devise‟ and „bequeath‟ denote different ways of leaving property
under English intestacy rules. In a situation where an English will is to be
translated into Danish, a legal translator might choose to make this two-
sided affair implicit because it represents a cultural difference compared
with Danish inheritance rules, the specification of which may represent an
unnecessary complication for the reader of the Danish translation.
Example 3a
Jeg testamenterer min Mercedes to Cindarella
I leave my Mercedes to Cindarella (our close translation)
The above examples all point to the existence of explicitation and

implicitation in both student translator and professional translations, but
without studies of the translation processes, it is not possible to determine
why these manoeuvres take place (cf. Dimitrova 2005a).
2. Research questions and assumptions
When comparing student translators and professional translators, the

schism between the cognitive and the legal considerations gives rise to the
following research question: What is the relationship between any mental
explicitation and linguistic explicitation and/or implicitation as translation
choices across these two groups of translators? This question is interesting
seen from a pedagogical/didactic point of view because professional
translators may be said to represent the norm that student translators are
working towards on the basis of norms already internalised when they
embark on the road towards mastering legal translation. But no studies
have as yet established in any detail the norms, if any, prevailing at either
end of the scale of expertise.
When considering this question, two possible and mutually exclusive
scenarios might reasonably be assumed:
based on the cognitive angle, the inference processes of student
translators will emerge in translations to a higher degree than those
of professional translators because of the presumed greater cognitive
effort undertaken by student translators. This would support Pym‟s
hypothesis.
based on the legal angle, the inference processes of student
translators will emerge in translations to a lesser degree than those of
professional translators because of the students‟ presumed greater
uncertainty as to the validity of the results. This would contradict
Pym‟s hypothesis.
Consequently, it is by no means clear whether experience in legal
translation has a bearing on linguistic explicitation or implicitation choices.
111
3. The experiments
3.1 The student translators
In the first phase of our project we focused on the student translators and
designed an experiment where eight MA students of English for Specific
Purposes (ESP) working in pairs translated into English a Danish extract of
a law report consisting of 103 words while thinking aloud. As teachers we
have often been faced with student translators expressing a high degree of
uncertainty when working with legal translation tasks (Hjort-Pedersen &
Faber 2009).
Several studies of explicitation in translation have focused on
connectors and how translators use connectors to make explicit in the TT
their own understanding of discourse relations between sentences that are
only implicit in the ST (see e.g. Englund Dimitrova 2005b; Denver 2002;
and Blum-Kulka 1986). In our case, because of the above-mentioned
characteristic linguistic features of legal texts, we deemed it relevant to
focus on the one hand on „slots‟ left open in the ST due to the use of
nominalisations, passives, legal terminology and elliptical phrases, and on
the other on role names and names of source culture institutions. These
focus points require some processing and filling in of slots either implicitly
by automatised processing or explicitly, in order for the source text
understanding to be as complete as is necessary to ascribe the intended
meaning to the source text.
To analyse the processing of the slots in the ST by the informants,
we have chosen to draw on two concepts from relevance theory: “reference
assignment” (RA) and “enrichment” (EN).2 For our purposes, reference
assignment involves accessing (that is retrieving) a mental representation
which uniquely identifies the intended referent (Blakemore 1992: 68).
Enrichment represents the process of filling in missing information in a
linguistically encoded semantic representation (Blakemore 1992: 61).
The student experiment is described in more detail in Hjort-Pedersen
and Faber (in press). In this article we will briefly describe the results of the
experiments involving professional translators and discuss the problems
2
“Disambiguation” did not apply to this particular translation task.
involved in eliciting and comparing explicitation and implicitation patterns

across the two groups of informants.
3.2 The professional translators
At this stage in our project we have supplemented the student experiments

with experiments involving two professional translators translating the
same Danish source text as the students. Care was taken to ensure that the
text would be a challenge for translators at all levels. It was also crucial that
it contained a good deal of the types of implicit elements that we were
interested in. The professionals were given the same dictionary and Internet
access as the student translators and the same translation brief specifying
the commissioner of the translation and the target group. Also, as with the
student translators, the professional translators were instructed to think
aloud while processing the Danish source and translating it into English.
The think aloud was audio-recorded, and the actual translation process was
logged in Translog (Jakobsen 1999). Unlike the student translators, the
professional translators worked alone (see 4.1 below). They were given as
much time as they wanted to complete the task. The actual translation
process was followed up by a retrospective interview, where the informants
were asked to comment in more detail on their deliberations during major
pauses logged in Translog.
The examples below illustrate verbalised mental explicitation in the
form of an enrichment process undertaken by one of the student translator
groups and one of the professionals in relation to the first part of the Danish
law report „retsplejeloven findes dog ikke at udelukke….‟ (the
Administration of Justice Act is found not to exclude….); here the implicit
agent of the passive „findes‟ is made explicit:
(T1: „findes dog ikke at udelukke‟ at (læser op)
T2: Jeg ved ikke en gang hvad de mener
T1: Jo, det er bare at retten finder ikke, at retsplejeloven udelukker. Forstår du
godt, hvad jeg mener – eller hvad de mener
T2: Ja
T1: Så er det bare lige, hvordan vi nu får skrevet det)
[T1: is found not to exclude (reads ST segment)
113
T2: I don’t even know what they mean
T1: Actually, the point is simply that the court does not find that the
Administration of Justice Act excludes – do you understand what I mean – or
what they mean
T2: Yes
T1: Then we just have to figure out how to translate it]
Professional
„Findes dog ikke at udelukke‟ – altså jeg forstår nok ikke helt den her sætning –
landsretten, det er sådan noget landsretten siger
[„is found not to exclude’ – I’m not sure I understand this sentence – the High
Court, it is something said by the High Court]
3.3 Summary of the experiments
Table 1 shows our analyses of the TAPs. It compares the number and nature
of mental explicitation processes undertaken by the group of student
translators and the professional translators, respectively, in relation to our
focus points, as evidenced by the think-aloud protocols. RA represents
reference assignment, and EN represents enrichment.
As can be seen in the table, professional 1 resembles the student
groups with respect to the degree of mental explicitation, whereas
professional 2 does not verbalise anything which can be understood as
mental explicitation of the focus points. There is no reason to think that
inferencing with respect to these points is not taking place with
professional 2, but we have no way of knowing without verbalised
manifestation. This represents one of the problems involved in gaining
access to the mental explicitation process through the use of TAPs (see 4.2
below).
For our purposes, linguistic explicitation covers two types:
Addition (A), which is quantitative and involves the inclusion in
the TT of extra lexical elements.
Specification (S), which is qualitative and adds meaning(s) by
using lexical elements that are semantically more informative.
Similarly, we consider linguistic implicitation to be of two kinds:
Reduction (R), which involves leaving out meaningful ST lexical

elements in the TT.
Generalisation (G), which involves using TL lexical elements that
are semantically less specific than the ST lexical elements.
Table 1. Mental explicitation
Linguistic unit/ Number of Verbalised Verbalised mental

Focus point verbalised mental explicitation
mental explicitation Professional 2
explicitations Professional 1
by 4 groups of
students
RA Retsplejeloven 0 – –
Administration of Justice
Act
RA Byretten 3 + –
The City Court
RA afgørelsen 2 + –
the decision
RA kærendes processkrift 3 + –
the appellant’s statement
of case
EN findes 3 + –
is found
EN som sket 3 – –
as was the case
EN fremsættelse (af 3 + –
afvisningspåstand)
submission (of motion of
dismissal)
EN indsigelse (mod 2 – –
værnetinget
objection (to venue)
EN (sagens ) forberedelse 4 + –
(for byretten)
preparation (of the
case)(before the City
Court)
EN det tiltrædes 4 + –
is accepted
27 out of 403 = 67.5% 7 out of 10 = 70% 0 out of 10 = 0%
Table 2 summarises the correlations between mental explicitation and

linguistic explicitation and implicitation as well as the time spent across the
two groups of informants on processing and translating both the text
sample and our focus points. Professional translator 1, for instance, spends
3
4 groups x 10 focus points.
115
26.20 minutes on the translation as a whole and 12.04 minutes on the focus
points, which represents 45.95 % of the overall translation time. The
notation used in the table is to be understood as follows. For example, A (4)
means that all four student translator groups chose linguistic explicitation,
and S (1) means that only one of the groups explicitated a focus point by
specification.
Table 2. Mental and linguistic explicitation
Linguistic 4 groups of Professional

Professional
Unit/Focus point students translator 1
translator
2
men- linguistic men- linguistic men- linguistic
tal expl impl tal expl impl tal expl impl
RA Retspleje- 0 A(4) - A - A
loven
Administration of
Justice Act
RA Byretten 3 A(1) + A -
The City Court
RA afgørelsen 2 S(1) + -
the decision
RA kærendes 3 S(1) + -
processkrift
the appellant’s
statement of case
EN findes 3 A(1) R(2) + - A
is found
EN som sket 3 A(1) R(1) - R - R
as was the case S(1)
EN fremsættelse (af 3 A(1) R(1) + R - R
afvisningspåstand)
submission of a
motion of
dismissal)
EN indsigelse (mod 2 S(1) - -
værnetinget)
objection (to venue)
EN sagens forberedelse 4 S(4) + - A
(for byretten )
preparation (of the
case)(before the
City Court)
EN det tiltrædes 4 A(1) + - A
is accepted
Overall time/ 56.06/41.18 73.46% 26.20/12.04  45.95% 15.42/6.13  39.75%
Focus point time 49.09/38.02  77.45%
 % of overall time 47.06/29.35  62.37%
34.55/12.01  34.76%
It should be noted that mental explicitation is not always reflected in the

translation, which is why the number of mental explicitations on the one
hand and linguistic explicitations or implicitations on the other is not
always identical. For instance, professional translator 1 verbalises her
mental explicitation efforts extensively, but has no more linguistic
explicitation and implicitation choices than translator 2, who verbalises
nothing at all in relation to the focus points, but still uses linguistic
explicitation and implicitation.
Table 2 shows that the student translator groups spend a lot of time
processing the focus points, ranging from 77.45 % to 34.76 % of the
overall time spent on completing the task of understanding and translating
the short text. The time spent by the professional translators on the focus
points is also fairly long, 45.95 % and 39.75 %, but it is nevertheless
noticeably less than that spent by three of the four groups. This difference
may partly have been caused by the different data elicitation methods
employed (i.e. dialogue protocols vs. think aloud; see 4.1 below).
Of course we cannot as yet draw any conclusions from the figures
presented in Table 2, one reason being that the number of both student and
especially professional informants is too small. But we can say that
processing and translating the focus points represent a substantial part of
the overall time spent by all the informants. As to the translation product, a
tentative observation is that „addition‟ and „reduction‟ are used as strategies
by both students and professionals. „Specification‟, on the other hand, is
only chosen as a strategy by the student translator group. And, somewhat
surprisingly, „generalisation‟ has not been opted for by any informants.
4. Discussion of set-up
Following the tradition which has been established in process-oriented

translation studies, we have used a combination of different data elicitation
methods. However, there are a number of problems connected with
choosing an experimental design that will allow us to elicit as much
information as possible while at the same time not interfere too much with
the informants‟ comprehension and translation processes. These problems
are described and discussed below.
117
4.1 Dialogue protocols and individual think-aloud protocols
In order for us to be able to understand how the complexities of a legal ST

are processed by translators working on the production of a TT, the crucial
point is of course to find a viable way to gain access to their inferencing
processes without interfering (too much) with these processes. One method
often used is think aloud. However, as is well known, there are some
problems connected with think-aloud verbalisations, as stated by, for
instance, Jakobsen (2005: 178):
If we assume that verbalization does reflect processing knowledge reliably, then
we cannot know if the verbalization we record is a complete or only a partial
reflection. Furthermore, there is the danger […] that the verbalization
requirement distorts the primary processes we want to investigate.
A comprehensive discussion of the validity of TA data can be found in

Göpferich (2008: 22 ff.). The existence of an external addressee of the
verbalization may be one such interfering factor. Tirkkonen-Condit (1997:
73) points out that ideally the addressee of a protocol text should be the
subject himself or herself, so that
we can assume that the less the presence of an external addressee is manifested
in the verbalisation, the better the protocol reveals the internal processes we are
investigating. In other words, the more abnormal the protocol is as a text, the
better it functions as a register of processes.
Whereas the dialogues clearly were interpersonal but still with the
occasional reference to the existence of the (external) researcher, one of the
individual TAPs repeatedly revealed the informant‟s awareness of the
importance of verbalization to the experiment, e.g. in the form of
explanations of pauses oriented towards the researchers (What I’m doing
now is that I’m reading the text I have written again/This is a long entry (in
the dictionary), so I have to orient myself a bit ….). Such remarks clearly
represent an interruption of the primary cognitive activity involved in
understanding and translating. However, even though there are a number of
instances of disruptions in the cognitive processing which involve
awareness of the addressee, the informant reverts to the processing of the
focus points and resumes the flow of deliberations. This means that
awareness of the addressee results in a lengthening of the process, but not
necessarily in a major distortion.
Some researchers have used eye-tracking methodology, which does

not interfere with the cognitive processes of the individual. Nevertheless,
we have decided not to use eye-tracking for this experiment because legal
texts require a good deal of dictionary consulting, and this method would
therefore be less suitable for us because of the resulting disruptions of the
tracking of the eye (O‟Brien 2006: 186). Even if the problem of data loss
resulting from the informant looking away from the screen could somehow
be eliminated through the use of electronic legal dictionaries, eye-tracking
would still have to be combined with TAPs and/or retrospective interviews,
since eye-tracking will provide data on cognitive effort, but will not tell us
whether the workload is connected with the enrichment, reference
assignment and disambiguation processes of the informants or with the
choice of specific legal terminology and phraseology (or both).
Obviously, another problem connected with thinking aloud is that the
propensity to verbalise ongoing thought processes may differ a good deal
among informants, as also evidenced by our two professionals and as
reported by others (for an overview, see O‟Brien 2005: 41–43). For
instance, our two professional informants seem to be at opposite ends of the
verbalisation scale. Some researchers (e.g. Barbosa & Neiva 2003: 141),
have attempted to solve this problem by setting up training sessions to
enable their (student) informants to provide the concurrent verbalisation
that the researchers needed. However, we find that such a method would
not be realistic with professionals for reasons of both time and money
(unlike students, professionals have a business to run).
A way of remedying the scarcity of verbalisation is to use dialogue
protocols for both groups of informants. There is no doubt that dialogues
do not reflect the immediate thought processes of the individual translator.
As with all dialogues, social interaction with a dialogue partner in a
translation scenario will influence the way that knowledge processing is
undertaken as well as represented in the TT. However, the great advantage
of dialogues is that they create a natural verbalisation situation, in that they
will force the partners to make clear to each other their understanding of
the text in order to move from ST understanding to a TT. The verbalisation
resulting from the need to communicate the translator‟s understanding of
the ST may therefore be said to be at least a partial reflection of knowledge
processing. But, while students often have a practice of working closely
119
together on the translation of a specific text (in groups or pairs), this is not
the case with professional translators, where dialogues may represent a
major deviation from their natural translation framework. Also, dialogue
set-ups with professional translators might be felt to be a somewhat face-
threatening situation which may result in the professionals keeping their
thoughts to themselves in order not to reveal any uncertainty about
understanding or translation.
4.2 Comparability of data
As described, our data come from two slightly different types of

experimental set-ups, which could conflict with the notion of
comparability. But, as our research focus has been narrowed down to the
correlation between mental explicitation and linguistic explicitation/
implicitation with regard to particular pre-defined points in the ST, it means
that we are not studying the inferencing process as a whole nor trying to
unearth the full scale of translation strategies employed. We are solely
preoccupied with isolated enrichment and reference assignment processes
in connection with the focus points, which represent particular complexities
and characteristics of legal language evidenced in the short text that we
have been working with. Furthermore, because this is a case study, which is
explorative and hypothesis generating, we believe that the experimental
set-ups can be allowed to differ for the two groups of informants and still
provide comparable data. The text variable and the focus point variable are
controlled, since the text used is identical for the two sets of experiments,
the direction of translation is the same, as are the dictionaries etc. provided,
and there is no time pressure for either category of translators. The main
difference is thus in the dialogue vs. individual TAP parameter.
4.3 Use of retrospective interviews
Retrospective sessions were held with the professionals to supplement the

individual TAPs because we suspected that working alone they would not
be as informative in their thinking aloud as the student translators working
in pairs. It was hoped that the retrospective interviews could remedy
potential gaps in thinking aloud during the source-text processing and
translating. We chose to base the retrospective interviews on the pauses that

were evident when the logged translation was replayed.
The following serves as an example of how the retrospective
interview supplemented the TAP of professional translator 2. The focus
point is a passive „tiltrædes‟ (is accepted):
„Det tiltrædes at‟ – jeg brugte tid på – jeg tænkte, hvad f…. står der? Til sidst
nåede jeg frem til at det der måtte stå var at landsretten giver byretten ret i at det
var i orden at byretten, at de godt måtte protestere mod stedet –
‘It is accepted that’ – I spent time on – I thought what (expletive) does it really
mean? Finally I concluded that it had to mean that the High Court agrees with
the City Court that they were allowed to object to the venue.
Unlike the TAP, the retrospective interview text shows that professional
translator 2 had in fact engaged in enrichment during the understanding
process of who the agent of the passive is, she simply did not verbalise it at
the time.
As pointed out by many scholars, e.g. Haastrup (1991) and Bernadini
(2001), the difficult issue is to find the right balance between the
informant‟s own reactions to the pauses and prompting on the part of the
researchers. The risk of influencing informants by asking them to think
back on the process is of course high, as is that of ex-post rationalization on
the part of the informant. Still, we had the impression that the retrospective
session was a desirable channel for professional translator 2 to voice
thoughts about choices and difficulties, arguably because of the greater
naturalness in having an addressee to report to.
4.4 Professionals as informants
The two professional informants we report on in this paper have both done
some legal translation, but it is not their field of expertise. Like most
Danish translators they work both into and out of Danish, which is their
first language. It should be noted that the non-literary translation market in
Denmark is particularly centred on translation into English because of the
international role of this language today.
The problem of comparing the performance of professionals with
that of non-professionals and, on that basis, trying to determine what
professionalism is has been discussed by Bernardini (2001). She criticises
121
the current practice of defining professionals on the basis of years of

experience and official certifications, because such external factors do not
automatically entail expertise or professionalism (2001: 252).
However, we are not at this stage concerned with defining
professionalism as such or with translation quality assessment, but with
what individuals with and without practical translation experience do and
do not do. We focus on the processes and products of student translators
compared with those of individuals who are paid for translating legal texts,
so we operate with such external factors as certification and experience
from working as a professional translator. We are aware that these factors
are not automatically markers of expertise in the translation of legal texts,
but they represent to us the best way of differentiating between informants
with and without substantial experience in understanding a ST and
rendering it in the target language for a particular target group and/or for a
particular purpose.
5. Concluding remarks and perspectives
Because of the early stage of this project we are able to make tentative
observations only about the correlations across the two types of informants
of mental explicitations and resulting linguistic explicitation/implication.
Thus the results will not as yet support or dismiss either of the two
scenarios described in section 2, i.e. establish whether the inference
processes of students will emerge in their legal translations to a higher
degree than in the case of professional translators.
Our very first results indicate that professional translators spend less
time than the student translator groups on translating the text. This is of
course not surprising, since the professionals do not have to reach an
agreement with a partner and they are more experienced working as
translators. As far as the focus points are concerned, the amount of time
spent on them by the professional translators is considerably lower than
that spent by three of the groups and slightly lower than that of the fourth
group. On the basis of the TAP of professional translator 2, it is not
possible to separate the time spent on inferencing efforts from the time
spent on finding the right terminology and phraseology, the reason being
that we have no verbalised mental explicitation. As to professional
translator 1 and the student translator groups, the TAPs and the dialogues
show that there is interaction between mental explicitation and search for
appropriate terminology and phraseology, so that the time percentage spent
on the focus points covers both these activities.
On the product side, it appears that the types of explicitation/
implicitation that are found are „addition‟ and „reduction‟ with both student
translators and professionals. „Specification‟ is only used by the student
translators and „generalisation‟ by none of the informants. It is too early to
say whether it is indicative of norms in student and professional conduct.
As discussed above, methodological challenges remain in solving the
issue of how best to obtain access to manifestations of inference processes
in legal translation, especially when professional translators are involved as
informants. The next step in the project will be to extend the present data
by conducting experiments with more professional translators to tackle the
problem of establishing explicitation/implicitation patterns or norms in
professional legal translation.
Also, an interesting spin-off from this project will be to correlate the
findings of the present project with legal translation/explicitation and
implicitation preferences and reasons for such preferences of the end user
of legal translations, i.e. lawyers.
References
Barbosa, H. G. & Neiva, A. M. S. 2003. Using think-aloud protocols to

investigate the translation process. In F. Alves (ed.). Triangulating
Translation. Perspectives in Process Oriented Research. Philadelphia, PA,
USA: John Benjamins. 32.
http://site.ebrary.com/lib/kbhnhh/Doc?id=10046611&ppg=151
Bernardini, S. 2001. Think-aloud protocols in translation research; achievements, limits,
future prospects. Target 13 (2): 241-263.
Blakemore, D. 1992. Understanding Utterances. An Introduction to Pragmatics.
Cambridge/Mass:. Blackwell.
Blum-Bulka, S. 1986. Shifts of cohesion and coherence in translation. In J.
House & S. Blum-Kulka (eds). Interlingual and intercultural Communi-
cation. Tübingen: Gunter Narr. 17-35.
Chromá, M. 2005: Indeterminacy in criminal legislation: a translator‟s
perspective. In V. Bhatia, J. Engberg, M. Gotti & D. Heller (eds).
Vagueness in Normative Texts. Bern: Peter Lang. 379-411.
Denver, L. 2002. On the translation of semantic relations: an empirical study.
Revista, Brasileira de Lingüística Aplicada 2 (2). Gunter Narr. 25-46.
123
Englund Dimitrova, B. 2005a. Combining product and process analysis:

explicitation as a case in point. Bulletin Suisse de Linguistique Appliquée
81: 25- 39.
Englund Dimitrova, B. 2005b. Expertise and Explicitation in the Translation
Process. Amsterdam/Philadelphia: John Benjamins.
Endicott, T. 2005. The value of vagueness. In V. Bhatia, J. Engberg, M. Gotti, D.
Heller (eds). Vagueness in Normative Texts. Bern: Peter Lang. 27-48.
Göpferich, S. (2008). Translationsprozessforschung. Stand – Methoden –
Perspektiven. Tübingen: Gunter Narr.
Haastrup, K. 1991. Lexical Inferencing Procedures or Talking about Words.
Tübingen: Gunter Narr.
Hjort-Pedersen, M. & Faber, D. (in press). Explicitation and implicitation in legal
translation – a process study of trainee translators. META 55 (2).
Hjort-Pedersen, M. & Faber, D. (2009). Uncertainty in the cognitive processing
of a legal scenario: a process study of student translators. Hermes 42: 189-
209.
Jakobsen, A. L. 1999. Logging target text production with Translog. In G.
Hansen (ed.). Probing the Process in Translation. (Copenhagen Studies in
Language 24). Copenhagen: Samfundslitteratur. Appendix 1: 1-36.
Jakobsen, A.L. 2005. Investigating expert translators‟ processing knowledge. In
H. V. Dam, J. Engberg & H. Gerzymisch-Arbogast (eds). Knowledge
Systems and Translation. Berlin: Mouton de Gruyter. 173-189.
Kjær, A. L. 2000. On the structure of legal knowledge: the importance of having
legal rules for the understanding of legal texts. In L. Lundquist & R.
Jarvella (eds). Language, Text and Knowledge. Mental Models of Expert
Communication. Berlin/New York: Mouton de Gruyter. 127-161.
Klaudy, K. 1998. Explicitation. In M. Baker and K. Malmkjær (eds). Routledge
Encyclopedia of Translation Studies. London/New York: Routledge. 80-84.
Klaudy, K. & Károly, K. 2005: Implicitation in translation: empirical evidence
for operational asymmetry in translation. Across Languages and Cultures 6
(1): 3-28.
O‟Brien, S. 2006. Eye-tracking and translation memory matches. Perspectives:
Studies in Translatology 14 (3): 185-205.
O‟Brien, S. 2005. Methodologies for measuring the correlations between post-
editing effort and machine translatability. Machine Translation 19: 37-58.
Pápai, V. 2004. Explicitation: a universal of translated text? In A. Mauranen
(ed.). Translation Universals. Do They Exist? Philadelphia: John
Benjamins. Retrieved from
http://site.ebrary.com/lib/kbhnhh/Doc?id=10052859
Pym, A. 2005. Explaining explicitation. In K. Károly and Á. Fóris (eds). New
Trends in Translation Studies. Budapest: Akadémiai Kiadó. 29-43.
Šarčević, S. 1997. New Approaches to Legal Translation. The Hague: Kluwer
Law International.
Tirkkonen-Condit, S. 1997. Who verbalizes what: a linguistic analysis of TAP
texts. Target 9 (1): 69-84.
Unique items in translations
Louise Denver
Abstract
This study addresses the issue of inferencing and translating logical-

semantic relations across sentence boundaries. Drawing on both the
explicitation hypothesis and the unique items hypothesis, the aim is to
examine a number of adversative-concessive relations in a Spanish text
which in a Danish translation can be marked by means of the unique ellers
(„else‟), which has an additional property of being able to express the
speaker’s attitude at the pragmatic level. Both production and process data
are studied. The latter include think-aloud protocols and log files of
keyboard activities, since they can shed light on the inferencing and
decision-making of the subjects. With respect to the product data, it was
hypothesised that the use of explicitations by means of the unique ellers
would be markedly lower than when the same connector was used at the
propositional level with alternative, disjunctive or conditional meaning.
Furthermore, since the inferencing of adversative relations was taken to
involve a lower cognitive load than that of concessive relations, it was
expected that explicitations would to some extent be made by means of
adversative connectors. Both assumptions were supported by the empirical
data. As far as the process data are concerned, it was expected that parts
of the argumentative structure of the ST would pass unnoticed and that,
consequently, a direct transfer of the relations would not necessarily be the
result of a strategic choice, since this presupposes that alternative
translations have been considered. The process data gave evidence of such
gaps in the inferencing of the argumentative ST structure.
126 Louise Denver
1. Introduction
This article addresses the use of the Danish connector ellers („else‟) when
used uniquely in translations from Spanish (L2) into Danish (L1). The
logical-semantic relations studied here are those adversative-concessive
relations where a translation using Danish ellers would be the ideal lexical
choice – at least if translators infer the relation correctly and decide to
render the meaning by means of an explicitation in their TT. Consequently,
the study relates to two wider issues: the unique items hypothesis and the
explicitation hypothesis.
The explicitation hypothesis is here understood as an asymmetry
hypothesis, according to which explicitations in translations from L2 into
L1 are not always counter-balanced by implicitations in translations from
L1 into L2 (Blum-Kulka 1986; Tirkkonen-Condit 1993, 2004; Klaudy
1998; Englund-Dimitrova 2005). In other words, the explicitation
hypothesis predicts that the level of explicitness of the TT will be higher
than that of the ST. The results of earlier studies designed to contribute to
the verification of the explicitation hypothesis have been somewhat
contradictory (Englund-Dimitrova 2005: 35). However, results of a study
carried out at the Copenhagen Business School (Denver 2007) showed that
the mean explicitation rate of adversative relations was as high as 48 %.
Furthermore, studies of monolingual text production carried out by
Källgren for Swedish (quoted in Englund-Dimitrova 2003: 25) and Díez
Prados (2003: 212, 223) for English suggest that these two languages have
a preference for the explicit marking of the adversative relation. Since
Danish is also a Germanic language, the same tendency can be expected to
emerge for this language.
As stated by Reiss (quoted in Chesterman 2007), translations may
not fully exploit the linguistic resources of the TL. Following this
assumption, Tirkkonen-Condit (2004: 177) has suggested that TL-specific
items are under-represented in translated texts. Tirkkonen-Condit uses the
term „unique items‟ to refer to lexical items which “lack straightforward
linguistic counterparts in other languages”. Unique items are items which
do not readily suggest themselves as translation equivalents since they are
not lexicalised in a similar way in other languages. In his discussion of the
Unique items in translation 127
appropriateness of the term, Chesterman (2007) points out that “the notion
of uniqueness seems to be too strong in several respects” and suggests that
unique items are perhaps no more than “formal source lacunas” or lexical
gaps. Bearing these reservations in mind, we will, nevertheless, use this
phrase in our study for lack of a more appropriate label.
In the Danish lexicon we find an array of relational meanings which
can be covered by ellers. In initial position, ellers functions as a connector
with alternative, disjunctive or conditional meaning. The lexicon of other
languages, such as Spanish and English, has equivalent lexical items (in
English: „besides‟, „on other occasions‟, „if not‟). In the position after the
finite verb, however, ellers can be used as an adversative-concessive
connector with an additional pragmatic value to express a speaker‟s
comment on something which has been stated in the previous context
(Jensen 2000: 153). The following example illustrates the unique use of
ellers:
Cruzado de brazos y con expresión entre irritada y suficiente, esperó a que le
despejaran el campo sin mover un dedo ni abrir la boca. ¿Quién mandaba allí
más que nadie? Él, allí y en el país entero. ¿Quién tenía un micrófono para
hacerse oír sobre el griterío e impedir los golpes? Ante él había varios. Habría
bastado una palabra suya, para que las agresiones hubieran cesado. Pero no
lo dijo, ni siquiera “¡Alto!”. (El País, 7 April 2003)
Danish translation (..) Et ord fra ham ville (ELLERS) have været nok til at
standse overgrebet. Men han sagde ikke noget, ikke engang “stop!”
English [With his arms crossed and a partly irritated, partly self-satisfied
expression he [José María Aznar] waited for the square to be cleared without
raising his arms or opening his mouth. Apart from him, who had the power
there, on that square, or in the entire country? Who had a microphone to drown
the shouting and stop the beating? In front of him there were several. One word
from him would have been sufficient to stop the aggression. But he did not
say anything, not even “stop!”]
Adversative relations are marked in Spanish by means of the adversative

conjunction pero („but‟) or connectors such as sin embargo („however‟)
and no obstante („nevertheless‟).1 Adversative connectors introduce strong
arguments while concessive connectors introduce weak arguments which
do not invalidate the inference to be drawn from the propositional content
1
During the process of grammaticalisation, these two connectors were used to mark
concessive meaning. Today they are used as adversative connectors, except in
academic registers (Garachana Camarero 1988).
128 Louise Denver
of the adjacent proposition even though they express violated expectation.

Contrastive meaning is inherent in adversative as well as concessive
connectors, but adversative connectors have a more restricted meaning than
concessive connectors (Garachana Camarero 1988). Consequently, it would
be reasonable to assume that the inferencing of adversative relations
involves a lower cognitive load than that of concessive relations. In Danish,
the unique ellers can furthermore be used at the pragmatic level to express
a subjective value, i.e. the speaker‟s attitude. The Spanish lexicon contains
no item or fixed phrase which, in addition to the adversative-concessive
relational meaning, can express the speaker‟s attitude.2 This does not mean
that it is always impossible to render the pragmatic meaning in Spanish.
Sometimes the pragmatic value can be expressed by means of context-
dependent, ad-hoc idioms or phrasal expressions. In the example above, the
adversative meaning could have been indicated in the Spanish ST by means
of sin embargo. Pero would be ungrammatical, since pero cannot be used
recursively (*Pero habría bastado una palabra suya (..). Pero no lo dijo
(..)). However, the concessive meaning and the speaker‟s disapproval of the
Prime Minister‟s failure to act (He should have stopped the aggression)
cannot be made explicit by means of a connector or a fixed phrase in
Spanish.
In this study, the following assumptions have been made:
1. Following the explicitation hypothesis, it is postulated that when the
adverb ellers is used as a prototypical connector at the propositional
level in initial position with alternative, disjunctive or conditional
meaning translators will raise the level of explicit marking of the
relation.
2. On the other hand, in contexts where ellers could be used in a unique
sense, it is assumed that the level of explicit marking of the relation
in Danish TTs by means of ellers will be low. It is predicted that the
unique items hypothesis „overrules‟ the explicitation hypothesis, so
to speak.
3. Furthermore, in contexts where the unique use of ellers would cover
the full relational meaning, we assume that explicitations will to
some extent be made by means of adversative connectors which
2
The connective phrase closest to the meaning of ellers when used pragmatically is y
eso que, which belongs to the spoken register (Flamenco García 2000: 3934).
cover the contrastive meaning of the relation, but which leave the
concessive meaning and the pragmatic value (i.e. the speaker‟s
comment) unmarked.
4. Finally, it is assumed that parts of the relational network of the
Spanish ST will escape the translators‟ attention, simply because the
relations are unmarked by means of connectors in the original text,
which, in this way, contains no direct visual stimulus to respond to.
As a result, we expect to find gaps in the inferencing of the logical-
semantic relations which form the argumentative structure of the ST.
2. The experiments
The assumptions above were tested in four experiments. Two experiments

focused on the translation product while two experiments also included the
translation process in order to shed light on the (conscious) inferencing and
the decision-making of the subjects in the translation process.
2.1. Experimental data on the not-unique ellers: translation products
To test the first assumption, an experiment was carried out with 19 MA

students and 34 BA students at the Copenhagen Business School. The
students were asked to translate a Spanish text written in a register with a
relatively low degree of complexity – a letter to the editor – into Danish.
The ST was translated by the subjects at home. The implicit ST relation in
example (1) is disjunctive:
(1)
Me preocupa escuchar, de boca de personas que siempre han sido solidarias
con la inmigración y con los desfavorecidos de este planeta, los comentarios
que últimamente estoy escuchando, que creo empiezan a rozar la intolerancia y
la xenofobia. (El País, 13 November 2005)
Danish translation. I den seneste tid er jeg blevet bekymret over at høre
mennesker som (ELLERS)3 altid har været solidariske med indvandrerne i
3
This sequence could be reformulated in Danish with ellers in initial position: I den
seneste tid er jeg blevet bekymret over at høre mennesker komme med udtalelser som
jeg mener tangerer intolerance og fremmedhad. Ellers har disse mennesker altid
været solidariske med indvandrerne i Spanien og de udstødte på denne jord.
130 Louise Denver
Spanien og de udstødte på denne jord, komme med udtalelser som jeg mener
tangerer intolerance og fremmedhad.
English [Lately it has grieved me to hear people who have always shown
solidarity with the immigrants and the outcasts of this world make
comments which I think border on intolerance and xenophobia.]
The product data are shown in Table 1.
Table 1. Product data. TT explicitations.

TT explicitations ST misinterpretations
ellers alternatives
MA students (19) 12 0 0
BA students (34) 8 1 0
(tidligere = „earlier‟)
Total (53) 20 1 0
As can be seen in Table 1, 20 out of 53 subjects opted for an explicitation

by means of ellers. The mean explicitation rate is 38 %, which supports the
assumption that translators tend to raise the level of cohesive explicitness in
their TTs. The difference between the explicitation rates of the two groups
is marked. It is as high as 63 % in the MA student group, but much lower in
the BA student group (23 %). Thus subjects with a generally higher level of
linguistic competence and translation experience clearly made more
explicitations than the less experienced group (cf. Blum-Kulka 1986). The
items ellers and altid („always‟) are not infrequently juxtaposed in Danish,
although it can hardly be said to be a collocation. This could perhaps, in
part, explain the very high explicitation rate among the MA students while
the markedly lower explicitation rate of the BA students could to some
extent be attributed to the generally more imitative approach to translation
characteristic of less experienced translators: a risk avoidance strategy. The
explicitation of ellers when used as a connector with disjunctive,
alternative or conditional meaning will not be addressed further below.
2.2 Experimental data on the unique ellers: translation products
To elicit product data on the use of the unique use of ellers in translations,
two experiments were carried out. The ST used in the first experiment
contained one implicit relation which could be made explicit in the Danish
TT by means of ellers. Fifteen MA students participated in this experiment,
the translation was performed at home and the translation brief contained
no information about the aim of this study, i.e. the subject of cohesion.
Example (2) is the sequence also found in the introduction, which is
repeated below for the reader‟s convenience:
(2)
Cruzado de brazos y con expresión entre irritada y suficiente, esperó a que le
despejaran el campo sin mover un dedo ni abrir la boca. ¿Quién mandaba allí
más que nadie? Él, allí y en el país entero. ¿Quién tenía un micrófono para
hacerse oír sobre el griterío e impedir los golpes? Ante él había varios. Habría
bastado una palabra suya, para que las agresiones hubieran cesado. Pero no
lo dijo, ni siquiera “¡Alto!”. (El País, 7 April 2003)
Danish translation (..) Et ord fra ham ville (ELLERS) have været nok til at
standse overgrebet. Men han sagde ikke noget, ikke engang “stop!”
English [With his arms crossed and a partly irritated, partly self-satisfied
expression he [José María Aznar] waited for the square to be cleared without
raising his arms or opening his mouth. Apart from him, who had the power
there, on that square, or in the entire country? Who had a microphone to drown
the shouting and stop the beating? In front of him there were several. One word
from him would have been sufficient to stop the aggression. But he did not
say anything, not even “stop!”]
The product data are shown in Table 2.
Table 2. Product data. TT explicitations

ellers alternatives
As can be seen from Table 2, not a single MA student produced an

explicitation by means of ellers. One subject opted for an explicitation by
means of the conjunction men („but‟), which shows that she inferred the
adversative relation, although the use of men in this context results in an
ungrammatical sequence because men in Danish is not recursive (neither is
pero in Spanish nor but in English). Four subjects succeeded in indicating
that the Spanish Prime Minister‟s attempt to stop the beating was minimal.
They did so by adding the adverb bare/blot („just‟) to et („one‟), which
132 Louise Denver
ensures that et is to be taken in its numeric sense („just one word‟) and not
as the indefinite article („a word‟).4
The ST used in the second experiment contained two implicit
relations of the same category. The text was translated by 30 MA students
at home and the translation brief contained no information about the subject
of cohesion. The first ST sequence is seen in example (3):
(3)
Los daneses rechazaron ayer integrar su moneda en el euro, dando la espalda a la
racionalidad económica, a su clase dirigente. El resultado tendrá repercuciones
negativas para Dinamarca, para un euro que atraviesa difíciles momentos, para el
conjunto de Europa.
Cuando el primer ministro Poul Nyrup convocó la consulta en marzo, el sí

parecía asegurado. La debilidad del euro, la subida del precio del petróleo, las
dudas sobre la preservación del modelo danés de protección social o los debates
sobre una Europa federal han incidido en contra de la entrada de la corona en
la moneda europea, pese a llevar pegada a ella o al marco alemán más de 18
años. (El País, 29 September 2000)
Danish translation. (..) Da statsminister Poul Nyrup Rasmussen udskrev valget

i marts måned, tegnede det (ELLERS) til et sikkert ja. (..)
English [Yesterday the Danish population refused to integrate Danish currency

into the Eurozone; in this way they turned their backs on economic rationality,
on their leading class. The result will have negative repercussions on Denmark,
on a Euro which is having a difficult time, on the whole of Europe.
When, in March, Prime Minister Poul Nyrup Rasmussen decided to hold a

referendum, the obvious outcome seemed to be a yes. The weak Euro, rising
oil prices, doubts about the preservation of the Danish welfare state or the
debates on a European Federation have contributed to the refusal to let the crown
join the European currency, although the Danish crown has been linked to it or
to the German Mark for more than 18 years.]
As can be seen from Table 3, two subjects opted for an explicitation, both
by means of the unique use of ellers.
4
This translation does not exclude the possibility of inserting ellers in the TT, but it
expresses the speaker‟s disapproval at the pragmatic level and, in this way, raises the
level of relational explicitness in the TT. Therefore, it has been included under
alternative ways of making an explicitation of the relational meaning.
Table 3. Product data. TT explicitations

ellers Adversative
connectors
The two sentence components of the contrast between the Danish no to join
the Eurozone and the expected yes at the time when the decision to hold a
referendum was made are not adjacent (i.e. discontinuous in the linear
sequence). An additional complicating factor is that the second sentence
unit initiates a new paragraph. Topological distance is assumed to increase
the cognitive load related to the inferencing of logical relations (Tirkkonen-
Condit 1993: 280). Consequently, it was to be expected that there would be
a lower degree of successful inferencing of the relation and a reduced
probability of explicitations in the TT. Ellers is used to bridge text
segments in Danish and, in fact, two subjects chose this solution. However,
no subject made the basic adversative meaning explicit by means of Danish
men („but‟), which could be due to the fact that men is rarely used to
introduce a new paragraph, but alternatives, such as imidlertid („however‟),
could have been used to mark the adversative relation in the TT. Although
the relation linking the two paragraphs was marked by two subjects only, it
is interesting to see that the next adversative ST relation was made explicit
by means of men by as many as nine subjects (When, in March, Prime
Minister Poul Nyrup Rasmussen decided to hold a referendum, the obvious
outcome seemed to be a yes. But the weak Euro, rising oil prices (..)). We
have no way of knowing to what extent this difference in the explicitation
rates can be attributed to the above-mentioned topological distance or the
basic character of the adversative relation compared to the adversative-
concessive relation.
The second sequence from this experiment can be seen in example
(4), where arguments in favour of a no and a yes to join the Eurozone are
contrasted:
(4)
Cuando el primer ministro Poul Nyrup convocó la consulta en marzo, el sí
parecía asegurado. La debilidad del euro, la subida del precio del petróleo, las
dudas sobre la preservación del modelo danés de protección social o los debates
134 Louise Denver
sobre una Europa federal han incidido en contra de la entrada de la corona en la

moneda europea, pese a llevar pegada a ella o al marco alemán más de 18 años.
Ingresar en el euro hubiera permitido a Dinamarca participar en las
decisiones, y no sólo sufrir las consecuencias de las medidas del Banco
Central Europeo o de los ministros de Finanzas de la eurozona. (El País, 29
September 2000)
Danish translation (..) Et dansk ja til euroen ville (ELLERS) have gjort det
muligt for Danmark at deltage aktivt i beslutningerne i stedet for blot at
affinde sig med de beslutninger som træffes af den Europæiske
Centralbank eller eurozonens finansministre.
English [When, in March, Prime Minister Poul Nyrup Rasmussen decided to

hold a referendum, the obvious outcome seemed to be a yes. The weak Euro,
rising oil prices, doubts about the preservation of the Danish welfare state or the
debates on a European Federation have contributed to the refusal to let the crown
join the European currency, although the Danish crown has been linked to it or
to the German Mark for more than 18 years. A Danish yes to join the Euro
would have permitted Denmark to participate in the decisions and not just
suffer the consequences of the policy measures taken by the European
Central Bank and the finance ministers of the Eurozone.]
As can be seen from the product data for example (4), no explicitation was
made by any subject of the basic adversative relation or the subtler
adversative-concessive relation.
Table 4. Product data. MA students: TT explicitations

ellers Adversative
connectors
However, it should be borne in mind that in the nine cases where an

explicitation was made of the preceding adversative relation by means of
men, the recursive use of this connector was ruled out. The subjects would
have had to use other lexical items to mark the relation in their TT.
In total, the product data on the explicitation of the adversative-
concessive relation by means of the unique use of ellers exhibit a very low
explicitation rate of less than 3 %. If we include alternative ways of raising
the level of explicitness in the TT, the overall explicitation rate was 9 %. In
comparison, the explicitation rate of the disjunctive relation by means of
non-unique ellers was as high as 63 % in the group of subjects with a
comparable level of language competence and translation experience, i.e.
the MA students. Even if account is taken of contextual factors favouring
the use of non-unique ellers in the first experiment, it can be concluded that
the empirical data strongly support the assumption that the level of
relational TT explicitness is markedly lower in contexts where ellers can be
used in a unique sense.
2.3 Experimental data on the unique ellers: translation products and

processes
The empirical data were elicited from two different experiments in which
the same Spanish ST was used. It is a long and difficult Spanish text
belonging to the genre of „literary journalism‟, commenting in a sarcastic
tone on the double standards of the Catholic Church. In two sequences, the
relation would ideally be made explicit by the unique use of ellers. It
should be noted that the original Spanish text was manipulated. In the first
example, the relation was marked in the original text by means of the
adversative connector pero („but‟), but it was deleted in the ST used for the
experiments (for experimental details on the first experiment, see Denver
2007: 225). In the second example, the relation was implicit in the original
text.
The first experiment was carried out with two groups of translators,
five professional translators with at least ten years of postgraduate
professional experience and seven MA students. A combination of
keystroke logging in Translog (Jakobsen 1999) and tape-recorded think-
aloud protocols (TAPs) was used to elicit data on the procedures followed
by the subjects during the translation process.
The second experiment was carried out with 22 MA students
attending a translation course in which text cohesion was an important part
of the curriculum. They were asked to translate the ST bearing in mind the
procedure outlined below. In phase I, they were asked to translate the ST
into Danish. In phase II, they were asked to read the Spanish ST very
carefully once again with the purpose of inferring the logical-semantic
relations across sentence boundaries. They were instructed to focus on
argumentative relations in their comment. In phase III, they were asked to
read their Danish TT once more with special reference to cohesion and add
connectors if they thought it would improve the text. They were given a
fortnight to perform the assignment at home.
136 Louise Denver
Below each ST example will be commented on separately with an

analysis of the empirical data on the translation product and process elicited
from both experiments.
2.3.1 Experimental data on example (5)

The Spanish ST is characterised by a sarcastic tone underlining the
speaker‟s general disapproval of the Catholic Church. In the ST sequence
related to example (5), the speaker objects to an eccentric bishop‟s answer
to some women who have approached him in order to complain about the
discriminatory treatment given to the victims of terrorism compared to that
given to the terrorists.
(5)
El noble prelado (..) respondió (..): “¿Y dónde está escrito que un padre deba
querer por igual a todos sus hijos?”.
En mi cualidad de hijo, y también de padre (..), esa pregunta suya me ha sumido

en graves cavilaciones. Hombre, si él, que sabe tanta teología, dice que no está
escrito en ninguna parte que los padres deban querer por igual a todos los hijos,
será porque no lo está. Cuando uno es hijo le gusta pensar que no es menos
querido que sus hermanos (..). Quizá si un padre alberga en su corazón esa
mezquindad debería esforzarse en no hacerla visible. Es posible que no haya
amargura más decisiva en una existencia humana que la de quien siente que no
fue querido por sus padres. (El País Semanal, 25 February 2001)
Danish translation (..) Som barn vil man (ELLERS) gerne tro at man ikke er
mindre elsket end sine søskende (..).
English [The noble prelate (..) answered (..): “And where does the Bible say that
a father must love all his children the same?”
As a son, and also as a father (..), his question has really given me food for
thought. Well, if he, who knows so much about theology, says that it is not
written anywhere that parents should love all their children the same, it is
probably because it is true. As a child you would like to think that you are not
less loved than your brothers and sisters (..). If a father is so mean-hearted,
perhaps he should try not to show it. It is possible that the greatest bitterness that
a human being can feel is that of not having been loved by his parents.]
(1) The first experiment

The explicitation data related to example (5) are shown in Table 5a.
Table 5a. The first experiment. Product data. TT explicitations
ellers Adversative
connectors
Professionals (5) 0 0 0
Total (12) 0 0 0
From the product data, it can be seen that all subjects understood the ST
correctly. Nevertheless, the semantic relation was not made explicit by any
of the translators. This is surprising, firstly because the adversative relation
– originally marked in the Spanish ST by means of pero („but‟) – seems to
be easy to infer and, secondly, because Danish – like other Germanic
languages such as Swedish and English – presumably has a preference for
the explicit marking of adversative relations.
The log files and the TAPs were examined in order to find out
whether or not the subjects had engaged in mental processing of the
logical-semantic relation linking the two sentence units together. Did the
students, before starting to translate the sentence, pause long enough to
allow them time for conscious mental activities concerning the inferencing
of the relation in question? According to the log files, the average pause
length of the professionals was 16 seconds – with considerable individual
variation (from 2 to 43 seconds). Two subjects exhibited short pauses of 2
and 3 seconds. In their case, the decision-making can be characterised as
automated: they followed routine procedures. The mental activities of the
remaining three professionals were to some extent explained in the TAPs:
apart from reading aloud the next ST sentence, initial pauses were mainly
occupied with processing the formulation of the initial clause: „When you
are a child‟. Support for this was found in the data in the log files
containing the subjects‟ keyboard activities. They showed that two
professionals rephrased their first literal translation into „As a child‟. Not
unexpectedly, the average pause length of the MA students was notably
longer, that is, 32 seconds (with individual variations from 10 to 78
seconds). This means that no MA student paused less than ten seconds,
which allowed them time for reflecting on the logical-semantic relation.
The process data show that the subjects had no problem understanding the
138 Louise Denver
ST sentence. Three MA students spent time reformulating the initial clause.

However, when it comes to the question of cohesion, the process data of all
the subjects, professionals as well as MA students, contain no trace of
mental activity in the form of verbalisations concerning the mental
processing of the semantic relation. The log files showed no keyboard
activities that revealed hesitation when direct transfers of the logical-
semantic relation were made to the TTs. And in no case had an
explicitation of the relation been deleted in a later phase during the
translation process.
(2) The second experiment

The second experiment was designed with the specific object of eliciting
process data on the subjects‟ ability to draw successful inferences of the
network of argumentative ST relations when specifically asked to focus on
textual cohesion. The data related to example (5) can be seen in Table 5b:
Table 5b. The second experiment. Product and process data

Phase I Phase II Phase III ST mis-
Explicitations Comments Explicitations interpretations
ellers Adversative ellers Adversative
connectors connectors
MA students (22) 0 3 4 0 3 0
The data show that three subjects made the adversative relation explicit in
their TT when first producing the translation, that is, when their attention
was not primarily directed towards the question of cohesion. None of them
commented on their strategic choice in phase II. Four subjects succeeded in
making a relational inference when they were explicitly asked to focus on
textual cohesion in phase II. Two of the four subsequently chose to revise
their TT, inserting a connector, and two chose not to make the relation
explicit in their TT (of the last two one thought that a connector would
make the TT “heavy”, while the second did not state any reason for making
a direct transfer). Finally, one subject decided to insert a connector when
revising his TT (phase III) without earlier (phase II) having commented on
the relation. In total, out of 22 subjects, eight – or about one third of the
subjects – succeeded in inferring the basic relational meaning and six
subjects chose to make it explicit in their TT.
Fourteen subjects – or about two thirds of the subjects – literally
transferred the implicit ST relation without any comment. In their case, it
would not be accurate to talk about (conscious) decision-making – or

strategic choice, which presupposes that alternative translations of an item
have been considered. These data support the indirect evidence found in the
process data from the first experiment. They suggest that there are gaps in
the inferencing of the network of the argumentative structure of long (and
complex) texts. Intuitively, this is not surprising. It would be reasonable to
assume that parts of the argumentative structure of (long) texts escape the
cognition of the reader. Apart from the relational category – some relations
being considered to be more basic to human thinking than others –
contextual factors in actual discourse seem to play an important role for the
inferencing of logical-semantic relations, e.g. topological distance, as
suggested by Tirkkonen-Condit (1993: 280). Other factors, such as
sentence length and complexity (syntactic and semantic), seem to be of
importance. It would be reasonable to assume that the higher the number
and complexity of the problem units to be addressed in the close context,
the higher the probability of a relation escaping the translator‟s attention.
Whatever the reason, we found that some ST relations are easier to infer
than others (and the number of explicitations higher). As mentioned above,
about one third of the subjects drew a successful inference and/or made an
explicitation of the basic adversative relation already commented on in
example (5). On the other hand, with respect to the causal relation between
the two last sentence units of example (5), it turned out that as many as
seventeen subjects – or 77 % of the subjects – succeeded in inferring the
relation and/or made an explicitation in their TT (If a father is so mean-
hearted, perhaps he should try not to show it. For it is possible that the
greatest bitterness that a human being can feel is that of not having been
loved by his parents.)
In four cases, the lexical items chosen by the subjects who made the
relation of example (5) explicit in the TT belong to the adversative
category: dog, imidlertid („however‟) and in two cases adversative-
concessive connectors have been used: alligevel, ikke desto mindre (the
English equivalent being „nevertheless‟). However, if we look at the
comment on the relation made by the same two subjects, they both
categorise the relation as adversative, one of them adding that the relational
meaning is that of „violated expectation‟. Furthermore, one of the subjects
who decided to transfer the relation directly, but made a comment on the
140 Louise Denver
relation, categorised it as adversative or concessive. In this way, the data

related to example (5) suggest some confusion as to the adversative-
concessive relation as such. Furthermore, they suggest that the adversative
relation is more prominent cognitively. Four of six explicitations left the
concessive meaning unmarked, not to speak of the pragmatic value, the
speaker‟s disapproval.
2.3.2 Experimental data on example (6)

(6)
A Gregorio Ordóñez, los buenos padres jesuitas de San Sebastián le negaron el
cobijo de sus sedes eclesiásticas para un funeral en el quinto aniversario de su
asesinato. Los padres (..) temían que se politizara impíamente el acto. Uno
pensaba que la muerte lo igualaba todo. Incluso después de muertos hay hijos
más queridos que otros. (El País Semanal, 25 February 2001)
Danish translation (..) Jeg troede (ELLERS)/Og her gik man og troede/og jeg
som troede at i døden er vi alle lige. (..)
English [The good Jesuits of San Sebastian refused to celebrate a requiem mass
in their church on the fifth anniversary of the murder of Gregorio Ordóñez. The
fathers were afraid that the act should be impiously politicised. I thought that in
death we are all equal. Even after we are dead, some children are more loved
than others.]
The subjective value that can be marked by means of ellers in example (6)
is of the same nature as that expressed in example (5). However, contrary
to example (5), example (6) allows for an alternative translation, namely an
explicitation by means of an idiomatic phrase Og her gik man og troede/og
jeg som troede („And here I was thinking/I thought/‟). Both translations
would mark the speaker‟s disapproval of the decision made by the Jesuits
in San Sebastian and have been included under the heading „explicitation‟,
since the level of explicitness of the TT is higher than that of the Spanish
ST. It should be noted that an explicitation of the adversative meaning in
the TT by means of the conjunction men („but‟) would result in an
ungrammatical sequence if the translator also decided to make the
adversative relation between the last two sentence units of the sequence
explicit by means of men (*But I thought that in death we are all equal.
But even after we are dead, some children are more loved than others.).
(1) The first experiment
Table 6a contains the empirical data for example (6).
Table 6a. The first experiment. Product data. TT explicitations.

ellers Idiom
Professionals (5) 0 3 0
Total (12) 2 4 1
As can be seen from the table, a total of six translators chose one of the two
types of marking in their TT. The explicitation rate was a little higher
among the professionals, where three chose to use an idiomatic phrase. Of
the seven MA students, one did not understand the ST sequence correctly.
Two of the six remaining MA students made an explicitation by means of
ellers and one used an idiomatic phrase. As many as eight subjects, in total,
chose to make the contrast between the last two sentence units of example
(6) explicit by means of Danish men (I thought that in death we are all
equal. But even after we are dead, some children are more loved than
others.). As mentioned, the adversative men is not recursive, so this
explicitation rules out the marking of the second last relation by means of
men. Three of those eight subjects chose to make the second last
adversative-concessive relation explicit by means of an idiom (two
subjects) or the unique ellers (one subject), while five subjects left the
second last relation unmarked in their TT.
What can the process data tell us about the inferencing and strategic
decisions concerning the transfer of the logical-semantic relation? As for
initial pauses, the log files show that the average length was more or less
the same in both groups – 16 and 18 seconds for professionals and MA
students, respectively – with wide individual variations in both groups
(from 3 to 40 seconds in the professional group and from 4 to 63 in the
MA-student group). One subject in each group paused less than four
seconds, or more or less the time it takes to read the sentence, which
indicates that these two subjects followed routine procedures. The
professionals had no problems with text comprehension. They understood
142 Louise Denver
that the initial ST item uno („one‟) should be taken in its generic sense as
I/you/we, not as a numeral with specific reference („one of the Jesuits‟).
And the concurrent verbalisation indicates that, in their case, the initial
pause was devoted to reading the next ST sentence aloud, oral translation
into Danish and reformulations before writing down their translation.
Furthermore, the log files show that the professionals stuck to their first
version; they made no corrections, except for one professional who made a
minor revision to his first version. In total, the translation process of the
professional group was characterised by a high degree of confidence and
the time was mostly taken up with questions concerning fluency.
By contrast, the process data for the MA students show signs of
uncertainty. They had problems with ST comprehension. In fact, it turned
out that, apart from the initial uno, which constituted a problem unit for two
MA students, the sentence contained two other problem units: namely, the
meaning of igualar („make equal‟) and the question of the reference of the
pronoun todo („all‟). They struggled with the formulation of their TT, but
this was to a great extent because they were not quite sure whether they had
understood the ST correctly. However, the two MA students who made an
explicitation by means of ellers did not hesitate. When the connector was
inserted, it was done without verbalisation or keyboard activities which
could be related to mental processing of the question. In fact, in no case do
the process data show evidence of mental activities which could be related
to the inferencing or translation of the logical-semantic relation. None of
the subjects corrected their first translation and never was a connector
which had once been inserted in the TT deleted in a later phase. In fact, the
data suggest that little or no conscious thought was given to cohesion
during the translation process.
(2) The second experiment

Table 6b contains the empirical data for example (6):
Table 6b. The second experiment. Product and process data

Phase I Phase II Phase III
Explicitations Comments Explicitations ST misinterpretations
ellers Idiom ellers Idiom
MA students (22) 1 5 2 0 0 6
It turned out that ST comprehension of the unit uno was a problem for as
many as six subjects in the second experiment. They interpreted the unit as
referring to a specific person, i.e. one of the Jesuits. All figures below refer
to the sixteen subjects who understood the ST correctly. Of the sixteen
subjects, six chose one of the alternative ways of expressing the relational
meaning, including that of the speaker‟s disapproval. Five subjects opted
for an explicitation by means of an idiom, one inserted unique ellers in her
TT. All explicitations were made when the subjects were first translating
the text without special focus on cohesion (phase I). They did not comment
on the relation (phase II), except for one subject who categorised the
relation as adversative. The same categorisation was made by another
subject who did not opt for explicitation in her TT. Unfortunately, she gave
no reason as to why she chose to make a direct transfer. This means that, in
example (6), only two subjects explicitly categorised the relation as
adversative. We assume here that when prototypical adversative connectors
have been used in the experiments to mark a relation, the subjects are
aware of the relational category even if the explicitations have not been
accompanied by an explicit categorisation of the relation. On the other
hand, it cannot be assumed that explicitations by means of an idiomatic
phrase necessarily presuppose the subjects‟ conscious awareness of the
relational meaning included in the idiom. In four cases, the use of an idiom
was not followed by a comment in phase II, so there is no way of knowing
for certain if these subjects were able to categorise the relation
metalinguistically. However, in the nine cases where no explicitation or
comment was made, we can conclude that the subjects failed to draw any
relational inference. These data support the assumption that there are gaps
in the inferencing of the argumentative network of the ST and that direct
transfers are far from always based on conscious decision-making.
As for the linguistic means chosen in the explicitations, no subject
used the adversative conjunction men. As regards the last relation of
example (6) (I thought that in death we are all equal. But even after we are
dead, some children are more loved than others), twelve out of the sixteen
subjects (i.e. 75 %) who understood the ST sequence correctly chose to
make this contrast explicit by means of men and, in addition, two subjects
categorised the relation as adversative, but chose not to make it explicit in
their TT. This is interesting, since it shows a high measure of successful
144 Louise Denver
inferencing and a marked preference for making the last adversative

relation explicit in this context. The rates concerning the two relations are
fourteen explicitations and/or categorisations of the last relation (87 %)
compared to seven explicitations and/or categorisations of the second last
relation (43 %). This marked difference could, at least in part, be attributed
to the prominence of the last relation in the ST. In this way, the data seem
to suggest that contextual factors in actual discourse are important for the
extent to which relations are successfully inferred by the translator and the
subsequent possibility of a strategic choice to make the relation explicit in
the TT.
3. Conclusions
In conclusion, it can be stated that the product data for example (1)
concerning the explicitation of the disjunctive relation which can be
marked by ellers in Danish supported our first assumption that the level of
relational explicitness would be raised in the Danish TTs, as predicted by
the explicitation hypothesis. We found a marked difference between the
explicitation rates of the two groups of subjects, the MA students and the
BA students – 63 % compared to 23 %. The subjects with a higher level of
linguistic competence and translation experience tended to make more
explicitations than the less experienced subjects. A similar tendency,
although less marked, was reflected in the data obtained on the
explicitation of the adversative-concessive relation from the two groups of
subjects who participated in the experiments involving ST example (5) and
(6), where the overall explicitation rates of the professional translators vs.
the MA students were 30 % and 23 %, respectively.5 Furthermore, it was
assumed that the rate of explicitations by means of ellers would generally
be low in contexts where the connector could be used pragmatically to
express the speaker‟s attitude in addition to the adversative-concessive
meaning of the relation. The empirical data for ST examples (2) to (6)
exhibited an overall explicitation rate by means of ellers of approximately
4 % in the MA-student group. Thus it would appear that the unique-items
hypothesis overrules the explicitation hypothesis. The assumption that
5
From the second experiment (see Tables 5b and 6b) only the figures related to phase
I, the translation phase, have been reported.
inferencing the adversative relation involves a lower cognitive load than

that involved in inferencing the adversative-concessive relation (and the
speaker‟s attitude at the pragmatic level) – an assumption which to some
extent would result in explicitations by means of adversative connectors –
was supported by the empirical data. The overall explicitation rate by
means of adversative and concessive connectors was (almost) 5 % and 4 %,
respectively.6 In any case, the process data from the second experiment (ST
examples 5 and 6) revealed some confusion as to the distinction between
adversative and adversative-concessive meaning.
As for the inferencing of logical-semantic relations, the process data
from the first experiment (ST example 5 and 6) showed that the same
procedural pattern concerning the mental processing of the question of
cohesion was followed by both groups of translators – professionals and
MA students. The initial pauses were not occupied with inferencing
implicit ST relations. When explicitations were made, they were not
accompanied by keyboard activities or verbalisation of strategic decision-
making. In addition, the log files showed that in no case had connectors
once inserted been deleted in a later phase. In this way, little or no attention
seemed to be consciously paid to the question of connection. What
distinguished the professional group from the MA students was a higher
degree of general procedural certainty. The professionals had a better
understanding of the ST, they followed routine procedures to a higher
extent, their translation process was characterised by shorter initial pauses
and fewer hesitations, and by and large they retained their first translation.
The process data from example (5) (the second experiment) showed that
three connectors were inserted in phase I, i.e. when the MA students
translated the ST without focusing on connection, while three were not
inserted until phase III. By contrast, in example 6, all six explicitations
were made in phase I. This difference could probably be explained by the
fact that five subjects translated the ST meaning at the pragmatic level by
means of an idiomatic phrase in example (6). In any case, we only found
two cases of metalinguistic categorisation of the relation in example (6)
compared to four in example (5). In total, the number of metalinguistic
comments on relational inferencing turned out to be rather low. Altogether,
6
The figure includes all the explicitations made in the TTs for examples (5) and (6)
from the second experiment.
146 Louise Denver
the majority of the subjects left no trace of explicitations and/or

categorisations of the two ST relations. In this way, the process data from
both experiments support the assumption that there are gaps in the
inferencing of logical-semantic relations across sentence boundaries and
that the translations of these relations are far from always the result of a
strategic choice.
Finally, the data suggest that the adversative-concessive relation
addressed in this study was not easily inferred. Compared to the purely
adversative relation linking the propositional content of the last sentence
pair in ST example (6), the number of categorisations and/or explicitations
of the adversative relation was very high, i.e. 87 % as compared with 43 %
for the adversative concessive relation. This difference could be explained,
at least in part, by the extent to which the relations stand out in actual
discourse. In fact, textual prominence seems to contribute to the successful
inferencing of logical-semantic relations.
References
Blum-Kulka, S. 1986. Shifts of cohesion and coherence in translation. In J.

House and S. Blum-Kulka (eds). Interlingual and Intercultural
Communication: Discourse and Cognition in Translation and Second
Language Acquisition Studies. Tübingen: Narr. 17-35.
Chesterman, A. 2007. What is a unique item? In Y. Gambier, M. Schlesinger &
R. Stolze (eds). Doubts and Directions in Translation Studies Amsterdam:
Benjamins. 3-13.
Denver, L. 2007. Translating the implicit: on the inferencing and transfer of
semantic relations. In Y. Gambier, M. Schlesinger & R. Stolze (eds).
Doubts and Directions in Translation Studies. Amsterdam: Benjamins. 223-
235.
Díez Prados, M. 2003. Coherencia y cohesión en textos escritos en inglés por
alumnos de filología inglesa (estudio empírico). Alcalá de Henares:
Universidad de Alcalá de Henares.
Englund Dimitrova, B. 2003. Explicitation in Russian-Swedish translation:
sociolinguistic and pragmatic aspects. In B. Englund Dimitrova and A.
Pereswetoff-Morath (eds). Swedish Contributions to the Thirteenth
International Congress of Slavists, Ljubljana, 15-21 August. Lund: Slavica
Lundensia Suplementa. 20-31.
Englund Dimitrova, B. 2005. Expertise and Explicitation in the Translation
Process. Amsterdam: Benjamins.
Flamenco García, L. 2000. Las construcciones concesivas y adversativas. In I.

Bosque and V. Demonte (eds). Gramática Descriptiva de la Lengua
Española. Madrid: Espasa. 3805-3878.
Garachana Camarero, M. 1988. La evolución de los conectores
contraargumentativos: la gramaticalización de no obstante y sin embargo.
In M. A. Martín Zorraquino and E. Montolío Durán (eds). Los marcadores
del discurso. Teoría y análisis. Madrid: Arco/Libros. 193-212.
Hansen (ed.). Probing the Process in Translation: Methods and Results
(Copenhagen Studies in Language 24.). Copenhagen: Samfundslitteratur. 9-
20.
Jensen, E. S. 2000. Sætningsadverbialer og topologi med udgangspunkt i de
konnektive adverbialer. In J. Nørgård-Andersen, P. Durst-Andersen, L.
Jansen, B. Lihn-Jensen og J. Pedersen (eds). Ny forskning i grammatik.
[New research on grammar] Fællespublikation 7. Pharmakonsymposiet
1999. Odense: Odense Universistetsforlag. 141-154.
Klaudy, K. 1998. Explicitation. In M. Baker (ed.). Routledge Encyclopedia of
Translation Studies. London: Routledge. 80-83.
Tirkkonen-Condit, S. 1993. What happens to a uniquely Finnish particle in the
processes and products of translation? In Y. Gambier and J. Tommola (eds).
IV Scandinavian Symposium on Translation Theory: Translation and
Knowledge, Turku, 4.6.1992. Turku: University of Turku. 273-284.
Tirkkonen-Condit, S. 2004. Unique items: over- or under-represented in
translated language? In A. Mauranen and P. Kujamäki (eds). Translation
Universals: Do they exist? Amsterdam: Benjamins. 177-184.
From Ántonia to My Ántonia:
Tracking self-corrections with Translog1
Brenda Malkiel
Months afterward, Jim called at my apartment one stormy winter

afternoon, carrying a legal portfolio. He brought it into the sitting-room with
him, and said, as he stood warming his hands,
“Here is the thing about Ántonia. Do you still want to read it? I
finished it last night. I didn’t take time to arrange it; I simply wrote down
pretty much all that her name recalls to me. I suppose it hasn’t any form. It
hasn’t any title either.” He went into the next room, sat down at my desk and
wrote across the face of the portfolio “Ántonia.” He frowned at this moment,
then prefixed another word, making it “My Ántonia.” That seemed to satisfy
him. (Cather 1918: 2)
Abstract
This paper reports on an experiment in which 16 first-year translation students

translated two Hebrew texts into English using Translog word-processing
software. The self-corrections recorded in the logs were categorized according
to the specific action taken, for example (a) self-corrections to grammar, (b)
self-corrections of meaning, and (c) instances in which the student typed a word
or phrase, deleted it, and retyped it verbatim. Analysis of the logs indicates that
beginning translation students have a professional attitude towards the
translation process and understand that being correct is not necessarily
enough. Based on the contrastive analysis of English and Hebrew, we would
expect that certain textual elements would create difficulties in Hebrew-English
translation, and by extension, that translators working from Hebrew into
English would make self-corrections to these areas. Indeed, we found that
approximately 20% of total self-corrections were in the areas predicted to be
difficult for this language pair.
1
I join my colleagues in thanking Arnt Lykke Jakobsen for his generosity in making
Translog available to researchers around the globe. My personal thanks go to Miriam
Shlesinger for introducing me to the workings of the “black box.”
150 Brenda Malkiel
Revision
Revision is an integral part of writing – and of the type of writing which is

translation. Revision involves a series of self-corrections, instances in which the
writer or translator makes an addition, a deletion, or a change to the text. Here
the term “self-correction” is used to refer to both online and end revision. A
self-correction is not necessarily a change from incorrect to correct, but can
involve a subtle alteration, such as from Ántonia to My Ántonia.
Researchers in Translation Studies have asked, for example, whether
there is a correlation between the difficulty of a source-text chunk and the
amount of revision it generates (Campbell 1991); when, how, and why
translators revise their texts (Shih 2006); and whether certain features of the
revision process are indicative of translation quality (Breedveld 2002a). Both
Jensen (1999) and Jakobsen (2002) compared the effort their subjects expended
on drafting as opposed to end revision, with Jensen focusing on individual
revisions and Jakobsen on translation time. One of Jakobsen‟s findings was that
the student translators produced more revisions during the drafting stage than
did the professionals. The author interprets this as “indicating much greater
uncertainty in this group” (202).
This is not surprising: The literature is replete with evidence that
translation students suffer from insecurity. While self-confidence is
“fundamental to effective and successful translating” (Hönig 1991: 88),
translation students demonstrate self-doubt in the classroom (Kussmaul 1997:
246-7), on exams (Toury 1992: 68), and when they participate in empirical
studies. Jääskeläinen (1989) reports that one first-year subject participating in a
think-aloud protocol (TAP) study constantly checked the meanings of words she
already knew. Fraser (1996), who underscores the similarity between her
findings and those of Jääskeläinen, observes that translation students “become
paralysed” (247) upon encountering an unfamiliar word or phrase, and therefore
are highly dependent on bilingual dictionaries. Livbjerg & Mees (1999) relate
that the subjects who took part in their TAP-Translog study had so little self-
confidence that they devoted a good deal of attention to problems for which
they already had a solution.
Translation students have good reason to be concerned about the quality
of their target texts. According to Toury (1979), interlanguage is a law of
translation: “The analysis of thousands of pages translated into Hebrew and
Tracking self-corrections with Translog 151
English allows me to claim that virtually no translation is completely devoid of

formal equivalents, i.e., of manifestations of interlanguage.” Toury goes on to
assert that this phenomenon is “in striking analogy to what we know about
second-language learning” (226).
Weinreich (1953) defines interference phenomena as those “instances of
deviation from the norms of either language which occur in the speech of
bilinguals as a result of their familiarity with more than one language, i.e. as a
result of language contact” (1). Grosjean (1989) extends the definition of
interference to include written language. Interference is considered to be a
normative feature of bilingual language use (Hoffmann 1991), one which may
be affected by factors such as stress and fatigue (Dornic 1978).
While interference is a frequent subject of investigation in both second-
language acquisition and Translation Studies, there is a fundamental difference
in approach. Mauranen (2004: 66) points out that while in second-language
acquisition interference is generally understood as working from L1 to L2, in
translation the source language (most often L2) influences the target language
(generally L1). In other words, a bilingual is largely immune to interference
when speaking or writing in L1, but highly susceptible when translating into L1.
The language learner‟s errors are both regular and consistent (Corder
1981: 66); hence, contrastive analysis can be used to predict where a language
learner will go wrong (Selinker 1992: 100). If interlanguage is a translational
universal, if certain language errors are predictable, and if by its very nature text
translation involves revision, then it is possible that certain revisions are
predictable. A classic example would be that of false cognates. Given that false
cognates are a recurrent source of word-level interference for translators and
language learners alike (Anderman 1998: 43), we would anticipate that on
occasion translators will mistranslate a false cognate; we would also anticipate
that sometimes they will recognize their mistake and revise their translation.
Less often, translators will correctly translate a false cognate and then “correct”
themselves to create an error where none had been (Malkiel 2005: 106).
According to the literature, particularly challenging textual elements
include neologisms (Lehrer 1996); culture-bound words (Lehmuskallio et al.
1991); voids (Landsberg 1976); terms that are lexicalized in the target but not
the source language (Shlesinger 1992), henceforth referred to as “lexicalizable
strings”; idioms (Baker 1992: 63-78); adjectives (Malmkjær 1993); and
prepositions (Pedersen 1988: 121). It would stand to reason that these elements
152 Brenda Malkiel
would be a magnet for self-corrections, regardless of language pair. This,

however, is also a function of the particular source text – a text studded with
neologisms should generate more self-corrections to neologisms than to idioms.
Expected self-corrections in Hebrew-English translation
Based on the contrastive analysis of English and Hebrew, we would expect that
certain textual elements would create difficulties in Hebrew-English translation,
and by extension, that translators working from Hebrew into English would
make self-corrections to these areas.
In English the adjective precedes the noun and in Hebrew it follows the
noun. If the translator working from Hebrew into English does not notice
that the source-text noun is modified by an adjective, she will first type
her translation of the noun and then insert the adjective.
English offers a far greater variety of tense and aspect than Hebrew.
When working into English, the translator is apt to try out different
options before settling on one.
Hebrew and English share the same marks of punctuation, but have very
different rules for the comma.
Hebrew has no capital letters. When working out of Hebrew, the
translator must decide which words to capitalize. Although most
capitalization is a matter of correct/incorrect, certain cases do require
decision making.
The translation of prepositions can be unusually troublesome because of
the repetition of the preposition in Hebrew. The normative Hebrew
sentence Dan halax la’doar, la’makolet ve’la’yarkan, for example, is
literally translated into English as Dan went to the post office, to the mini-
market, and to the fruit store, but the more normative English form would
be Dan went to the post office, the mini-market, and the fruit store. When
translating from Hebrew to English, the translator will sometimes
translate the preposition each time it appears in the source text, and then
revise the target text to eliminate the extraneous prepositions;
alternatively the translator might translate the preposition once, and then,
influenced by the source text, revise the translation by adding
prepositions.
Because there is no indefinite article in Hebrew, the indefinite article can

be a source of difficulty.
Certain important concepts take the plural in Hebrew but the singular in
English, for example water (mayim), life (xaim), and sky (shamaim). A
native Hebrew speaker might say “the water are”; it is unlikely for a
native English speaker to make the same mistake.
Categorization of self-corrections
This paper reports on an experiment in which 16 beginning translation students

translated two texts using Translog word-processing software, with half
working into their L1 and half into L2. In an effort to optimize ecological
validity, the experiment had no think-aloud component and therefore all insights
regarding the revision process are based on the logs. The self-corrections
recorded in the logs were categorized according to the specific action taken, for
example (a) self-corrections to grammar, (b) self-corrections of meaning, and
(c) instances in which the student typed a word or phrase, deleted it, and retyped
it verbatim. This project did not analyze whether the self-corrections improved
or detracted from the translation.
The categorization scheme emerged from the data rather than from a pre-
established list. The categorization and classification process was to a great
extent subjective, and it is certainly within the bounds of possibility that another
researcher would have broken down the data differently. Nonetheless, a list of
discrete self-corrections is of little value – analysis required categorization.
Two points illustrate how much decision making went into classification
and categorization:
Prepositions are difficult regardless of language pair, and at times the
students revised their translations by deleting the original preposition and
substituting another (arrive to / arrive at).2 These self-corrections appear
in Table 1 (Correcting the Target Text). Self-corrections which involved
adding or deleting extra prepositions – in other words, those self-
corrections to prepositions that we would predict to find in Hebrew-
English translation – were included in Table 2 (Refining the Target Text).
2
In each example from the logs, the original translation appears in Roman and the self-
correction in italics, separated by a slash: more and more of our time / more and more time.
154 Brenda Malkiel
Although one might expect that all self-corrections to prepositions would

belong in one category, this is not necessarily true for this particular
language pair.
After lengthy deliberation, self-corrections of punctuation were placed in
Table 2. Based on my own experience in translation and translator
training, I believe that most often a change to punctuation is intended to
hone the target text rather than to correct an actual error. [Ex post facto it
is clear that Tables 1 and 2 would have been imbalanced even had self-
corrections to punctuation been included in Table 1.]
Method
Subjects
The subjects (N = 16) are members of the first-year translation classes at Beit
Berl College, Kfar Saba, Israel and Bar-Ilan University, Ramat Gan, Israel.
Both institutions offer two-year programs in Hebrew-English and English-
Hebrew translation. The admission process at both involves an entrance exam
and an interview.
Eight of the subjects are native English speakers and eight native Hebrew
speakers. The subjects compose a heterogeneous group in terms of birthplace
(Israel, the United States, Canada, and Great Britain), age (early 20‟s to early
70‟s), and education (from high school to M.A.). Some of the subjects had some
very limited experience in translation before entering the program, but none had
studied translation or worked professionally in the field.
Materials
Texts
This experiment is based on two Hebrew texts with a similar length
(approximately 330 words) and a similar form, an op-ed piece ostensibly written
for an Israeli newspaper, to be translated for an American one. Each contains
ten Hebrew-English false cognates, five lexicalizable strings, and five culture-
bound expressions.
Disks
Each student received a disk with Translog User and four blank source-text
files, two for the trial translations and two for the actual translations. (The trial
translations will be discussed below.) In truth, the students could have done all
four translations using the same source-text file, but it was felt that they would
find it psychologically reassuring to have four separate files, especially since
this was their first exposure to Translog.
Instruction sheets
The instruction sheet set out in detail how to use the software and listed the
various dos and don‟ts, e.g. they could use any dictionaries or reference books
of their choice but were not to do a rough draft.
Procedure
At the start of their first year of training, the students were instructed in how to
load, use, and save files on Translog. They were then given an envelope
containing all the materials they would need for the experiment.
All sixteen students translated the same two source texts. For variability,
however, half first translated Text A and half started with Text B. They were
given hard copies of the texts in separate sealed envelopes in the hopes that this
would discourage them from examining the second text before translating the
first. The students translated the texts on Translog on their own computers,
either at home or at the office. The experiment itself had two sessions, one per
text, and the students could choose the day and time they found most
convenient, provided the two sessions took place approximately a week apart.
Each session began with a trial translation, where the students translated
two or three sentences from a newspaper or book into English. The purpose of
the trial translation was to allow the students to become comfortable with how
Translog looks and works before they began the task of translating the source
texts. The second session took place one week after the first. The second session
was identical in format to the first, but this time the students translated the
second text.
Data analysis
With the Translog replay feature, the researcher can “watch” the translator in
action, as she types, deletes, pauses, moves the cursor, etc. A self-correction
156 Brenda Malkiel
was identified every time a student either deleted something she had written or
moved through the translation to insert a character, word, or phrase.
When the student went on to correct the self-correction, each change was
counted as a self-correction:
all kinds of / many sorts of / many kinds of – two self-corrections
and mayonnaise / and even mayonnaise / and over less mayonnaise / to mayonnaise /
including mayonnaise – four self-corrections
When one change demands another change of the same type, this was counted
as a single self-correction, as with:
Commas: where Rachel the biblical Mother is buried / where Rachel, the biblical
Mother, is buried
Plural: this social dynamic / these social dynamics
In certain cases a student changed more than one element of a phrase. Here each
change was counted as a self-correction. For example:
to solve / to find ad hoc solutions – two self-corrections
which are considered healthy / which are supposed to be good – two self-corrections
After the self-corrections were identified, they were divided into categories and
then into two umbrella classes, those presumably intended to correct an actual
or perceived error in the target text (Table 1) and those that fine-tune the target
text (Table 2).
Findings
The eight native English speakers and eight native Hebrew speakers each
translated two texts on Translog, generating a total of 32 logs. Statistical
analysis of the data (two-tailed t-tests) showed no significant effect for either
source text or mother tongue. The fact that there was no significant effect for
source text indicates that the two source texts were evenly matched. Translation
into L1 versus translation into L2 (here referred to as “directionality”) is
addressed below in the Discussion section.
Because there was no significant effect for source text or directionality,
the 1257 self-corrections in the 32 logs were treated as a single group. Tables 1
and 2 present the categories of self-corrections with two examples of each.
Table 1. Correcting the Target Text

Type of self-correction Examples from the logs Instances
Insertion of an omitted in spite the fact / in spite of the fact 69
element bestseller / idiotic bestseller
Substitution of one appear in / appear on 45
preposition for another starting with / starting from
Meaning dramatic rise / dramatic decrease 44
hoards / hordes
Grammar excessive aggressive behavior / 7
excessively aggressive behavior
clothes that was so / clothes that were
so
Total 165
These 165 self-corrections include, inter alia, changes to the translation of false
cognates (tights / stretch pants), the retranslation of polysemous words found in
the source text (undershirt / tank top), and the substitution of one homonym for
another (are / our). Even given the variety of self-corrections included in Table
1, actual corrections to the target text account for only about 13% of total self-
corrections. The remaining 87% relate to the students‟ efforts to hone their
translations (see Table 2).
Synonyms – those instances when the student restates a word or phrase –
account for more than half of all self-corrections. This category includes not
only a simple substitution of noun for noun or adjective for adjective (silly /
idiotic), but also the addition of a clarification (impossible deadlines /
impossible deadlines to meet) and the reshuffling of elements (We once used to
/ Once, we used to). The 158 self-corrections to punctuation include changes to
sentence breaks and upper/lower case as well as the addition or deletion of
commas, hyphens, quotation marks, colons, semi-colons, dashes, and
parentheses. The remaining 249 self-corrections include everything from the
change from the passive to the active voice (are being chosen / are choosing) to
pluralization (this social dynamic / these social dynamics).
In addition to the self-corrections presented in Tables 1 and 2, the
subjects made 156 changes to the spelling of a word. These changes are not
included in the calculation of total self-corrections. Typos shed little or no light
on the translation process, but are merely a reflection of the subject‟s
competence at the keyboard, and it is difficult, if not impossible, to distinguish
between self-corrections of typos and self-corrections of spelling mistakes.
158 Brenda Malkiel
Coupled with this, self-corrections of spelling are almost entirely irrelevant,

since had the subjects been working on a word-processing program with a spell-
check feature, presumably they would have used the spell checker after
completing their translation.
Table 2. Refining the Target Text

Type of self-correction Examples from the logs Instances
Replacing a word or registration / enrollment 685
phrase with a cause irreversible damage to our blood
synonymous one vessels / cause our blood vessels
irreversible damage
Punctuation Once we were / Once, we were 158
Universities / universities
Articles in the country clubs / in country clubs 80
exclusive neighborhoods / the
exclusive neighborhoods
Tense and aspect we replaced / did we replace / have we 59
replaced
is / will
False starts and retyping athletic cl / athletic styles 57
the identical word or The State of Education Today / The
phrase State of Education Today
Singular to plural and an ad hoc solution / ad hoc solutions 37
vice versa country clubs / country club
Adding or deleting an in the Stone Age or the Middle Ages / 16
extra preposition in the Stone Age or in the Middle Ages
with Agnon‟s novellas and with
Shakespeare‟s dramas / with Agnon’s
novellas and Shakespeare’s dramas
Total 1092
Together, the two source texts contain 20 Hebrew-English false cognates,

10 lexicalizable strings, and 10 culture-bound expressions. Table 3 presents the
number of self-corrections made to these elements.
The logs contain a similar number of self-corrections to lexicalizable
strings and false cognates; there are no self-corrections to culture-bound terms.
Table 3. Self-corrections to Lexicalizable Strings, False Cognates, and Culture-bound

Terms
Type of self-correction Instances
Lexicalizable strings 16
False cognates 15
Culture-bound terms 0
Total 31
The 251 self-corrections made to the areas predicted to be problematic in

Hebrew-English translation are presented below in Table 4.
Table 4. Language-pair-predictable Self-corrections

Type of self-correction Instances
Insertion of an adjective before the noun 62
Tense and aspect 59
Commas 57
Capitalization 43
Addition or deletion of an extra preposition 16
Indefinite article 11
Changing “life are” to “life is” 3
Total 251
The majority of these self-corrections came from the insertion of the adjective,
changes to tense and aspect, and changes to punctuation. Approximately 20% of
total self-corrections are predictable in Hebrew-English translation.
Discussion
According to conventional wisdom, translators produce higher-quality

translations when they work into their mother tongue. The jury is still out with
regard to the effect of directionality on the translation process. The language
used in the literature is exceedingly tentative (emphasis added):
Translation into L2 and L1 “are assumed to differ, at least partially”
(Krings 1987: 161).
“Both the social and cognitive models of translation processes suggest
that similar skills are necessary to produce an adequate translation into
one‟s mother tongue and into one‟s second language” (Kiraly 1995: 109).
“translation into a second language is likely to be more difficult than into
the first language” (Campbell 1999: 34).
160 Brenda Malkiel
An analysis of the target texts is beyond the purview of this study; nonetheless I
would comment that it is immediately apparent (and not surprising) that the
translations of the native English speakers were far superior to those of their
classmates working into L2. As regards the translation process, statistical
analysis of the self-corrections showed no significant effect for directionality. It
is possible that the native speakers‟ superior ability in the target language was
somehow balanced out by the non-native speakers‟ command of the source
language. Perhaps the students working into L1 held themselves to a higher
standard than their classmates translating into L2. It is also possible that
directionality plays a more important role in the revision process of professional
translators and even advanced-level translation students. As Séguinot (1991)
notes, classroom time has an important effect on revision strategies.
Based on the principles of contrastive analysis, the universality of
interlanguage, and the fact that text translation by definition involves revision,
we would anticipate that certain revisions would be predictable. This proved to
be the case: 20% of total self-corrections occurred in those areas posited to be
difficult in Hebrew-English translation. The finding that certain self-corrections
are predictable lends support to the claims put forth by Selinker (1992) and
Corder (1981).
It is important to underscore that the translator is not subject to
interference from the source language per se, but from the source language as
represented in a particular source text or texts. The source texts used in this
study contained numerous lexicalizable strings, false cognates, and culture-
bound terms. We would predict that these elements would pose a challenge for
the translator and in fact the logs contain 16 self-corrections to lexicalizable
strings and 15 to false cognates. In contrast, the students did not make a single
self-correction to the culture-bound terms.
The source texts contained, e.g., a reference to the Israeli Nobelist S.I.
Agnon; the expression shishi-shabbat (Friday-Saturday), which is the Hebrew
equivalent for the weekend; and the names of two upper-crust suburbs of Tel
Aviv. The fact that the students deliberated about matters such as punctuation
and prepositions but made no self-corrections to culture-bound elements
indicates that they do not yet appreciate the interrelationship of culture and
translation and the need to mold the target text to meet the needs of the target
audience.
And in a similar vein: The unit of translation is said to be an indication of

proficiency (Lörscher 1996: 30; Bernardini 2001: 249). All 16 subjects on
occasion moved the cursor to insert the adjective before the noun. This
demonstrates that, at least some of the time, they were translating at the level of
individual words rather than phrases or sentences – evidence of their status as
beginning translators.
It seems self-evident that some revisions are more far-reaching than
others, i.e. that the decision to replace “lecturers” with “graduates” is a more
radical change than the addition of an Oxford comma. It is striking that the
lion‟s share of self-corrections in the logs were not strictly speaking corrections
at all. In fact, there were more self-corrections that involved replacing a word or
phrase with a synonymous one than all other self-corrections combined. We
found that the students devoted a great deal of time and energy to refining their
translations, making changes that had little or nothing to do with right and
wrong. Thus a contradictory portrait emerges from the data. These students
make no self-corrections to culture-bound terms and will translate at the level of
the word, but nonetheless seem to view translation as a decision-making
process.
An understanding of the difficulty inherent in translation is one stamp of
a good translator. Indeed, some of the most highly regarded professional
translators are tremendously self-critical. The type of deliberations reflected in
the logs indicates that beginning translation students understand some of the
challenges of translation. In this, they have a mature attitude towards the
translation process, where being correct is not enough. Above we discussed the
idea that insecurity is common among translation students. The energy devoted
to refining the target text is evidence that beginning translation students lack
self-confidence precisely because they have a rudimentary appreciation of what
professional translation entails.
Concluding remarks
Because TAPs are so time-consuming to administer and code, most TAP studies
examine only a small group of subjects, who generally translate a single short
text. Another problem is that the data from TAPs can be difficult to categorize
and analyze, and the researcher runs the risk of overinterpreting individual
utterances (Breedveld 2002b: 98). In addition to the fact that a Translog study
162 Brenda Malkiel
provides a far better facsimile of the “authentic” translation process than a TAP
study can, thanks to the relative ease of data analysis, Translog experiments can
have more subjects, more texts, and longer texts than is customary in TAP
studies.
It was my hope that the methodology used in the present study would
provide more information about the revision process than we can obtain from a
global calculation of keystrokes. The additional insight, however, comes at a
great cost. While the researcher can obtain objective measures of time and
keystrokes from Translog with the click of the mouse, an analysis of self-
corrections requires a tremendous expenditure of time and labor. Hand-in-hand
with this, the researcher is forced to categorize and classify the data. In terms of
both its subjectivity and the time commitment it involves, this study shares
some of the drawbacks of TAP studies.
It is almost de rigueur to close with suggestions for future research, and it
is tempting to propose a larger study, one which investigates the types of self-
corrections translators make as well as the relationship between self-corrections
and text quality. What types of errors are caught during online as opposed to
end revision? Do self-corrections actually improve the text? How many
mistakes in comprehension or grammar survive the self-correction process? But
given the methodological drawbacks of the present study, I would suggest that
such a study be undertaken only as a collaborative effort.3 If several researchers
were to work together in designing the study and defining categories of self-
corrections, they would be able to conduct a large-scale, multi-lingual
experiment comparing students to professionals, examining the differences
between translation into L1 and translation into L2, and determining the effect
of source and target languages. This would certainly help us to better
understand the nature of self-corrections as well as the translation process itself.
References
Anderman, G. 1998. Finding the right words: Translation and language teaching. In K.
Malmkjær (ed.). Translation and Language Teaching: Language Teaching and
Translation. Manchester: St. Jerome. 39-48.
Baker, M. 1992. In Other Words: A Coursebook on Translation. London: Routledge.
3
For a detailed treatment of collaborative research in Translation Studies,
see Göpferich (2008: 254).
Bernardini, S. 2001. Think-aloud protocols in translation research: Achievements,

limits, future prospects. Target 13 (2): 241-263.
Breedveld, H. 2002a. Translation processes in time. Target 14 (2): 221-240.
Breedveld, H. 2002b. Writing and revising processes in professional translation.
Across Languages and Cultures 3 (1): 91-100.
Campbell, S. 1991. Towards a model of translation competence. Meta XXXVI (2/3):
329-343.
Campbell, S. 1999. A cognitive approach to source text difficulty in translation.
Target 11 (1): 33-63.
Cather, W. 1918. My Ántonia. Reprinted in 1988. Boston: Houghton Mifflin
Company.
Corder, S.P. 1981. Error Analysis and Interlanguage. Oxford: Oxford University
Press.
Dornic, S. 1978. The bilingual‟s performance: Language dominance, stress, and
individual differences. In D. Gerver & H.W. Sinaiko (eds). Language
Interpretation and Communication. New York/London: Plenum. 259-271.
Fraser, J. 1996. Professional versus student behavior. In C. Dollerup & V. Appel
(eds). Teaching Translation and Interpreting 3: New Horizons. Amsterdam /
Göpferich, S. 2008. Translationsprozessforschung: Stand – Methoden – Perspektiven.
Translationswissenschaft 4. Tübingen: Narr.
Grosjean, F. 1989. Neurolinguists, beware! The bilingual is not two monolinguals in
one person. Brain and Language 36: 3-15.
Hoffmann, C. 1991. An Introduction to Bilingualism. London: Longman Linguistics
Library.
Hönig, H.G. 1991. Holmes‟ “Mapping Theory” and the landscape of mental
translation processes. In K.M. van Leuven-Zwart & T. Naaijkens (eds).
Translation Studies: The State of the Art: Proceedings of the First James S
Holmes Symposium on Translation Studies. Amsterdam / Atlanta: Rodopi. 77-
89.
Jakobsen, A.L. 2002. Translation drafting by professional translators and by
translation students. In G. Hansen (ed.). Empirical Translation Studies: Process
and Product. Copenhagen: Samfundslitteratur. 191-204.
Jääskeläinen, R. 1989. The role of reference material in professional vs. non-
professional translation: A think-aloud protocol study. In S. Tirkkonen-Condit &
S. Condit (eds). Empirical Studies in Translation and Linguistics. Joensuu:
University of Joensuu. 175-200.
Jensen, A. 1999. Time pressure in translation. In G. Hansen (ed.). Probing the Process
in Translation: Methods and Results. Copenhagen: Samfundslitteratur. 103-119.
164 Brenda Malkiel
Kiraly, D.C. 1995. Pathways to Translation: Pedagogy and Process. Kent, Ohio /
London, England: Kent State University Press.
Krings, H.P. 1987. The use of introspective data in translation. In C. Færch & G.
Kasper (eds). Introspection in Second Language Research. Clevedon /
Kussmaul, P. 1997. Comprehension processes and translation. A think-aloud protocol
(TAP) study. In M. Snell-Hornby, Z. Jettmarová, & K. Kaindl (eds). Translation
as Intercultural Communication: Selected Papers from the EST Congress –
Prague 1995. Amsterdam / Philadelphia: John Benjamins. 239-248.
Landsberg, M.E. 1976. Translation theory: An appraisal of some general problems.
Meta XXI (4): 235-251.
Lehmuskallio, A., Podbereznyj, V. & Tommola, H. 1991. Towards a Finnish-Russian
dictionary of Finnish culture-bound words. In S. Tirkkonen-Condit (ed.).
Empirical Research in Translation and Intercultural Studies. Tübingen: Narr.
157-164.
Lehrer, A. 1996. Problems in the translation of creative neologisms. In B.
Lewandowska-Tomaszczyk & M. Thelen (eds). Translation and Meaning, Part
4. Maastricht: Universitaire Pers Maastricht. 141-148.
Livbjerg, I. & Mees, I.M. 1999. A study of the use of dictionaries in Danish-English
translation. In G. Hansen (ed.). Probing the Process in Translation: Methods and
Results. Copenhagen: Samfundslitteratur. 135-147.
Lörscher, W. 1996. A psycholinguistic analysis of translation processes. Meta XLV
(1): 26-32.
Malkiel, B. 2005. The transition from “natural” to professional translation: The
contribution of training and experience. PhD dissertation, Bar-Ilan University.
Malmkjær, K. 1993. Who can make nice a better word than pretty?: Collocation,
translation, and psycholinguistics. In M. Baker, G. Francis & E. Tognini-Bonelli
(eds). Text and Technology: In Honor of John Sinclair. Philadelphia/Amsterdam:
John Benjamins. 213-232.
Mauranen, A. 2004. Corpora, universals and interference. In A. Mauranen & P.
Kujamäki (eds). Translation Universals: Do They Exist? Amsterdam /
Pedersen, V.H. (1988). Essays on Translation. Copenhagen: Erhvervsøkonomisk
Forlag S/I.
Séguinot, C. 1991. A study of student translation strategies. In S. Tirkkonen-Condit
(ed.). Empirical Research in Translation and Intercultural Studies. Tübingen:
Narr. 79-88.
Selinker, L. 1992. Rediscovering Interlanguage. London / New York: Longman.
Shih, C. Y-Y. 2006. Revision from translators‟ point of view. Target 18 (2): 295–312.
Shlesinger, M. 1992. Lexicalization in translation: An empirical study of students‟

progress. In C. Dollerup & A. Loddegaard (eds). Teaching Translation and
Interpreting: Training, Talent and Experience. Amsterdam/Philadelphia: John
Benjamins. 123-127.
Toury, G. 1979. Interlanguage and its manifestations in translation. Meta XXIV (2):
223-231.
Toury, G. 1992. „EVERYTHING HAS ITS PRICE‟: An alternative to normative
conditioning in translator training. INTERFACE: Journal of Applied Linguistics
6 (3): 60-72.
Weinreich, U. 1953. Languages in Contact: Findings and Problems. New York:
Linguistic Circle of New York.
Typos & Co.
Ricardo Muñoz Martín
Abstract
An orientational, quantitative analysis was undertaken of 44 log files of

different tasks carried out by four subjects. They were examined with a
view to attention and typos, since it was thought that these phenomena
could indicate attentional lapses. Results suggest that typos might be a
useful means of profiling subjects and discriminating between different
tasks (translation, self-revision, revision of another person’s text). Other
findings point to the possibility of computing intervention sequences to
measure recursiveness and the effects stress may cause on typing and
translating.
1. Introduction
Many of the studies on translation processes that have collected data with
computer programs such as Translog, Camtasia, and KGBSpy have
concentrated on problem solving and, more often than not, time has been
the main criterion for the analysis. Consequently, translation problems have
operationally been associated with pauses of a certain length. But time
analysis has its limitations, since we find intersubject variation as a result
of differences in both mental processing speed and typing skills.
Translation process protocols, such as those of Translog, provide rich
information, so there might be other ways of studying them to obtain
valuable insights into the activities and the subjects‟ mental processes. This
paper is a preliminary attempt to try out one possible way to do so. In order
to test a feature which is frequently overlooked, this research focuses on
typos, since mistypings might not only be motivated by feeble typing skills,
keyboard size and the like, but also by attentional lapses.
Typing consists of a strict serial execution of movement sequences,
coordinated by spontaneous pattern generation (Kelso and Schoner 1988),
168 Ricardo Muñoz Martín
rather than by following pre-established generalized motor programs

(Schmidt 1975) and time-controlled internal representations (Gentner
1982), since perception and action are taken to be reciprocally causal
(Lombardo 1987). The decision to type a word pre-activates the typing of
its letters, which are sequentially and mutually inhibited in such a way that
the first letter is the strongest at the beginning of the sequence (Rumelhart
and Norman 1982). The extent of the influences of each letter on the hand
and finger it directs depends on the extent of the activation of the letter
(Rumelhart et al. 2004: 81). Logan (1982) found that typists took about
200 ms to stop typing after a signal instructing them to do so. The delay
usually involved entering one or two additional letters, because there is a
point of no return once the movement has been commanded (around
166 ms). Some overlearned sequences, such as the one for the word the,
may surpass this limit and get fully typed (Schmidt and Lee 2005: 180-
181). But “the actions do not appear to be structured with feedback to
„verify‟ that a certain finger movement has been made before the next one
is commanded” (Schmidt and Lee 2005: 178). Thus, typists rely on visual
monitoring to discern whether what they entered was correct. To do so,
they need to divert their attention, especially when they are not touch
typists.
Attention is awareness of the here and now in a focal and perceptive
way, i.e. it is a state of current awareness. Attention is not a single entity,
but a set of mental processes which interact with each other and with other
brain processes when carrying out perceptual, cognitive, and motor tasks.
People cannot focus on everything, so instead they choose a subset of
events in phenomenal experience, and attention can be intentionally
aroused to prioritize stimuli which satisfy certain needs. Typists perceive
limited cues in the environment and select stimuli according to what they
think is important, and on the basis of their experiences, their knowledge,
and their expectations. Thus, focused and selective attention serve
coherent, goal-directed behavior. When attention needs to be maintained
over a long period of time, typists enter in a state of sustained attention, or
vigilance. Sustained attention keeps control over long-term goals and
depends on the processing capacity of the subjects. Selective and sustained
attention seem to be opposing processes so that when focused attention
increases, sustained attention decreases (Parasuraman 2000: 7). Translators
Typos & Co. 169
need to switch behaviors with different cognitive requirements

continuously throughout their tasks, an ability which has been described as
alternating attention (Sohlberg & Mateer 1989: 129). Alternating attention
involves working memory processes and may improve through training.
The level of attention arousal varies with the changing demands of a
task. When attention-drawing events are frequent, performance improves.
Typists may also invoke full series of actions without being aware of them
when they have been trained or over-learned. To do so, typists establish
very low sensitivity thresholds for certain kinds of stimuli, which are thus
attenuated (basically, unattended) but can nevertheless activate the
perceptual systems. Current perceived experience is encoded into patterns
of expectation, and awareness is aroused when these expectations are
violated, so a minimum of focal attention must remain. Hence, assessing
the adequacy of signals is a key step in the process because it entails
shifting the level of arousal and devoting more attentional resources to it.
Since processing capacity is limited, stimuli compete for attention.
People can successfully monitor many sources of information at once, as in
the case of translating, but when one of the sources comes into focus,
attention to the other ones is drastically lowered (Duncan 1980). Highly
demanding tasks reduce spare cognitive resources, and at peaks they may
dramatically lower attention to monitoring the environment. Time pressure
and task complexity may lead to performance failure in sustained attention.
Attention is not always controlled, and unexpected experiences may draw a
subject‟s attention involuntarily. Furthermore, attentional lapses may also
be caused by “internal” factors, such as having a sudden idea.
This brief account of typing and attention points to the possibility
that typos might hint to situations where cognitive resources have been
reallocated to support other mental activities, such as evaluating and
problem solving. Since translating and revising are likely to entail different
combinations and proportions of cognitive activities, interventions (see
Section 3), and typos in particular, might show consistent patterns for
different tasks. If the hypothesis were to be confirmed, future software
applications might be developed which are able to detect what kind of text
it was and also who the author might have been. On the other hand, since
interventions are modifications in the draft to approximate the subjects‟
desired state of the final text, they might correspond with each subject‟s
particular view on quality and/or on the task itself, and thus be more or less
consistent from one text to the next. This means that they might also be
considered independently of the texts. Typing skills and cognitive styles
tend to differ between subjects, so typos might be used to characterize
them. These were the main hypotheses of this work: that categorized
actions, and typos in particular, might be capable of doing so. As an
orientational study, it was open, however, to focus on other findings.
2. Materials and methods
Four advanced translation students at the University of Granada School of

Translation and Interpreting were requested to revise three drafts translated
from English into Spanish by a person other than themselves (“other-
drafts”). They also translated four original English texts into Spanish, and
then revised them.
Subjects were enrolled in an advanced translation workshop on texts
for mass media, and they were all 22 years old when data were collected.
Andrés and Bruno are male, Carla and Diana are female (fictive names),
and all of them are right-handed. Subjects were profiled with the indexes
for verbal comprehension (in Spanish), working memory capacity and
processing speed of the WAIS III test (administered Jan. 18, 2007), and
also with the sections on structure/grammar and reading of the paper-based
TOEFL test (administered Jan. 23, 2007), as presented in table 1.
Table 1. Subjects‟ profiles in the WAIS III and TOEFL tests
Andrés Bruno Carla Diana
Verbal comprehension 109 100 106 112
Working Memory 116 100 129 104
Processing Speed 125 114 84 120
Structure (English) 95 85 83 95
Reading (English) 92 86 71 95
Andrés and Carla frequently read creative literature and also read
newspapers virtually on a daily basis. Bruno is especially interested in
localization, and his typing skills are better than those of the rest. Diana
considers herself bilingual, her father being a British national, and was the
only subject with no previous training in translating against the clock.
Typos & Co. 171
Original texts were all taken from The Economist and are still available
online at the time of writing. Codes for original texts and other-drafts,
titles, word counts, and dates of publication, translation, self revision, and
other-revision are shown in Table 2.
Test sessions were carried out in regular, two-hour long class
meetings, and took place in the usual classroom, which had computers with
Windows XP, Office 2003, Translog 2000 and free access to the Internet
during the trials. Subjects were asked (a) to try to complete their draft
translations in one hour, and (b) to revise their own translations in 30
minutes. Revisions on other-drafts were also allotted 30 minutes. Time
constraints were established to cause stress in the subjects, in the hope that
their behavior would be affected in a way that might yield more
information.
Table 2. Codes, titles, word counts and publication dates of source texts and
dates of experiments
Translations and self- Other-revisions
revisions
ST Draft
Code Titles Publ. Session Session
words Words
T1 Politics this week 465 11/11/06 11/16/06
T2 Without prejudice 470 11/11/06 11/23/06
T3 South Africa rethinks 544 11/11/06 12/07/06
T4 Powering up (1st part) 490 09/14/06 01/11/07
R1 All creatures great and 453 10/26/06 514 11/16/06
small
R2 Eyeing up the collaboration 549 11/02/06 586 12/15/06
R3 Powering up (2nd part) 507 09/14/06 556 01/11/07
Data were collected by means of Translog 2000. Originals to be translated

had been entered into the program. Revisions were performed in Translog
2000 as well by pasting translations from MS-Word into the program at the
beginning of the task. Other-revisions were performed as fresh
simultaneous looks at originals and drafts, i.e., subjects accessed both texts
at the same time. All originals and other-drafts were also made available as
RTF documents. Test session arrangements and mechanics had been
explained and trained in previous class meetings.
Translog coded texts were recoded, as shown in the examples, by
means of a Word 2003 macro, in order to process them with WordSmith
Tools 5.0 and also to ease reading. Every draft was analyzed and coded
separately by two analysts in two steps to ensure inter-rater consistency.

This was especially important for missed phenomena, i.e., for anomalies in
the copy (as perceived by analysts) which might be considered problems
missed or skipped by the subjects.
3. Concepts and terms
Since not all of the subjects‟ activities on written portions of texts can be
thought of as corrections, interventions were defined as “interruptions of
the typing stream followed by any keyboard activity not aimed at adding
information to the draft after the rightmost point” (the text being
metaphorically viewed as a single line proceeding smoothly from the
beginning of the text to its end.) Interventions usually consist of deletions
and/or additions. They will appear isolated now and then but, usually, a
deletion and a subsequent addition (or vice versa) constitute a pair of
interventions applied to the same linguistic unit. However, the distinction
between single and double interventions does not seem productive.
Therefore no difference was established between isolated deletions or
additions, and combinations of deletion+addition in this study, and the term
intervention will here be used to mean both kinds.
A difficult point was that tasks and subjects might perhaps be better
characterized by what they did not show or do than by their interventions.
To account for that, missed phenomena (MPs) were defined as “an anomaly
which should have prompted a subject to act on an already written text
segment, as inferred from the data by the analyst.” Missed phenomena were
envisioned as missing interventions, and they were categorized and
computed accordingly. The next section is devoted to categorizing
interventions and missed phenomena.
4. Categories of interventions and missed phenomena
Interventions and missed phenomena as defined above were classified into

several categories. Since the study focuses on typos, the categorization
scheme adopted is more detailed than those for phenomena of other kinds.
Table 3 provides a chart summarizing them. The first criterion was adopted
to enable us to distinguish between phenomena exclusive to translating and
Typos & Co. 173
those which might be shared by other types of writing. Only two categories
were considered to be translation-specific: (1) interventions and missed
phenomena where subjects added or could have added information present
in the original; and (2) cases where subjects changed their minds as to the
meaning of a source text segment. Here, they are reported together as T
phenomena. When subjects are translating, most traces of their behavior
and mental activities are probably related to translating, but only in these
two cases did there seem to be an obvious link to the original text. All other
categories are shared with other kinds of computerized drafting and/or
writing, and their relationships to the translation task and the original text
need to be inferred by the analyst. Hence, what is translation-specific is the
cause or explanation, not the observed behavior.
Table 3. Summary and examples of categories

No. Example Categ.
1 {×}{«1},_•••••el_ programa_gener{«5}del_ordenador_genera_
T
2 [04.20]port•ador•es_ del_•VIG{«1}H{«18}con_SIDA•••_)
3 de_Sudáfrica,_pero_está_pla{«4}n_plan•eando_•••
4 •••sol•icitan_un_seguro••,••_[29.01]{«2}_pero_que_•••mueren_
SL
5 [11.14]vot•ó_•legalizar•_l•os{«13}la_legalizaci•ón_de_los_
6 _de_Pusan_•Nationa_Un{«3}l_University{-26}_la_•Universidad_de
7 {×} _asciende_u {«1} a_unos •• _tres_millones_de_rands_ • _( {«3} _ •
L
([05.12]9 •••• {«1} •• 9 •• {«1} 392 •••• . • 87 • 0 ••• € • )
8 personas_que_qui{«3}[04.69]quieren_co•ntrat{«7}seleccionar._
9 m•ísm{«3}ismos•_
10 NO_CU••{«5}no_cubre_[08.31]{»2}
ILC
11 ••ass••o{«4}Ass•ociaton{-2}i
12 proteje{«2}ger_sus_
13 ••••absorbe••_•un_for{«1}••t•ón_ 1W
14 d•e_mae{«1}teriales_más_ 1M
15 {+2}{×}{«3}os_paneles_soalres_{«6}••lares_ 1K
16 [05.30],_crear_una_corriente_ec{«1}••l•éctrica 1+
17 del_prime_{«1}r 1S
18 ••••el_mercer{«2}ado_de_energ•ía_ 1?
Codes: • ½ second pause, [longer pauses], {«erase left, # of spaces}, {x}
mouseclick,{+move right, # of spaces}, {sc-# Shift+Ctrl, move left # of spaces}, etc.
Secondly, a distinction was made between interventions aimed at

modifying infralexical (IL), lexical (L), and supralexical (SL) phenomena.
This was relevant to contrast the interaction of typos with other kinds of
interventions.
SL phenomena comprised cases such as grammatical changes and
adjustments, such as changing noun gender, number, verb person, etc., to
match it with another word or to make the text conform to some
grammatical norm, such as time/tense correlations (example 3); additions
and/or deletions of punctuation marks (example 4), where blank spaces and
punctuation marks were taken to belong to the linguistic unit they were
attached to; and syntactic changes which also implied replacing at least one
full word (example 5).
Lexical phenomena (L) covered those related to proper names, e.g.
sanctioned or customary translations for institutional names or acronyms
(e.g. NATO/OTAN), transliteration (Mahmoud Ahmadinejad/Mahmud
Ahmadineyad), and those derived from miscopying a name intended to look
exactly the same as in the original (example 6); phenomena related to
weights and measures, units and conversions (example 7); all other
instances where subjects simply substituted one word for another and did
not imply a variation in the interpretation (example 8).
As for infralexical phenomena (IL), a distinction was made between
sanctioned phenomena, i.e. corrections (ILC), and typos. Sanctioned
phenomena included accents (example 9); uppercase instead of lowercase
(example 10);1 lowercase instead of uppercase (example 11); and other
orthographic phenomena (example 12). The first three (IL) categories were
singled out because they involved an additional mechanical effort to
combine two keys in one stroke, but no detailed account will be offered
here. Moreover, accents have minimal visual saliency and most native
Spaniards, including some professional writers, have problems with the
rules governing their use. Typing lowercase instead of uppercase letters can
often be attributed to mechanical reasons, but the opposite cannot, since
pressing and holding the caps key is deemed intentional, and may thus be
due to other reasons, e.g. interference from the original text (except when
accidentally hitting the caps lock key, which usually leads to capitalizing
more than just one letter).
1
Note that uppercase and lowercase key values do not only correspond to letters, as
shown in example 1 (see also standard Spanish qwerty keyboard layout in figure 1):
d•e_1••50[03.17]ª••{«1}º{«1}.
Typos & Co. 175
Typos were classified according to their behavioral/mechanical

nature into six groups: 1W, when subjects pressed a neighboring key
instead of, or together with, the targeted one, such as pressing 3, 4, w, r, s
or d instead of letter e (example 13; see also Fig. 1); 1M, for phenomena
where subjects simply missed one or more keys, including the space bar, as
in example 14. 1K included cases where keys were typed in the wrong
order or where the correct key was wrongly repeated (example 15); 1+
comprised cases of pressing a non-neighboring key (example 16). 1S
referred to introducing a blank space in the middle of a word, or between a
word and a punctuation mark when the space bar was not next to the
targeted key (example 17). Finally, 1? covered all other IL non-correction
phenomena not attributed to any of the preceding categories (example 18).
Figure 1. Neighbor keys for letter e on Spanish

qwerty keyboard.
Of course, phenomena are interpreted by the text analyst, who sometimes

will have no problem assigning a cause to an intervention but in other cases
may be clueless as to the motivation for it. This explains why L and SL
categories are so broad. Furthermore, the principle was established that the
simplest explanation was always to be preferred. Hence, cases where
subjects might have mistyped a word or changed their minds when writing
another word, were interpreted as typos, and not as L, unless fully justified
by the co-text. Note that, in order to modify a typo or a spelling mistake,
subjects will often erase many words until they reach the point where they
want to intervene, fix their target, and then retype what they erased to get to
the point of intervention. These cases will count here as IL phenomena.
Also, now and then, subjects‟ interventions do not correspond directly to
the phenomenon. For instance, in 19 the subject mistyped the word
prohibitiva (“prohibitive”) but, when he moved to fix it, he replaced it by

extremadamente cara (“extremely expensive”):
(19) [03.33]prohibitivca{«12}••estremed{«2}adamente_cara
This might point to cases where different options were competing in the
subject‟s mind and either caused a drop in attention before he intervened to
change his preference, or to situations where an accidental typo gave the
subject more time to decide between two options while fixing the typo.
5. Results and discussion
5.1 Differences between tasks
Subjects always intervened much more and had many more missed
phenomena (MPs) when translating than when they revised texts (Table 4).
They also always intervened more in other-revisions than in self-revisions,
but their MPs were higher in self-revisions than in other-revisions. This
might have a cognitive explanation. People do not translate texts but their
interpretations, so self-revisions may have demanded shallower mental
processing, since subjects had already formed their interpretations when
composing their drafts. Once an interpretation has been reached, translators
seem to focus less on the signal that prompted it. On the other hand, in
other-revisions subjects were forming their interpretations while carrying
out the task, and so they might have been able to spot more (surprising)
phenomena. Thus, other-revisions might be considered a sort of middle
ground, since subjects intervened more in the texts than when they revised
their own and, crucially, were also more successful. In any case,
categorizing and contrasting interventions and MPs might be a possible
way to differentiate the three tasks. The difficulty of the original text might
motivate systematic variations in the number of interventions, as evidenced
by the higher rates found for both interventions and MPs in T3, which
Bruno, Carla and Diana seem to have found particularly difficult to render.
So, this potential characterization should in principle always correlate with
the difficulty of the original text.
Typos & Co. 177
Table 4. Average interventions and MPs in other-revisions, translations, and self-

revisions by text
Other-revs Trans Self-revs

Text Int MPs Int MPs Int MPs
1 35 4 137 27 29,3 13
2 37,25 6,75 153,75 24 15,3 12
3 39 5 218 37 22,3 23
4 167,25 33 19,5 14
Table 5 displays draft averages for the number of interventions and MPs
divided into typos and the rest of the categories (NT: ILC+L+SL+T).
Subjects were also clearly far better at fixing typos than non-typing related
(NT) phenomena in all tasks. This was probably so because there is a
bottom-up bias which may draw attention to local phenomena while
searching for information (Sagi & Julesz 1984). Also, research on visual
attention has shown that searching for targets defined by individual features
is much less effortful than searching for targets defined by conjunctions of
features (Treisman & Gelade 1980). Furthermore, the status of typos as
mistakes is not usually contested, even by translation teachers, clients and
addressees, who will overlook them when they are few in number, though
this was not the case with the teacher in this class. Interventions on typos
ranked highest in draft translations (nearly 52 %), and were even slightly
higher than on NT phenomena. This result was expected, since subjects
need to type far more text when translating than when revising, and the
chances of mistyping were therefore much higher. Also as before,
intervention rates on both typos and NT phenomena were higher in other-
revisions than in self-revisions. This is probably due to the fact that
subjects had already performed many interventions when translating. In
self-revisions, differences between typos and NT interventions were very
small. A possible explanation for this is that subjects—advanced translation
students—had already developed a sense of correctness which was not only
connected with grammar and orthography. Therefore they would not
distinguish between phenomena on those grounds and would try to fix
everything, thereby achieving a comparable rate. This could lead to a
hypothesis that there might be a tendency to find systematic rates of MPs
for translators, something which deserves to be checked. Non-typing
related MPs in other-revision and self-revision tasks were similar, which

seems to support this explanation.
revisions, divided into typos (T) and other phenomena (NT=ILC+L+SL+T)
Interventions Missed phenomena

Other- revs Translations Self revs Other- revs Translations Self-revs
Text T NT T NT T NT T NT T NT typos NT
1 7.5 28.5 81.2 65.5 9.5 8.75 1 3 4.5 31.5 3.5 4.7
2 8 29.25 80 78.5 0.5 4 0 6.7 4.75 24.5 0.25 4.7
3 8.6 32.6 111.3 123.3 3.3 3.3 0 6.8 4.7 39.4 2.25 7
4 96.5 78.2 3.2 3.25 6 37 0 6.2
5.2 Subjects’ profiling
Andrés‟ average interventions were usually the lowest in other-revision and

translation tasks (Table 6). In other-revisions, he also managed to miss the
fewest phenomena in two of the three texts. When translating, Andrés
performed the fewest interventions of the four students, but he had the
highest rate of MPs in all texts except T1. Bruno‟s interventions and MPs
were above average in other-revisions, but when translating and self-
revising they were always below average and his MPs were the lowest.
Carla‟s interventions in other-revisions tended to be the fewest, but she
missed very few phenomena in two of the three texts. On the other hand,
Carla was always above average in her translation MPs. Diana intervened
way above average in all of her other-revisions and translations, with poor
to average results. Also, when she revised her own work, Diana was below
average in both interventions and MPs except in T4.
revisions by subject.
Other-revs Trans Self-revs

Subject Int MPs Int MPs Int MPs
Andrés 25.00 3.33 123.75 35.25 21.25 14.50
Bruno 39.33 5.33 147.50 20.50 24.50 17.50
Carla 30.33 4.33 165.25 34.75 21.75 16.50
Diana 53.67 8.00 239.50 30.50 18.75 12.25
Curiously, subjects seemed to miss more phenomena when they performed

more interventions in a particular text. This is particularly noticeable in
Typos & Co. 179
other-revisions R2 and R3, in online changes in T1, and in self-revisions in

T2 and T4. No explanation can be provided for this as yet, but tentatively
we can toy with the notion that subjects were aware of their own
weaknesses in each task, and worked more (but unsuccessfully) to
compensate for this, for too much attention devoted to the execution of a
well-learned skill can disrupt performance. Andrés is a case in point.
Although he was aware of the mechanics of the session (where revision
immediately followed the translation) and he also knew that only the final
draft (that is, after self-revising it) would be checked for class purposes,
Andrés always read his drafts a second time and intervened to improve
them before he closed the log session. He did so in a peculiar way: he
placed the cursor at the beginning and moved it along the lines (sometimes
also backwards) as he was reading, probably searching for IL phenomena
(his weakest point, see below). When questioned about it, he stated that he
always behaved that way and was surprised to learn that other students did
not proceed that way. This might be an example of a compensatory strategy
to overcome a problem.
Data show that subjects‟ behaviors tend to be consistent within each
task, and different from those of other subjects throughout the three tasks.
Categorizing and computing interventions and MPs, then, might help to
profile subjects.
revisions, divided into typos (T) and other phenomena (NT=ILC+L+SL+T), by
subject
Subject Interventions Missed phenomena
Other- revs Translations Self-revs Other-revs Translations Self-revs
T NT T NT T NT T NT T NT T NT
Andrés 5.00 21.33 78.25 51.50 4.25 8.25 0.67 2.67 13.25 25.75 1.25 7.75
Bruno 8.00 31.33 71.75 83.75 3.50 5.50 0.33 5.00 0.75 32.00 1.50 6.75
Carla 4.33 26.00 85.00 91.75 5.25 3.75 0.33 4.33 4.25 44.50 2.25 4.50
Diana 14.67 39.00 133.50 113.25 4.25 2.25 0.00 8.33 3.75 33.50 1.50 5.25
Table 7 shows average interventions and MPs divided into typos and other
phenomena (NT), by subject. In other-revisions, Andrés tended to intervene
below average on both typos and NTs (his MPs were the lowest in R2 and
R3). When translating, Andrés‟ interventions on typos were below average
in T1, T2, and T3, but he had the largest numbers of MPs in the four texts.
For NT phenomena, his interventions were always the lowest and so were
his MPs, except for T3, the “difficult” text. When he revised his own drafts,
he tended to intervene above average on typos and to miss fewer of them,
but for NT his interventions and MPs were the highest in three of the four
texts. Hence, data portray Andrés as an efficient reviser of somebody else‟s
drafts, who will mistype a lot while he is translating but will have the
fewest problems of other kinds. He would not fix many typos online but,
when revising, he concentrated on them at the expense of missing other
phenomena. No explanation can be offered for Andrés‟ high rate of typos.
He was questioned about it a long time after the experiment, and his views
on translation might play a role, since he considered that translating has to
do mainly with meaning, and thought the final physical product to be of
lesser interest. Nevertheless, he was thorough with regard to other-
revisions, so the question remains open.
Bruno made almost no typos in other-revisions, but for NT he
intervened above average in R2 and R3, and his MPs were above average
in R1 and R2. When he was translating, he had to fix fewer typos than
anybody else and had nearly no MPs of this kind. He tended to intervene on
NT phenomena, and his MPs were below average as well, except for T1. In
self-revisions, Bruno nearly always had the lowest number of interventions
and MPs with respect to typos (he had almost no MPs, except in T1), but
with NT phenomena, half the time he had the highest number of
interventions and also many MPs in half of the texts. Bruno‟s meticulous
drive and typing skills are demonstrated in all tasks (he was the only one to
bother about double blank spaces between words), but he had more
problems with NT phenomena, perhaps due to his limited “world
knowledge” (he missed more translation-specific phenomena than the other
subjects).
Carla was very efficient at fixing typos in other-revisions, since she
left nearly none, but she also intervened very little, and for NT phenomena
her interventions were also below average, although she had more MPs
than the average in R2. She also intervened below average on typos in all
of her translations except T2, but had more MPs in half of the texts. As for
NT phenomena, she tended to intervene above average but had the highest
number of MPs in all translations except in T2. In her self-revisions, she
intervened more than the other subjects on typos in two of the texts, and her
MPs were above average in two texts as well. NT interventions and MPs
Typos & Co. 181
were the lowest in two texts, and below average in yet another text in each
category. In brief, Carla was good when she worked on other-translations.
When translating, it seems that she had to pay a price for her poor
command of English, and perhaps because of this, when she revised her
own translations, she concentrated on typos, but was not very successful.
Diana had the highest number of interventions in two other-revisions
and the third was above average too, but while she simply left no typos, her
NT MPs were the highest in R1 and R2. She also intervened more than
anybody else on typos in the four translations and left MPs below average
in three of the four texts. For NT phenomena, she also had more
interventions than the rest in three of the four texts, but now she was only
below average in T3 and T4. When she revised her own drafts, she
intervened above average in T1 and T2, and below in T3 and T4, and left
few MPs in three of the texts. Diana‟s interventions on NT phenomena
were always below average, and MPs were also below average in T1 and
T3. All subjects except Diana had participated in another semester
workshop the year before, where they also had to translate against the clock
for an hour once a week, although the targeted word count was smaller.
This might explain why Diana mistyped more frequently than her
classmates. Diana‟s performance improved throughout the sessions, so she
seems to have undergone a process of adaptation to working against the
clock during the tests. This interpretation is supported by the fact that typos
were higher for all subjects except Bruno in the first text. This was
probably due to stress, for at least two reasons: (1) all students in the class
had to translate against the clock one hour per week, but the first test
session was near the beginning of the semester; and, crucially, (2) they had
been informed in advance that this time their work would also be used for
research purposes.
6. Other findings
6.1 Typos in sequences when translating
Non-translation interventions had a correlation coefficient with typos of

0.999. There seemed to be a positive correlation between typos and the rest
of the phenomena in the case of Bruno and Carla, and a negative
correlation with Andrés. This was especially noticeable in online

interventions (when translating), where typos seemed to appear very often
combined with other interventions. In order to account for this, intervention
sequences were defined as “sets of consecutive, uninterrupted interventions
performed on related complete or incomplete linguistic units already
typed.” These units usually belong to the same text segment. To illustrate
the concept of sequence, let us consider example 20. In this example, the
subject was typing aparición when she changed her mind, and (a) replaced
it with fundación, but she then (b) deleted it to type creación instead. She
also mistyped creación and (c) corrected it, and finally introduced a blank
space at the end, which she (d) deleted to insert a comma. These actions
were taken to be a sequence with four interventions: L+L+1M+1S.
Arrows mark the beginning and end of each sequence:
(20) [08.02] A_menos_de_un_año_desd•_{«1}e_su_

apar•c{«5}funda{«5}cra•c{«2}ea•ción_{«1},•_
As with isolated interventions, sequenced interventions may happen either

at the rightmost point of a draft (local interventions) or else in previous
segments (backward interventions). The minimal text segment that can be
used to distinguish local from backward interventions are syntagmas, on
the assumption that subjects usually hold at least a full syntagma or phrase
in their working memories at any one time. Thus, an intervention on a
linguistic unit (letter, word) removed from the rightmost point of insertion
but within the same text segment (syntagma) is considered a local
intervention as well. Sequences were computed as to their length (by
number of interventions) and a distinction was also made between
sequences which did not contain any intervention on infralexical typos but
had interventions at other levels; those which solely contained typos; and
those which contained both kinds. Isolated backward interventions in the
middle of a sequence were here considered to break the sequence.
Backward sequences were computed independently from previous
interventions in the same text segment, whether isolated or within other
sequences, since there were often no text cues to decide whether subjects
were returning to an unsatisfactory solution or had changed their minds
Typos & Co. 183
with respect to that solution in the light of new information of later

passages in the original text or their renderings.
There were 2970 online interventions on draft translations included
in the study, out of which 1515 happened in sequences as defined above
(51 %). Typos had the highest figures (956 free, 881 sequenced), and SL
phenomena had the highest relative proportion of sequenced interventions
(544 sequenced vs. 392 free).
Out of a total of 523 sequences, 338 comprised two interventions
(64.5 %), 99 had three (19 %), 31 had four (6 %), 20 had five (4 %), and
the rest were above five (6.5 %), up to 18 interventions. Data show that the
most usual sequence contained two interventions, one of which was on an
IL phenomenon, whether sanctioned (subject to norms) or not.
Investigating whether fixing a typo led to another type of intervention, or
focusing on a phenomenon made subjects drop their attention and mistype
something and fix it, was beyond the possibilities of this study. A
preliminary qualitative analysis shows that the longer the sequence, the
higher the possibilities that subjects were dealing with translation
problems.
As for their composition, 126 sequences contained interventions only
on IL phenomena, 65 only on phenomena of other kinds, and 206 mixed
interventions on both IL and other phenomena. Qualitative appreciation of
the series shows that sequences which did not have any typos are very good
candidates for signaling translation problems, but so are mixed sequences,
for many typos seem to occur when subjects divert their attention from
typing to problem solving.
From the point of view of the subjects, Diana had the largest ratios
(630 sequenced, 410 free), and Andrés the lowest (250 sequenced, 294
free). Bruno had 300 sequenced interventions vs. 339 free, and Carla, 335
sequenced and 412 free interventions. Interestingly, subjects who were
used to working against the clock had very close rates, between 45 % and
47 %. This point definitely deserves further research.
When typos were analyzed with respect to their nature, 1W (pressing
a neighboring key) turned out to be the most usual (196 free; 192
sequenced), followed closely by 1M (missing a key or keys: 171 free, 155
sequenced), then by 1K (typing keys in the wrong order: 123 free, 112
sequenced). As expected, 1S (introducing a blank space: 27 free, 25
sequenced) was the lowest kind. Average percentages of free vs. sequenced
interventions were very close in all categories, so the distinction between
typos in sequences turned out to be uninteresting except for differences
between subjects (Table 8).
Table 8. Free (F) vs. sequenced (SQ) typos by subject and type
Andrés Bruno Carla Diana
F SQ F SQ F SQ F SQ
1W 65 39 34 29 62 52 35 72
1M 41 35 43 29 23 12 64 79
1K 27 14 23 27 25 21 48 50
1S 7 6 4 5 2 6 14 8
Andrés‟ sequenced interventions were usually below average, but his

interventions on free 1W phenomena were the highest. Bruno had the
lowest free and sequenced interventions in half of the texts, but was above
average in free 1K interventions. Carla had the fewest 1M interventions on
both counts, but all her 1W interventions were above average. Diana had
the highest rates in nearly all kinds and circumstances but, curiously, her
free 1W interventions were the lowest. The highest proportions of typos in
sequenced interventions were also interesting, since they fell on 1M for
Andrés, 1K for Bruno, 1S for Carla, and 1W for Diana. Sequence length
might also hint at learning processes, for Diana had the longest sequences
on average, one of them with 23 interventions. As for statistical correlation
coefficients, the most interesting one matched 1K with SL interventions, at
0.993. The sample was too small to conclude that any of these facts are
significant, but they should be taken into account in future research.
Sequenced intervention indexes might provide a complement for
recursiveness (Buchweitz and Alves 2006).
6.2 Peculiarities associated with typos
Differences between typo categories did not seem to yield very interesting
results, but data raised some questions which deserve further consideration.
The first is whether typos should also be categorized according to other
criteria which might be more relevant to cognitive approaches. Some
peculiarities have emerged from our data which support that notion. This is
the case with typos performed between the moment subjects spot a
previous typo and the moment they actually stop typing to fix the first one,
Typos & Co. 185
as in examples 21-28. This might be a symptom of attention dropping once

the subjects have passed the point of no return in motor activity (second
typo in bold; correct word between brackets).
(21) [08.01]de_los_sectores_p•úblico_y_pria¡••{«2}vado. [privado]

(22) ••coma•´{«1}ñías_ [compañías]
(23) cog•a{«2}nt•agiados_•••de•l_•HIV [contagiados]
(24) directmanete{«6}amente_par•a_•gener•ar_ [directamente]
(25) [54.71]p•or_bombardear_a_la_comunidad_jucín_••en{«6}día_ [judía]
(26) {×}perono{«3}sonal.•¶¶ [personal]
(27) [05.55]la_maypr•´{«3}oría_mueren_•en•_el_momento_ [mayoría]
(28) dos_comap••a•ía{«23} [compañías]
A second case in point are typos performed when a previous typo has just
been fixed, as in examples 29-34, for they might point to late or defective
reallocation of cognitive resources to the typing task (new typos in bold).
(29) [38.73]los_insintos_de_una_persona_en_eun_refl{«8}uhn_{«3}n_n•ítido_
reflejo
(30) •cosi{«2}ndi•e{«3}sidere{«1}ada_
(31) israe•lita_•es_hasesi•nado{«10}asesionado_y_otros_
(32) [11.50]a_•una_célula_[15.52]y_deso{«1}py•´{«2}ués_
(33) el_pro•metido_{-18}{+17}••_•••est{«1}pñi{«2}í•rit•u_de_
(34) ••para_ge•nera•r_•energía_dier{«2}rac{«2}ectament•e,_
Another perspective on data relates to the prominence of words held in the

subjects‟ working memories when they are typing, since sometimes typos
seem to advance letters of other words which are to be entered later, as in
examples 35-41 (advanced letters in bold).
(35) ••••el_mercer{«2}ado_de_energ•ía_
(36) [03.11]a_•los_an{«2}ci{«2}años_ cincuenta,
(37) seq{«1}eres{«4}res_quer•idos_
(38) di_di{«5}••su_dis•tribuc•i•´n,_
(39) la_opinión_•que_tiene_su{«1}obre_su
(40) ••••que[03.04]_•perd•ió_por_co{«2}[03.04]m{«1}pco{«8}_
(41) {+24}.•••_[03.20]All•Life•_mand{«4}se_env{«1}carga_de_env•iá{«1}ar_
This might support the finding of Dragsted and Hansen (2009) that problem
words tend to be fixated long before production. Subjects will in such cases
read beyond what they consider an appropriate translation unit to ensure
coherence in their interpretation. When a problematic word is found after
this translation unit, or when a translation which is considered good pops
up, they will keep it in mind and the additional effort may interfere with
current typing.
Finally, stress and/or cognitive load may have an influence which
can be established by means of this kind of analysis, for instance, by
identifying typos due to phonetic interference with orthography. Andrés,
Bruno and Carla are Andalusian, where C and Z will be pronounced /s/, as
most native Spanish speakers in the world do. Examples 42-45 show such
interferences in high-frequency words, which means that there can be no
doubt that these subjects knew how to spell them.
(42) una_••parri{«1}••illa_tridimencional_[05.07]{«24}tabla_•tridimensional_
(43) [08.41]que_analiz{«1}cen
(44) fuera_del_alcanz{«1}ce_•••de_millones_de_
(45) pequeña_•persuaci[09.27]{«2}sión._
Examples 46-47 show similar typos, but here the option of writing C was
ruled out, since it is read as /k/. Thus, orthographic concerns and phonetics
may have a separate or combined influence. Phonetic interference is not
limited to different Spanish pronunciations. In 48-49, English phonetics is
reflected in the drafts with Spanish spelling. In 50-51, English and Spanish
spellings are mixed in cognates (typo in bold).
(46) {×}[31.74]{×}[03.09]Ellos_amenaca{«1}z{«2}zan_
(47) atrasada_y_suavic{«1}zada,_
(48) [05.49]Pero_O{«1}All•Life_está_
(49) Martin Luc{«1}ther_King
(50) {×}_•que_•••Isra•el_•••••nunca_acc{«1}ep•taría•_••••
(51) A•ssociac{«1}tion•,_
It is worth pointing out that examples 49-51 were performed by Diana, who
considers herself bilingual. Again, these typos happen in words so common
Typos & Co. 187
that it is hard to believe that any of these subjects had problems with them.
Rather, what might be at stake is a sort of suspension of rules (including
those which keep languages apart), either due to the task itself or, more
likely, to the stressing situation subjects were in, since they were translating
against the clock and conscious that their products would be evaluated for
class purposes, and also studied for research. At this stage, only such a
suspension can explain why one of the subjects performed the following:
(52) •virñus{«3}ús
(53) [17.50]kk{«2}
(54) ••••son_un_•oo{«7}••n_m{«1}algo_más_
(55) [04.46],_dd{«2}dxg[06.27]{«3}lo_está_usando_
In 52, Carla fixed a typo, but introduced a spelling mistake (virus has no
accent in Spanish). The problem is that the word is far too frequent to have
been misspelled. Carla stopped to correct the typo, so she had some extra
time to notice it, but did not. When stimuli are similar in nature, the
competition for attention is biased towards information relevant to the
current behavior (Desimone & Duncan 1995). Carla, with the slowest
processing speed, might have been particularly stressed, so that she might
have forgotten what she was writing once the typo was identified. She also
typed nonsense sequences (60-62), which might be related to cognitive load
and stress.
7. Conclusions and further research
Typos enjoy a privileged status within phenomena, for they tend to be fixed
online in local interventions more than any other category, including IL
correction phenomena such as accents and orthography. Quantitative
analysis of categorized phenomena and interventions, especially typos, may
be useful for distinguishing between tasks as to the amount of mental
processing they entail, which might correlate with the number of MPs and
interventions. Subjects‟ behaviors seemed to be consistent across tasks and
to be different from those of other subjects. Hence, this analysis might also
prove useful for profiling subjects as well. A simple computer application
with a speller might be able to identify the type of task carried out on a text
and who the author was.
About half of the interventions happened in sequences, and their
length and composition might relate to the (mental) importance of the
problem at hand. Typos also seemed to play an important role as markers
of other kinds of phenomena when in sequence, thereby hinting that a
subject might have experienced some problem or delay when reallocating
mental resources to meet current demands in the task.
This laborious kind of analysis could be much easier when partially
automated. Meanwhile, the goal needs to be restricted to ascertain whether
typos are informative by themselves and in sequences. Achieving this first
goal would pave the way to study the questions just raised, and perhaps
contribute to protocol analyses in general.
References
Buchweitz, A. & Alves, F. 2006. Cognitive adaptation in translation: an interface

between language direction, time, and recursiveness in target text
production. Letras de hoje 41/2: 241-272.
Desimone, R. & Duncan, J. 1995. Neural mechanisms of selective visual
attention. Annual Review of Neuroscience 18: 193-222.
Dragsted, B. & Hansen, I.G. 2009. Comprehension and production in translation:
a pilot study on segmentation and the coordination of reading and writing
processes. In S. Göpferich, A.L. Jakobsen & I.M. Mees (eds) Looking at
Eyes. Eye-Tracking Studies of Reading and Translation Processing.
Copenhagen: Samfundslitteratur. 9-30.
Duncan, J. 1980. The locus of interference in the perception of simultaneous
stimuli. Psychological Review 87/3: 272-300.
Gentner, D.R. 1982. Evidence against a central control model of timing in typing.
Journal of Experimental Psychology: Human Perception and Performance,
8: 793-810.
Kelso, J.A.S. & Schoner, G. 1988. Self-organization of coordinative movement
patterns. Human Movement Science 7: 27- 46.
Logan, G.D. 1982. On the ability to inhibit complex movements: a stop-signal
study of typewriting. Journal of Experimental Psychology: Human
Perception and Performance 8: 778-792.
Lombardo, T.J. 1987. The Reciprocity of Perceiver and Environment. Hillsdale
(NJ): Erlbaum.
Parasuraman, R. (ed.) 2000. The Attentive Brain. Boston (MA): MIT.
Rumelhart, D.E. & Norman, D.A. 1982. Simulating a skilled typist: a study in
skilled motor performance. Cognitive Science 6: 1-36.
Typos & Co. 189
Rumelhart, D.E., McClelland, J.L., Hinton, G.E. 2004. The appeal of parallel
distributed processing. In D.A. Balota & E.J Marsh (eds) Cognitive
Psychology. Key Readings in Cognition. New York: Psychology Press. 75-
99.
Sagi, D. & Julesz, B. 1984. Detection versus discrimination of visual orientation.
Perception 13 (5): 619-628.
Schmidt, R.A. 1975. Motor skills. New York: Harper and Row.
Schmidt, R.A. & Lee, T.D. 2005. Motor control and learning. A Behavioral
Emphasis. 4th edn. New York: Barnes & Noble.
Sohlberg, M.M. & Mateer, C.A. 1989. Introduction to Cognitive Rehabilitation:
Theory and Practice. New York: Guilford Press.
Treisman, A. & Gelade, G. 1980. A feature integration theory of attention.
Cognitive Psychology 12: 97-136.
Translation technology in time: investigating the impact of
translation memory systems and time pressure on types of
internal and external support1
Fabio Alves and Tânia Liparini Campos
Abstract
This paper reports on a process-oriented study which investigates the types

of support used by professional translators and the impact of a translation
memory system (TMS) and time pressure on the use of those types of
support. Drawing on the notions of internal and external support (Alves
1997), chains of cognitive implication (PACTE 2005), and phases of the
translation process (Jakobsen 2002), the performance of 12 professional
translators was analysed in terms of the types of support they used for
orientation, drafting, and revision. The presence of a TMS and time
pressure were the independent variables while the contrast between the
language pairs English/Brazilian Portuguese and German/Brazilian
Portuguese was a dependent variable. The text samples, language
direction, subjects’ experience as professional translators, and their
familiarisation with the TMS were controlled variables. The study
considers data previously analysed by Machado (2007), Batista (2007),
and Alves & Liparini Campos (2008) and adds time pressure as an element
of scrutiny. The results highlight the role played by internal support in all
task combinations with respect to problem-solving and decision-making.
Translation memory systems seem to affect the types of support used by
increasing the number of occurrences of combined use of internal and
external support in the drafting phase. The results also show that time
pressure reduces the occurrence of revision pauses both in drafting and
revision phases, and indicate that time pressure affects mostly revision
processes both with and without the use of a translation memory system.
1
The authors would like to thank Igor A.L. Silva for a careful reading of this paper and
for the invaluable suggestions provided.
192 Fabio Alves and Tânia Liparini Campos
1. Introduction
The impact of translation technology on translator‟s behaviour has been an

object of scientific interest in translation process research in recent years.
Dragsted (2004) was one of the first researchers to investigate the impact of
a translation memory system (TMS) in the segmentation patterns of
translators. She found evidence that natural segmentation patterns of
translators working in the Danish-English language pair is affected by the
use of a TMS. Inspired by Dragsted‟s research design, the present paper,
developed within the framework of the SEGTRAD2 project and carried out
at LETRA, the Laboratory of Experimentation in Translation at the Federal
University of Minas Gerais, looks at the impact a TMS may have on the
types of support resorted to by translators to solve translation problems.
More specifically, this paper draws on the distinction between internal and
external support (Alves 1997) and on a rank scale adapted from PACTE
(2005), called chains of cognitive implication. Besides building on data
previously analysed by Batista (2007), Machado (2007) and Alves &
Liparini Campos (2008), which focused on types of support used in
orientation, drafting and revision phases (Jakobsen 2002) of the translation
process and the impact of the use of a TMS, it also introduces a new
variable to be studied in conjunction with translation technology, namely
time pressure. From a process-oriented perspective, this paper analyses
pause patterns and resources accessed during a translation carried out by 12
professional translators – six translating from English into Brazilian
Portuguese and six translating from German into Brazilian Portuguese –
who carried out four different tasks: (i) translations performed without any
interference, (ii) translations carried out with the aid of a TMS, (iii)
translations performed under time pressure, and (iv) translations rendered
with the aid of a TMS under time pressure.3 In the following sections, data
collected during translation processes with and without time pressure and
2
The SEGTRAD Project (Cognitive Segmentation and Translation Memory Systems:
investigating the interface between translators‟ performance and translation
technology) was supported by the Brazilian Research Council (CNPq) grant n°
301270/2005-8.
3
The data and results presented here draw on Liparini Campos‟s (2008) research paper
submitted and approved for the final qualifying exam towards the PhD Degree at the
Graduate Program for Lingusitics and Applied Linguistics, Federal University of
Minas Gerais, Brazil.
Translation technology in time 193
the use of translation technology will be analysed with a view to

elucidating the impact they may have on types of internal and external
support in the drafting and revision phases of the translation process.
2. Theoretical underpinnings
The different types of support used by translators in the course of the

translation process have been investigated in many respects. Alves (1997)
proposed the categories of internal and external support as a way of
differentiating between two types of strategies employed by translators, i.e.,
those which rely on the translators‟ personal worldview (internal support),
such as the translator‟s encyclopaedic knowledge, and those which draw on
documentation sources (external support), such as dictionaries, reference
materials, online resources, etc. In other words, internal support implies the
use of automatic and non-automatic existing cognitive resources whereas
external support involves the use of any source of documentation to make
up for information which is not immediately available to the translator.
PACTE (2005) divided the categories of internal and external
support into two subcategories each. Building on Alves (1997), they
developed a rank scale termed “chains of cognitive implication”, to account
for differences observed in the use of translation strategies, in which one
form of support was always dominant over the other. The four categories of
cognitive support in the translation process which PACTE suggests are: (i)
simple internal support (SIS), (ii) internal support dominant combined with
external support (DIS), (iii) external support dominant combined with
internal support (DES), and (iv) simple external support (SES). SIS occurs
when a definitive translation solution is reached by using internal support
alone, without any external consultation. DIS involves complex
documentation searches which contribute but do not lead to a definitive
translation solution. The definitive solution is mainly the result of internal
support. DES involves complex consultations which are the basis for a
definitive translation solution with the definitive solution being mainly the
result of external support. SES takes place when a bilingual dictionary is
consulted and the solution is accepted.
The main goal of PACTE‟s research is to investigate translation
competence (TC) and its acquisition in written translation. In their TC
model, PACTE (2003) postulates five subcompetences, namely: (i)

bilingual subcompetence comprising pragmatic, socio-linguistic, textual,
grammatical and lexical knowledge; (ii) extra-linguistic subcompetence
involving bicultural, encyclopaedic, and subject knowledge; (iii)
knowledge about translation subcompetence including knowledge, both
implicit and explicit, about translation and aspects of the profession; (iv)
instrumental subcompetence related to the use of documentation resources
as well as information and communication technologies applied to
translation; (v) strategic subcompetence which ensures the efficiency of the
translation process, creating links between the different subcompetences
and controlling the flow of the translation process. Additionally, PACTE
also considers the role of psycho-physiological components, including
different types of cognitive and attitudinal components and psycho-motor
mechanisms. Aiming at the validation of their TC model, PACTE (2000,
2003, 2005) developed an experimental design with a series of
methodological steps geared to the investigation of each of the five
subcompetences. Instances of internal and external support were adopted to
investigate the translators‟ decision-making processes which, according to
PACTE, are related to the instrumental and strategic subcompetences.
Drawing on PACTE‟s work, Machado (2007) and Batista (2007)
differentiated between two additional subtypes of support, namely support
for orientation and support for revision. They built on Jakobsen‟s (2002)
division of the translation process in three phases: orientation, drafting, and
revision. In translation process research, orientation begins when the
translator sees the source text for the first time and ends when the first key
for typing the target text is pressed on the keyboard. Jakobsen points out
that orientation may also be transferred to the drafting phase, being called
online orientation (in contrast to the initial orientation phase). Drafting
begins immediately after the orientation phase and ends when the full stop
corresponding to the end of the source text is typed in the target text.
Revision starts immediately after the drafting phase and ends when the
translator considers that his/her task has been completed. While text
production is underway, the translator may also revise the text extensively:
this is called online revision to differentiate it from the end-revision phase.
Jakobsen (2005) distinguishes orientation pauses (OP) from revision
pauses (RP) during the drafting phase according to the following criteria: if
a pause occurs before the space bar is pressed, it is a revision pause;

otherwise, if it occurs after the translator has pressed the space bar it is
considered an orientation pause. However, it is also important to consider
what the translator does after the pause: if some part of the text which has
already been translated is modified, the pause immediately before the
modification is most probably a revision pause; if a new part of the text is
translated, the pause directly before this text production is most probably
an orientation pause.
Blending PACTE‟s (2005) and Jakobsen‟s (2002, 2005)
classifications, Machado (2007) and Batista (2007) proposed jointly the
following subtypes of internal and external support: simple internal support
for orientation (SISO), simple external support for orientation (SESO),
dominant internal support for orientation (DISO), dominant external
support for orientation (DESO) as well as simple internal support for
revision (SISR), simple external support for revision (SESR), dominant
internal support for revision (DISR), dominant external support for revision
(DESR). Aiming to provide relevant insights into the kind of impact a TMS
may have on the performance of professional translators, Machado‟s and
Batista‟s works were based on data collected from five professional
translators working first without and later with the aid of a translation
memory system in translations from English into Brazilian Portuguese.
The first set of four categories (SISO/SESO/DISO/DESO) were
applied in the research carried out by Machado (2007) for the analyses of
initial orientation and online orientation processes. The second set of four
categories (SISR/SESR/DISR/DESR) were investigated in Batista‟s (2007)
research on online revision and end-revision processes. Machado‟s and
Batista‟s results provided the foundations for Alves & Liparini Campos
(2008) who also investigated the impact of a TMS applying the same eight
categories of support (SISO/SESO/DISO/DESO/SISR/SESR/DISR/DESR)
on the process of 12 professional translators, six translating from English
into Brazilian Portuguese and six translating from German into Brazilian
Portuguese.
Alves & Liparini Campos (2008) confirmed the results of Machado
(2007) and found that SISO was the most productive type of support for
online orientation when translators carried out their work without the aid of
a TMS. When a TMS was used, SISO was still the most productive type of
support, but there was a major increase in the use of DISO, often employed
by translators to check their renderings against the solutions contained in
the translation memory (TM). Moves such as acceptance, refusal or
modifications of TM suggestions during online orientation can be
accounted for in terms of the translators‟ instrumental competence.
Contrary to Machado‟s results, DISO proved to be the most productive
form of support for orientation in the German subgroup in the translations
rendered with the aid of a TMS. The authors believe that this discrepancy
may be explained in terms of individual features, which must be further
investigated.
As far as support for revision is concerned, Batista (2007) identified
a predominance of SISR both in online and in end-revision processes, a
finding also corroborated by Alves & Liparini Campos (2008). When a
TMS was used, subjects showed a tendency to reduce the numbers of
revision pauses during the drafting phase as well as during the final
revision phase. In online revision, SISR reduced in the presence of a TMS
as a consequence of the reduction of revision pauses (RP), and DISR and
DESR tended to increase. As far as both types of revision (end-revision and
online revision) are concerned, Alves & Liparini Campos (2008) confirmed
the results in Batista (2007).
Another important yet scarcely investigated variable concerning the
translation process is time pressure. Jensen (2001) carried out an empirical
investigation of the effects of time pressure on the cognitive process of
professional and novice translators and found that this variable caused
subjects‟ translation process to become faster and less reflexive. When
translating under time pressure, both novice and professional translators
also spent less time on both initial orientation and end-revision and resorted
to less external consultations. Liparini Campos (2005) investigated the
effects of time pressure on the cognitive process of five novice translators
translating from German into Brazilian Portuguese and corroborated
Jensen‟s (2001) findings with respect to time spent on orientation and
revision phases. Liparini Campos‟ (2005) results indicate that time pressure
has an impact both on the translation process and product. When translating
under time pressure, novice translators showed a tendency to reduce
recursiveness during online revision and had no end-revision phase, which
may explain why some texts produced under time pressure presented
problems related to changes in their cohesive ties and to the maintenance of

the thematic progression.
The present study builds on Machado (2007), Batista (2007), and
Alves & Liparini Campos (2008) and adds time pressure as an independent
variable. The impact of time pressure in conjunction with translation
technology is investigated with respect to the types of support used by 12
professional translators in the language pairs German-Brazilian Portuguese
and English-Brazilian Portuguese.
3. Methodology
3.1. Research design
The experimental design used in the present paper builds on Alves &
Liparini Campos (2008) and is an extension of the original research design
by Machado (2007) and Batista (2007). The presence of a TMS and time
pressure were the independent variables while the contrast between the
language pairs English/Brazilian Portuguese and German/Brazilian
Portuguese was a dependent variable. The text samples, language direction,
subjects‟ experience as professional transtalors, and their familiarisation
with the TMS were controlled variables in the study.
Each source text in English and German consisted of extracts of
approximately 500 words, collected from technical manuals. The eight
source texts were comparable not only in terms of size but also in terms of
linguistic complexity/density and level of difficulty. They contained
instructions for the use of a blood sugar meter in English (T1) and in
German (T2), instructions for the use of an electric toothbrush in English
(T3) and in German (T4), instructions for the use of a heart rate monitor in
English (T5) and in German (T6) and instructions for the use of a
thermometer in English (T7) and in German (T8). As complementary
stimuli, subjects also received printed versions of the original source texts
accompanied by illustrations to have full access to the original texts and not
only to the versions edited for use in Translog and in Trados Translator‟s
Workbench 7, the TMS used in the study.
T1 and T2 were translated without the aid of a TMS and without
time pressure. T3 and T4 were translated using Trados Translator‟s
Workbench 7 without time pressure. T5 and T6 were translated with time

pressure and without the aid of a TMS, while T7 and T8 were translated
with time pressure and with the aid of a TMS. None of the texts had been
translated into Portuguese before, a requirement we insisted on to avoid the
risk that existing translations could be found on the web. Time pressure
was established separately for each subject in that we allowed 70 % of the
time spent by each individual translator in the analogous task carried out
without time pressure. First, translations performed without the use of a
TMS were recorded with Translog. Next, for translations carried out with
the aid of a TMS, the onscreen recording software Camtasia registered the
unfolding of the translation process. Direct observation allowed that notes
on translator‟s behaviour and consultations during the translation task were
registered by the researcher in pre-elaborated observation charts.
Tables 4 to 19, provided in the appendix, contain data from subjects
divided into two groups of six translators each. Subjects identified as E1,
E2, E3, E4, E5 and E6 translated the English source texts T1, T3, T5 and
T7. Subjects identified as G1, G2, G3, G4, G5 and G6 translated the
German source texts T2, T4, T6 and T8. All subjects had at least six years
of experience as professional translators and had been working with a TMS
for two years or longer. Once they had finished their translations, they
watched their work being replayed, using either Translog (for translations
carried out without a TMS) or the Camtasia replay function (for
translations carried out with the aid of a TMS) to prompt their retrospective
protocols recorded immediately thereafter.
Translation memories were built by aligning texts which were
similar in content to T3/T4 and to T7/T8 with their already existing
translations. The TMS was programmed not to incorporate into the TM the
options which were produced by the 12 subjects, ensuring that all subjects
had access to the same identical TM content throughout the experiment. To
make certain that translation memories would not be a source of
interference in the comparison of data from English and German target
texts, we set the TM at the same parity levels, i.e., they retrieved TM
suggestions which were at least 70 % similar to the input being processed,
and had the approximate numbers of matches aligned similarly in both
language pairs.
All procedures followed the methodological approach known as data

triangulation (Alves 2003), which attempts to map the translation process
using data collected from different vantage points (see also Jakobsen 1999
for a discussion of this technique originally used in the social sciences). In
this paper, the sources for triangulating translation process data were the
recordings of target text production in real time, direct observation charts
registering notes on translator‟s consultation and behaviour, and
retrospective protocols.
3.2. Methodology for data analysis
Data generated in the experiment consisted of 48 target texts. Pauses which

occurred during their production were classified on the basis of the four
subtypes of support for orientation (SISO/SESO/DISO/DESO) and the four
subtypes of support for revision (SISR/SESR/DISR/DESR) described in
Section 2. Target text protocols were divided into segments delimited by
pauses of five seconds or longer. Classification for translations rendered
without a TMS (T1/T2 and T5/T6) was based on the semi-automatic
analysis of pauses retrieved with the aid of Translog4 logfiles. For
translations rendered with the use of a TMS (T3/T4 and T7/T8),
classification was made manually by replaying Camtasia files; delimited
intervals with pauses of five seconds or longer were identified manually
using the clock provided by the software.
Classification of pauses and their corresponding type of support
involved replaying the translation process either with Translog or Camtasia
and triangulating it with data from the observation charts and retrospective
protocols. Each pause was analysed separately by two researchers who
classified it according to the type of pause and the type of support used. If
there was 100% agreement on their judgments, the classification of the type
of pause/support was final; if there was disagreement in the analyses, the
pause was reanalysed by a third researcher and discussed until a unanimous
decision was reached as to what type of pause/support it corresponded
with.
4
Translog provides visual representations in which pauses are identified by asterisks.
The pause value is defined by the researcher. As a methodological decision for the
identification of segments, we opted for a five-second pause interval which has been
the standard value used at LETRA.
Classification of the type of pause was carried out using Jakobsen‟s

(2005) criteria for the identification of orientation pauses (OP) and revision
pauses (RP). Most of the time, it was quite clear whether a pause was used
for orientation or for revision. However, there were a few instances where
processes of orientation and revision seemed to overlap, particularly when
there was a long pause and there could be mental activity going on in terms
of translating a new segment and revising a translated segment at the same
time. As a methodological decision, those pauses were classified as
instances of revision only if the translator modified a previously typed
rendering of the source text after the pause; otherwise, it was considered an
orientation pause.
After the classification of pauses in OP and RP, each pause was
classified according to the type of support used to solve the translation
problem. Whenever the translator did not carry out any kind of search
whatsoever within a given pause interval, it was classified as simple
internal support (SIS). The reading of the printed source text version and
the scrutiny of the accompanying illustrations were not considered as a
kind of search since they had been provided as stimuli for the source text.
Whenever a single external source was looked up, such as a dictionary, a
glossary, or a website, this was considered as a simple search. If this
consultation played a vital role for problem-solving and decision-making, it
was classified as simple external support (SES). Whenever there was a
single search or a complex search within a pause interval followed by a
translation choice which was not provided by the consultation itself, the
type of support was classified as dominant internal support (DIS). Finally,
when there was a complex search within a pause interval and the
suggestions provided were accepted, the type of support was classified as
dominant external support (DES). Observation charts were instrumental in
determining the type of support used in the texts translated in Translog (T1,
T2, T5 and T6). For texts T3, T4, T7 and T8, external searches were
documented in Camtasia with observation charts providing only
confirmation of the type of support used by the translator.
When a translator came across a suggestion provided by the TM and
paused to think about it or revise it, independently of accepting or rejecting
it, the type of support was classified as DIS. Input provided by the TM was
considered in itself an additional source of information, but as the translator
did not make the choice to search for external sources of documentation,
those instances were not considered to be simple external support. A pause
was classified as SES or DES when the translator accepted a suggestion
provided by an external source other than the TM. The Trados
Concordancer was considered to be a source of external support since it has
to be looked up by the translator and works differently from the standard
solutions provided by the TM. The spell checker of the word processor was
also considered as a type of external support.
Although each pause was analysed on its own, the solution of a
particular translation problem can also be achieved through a sequence of
pauses. Quite often, a single pause accounts only for the partial solution of
a translation problem, which will then be completed or revised at a later
stage. Segments do not coincide across the samples of the 12 subjects.
These vary widely among the subjects, some segmenting the text in larger
chunks, others showing a pattern of shorter segments. This explains why
some subjects show a larger number of pauses – and consequently a larger
number of segments, which are those strings of text produced between two
pauses – than others when translating the same source text.5 Therefore,
differently from Machado (2007), Batista (2007), and Alves & Liparini
Campos (2008), data was analysed in relative numbers using percentages
and averages, and presented in Tables 1 and 2 in Section 4. For a more
detailed analysis, Tables 4 to 19 in the appendix present the occurrences
quantified in terms of numbers of pauses and types of support.
Table 1 contains data on the effect of the use of a TMS and time
pressure on the frequency of types of support used by translators in the
drafting phase. Table 2 shows data on the effect of both the use of a TMS
and time pressure on the frequency of types of support used by translators
in the end-revision phase. In Tables 1 and 2 data were consolidated as
follows: (i) data on T1/T2 were presented separately at first and followed
by an average consolidating the occurrences of types of support for the
tasks carried out without any additional technological support or processing
effort; (ii) data on T3/T4 were presented next, also separately at first and
then followed by an average consolidating the occurrences of types of
5
For a detailed study on segmentation patterns of the 12 subjects, see Rodrigues
(forthcoming) which replicates Dragsted‟s (2004) PhD study for translations from
English and German into Brazilian Portuguese.
support for the tasks carried out with additional technological support,
namely the use of a TMS; (iii) data on T5/T6 followed for the tasks carried
out without any additional technological support but with time pressure as
an element of extra processing effort; and (iv) data on T7/T8 were
presented for the tasks carried out with additional technological support
(TMS) and extra processing effort, i.e., time pressure.
4. Data analysis
In accordance with the results found by Alves & Liparini (2008), initial
orientation is not a significant factor in the analyses of types of support
used by professional translators. Although there is a slight increase in the
time spent on initial orientation when a TMS is used, as a result of
calibration of the TMS settings (Machado 2007), not much time is spent on
initial orientation (0% to 8.5%). In line with comments made by Jakobsen
(2002), initial orientation is quite often transferred into the drafting phase
and more seldom into the end-revision phase. These results also
corroborate PACTE (2008), who found similar evidence observing the
relatively small percentage of time spent on the orientation stage (7.2%) by
all subjects in their experiment on TC and even less time (5%) among their
„best‟ subjects. This tendency was also observed when time pressure was
included in our experimental design both with and without the use of a
TMS. Therefore, data from initial and online orientation were merged and
initial orientation processes will be analysed together as online orientation
within the drafting phase.
4.1. Types of support in the drafting phase
Table 1 below shows the mean values for the types of pause/support used in
the drafting phase. Data is presented in percentages and grouped together
for similar tasks. Data on T1/T2 tasks, carried out without the technological
support of a TMS or the impact of time pressure, reveal that 70 % of the
pauses in the drafting phase are instances of orientation. Although the
English subgroup (ESG) shows a lower occurrence of OP (65 %) than the
German subgroup (GSG) (74 %), both subgroups put a lot more effort into
orienting themselves than revising the texts they produced. SISO is the
most productive type of support for orientation, followed by SESO, with

very few occurrences of DISO and DESO. Most instances of external
support involve web searches or dictionary look-ups to find translation
alternatives for specific terms.
This indicates that most professional translators rely predominantly
on their previous knowledge and mainly use internal support to solve
problems and make decisions. As for online revision, data on T1/T2 tasks
show that SISR was the most predominant type of support and accounted
for the vast majority of occurrences. SESR, DISR and DESR were very
few in numbers.
Table 1. Mean value (in percentages) for types of pause/support in the drafting
phase (ESG – English subgroup; GSG – German subgroup; OP – orientation
pause; RP – revision pause)
OP SISO SESO DISO DESO RP SISR SESR DISR DESR Total
ESG/T1 65% 54% 9% 0% 2% 35% 32% 3% 0% 0% 100%
GSG/T2 74% 64% 7% 1% 2% 26% 23% 1% 1% 1% 100%
Average 70% 59% 8% 1% 2% 30% 27% 2% 1% 0% 100%
ESG/T3 73% 35% 9% 27% 2% 27% 19% 3% 4% 1% 100%
GSG/T4 85% 27% 6% 49% 3% 15% 13% 1% 1% 0% 100%
Average 79% 31% 8% 38% 2% 21% 16% 2% 3% 0% 100%
ESG/T5 78% 62% 14% 0% 2% 22% 18% 3% 1% 0% 100%
GSG/T6 75% 67% 5% 1% 2% 25% 23% 2% 0% 0% 100%
Average 76% 64% 9% 1% 2% 24% 20% 3% 1% 0% 100%
ESG/T7 79% 37% 12% 28% 2% 21% 16% 3% 2% 0% 100%
GSG/T8 80% 48% 9% 22% 1% 20% 18% 0% 1% 1% 100%
Average 80% 42% 11% 25% 2% 20% 17% 1% 2% 0% 100%
On the other hand, technological support in tasks T3/T4 yielded an increase

in the occurrence of orientation pauses. OP went up from 70 % in tasks
T1/T2 to 79 % in tasks T3/T4. Consequently, there was a reduction in the
occurrence of revision pauses, RP moving down from 30 % in tasks T1/T2
to 21 % in tasks T3/T4. There was also a significant reduction in instances
of SISO in T3/T4. In the German subgroup, the average percentage of
SISO went down from 64 % in T2 to 27 % in T4, whereas in the English
subgroup occurrences of SISO decreased on average from 54 % in T1 to
35 % in T3. Although the German subgroup shows a higher decrease in the
occurrences of SISO than the English subgroup, both groups showed the
same overall tendency. Reduction in the instances of SISO may be due to a
significant increase in instances of DISO in the presence of a TMS.
Overall, DISO moved from 1 % in T1/T2 to 38 % in T3/T4. DISO

increases as a direct result of the impact of the TMS on the types of
support. When a translation alternative is offered by the TM, the translator
is led to consider the options provided before moving on and has to
interrupt the writing process to think of possible changes for the solutions
offered by the TM. With the use of a TMS, most instances of DISO are
related to those types of pause occurrences.
For online revision pauses, SISR remained the most productive type
of support in tasks T3/T4. However, it was employed less frequently than
in tasks T1/T2. Differently from what happened in T1/T2, where SISO and
SISR were the most frequent types of support, instances of DISO in T3/T4
were higher than SISR and, in the German subgroup, also higher than
SISO. We may assume that the overall reduction of RP, and of SISR in
particular, are related to the impact of DISO in the translation process. The
German subgroup, which showed a higher reduction in the number of RP
was also the subgroup with a higher count of DISO. In T3/T4, the presence
of a higher count of DISO optimised online orientation processes and
reduced the need for revisions, a clear indication of the positive impact of a
TMS on the allocation of effort in the translation process.
The use of a TMS does not seem to have had an impact on external
support. However, the nature of the searches varied. Whereas in T1/T2
external support involved web searches for parallel texts and dictionary
look-ups, in T3/T4 the Trados Concordancer was the most frequent type of
external support. In T3/T4, external support was also used to accept or
reject solutions offered by the TMS.
With the introduction of time pressure in tasks T5/T6, there was a
minor increase in the occurrence of OP, which went up from 70 % in T1/T2
to 76 % in T5/T6. There was also a minor reduction in the occurrence of
RP, which moved down from 30 % in T1/T2 to 24 % in T5/T6. Similar to
what was observed for tasks T3/T4, carried out with the aid of a TMS, time
pressure seems to have had an impact on online revision. All 12 subjects
showed a smaller number of RP when translating under time pressure.
Although, on average, time pressure had the same effect in both subgroups,
this impact was greater on the English subgroup, in which the occurrence
of RP moved down from 35 % in T1 to 22 % in T5. As far as the types of
support used for problem-solving in the drafting phase are concerned, there
was no effect of time pressure. These results corroborate some of the

findings of Liparini Campos‟s (2005) study which, although using a
different methodology, also showed that time pressure has an impact on
online revision, resulting in a reduction in the keystroke movements related
to revision during the drafting phase.
Finally, tasks T7/T8 were carried out with the aid of a TMS and
under time pressure. If compared with T1/T2, there was an increase in the
occurrence of OP, which went up from 70% to 80%. Consequently, there
was a reduction in the occurrence of RP, which moved down from 30% to
20%. This difference was also observed in the comparison of data from
tasks T1/T2 and T3/T4, as well as from tasks T1/T2 and T5/T6. However,
in T7/T8 the reduction in the occurrence of RP was even more striking for
both the English subgroup and the German subgroup. The reduction in the
occurrence of RP is likely to have been motivated by the same factors
mentioned earlier for T3/T4. The increase in instances of DISO modifies
the process of online revision as online orientation becomes more effective
as a result of the use of the TMS. DISO, however, always occurred less
frequently than SISO, which remained the most productive type of support
for both subgroups in T7/T8. We can argue that time pressure had an
impact on online revision processes and also reduced the occurrence of
revisions for solutions provided by the TMS.
As far as external support is concerned, SESO was the most
productive type of support for T7/T8, with a minor increase in numbers
when compared to the other tasks. DESO showed the same tendency
observed in the previous tasks. SESR, DISR, and DESR were also
negligible in terms of the support they provided.
Overall, the use of a TMS in conjunction with time pressure had a
greater impact on online revision processes. However, the impact appears
to be different for each subgroup. For the German subgroup, time pressure
reduced revision associated with solutions offered by the TMS, observable
in the reduction in the occurrence of DISO in relation to the task carried out
with the use of a TMS and without time pressure (T4). Alternatively, for
the English subgroup there was a reduction in the occurrence of RP, an
effect similar to the one observed when the subjects performed task T5.
The impact of time pressure for the English subgroup was similar both
when translating with or without the aid of a TMS.
4.2. Types of support in the end-revision phase
Table 2 below shows the mean values for the types of pause/support found
in the end-revision phase. Data is presented in percentages and grouped
together for similar tasks.
Table 2. Mean value (in percentages) for types of pause/support in the revision
phase6
OP SISO SESO DISO DESO RP SISR SESR DISR DESR Total

ESG/T1 1% 1% 0% 0% 0% 99% 93% 5% 0% 1% 100%
GSG/T2 0% 0% 0% 0% 0% 100% 96% 4% 0% 0% 100%
Average 0% 0% 0% 0% 0% 100% 95% 5% 0% 0% 100%
ESG/T3 0% 0% 0% 0% 0% 100% 95% 2% 3% 0% 100%
GSG/T4 0% 0% 0% 0% 0% 100% 95% 1% 4% 0% 100%
Average 0% 0% 0% 0% 0% 100% 95% 2% 3% 0% 100%
ESG/T5 10% 10% 0% 0% 0% 90% 88% 1% 0% 1% 100%
GSG/T6 0% 0% 0% 0% 0% 100% 94% 6% 0% 0% 100%
Average 6% 6% 0% 0% 0% 94% 91% 3% 0% 0% 100%
ESG/T7 9% 5% 0% 4% 0% 91% 91% 0% 0% 0% 100%
GSG/T8 0% 0% 0% 0% 0% 100% 96% 2% 0% 2% 100%
Average 4% 2% 0% 2% 0% 96% 94% 1% 0% 1% 100%
Orientation pauses are rare in end-revision of all eight tasks, and the most
predominant type of support is SISR, which corroborates the results of
Batista (2007) and Alves & Liparini Campos (2008). During the end-
revision phase, translators tend to check the target texts produced so far to
solve remaining problems; for this they predominantly resort to internal
support.
The second most frequent type of support in the revision phase is
SESR. Nevertheless, it occurs much less often than SISR. Whereas, on
average, instances of SISR vary between 91 % and 95 % of the total
occurrences of support for end-revision in all eight tasks, the mean value
for SESR varies between 1 % and 5 %. Additionally, not all subjects used
SESR, most of them relying exclusively on SISR to support changes made
in the end-revision phase.
6
Not all subjects had an end-revision phase, particularly when translations were
rendered under time pressure (tasks T5/T6 and T7/T8). Data of those subjects who
had no end-revision were excluded from the calculation of the mean values shown in
Table 2.
Instances of SESR correspond mainly to searches for specific terms

on the web or in dictionaries. In tasks T3/T4 and T7/T8, carried out with
the aid of a TMS, SESR also involved spell-checking. Searches related to
translation problem-solving are practically non-existent since most
problems were dealt with in the drafting phase. If one examines the pause
patterns and the types of support involved, it emerges that neither the use of
a TMS nor time pressure had an impact on the end-revision phase.
5. Concluding remarks
Table 3 below summarises the main characteristics of the process of the 12

professional translators with respect to the type of pauses and the type of
support resorted to in translation problem-solving in each phase of the
translation process.
Table 3. Characteristics of the translation process with respect to types of pauses

and suppport
Orientation phase Irrespective of technological support and/or time pressure,
orientation seldom occurs as a separate phase and most often
takes place as online orientation during the drafting phase.
Drafting phase Irrespective of technological support and/or time pressure,

orientation pauses (OP) are more frequent than revision pauses
(RP)
Irrespective of technological support and/or time pressure, SIS is

the most prevalent type of support.
With the use of a TMS, there was an increase in the occurrence of

orientation pauses (OP) and reduction in the occurrence of
revision pauses (RP).
With the use of a TMS, there was a significant increase in the

occurrence of DISO.
With time pressure, there was a reduction in the occurrence of

revision pauses (RP).
With the use of a TMS together with time pressure, there was a
reduction in the occurrence of DISO in the German subgroup.
With the use of a TMS together with time pressure, there was a
reduction in the occurrence of revision pauses (RP) in the English
subgroup.
Revision Phase Irrespective of technological support and/or time pressure, the

occurrence of orientation pauses (OP) is very rare.
Irrespective of technological support and/or time pressure, SISR

is the most frequent type of support.
The use of a TMS did not have an impact on the types of support.
Time pressure did not have an impact on the types of support.
With time pressure, there was an overall reduction in the number

of pauses.
The findings of this paper corroborate the studies of Batista (2007),

Machado (2007), and Alves & Liparini Campos (2008). Overall, a separate
orientation phase seldom occurs in the process of professional translators,
irrespective of technological support and/or time pressure. Orientation
occurs mainly during the drafting phase. There is also extensive online
revision, but orientation pauses are much more frequent than revision
pauses during the drafting phase. Contrary to orientation, revision occurs
also as a separate phase at the end of the translation process of the
professional translators investigated here.
Irrespective of technological support and/or time pressure, SIS was
the most predominant type of support for orientation and for revision in
both drafting and revision phases. Although documentation is an important
source of support during the translation process, since all the subjects
looked for help on the web, in dictionaries, and in technologies such as the
spell checker of the word processor and the TMS itself, professional
translators rely mostly on their own knowledge to solve translation
problems.
Concerning the effect of the use of a TMS, results show that
translation technology does change the way professional translators behave
and optimises sources of external support such as the TMS Concordancer.
Translation technology also optimises online orientation processes,
reducing the need for revisions during the drafting phase. The use of a
TMS does not reduce the important role played by internal support in terms
of problem-solving or decision-making, but results show that the dominant
type of internal support becomes more prevalent when a TMS is used. The
scrutiny of the suggestions offered by the TMS, the enquiry about the
reliability of the TM and the search for consistency are clear indicators that
internal support is vital for the management of support in the translation
process.
There was no effect of time pressure on the types of support used by
professional translators, but there was a reduction in the occurrence of
revision pauses both in drafting and revision phases, indicating that time
pressure affects mostly revision processes. The need to rely on solutions
offered by the TMS increases when translations are rendered under time
pressure, as professional translators tend to accept TMS translation
solutions without revising them to cope with difficulties caused by time
constraints.
In general terms, time pressure and translation technology require
that translators be more aware of their behaviours. These findings should
have an impact on the training of translators, which should take into
account the importance of internal support as the most productive type of
support in all task combinations.
References
Alves, F. 1997. A formação de tradutores a partir de uma abordagem cognitiva:

reflexões de um projeto de ensino. TradTerm 4(2): 19-40.
Alves, F. (ed.) 2003. Triangulating Translation: Perspectives in Process-
oriented Research. Amsterdam: John Benjamins.
Alves, F., Liparini Campos, T. 2008. Chains of cognitive implication in
orientation and revision during the translation process: investigating the
impact of translation memory systems in the performance of professional
translators. Proceedings of the XVIII FIT World Congress. Xangai: Foreign
Language Press. CD-ROM.
Batista, B. 2007. O Impacto dos Sistemas de Memória de Tradução nos
Processos de Revisão de Tradutores Profissionais Brasileiros. Unpublished
MA Thesis. Federal University of Minas Gerais, Brazil.
Dragsted, B. 2004. Segmentation in Translation and Translation Memory
Systems. An Empirical Investigation of Cognitive Segmentation and Effects
of Integrating a TM System into the Translation Process. Unpublished PhD
thesis. Copenhagen Business School.
Hansen (ed.) Probing the Process in Translation: Methods and Results
(Copenhagen Studies in Language 24). Copenhagen: Samfundslitteratur. 9-
20.
Jakobsen, A. L. 2002. Translation drafting by professional translators and by
translation students. In G. Hansen (ed.) Empirical Translation Studies:
Process and Product (Copenhagen Studies in Language 27). Copenhagen:

Samfundslitteratur. 191-204.
Jakobsen, A. L. 2005. Investigating expert translators‟ processing knowledge. In
H.V. Dam, J. Engberg & H. Gerzymisch-Arbogast (eds). Knowledge
Systems and Translations. Berlin: Mouton de Gruyter. 173-189.
Jensen, A. 2001. The Effects of Time on Cognitive Processes and Strategies in
Translation. Copenhagen: Working Papers in LSP, 2001/2.
Liparini Campos, T. 2005. O efeito da pressão de tempo na realização de tarefas
de tradução: uma análise processual sobre o desempenho de tradutores em
formação. Unpublished MA Thesis. Federal University of Minas Gerais,
Brazil.
Liparini Campos, T. 2008. O efeito do uso de um sistema de memória de
tradução e da pressão de tempo sobre o processo de tomada de decisão de
tradutores profissionais. Final qualifying exam paper towards the PhD
Degree at the Graduate Program for Lingusitics and Applied Linguistics,
Federal University of Minas Gerais, Brazil.
Machado, I. 2007. Processos de Orientação Inicial e em Tempo Real e sua
Interface com Sistemas de Memória de Tradução. Unpublished MA Thesis.
Federal University of Minas Gerais, Brazil.
PACTE 2000. Acquiring translation competence: hypotheses and methodological
problems in a research project. In A. Beeby, D. Ensinger & M. Presas (eds).
Investigating Translation. Amsterdam: John Benjamins. 99-106.
PACTE 2003. Building a translation competence model. In F. Alves (ed.)
Triangulating Translation: Perspectives in Process Oriented Research.
Amsterdam/Philadelphia: John Benjamins. 43-66.
PACTE 2005. Investigating translation competence: conceptual and
methodological issues. Meta 50(2): 609-619.
PACTE 2008. First results of a translation competence experiment: “knowledge
of translation” and “efficacy of the translation process”. In J. Kearns (ed.).
Translator and Interpreter Training: Issues, Methods and Debates.
London: Continuum International Publishing Group. 104-126.
Rodrigues, R. (forthcoming). Redação e segmentação em tradução: uma análise
do processo tradutório de tradutores profissionais nos pares lingüísticos
alemão-português e inglês-português com e sem o auxílio de sistemas de
memória de tradução. MA Thesis. Federal University of Minas Gerais,
Brazil.
APPENDIX
Table 4: Occurrences of types of support in the drafting phase in T1/T2
Subject OP SISO SESO DISO DESO RP SISR SESR DISR DESR Total
E1 31 20 10 0 1 23 19 4 0 0 54
E2 39 33 4 0 2 9 9 0 0 0 48
E3 40 33 6 0 1 35 33 2 0 0 75
E4 40 39 1 0 0 58 56 1 0 1 98
E5 35 35 0 0 0 7 7 0 0 0 42
E6 31 20 8 0 3 12 8 3 1 0 43
G1 28 28 0 0 0 15 14 1 0 0 43
G2 38 34 3 0 1 9 8 1 0 0 47
G3 72 61 8 0 3 33 25 1 4 3 105
G4 85 68 15 0 2 16 15 1 0 0 101
G5 32 29 1 2 0 18 18 0 0 0 50
G6 74 56 10 3 5 19 15 1 3 0 93
Table 5: Percentages of types of support in the drafting phase in T1/T2
E1 57% 37% 18% 0% 2% 43% 35% 8% 0% 0% 100%
E2 81% 69% 8% 0% 4% 19% 19% 0% 0% 0% 100%
E3 53% 44% 8% 0% 1% 47% 44% 3% 0% 0% 100%
E4 41% 40% 1% 0% 0% 59% 57% 1% 0% 1% 100%
E5 83% 83% 0% 0% 0% 17% 17% 0% 0% 0% 100%
E6 72% 46% 19% 0% 7% 28% 19% 7% 2% 0% 100%
Average 64% 53% 9% 0% 2% 35% 32% 3% 0% 0% 100%

G1 65% 65% 0% 0% 0% 35% 33% 2% 0% 0% 100%
G2 81% 73% 6% 0% 2% 19% 17% 2% 0% 0% 100%
G3 69% 58% 8% 0% 3% 31% 23% 1% 4% 3% 100%
G4 84% 67% 15% 0% 2% 16% 15% 1% 0% 0% 100%
G5 64% 58% 2% 4% 0% 36% 36% 0% 0% 0% 100%
G6 80% 60% 12% 3% 5% 20% 16% 1% 3% 0% 100%
Average 74% 64% 7% 1% 2% 26% 23% 1% 1% 1% 100%
Final Average 69% 58% 8% 1% 2% 30% 27% 2% 1% 0% 100%

E1 48 36 3 9 0 11 7 1 3 0 59
E2 30 17 8 4 1 8 6 1 1 0 38
E3 17 12 2 3 0 10 7 1 2 0 27
E4 48 25 4 17 2 29 24 2 1 2 77
E5 51 10 1 38 2 10 9 0 1 0 61
E6 63 11 15 35 2 27 13 6 8 0 90
G1 53 13 4 35 1 6 3 0 3 0 59
G2 26 8 0 17 1 1 1 0 0 0 27
G3 55 20 8 26 1 9 8 1 0 0 44
G4 84 27 6 46 5 21 18 1 2 0 105
G5 48 7 6 35 0 19 18 1 0 0 67
G6 112 55 5 44 8 22 19 0 3 0 134
E1 81% 61% 5% 15% 0% 19% 12% 2% 5% 0% 100%
E2 80% 46% 21% 10% 3% 20% 16% 2% 2% 0% 100%
E3 63% 45% 7% 11% 0% 37% 26% 4% 7% 0% 100%
E4 62% 32% 5% 22% 3% 38% 31% 3% 1% 3% 100%
E5 84% 16% 2% 63% 3% 16% 14% 0% 2% 0% 100%
E6 70% 12% 17% 39% 2% 30% 14% 7% 9% 0% 100%
Average 73% 35% 9% 27% 2% 27% 19% 3% 4% 1% 100%

G1 90% 22% 7% 59% 2% 10% 5% 0% 5% 0% 100%
G2 96% 30% 0% 63% 3% 4% 4% 0% 0% 0% 100%
G3 86% 31% 12% 41% 2% 14% 12% 2% 0% 0% 100%
G4 80% 26% 6% 44% 4% 20% 17% 1% 2% 0% 100%
G5 71% 10% 9% 52% 0% 29% 27% 2% 0% 0% 100%
G6 84% 41% 4% 33% 6% 16% 14% 0% 2% 0% 100%
Average 85% 27% 6% 49% 3% 16% 13% 1% 2% 0% 100%
Final Average 79% 31% 8% 38% 2% 22% 17% 2% 3% 0% 100%

E1 13 11 1 0 1 2 2 0 0 0 15
E2 24 16 8 0 0 3 3 0 0 0 27
E3 26 25 0 0 1 8 8 0 0 0 34
E4 41 34 5 2 0 16 15 0 1 0 57
E5 24 23 1 0 0 6 5 1 0 0 30
E6 19 9 10 0 0 12 5 5 2 0 31
G1 19 18 1 0 0 17 15 2 0 0 36
G2 16 15 0 1 0 6 4 2 0 0 22
G3 55 54 0 0 1 15 15 0 0 0 70
G4 58 49 8 0 1 12 12 0 0 0 70
G5 35 35 0 0 0 18 18 0 0 0 53
G6 65 50 9 2 4 5 4 1 0 0 70
E1 87% 73% 7% 0% 7% 13% 13% 0% 0% 0% 100%
E2 89% 59% 30% 0% 0% 11% 11% 0% 0% 0% 100%
E3 76% 73% 0% 0% 3% 24% 24% 0% 0% 0% 100%
E4 72% 60% 9% 3% 0% 28% 26% 0% 2% 0% 100%
E5 80% 77% 3% 0% 0% 20% 17% 3% 0% 0% 100%
E6 61% 29% 32% 0% 0% 39% 16% 16% 7% 0% 100%
Average 78% 62% 14% 0% 2% 22% 18% 3% 1% 0% 100%

G1 53% 50% 3% 0% 0% 47% 42% 5% 0% 0% 100%
G2 73% 68% 0% 5% 0% 27% 18% 9% 0% 0% 100%
G3 79% 77% 0% 0% 2% 21% 21% 0% 0% 0% 100%
G4 83% 70% 11% 0% 2% 17% 17% 0% 0% 0% 100%
G5 66% 66% 0% 0% 0% 34% 34% 0% 0% 0% 100%
G6 93% 71% 13% 3% 6% 7% 6% 1% 0% 0% 100%
Average 75% 67% 5% 1% 2% 25% 23% 2% 0% 0% 100%
Final Average 76% 64% 9% 1% 2% 24% 20% 3% 1% 0% 100%

E1 34 15 3 15 1 10 10 0 0 0 44
E2 23 14 3 6 0 3 2 1 0 0 26
E3 26 11 5 8 2 3 2 0 1 0 29
E4 28 16 2 10 0 6 6 0 0 0 34
E5 37 17 0 20 0 16 14 1 1 0 53
E6 35 8 16 10 1 18 8 5 5 0 53
G1 20 7 5 8 0 2 2 0 0 0 22
G2 16 12 0 4 0 4 4 0 0 0 20
G3 29 21 3 5 0 8 8 0 0 0 37
G4 39 20 6 12 1 6 5 1 0 0 45
G5 25 13 3 9 0 15 14 0 1 0 40
G6 60 44 3 9 4 12 9 0 1 2 72
E1 77% 34% 7% 34% 2% 23% 23% 0% 0% 0% 100%
E2 88% 53% 12% 23% 0% 12% 8% 4% 0% 0% 100%
E3 90% 38% 17% 28% 7% 10% 7% 0% 3% 0% 100%
E4 82% 47% 6% 29% 0% 18% 18% 0% 0% 0% 100%
E5 70% 32% 0% 38% 0% 30% 26% 2% 2% 0% 100%
E6 66% 15% 30% 19% 2% 34% 15% 9,50% 9,50% 0% 100%
Average 79% 37% 12% 28% 2% 21% 16% 3% 2% 0% 100%

G1 91% 32% 23% 36% 0% 9% 9% 0% 0% 0% 100%
G2 80% 60% 0% 20% 0% 20% 20% 0% 0% 0% 100%
G3 78% 57% 8% 13% 0% 22% 22% 0% 0% 0% 100%
G4 87% 45% 13% 27% 2% 13% 11% 2% 0% 0% 100%
G5 62% 32% 7% 23% 0% 38% 35% 0% 3% 0% 100%
G6 83% 61% 4% 13% 5% 17% 13% 0% 1% 3% 100%
Average 80% 48% 9% 22% 1% 20% 18% 0% 1% 1% 100%
Final Average 80% 42% 11% 25% 2% 20% 17% 1% 2% 0% 100%

Table 12: Occurrences of types of support in the revision phase in T1/T2
E1 0 0 0 0 0 21 19 2 0 0 21
E2 0 0 0 0 0 1 1 0 0 0 1
E3 0 0 0 0 0 14 14 0 0 0 14
E4 0 0 0 0 0 25 25 0 0 0 25
E5 2 2 0 0 0 54 47 7 0 0 56
E6 0 0 0 0 0 19 16 2 0 1 19
G1 0 0 0 0 0 59 59 0 0 0 59
G2 0 0 0 0 0 5 5 0 0 0 5
G3 0 0 0 0 0 0 0 0 0 0 0
G4 0 0 0 0 0 20 17 3 0 0 20
G5 0 0 0 0 0 11 11 0 0 0 11
G6 0 0 0 0 0 28 26 2 0 0 28
Table 13: Percentages of types of support in the revision phase in T1/T2
E1 0% 0% 0% 0% 0% 100% 90% 10% 0% 0% 100%
E2 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E3 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E4 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E5 4% 4% 0% 0% 0% 96% 84% 12% 0% 0% 100%
E6 0% 0% 0% 0% 0% 100% 84% 11% 0% 5% 100%
Average 1% 1% 0% 0% 0% 99% 93% 5% 0% 1% 100%

G1 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G2 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G3 x x x x X x x x x x x
G4 0% 0% 0% 0% 0% 100% 85% 15% 0% 0% 100%
G5 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G6 0% 0% 0% 0% 0% 100% 93% 7% 0% 0% 100%
Average 0% 0% 0% 0% 0% 100% 96% 4% 0% 0% 100%
Final Average 0% 0% 0% 0% 0% 100% 95% 5% 0% 0% 100%

E1 0 0 0 0 0 5 5 0 0 0 5
E2 0 0 0 0 0 1 1 0 0 0 1
E3 0 0 0 0 0 10 10 0 0 0 10
E4 0 0 0 0 0 15 15 0 0 0 15
E5 0 0 0 0 0 29 28 1 0 0 29
E6 0 0 0 0 0 23 17 2 4 0 23
G1 0 0 0 0 0 49 48 0 1 0 49
G2 0 0 0 0 0 11 11 0 0 0 11
G3 0 0 0 0 0 0 0 0 0 0 0
G4 0 0 0 0 0 27 27 0 0 0 27
G5 0 0 0 0 0 22 17 1 4 0 22
G6 0 0 0 0 0 33 33 0 0 0 33
E1 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E2 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E3 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E4 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E5 0% 0% 0% 0% 0% 100% 97% 3% 0% 0% 100%
E6 0% 0% 0% 0% 0% 100% 74% 9% 17% 0% 100%
Average 0% 0% 0% 0% 0% 100% 95% 2% 3% 0% 100%

G1 0% 0% 0% 0% 0% 100% 98% 0% 2% 0% 100%
G2 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G4 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G5 0% 0% 0% 0% 0% 100% 77% 5% 18% 0% 100%
G6 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
Average 0% 0% 0% 0% 0% 100% 95% 1% 4% 0% 100%
Final Average 0% 0% 0% 0% 0% 100% 95% 2% 3% 0% 100%

E1 0 0 0 0 0 13 13 0 0 0 13
E2 0 0 0 0 0 0 0 0 0 0 0
E3 0 0 0 0 0 4 4 0 0 0 4
E4 0 0 0 0 0 24 22 2 0 0 24
E5 0 0 0 0 0 31 29 1 0 1 31
E6 1 1 0 0 0 1 1 0 0 0 2
G1 0 0 0 0 0 5 5 0 0 0 5
G2 0 0 0 0 0 0 0 0 0 0 0
G3 0 0 0 0 0 0 0 0 0 0 0
G4 0 0 0 0 0 1 1 0 0 0 1
G5 0 0 0 0 0 0 0 0 0 0 0
G6 0 0 0 0 0 17 14 3 0 0 17
E1 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E2 x x x x X x x x x x x
E3 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
E4 0% 0% 0% 0% 0% 100% 92% 0% 0% 0% 100%
E5 0% 0% 0% 0% 0% 100% 94% 3% 0% 3% 100%
E6 50% 50% 0% 0% 0% 50% 50% 0% 0% 0% 100%
Average 10% 10% 0% 0% 0% 90% 88% 1% 0% 1% 100%

G1 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G4 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G6 0% 0% 0% 0% 0% 100% 82% 18% 0% 0% 100%
Average 0% 0% 0% 0% 0% 100% 94% 6% 0% 0% 100%
Final Average 6% 6% 0% 0% 0% 94% 91% 3% 0% 0% 100%

E1 0 0 0 0 0 0 0 0 0 0 0
E2 0 0 0 0 0 0 0 0 0 0 0
E3 0 0 0 0 0 0 0 0 0 0 0
E4 1 0 0 1 0 12 12 0 0 0 13
E5 1 1 0 0 0 9 9 0 0 0 10
E6 0 0 0 0 0 0 0 0 0 0 0
G1 0 0 0 0 0 29 29 0 0 0 29
G2 0 0 0 0 0 0 0 0 0 0 0
G3 0 0 0 0 0 0 0 0 0 0 0
G4 0 0 0 0 0 22 21 1 0 0 22
G5 0 0 0 0 0 0 0 0 0 0 0
G6 0 0 0 0 0 33 31 0 0 2 33
E1 x x x x x x x x x x x
E4 8% 0% 0% 8% 0% 92% 92% 0% 0% 0% 100%
E5 10% 10% 0% 0% 0% 90% 90% 0% 0% 0% 100%
Average 9% 5% 0% 4% 0% 91% 91% 0% 0% 0% 100%

G1 0% 0% 0% 0% 0% 100% 100% 0% 0% 0% 100%
G2 X x x x x x x x x x x
G3 X x x x x x x x x x x
G4 0% 0% 0% 0% 0% 100% 95% 5% 0% 0% 100%
G5 x x x x x x x x x x x
G6 0% 0% 0% 0% 0% 100% 94% 0% 0% 6% 100%
Average 0% 0% 0% 0% 0% 100% 96% 2% 0% 2% 100%
Final Average 4% 2% 0% 2% 0% 96% 94% 1% 0% 1% 100%

Cognates in language, in the mind and in a prompting
dictionary for translation
Maxim I. Stamenov
Abstract
Cognates are words that are shared by a particular pair of languages. Even
though they receive some attention in second-language learning and
translation training, the scope of the phenomenon and its sources is not
usually taken into account. Cognates are either true or false (the so called
“false friends” of the translator). However, the great majority in any pair of
languages is constituted by the “partial cognates”, i.e., cognates that share
one sense, but differ with respect to others. These are the cognates that offer
the greatest challenge to bilinguals, translators and compilers of dictionaries.
The main problem is how to assess, code, and make the differences in their
meanings accessible in a convenient way to the bilingual learner/user. The
most promising approach is to take into account the specificity of access to the
bilingual mental lexicon as studied in psycholinguistics. Both false and partial
cognates should be incorporated when developing computerized dictionaries,
notably prompting dictionaries. Prompting is assumed to alleviate translation
difficulties by offering a shortcut to those translation equivalents in the target
language that deviate most from the shared meaning components in a partial
cognate pair of words.
1. Introduction
Cognates are words in one language “that have the same origin as a word in
another language” (Longman 2005). False cognates refer to pairs of words in
the same or different languages that are similar in form and meaning but have
different roots, i.e., do not share a common origin. They are juxtaposed, on the
220 Maxim I. Stamenov
one hand, to cognates and, on the other, to false friends (or faux amis) that are
pairs of words in two languages that look and/or sound similar, but differ in
meaning. The asymmetric relationships between these three concepts result in
different usage and sometimes also in misunderstandings.
In the broader perspective adopted here, (true) cognates are words in
two languages that are identical or similar in form and meaning while not
necessarily sharing the same origin. False cognates, on the other hand, are
words in two languages that are identical or similar in form but which differ in
meaning and consequently may mislead the bilingual to think that they have
the same or similar meaning. In this broader sense, cognates and false
cognates (including false friends) are treated as a byproduct of the shared
genealogy of languages and/or of language contacts.
For English and German, for example, some of the frequently cited true
cognates, in terms of pronunciation and meaning, are:
(1) compatible kompatibel (tech., lit.)
competence die Kompetenz
diff. - Kompetenz in German refers more to “authority” or “jurisdiction”
competent (adj.) kompetent
diff. - Also “authorized” or “having jurisdiction”
drink trinken
edit (computing) editieren
editor (computing) der Editor [term]
learn lernen/erlernen
lie (recline) liegen
march die Mark (geog.) [term]
march v. marschieren
March der März (month)
operate operieren (medical)
A mere inspection of this short list allows us to distinguish between cognates

on several counts, e.g., on the basis of complete vs. partial overlap in meaning
or general vs. specialized usage in the two languages, as well as to discover
regular differences in similar word forms that are due to the
phonetic/phonological and morphological characteristics of the lexical units in
the two languages.
Although cognates are well known and are sometimes taught in a
systematic way in foreign language classes, it is more usual to find lists of
false cognates compiled for different language pairs. In (2) I offer a set of
Cognates in language, in the mind, and in a prompting dictionary 221
verbs in German-English-German presented in such a way as to make more

obvious what makes them “false” for a native German speaker who knows
English, i.e., in a direction-specific way:1
(2) absolvieren to complete a course or exam | absolve = lossprechen
bangen to be afraid/concerned | bang = knallen, r Knall
bannen to bewitch, captivate; excommunicate | ban = verbieten, s Verbot
bekommen to get, receive | to become = werden
blenden to dazzle; fade (cinema); blind, hoodwink | blend = mischen
irritieren to confuse, distract, put off | to irritate, annoy = ärgern, auf die Nerven gehen
(get on s.o.‟s nerves), reizen (skin), irritieren (bother)
konkurrieren compete | concur = übereinstimmen | conquer = erobern
mobben to harass, bully (at work) | to mob = herfallen über, belagern
picken to stick, be sticky; peck | pick = (aus)wählen/aufstellen (choose), entfernen
(remove), pflücken (flowers)
spenden to donate | to spend = ausgeben (money); verbringen (time)
synchronisieren dub (a film) | synchronize = abstimmen, gleichstellen
tasten to touch | to taste = kosten, schmecken
übersehen to overlook, miss (something) | to oversee = überwachen, beaufsichtigen
winken to wave | to wink = blinzern, zwinkern
The phenomenology of cognates and false cognates in different pairs of

languages varies a great deal (for overviews, see Granger and Swallow (1988),
Kileva-Stamenova and Dentscheva (1997), Friel and Kennison (2001), Lalor
and Kirsner (2001), Sherkina (2003), Gouws et al. (2004), Szpila (2005) and
Chamizo-Domínguez (2007).2
Two functions of cognates are often discussed, the first being that it is
possible to learn an L2 faster if it contains a large number of cognates from
the L1. The second is the problem of how to deal with “false cognates” in the
L2/L1 of a bilingual or translator. Even linguists, translation studies specialists
and psycholinguists are often not fully aware of the scope of the phenomenon.
2. Language history and language contacts in the proliferation of cognates
There are two ways in which a pair of languages may be close to each other or
grow closer to each other with respect to the form and meaning of vocabulary.
1
http://german.about.com/library/blcognates_C.htm, consulted 07.01.2009.
2
The reader is also referred to the online bibliography of false friends:
http://www.lipczuk.buncic.de/
The first is the shared origin of a pair of languages, e.g., English and German.
For example, English good and German gut are similar in form and have the
same meaning. The second is related to extensive contacts between two
languages, e.g., German and French, throughout the history of the two
cultures. When languages and cultures display trends of convergence in the
developments of certain features, we speak of so-called “language unions”. It
is worth mentioning here that the first example which was noticed by
European linguists and led to the formation of the concept was that of the
Balkan language union (Sprachbund).3 The degree of similarity of the
languages that are part of this union was studied both with respect to
convergent grammatical features and shared lexical resources. From what we
can see today, it seems that we have good reason to consider the European
area as a whole as developing into a language union, even including languages
that belong to different language families, e.g., Indo-European and Finno-
Ugric. The main reason for the higher degree of convergence (compared to
diversification) of the languages of Europe is the globalization of social,
political and economic life both within Europe and in the world in general.
This globalization calls for unification and standardization of different aspects
of communication and information exchange, including:
proliferation of internationalisms;
exchange and unification of terminology in all spheres of social, political
and economic life;
increased cultural borrowing and adoption of alternative styles of life from
other cultures that involve borrowing of appropriate vocabulary;
an increasing percentage of the European population becoming bilingual
or multilingual as a result of the need for intensive social, political and
economic contacts.
All these trends offer fertile ground for the further development of cognates.
This is especially the case in the context of the European Union as a unique
attempt at unification of a large number of economies, cultures and languages.
Cognates in the European languages are the result of:
1. shared language history;
3
By language union (Sprachbund) is meant a set of languages, geographically close but
not necessarily genetically related, in which similar phenomena can be found.
2. language contacts for general purpose interaction;

3. shared European cultural history and the tradition of ancient Greek and
Latin as its linguistic background;
4. shared languages of jurisdiction (Latin) and diplomacy (French) (for the
situation in Bulgarian (see Balkandgieva 1999, 2000; Yankova 2003,
2005);
5. shared scientific terminology (with their Greek and Latin background and
basis);
6. shared language of high technologies (nowadays mainly English);
7. language and communication via the mass media and Internet;
8. English as the lingua franca of the global village;
9. the creation and maintenance of multilingual institutions in Europe, e.g.
the European Commission;
10. the translation of large numbers of texts on a regular basis from/into
different languages within the EU, which currently has 23 official
languages;4
11. the shared territory of the EU, the common European identity and the
movement toward “common European” or European language union, i.e.,
the trend toward convergence of the languages spoken in the EU and in
Europe in general (cf. Haspelmath 2001; Ramat 2000).
It is not difficult to predict that with increasing intercultural and

interlanguage contacts in the EU, and in Europe as a whole, the borrowing of
words and expressions will accelerate in the future. Thus the study of cognates
and their functions becomes a window into the present and future of the
European cultural and linguistic area, as well as into the brains/minds of the
multilingual European citizen.
In this context it is perhaps appropriate to ask what the relationship is
between cognates and loan words in a pair of languages. The identification of
a certain word as a cognate or loan (if it is borrowed from a foreign language)
4
The 23 official languages in question imply 506 possible pairs. If we add the reversal of
the direction as a factor, the total number of possibilities amounts to 1012 pairs of source
and target languages for the purposes of translation in the EU (especially for the
European Commission). Thus ideally we have to develop and maintain in a unified
database 1012 bilingual dictionaries, including information on the use of false cognates.
In practice, however, it is well known that translations from certain languages to others
are done through the mediation of one of the major European languages, e.g., English,
French or German, and not directly.
is a matter of perspective. Loan words are treated as cognates as soon as L1

speakers know the L2 from which the loans have been derived.
3. Number of cognates and their importance for specific purposes
It is obvious that languages that are genetically close to each other and/or have
an extended history of language contacts in the past will have more cognates
compared to languages that are distant from each other on both of these
counts. Thus we may expect that language pairs like Danish-Norwegian or
Spanish-Portuguese will share more cognates than, say, French-Hungarian or
Italian-Finnish. Note that cognates that are loans in one direction, e.g., from
English to Slovenian, may be present in much larger numbers than in the other
direction, e.g., from Slovenian to English. In other words, the distribution of
cognates is not only specific to a certain language pair but also depends on the
language direction.
How many cognates there are in a particular pair of languages is a more
nontrivial question than usually conceived. For example, the recent
publication of a dictionary of “shared and similar words” in Bulgarian and
Romanian (cf. Kaldieva-Zaharieva 2007) has shown that the two languages
share some 12,000 to 13,000 lexical entries of this type (including
terminology introduced in standard dictionaries, internationalisms, etc.). Thus
an impressive portion of the lexicon (some 20 % to 25 %) familiar to a
speaker of Bulgarian or Romanian (actively and/or passively) consists of
cognates (both true and false in the broader sense adopted here). Note that
these languages are not genetically closely related to each other – Romanian is
a Romance language while Bulgarian is a Slavic language.
The complete, or close to exhaustive, set of cognates for a pair of
languages, as the list of entries in Kaldieva-Zaharieva (2007) appears to be for
Bulgarian-Romanian, can help us become acquainted with the nature and
scope of the phenomenon in general, as well as select pre-defined types of
cognates that form subsets of the overall set. For example, quite a lot of words
from the complete set are internationalisms or specialized terminology which
may not be appropriate for translation for general purposes, but only for
translation in specialized fields, e.g., in the field of medicine. Accordingly, it

makes sense to develop one dictionary of false cognates with general meaning
and specialized dictionaries for the terminology in the fields in which a great
deal of translations are carried out within the EU and in Europe in general.
Alternatively, it may be useful to distinguish between different types of
cognates from the point of view of similarities and differences in their form
and meaning (Sections 6 and 7). The most problematic type is not that of false
friends that share no meaning whatsoever in a pair of languages, e.g., the
Bulgarian евентуално that sounds like eventually and means “possibly,
perhaps” (eventually is a false friend of eventuell on the same basis in English-
German), but the so-called partial cognates, i.e., cognates that share certain
senses only. Partial cognates are the most frequent type of cognate in general
usage. This is also the type that is most likely to proliferate in the future for
the simple reason that in the majority of cases of word borrowing the meaning
taken in the target language (especially if we are dealing with a word that is
both in general and specialized use) happens to be one among several
available meanings in the source language.
The cognates in general usage form the most significant subset of the
overall set in terms of possibilities and frequency of use. For English-
Bulgarian, in my own count, this subset consists of 3,046 words (2,868 of
which are partial false cognates and 178 false cognates from a dictionary of
60,000 entries taken as a base); the majority of these are partial cognates. A
comparable set of false cognates is available for English-German (2,500
entries in Bennemann et al. 1994) and for French-German (2,800 entries in
Kühnel 1995). It should be mentioned that I have excluded here most of the
terminology that is usually included in general purpose bilingual dictionaries
and related to fields such as medicine, law, legislation and jurisprudence, and
business and finance. In my opinion, the cognates in terminologies and
languages for special purposes should be dealt with in specialized dictionaries
and/or translation memories.
A comparison of the distribution of false (full and partial) cognates in
different dictionaries shows both comparable trends and certain differences
with respect to decisions as to what to include. For German-English, Barnickel
(1992) includes 747 partial cognates and 108 false cognates that share no
meaning whatsoever. If we compare this dictionary with Bennemann et al.
(1994), there is an obvious difference in terms of coverage of the overall set of
cognates. What is at stake here is a difference in the selection of false cognates
and the purpose of the dictionary in question. Although it seems obvious that a
dictionary of false friends implies by definition a clear criterion for selection
and inclusion of the entries, the difference in the two cited above shows that
this is not yet the case. Barnickel (1992) is not an exception in this respect.
Dictionaries of “false friends” for other language pairs are also selective in
their coverage. Gottlieb (1985) includes about 400 entries for German-Russian
– with different denotations, different aspects of meaning, different styles of
use and, finally, cognates that are restricted in different ways on the basis of
their phraseological relations. Schwarz (1993) in his dictionary of false friends
in Danish-English has 1,610 entries which would appear to be a much more
representative sample compared to Barnickel (1992) or Gottlieb (1985),
although again, it probably does not come close to being truly representative.
While aiming at an exhaustive coverage of false cognates may appear to be an
ideal that in practice is never achieved fully, the point of such a goal is more a
matter of functional justification – from what perspective and for what
purpose we intend to make a selection of false cognates from the overall set.
4. True and false cognates in the terminology of law, legislation and

jurisprudence in English-Bulgarian
Ideally, terminology and terminological use of words in a pair of languages

must be based on the similarity of their meanings (as terms) even in cases
where there are considerable deviations in their word forms. In other words,
when it comes to borrowing and cognates, terminology should be considered,
by definition, an area where we must have true cognates only. In practice,
however, there are areas of terminological usage that in a certain language
may depend on borrowing from different culture-specific traditions and where
we may consequently find a different conceptualization and use of the same
words as terms. This is the case, for example, in fields like law, legislation and
jurisprudence, medicine, or business and finance, where there are considerable

language and-culture-specific differences. Such differences justify the
development of specialized dictionaries of false cognates, having in mind the
importance of the subject matter they refer to and the large number of texts
that are translated on a regular basis, e.g., within the EU. In order to illustrate
the nature of the problems, some English-Bulgarian examples from the fields
of law, legislation and jurisprudence will be given below (related to these are
the fields of business and finance as represented in Balkandgieva, 1999,
2000).
In dealing with false and partial cognates there are many cases where
we have words in the source language that have both common-usage and
specialized/terminological meanings. It is the specialized meanings of such
words that nowadays tend to be borrowed and turned into cognates. The
general-usage senses are not borrowed. In this way, we introduce into the
target language a set of partial cognates with true cognate relationship in the
terminological or special language sense that are false cognates in general use:
(3) a. to discount v
1. “to regard something as unlikely to be true or important”;
2. “to reduce the price of something” = дисконтирам “to reduce the price
or charge of something”;
b. discount n “a reduction in the usual price of something” дисконт(o) “a
reduction in the interest rate that the bank makes in the case of policy or bill
payment before maturity (in the case such a reduction applies according to
the policy‟s conditions)”
As can be seen in (3a), sense 2 of the English word discount as a verb is

identical to the one in use in Bulgarian with the restriction that in Bulgarian
the cognate in question is used only in the specialized discourse of commercial
law and business. The other sense of the verb to discount in English (“to
regard something as unlikely to be true or important”) is a false cognate. This
example looks, at first glance, as a straightforward case of a deviation along
the lines described above. Even such a “simple” case, however, can offer
additional challenges if we consider that the word discount in English is both a
noun and a verb. As a noun (3b) it has only one sense, the one that matches the
meaning of the Bulgarian word as a verb. As a noun, however, it turns into a
false cognate to the corresponding English noun. This is the case because the
meaning of the noun in Bulgarian is used in an even more specialized sense

(due to having been borrowed originally from Italian).
Another problem, this time closer to linguistic matters, can be illustrated
by the word (and concept of) clause in English compared to Bulgarian:
(4) clause n
a. “a group of words that contains a subject and a verb, but which is usually
only part of a sentence” = изречение “sentence”;
b. “a part of a written law or legal document covering a particular subject of
the whole law or document” = клауза “an item in a legal document,
stipulation that says that a particular thing must or must not be done”
In Bulgarian, clause in (4a) is translated as изречение “sentence“. In this

language it is not possible to differentiate between clause and sentence. As
soon as a string of words has a subject and a finite verb form, it is
automatically classified/identified as a sentence. The difference between
clause and sentence must be made descriptively in juxtaposing a simple to a
complex sentence. Thus it is difficult for a Bulgarian bilingual to associate the
second meaning of clause in English, which is a true cognate in Bulgarian
(4b), with the first, which must be obvious and appear logical only to a native
speaker of English (as it is in common usage). This circumstance makes the
semantic motivation of clause based on its meaning in (4a) opaque to a
Bulgarian bilingual in its terminological usage in (4b).
Terminological cognates may become potentially even more misleading
in the case of phraseological units which are composed of them, e.g., when
one of the words is a true and the other a false cognate. For example, a
Bulgarian bilingual would be inclined to make an error in translating the
Bulgarian конфискувам документ “to seize a document” as *to confiscate a
document, because both конфискувам and документ are cognates, but the
first, in this case, is a false cognate whereas the last is a true cognate. There
are quite a lot of such cases both in general and specialized usage.
5. False cognates in a pair of languages vs. homonyms in a single language
Longman (2005) defines a homonym as follows: “a word that is spelt the same
and sounds the same as another, but is different in meaning or origin. For
example, the noun „bear‟ and the verb „bear‟ are homonyms”. A false cognate
could be defined in the same way – “word that is spelt the same and/or sounds
the same as another, but is different in meaning” – though not in a single, but
in two different languages. There are, however, not only similarities but also
important differences between homonyms and cognates that may help us
distinguish between them and that also may result in differences in the
behavior of the monolingual vs. bilingual mind from a psycholinguistic point
of view.
To start with, it is remarkable how small a number of homonyms really
matter, i.e., how few may be confused and lead to misunderstanding during
monolingual language use. Even if, nominally, there may appear to be a great
deal of homophones and homographs in a single language (e.g., Hobbs 2006
lists 9,040 homophones and 2,133 homographs in American English and one
would expect a comparable, if not identical number in British English), these
are words whose “other”, i.e., nondominant, meanings are, as a rule, quite
exotic and of low frequency compared to the dominant one. Homonyms
whose two meanings occur at a ratio of 5:95 or more in favor of the
nondominant meaning in English are restricted to a few hundred occurrences.
In addition, one should, for practical purposes, exclude from consideration the
phenomenon of grammatical homonymy in English where the basic word
forms of nouns and verbs like cut or flirt are identical. This is a kind of
homonymy which is tolerated because it is easy to disambiguate during actual
comprehension and production because they belong to two different word
classes and are automatically identified as one or the other by the
speaker/listener processing the sentence.
If we attempt to find cognates that have characteristics similar to the
homonyms within a single language, best fit seem to be the ones which have
here been termed “chance homonyms”, i.e. that are not considered cognates,
but words where there is a chance coincidence of the form, which can happen
in any pair of languages. In cases like these we have meanings that have
nothing to do with each other. In English-Bulgarian, for example, chance
homophones that overlap completely with certain English words are, e.g., boy
(бой “fight1; height2”), call (кол “stake, post”), crust (кръст “cross”), dim
(дим “smoke, fume”), job (джоб “pocket”), list (лист “sheet”), lost (лост
“bar, lever”), luck (лък “bow”), mass (мас “fat; ointment”), most (мост
“bridge”), must (мъст “retaliation”), star (стар “old”), sun (сън “dream”),
talk (ток “current1; heel2”), vest (вест “news”). There are not very many in
English-Bulgarian, as the proliferation of such “homonyms” is prevented by
differences in sound and syllable structure (pronunciation), the difference
between the Latin vs. the Cyrillic alphabet (reading and writing) and the
phonetic vs. morphological conventions in the writing systems of the two
languages; they amount overall to some 60 words (from a bilingual dictionary
of 60,000 entries). They are the best representatives of interlingual
homonymy, if we take as our criterion the correspondence in form and the
incommensurability of meaning. But they are a curiosity of no practical
consequence whatsoever, e.g. talk (as a pronounced word form) in Bulgarian
means either “current” or “heel” (because in Bulgarian itself it happens to
have two homonymic meanings). The cognates on the other hand are of
interest for us inasmuch as they may provide a bridge to the meaning in
another language or mislead us to think that there is such a link. In other
words, they are not based on the strict incommensurability of meaning that
forms the core of the concept of homonymy.
6. The form of cognates
In lexicological and lexicographic practice, dictionaries of true and false

cognates have so far been compiled without computation of any quantitative
measures of the degree of formal similarity of the words (see above). Thus, for
example, for etymologists and historians of language it seems obvious that to
think in English and denken in German are similar in form. When they are
matched formally, however, English shares only two letters and phonemes
with the German word. Thus, from a quantitative point of view, the similarity
of these two words is not very high. So far the selection of cognates in
dictionaries has been made on the basis of the linguistic intuitions, and at the
discretion of, the compilers in charge. On the other hand, however, it is well
known from psycholinguistic research that the recognition of a word form as a
cognate in a particular pair of languages by a bilingual is, in general, a

function of the degree of form match/similarity (e.g., Dijkstra et al. 1999), i.e.,
the higher the form match, the higher the probability that the bilingual will
react to it as a cognate during different linguistic tasks involving lexical access
and retrieval.
It is worth noting, especially in languages that differ widely with regard
to their rules of word form derivation and inflection, that bilinguals appear to
be sensitive to the inner morphological structure of word forms and to
recognize cognates on the basis of their stems (and their respective meanings)
even if, nominally, their overall forms may deviate considerably from each
other in terms of a strict sound/shape match (Sanchez-Casas and Garcia-Albea
2005) as a result of the addition of prefixes, postfixes and endings.
If we consider the overall chances of match of word forms in two
different languages, the possibilities are as follows:
1. in the phonetic and phonological structure;
2. in the syllable structure;
3. in the morphological structure (derivation and inflection) via regular/
irregular means (the generative lexicon);
4. in the orthographic codification of the afore-mentioned levels of structure
(phonetic, syllable, morphological) of a word or word-like expression.
As can be seen from this list of options, the chances of complete formal
match between words in two lexicons should be rated as very low. Even in
language pairs like English and German with their common genetic origin and
extensive history of language contacts, there are not many words that sound
and mean exactly the same. According to Barnickel‟s (1992) count, there are
only a few hundred words that are both written and pronounced with stress on
the same syllable in both languages. These are English words like bonus,
clinic, ego, embargo, hibiscus, balustrade or blockade. It should be noted that
even in these cases of “absolute” fit, in German the words are written,
following the orthographic conventions, with a capital letter. Thus at the level
of pronunciation and spelling English and German tend to be differentiated in
a quite well-defined way, even in the case of completely true cognates, which
is the most favorable condition for having “the same word” in two lexicons.
Once we are reminded of how different even closely related languages

are in their sounds and/or orthography, it comes as a surprise that the mind
apparently disregards the full potential of taking advantage of the difference in
word forms in real life when accessing word representations in the bilingual
lexicon. There is a large body of literature in psycholinguistics documenting
this point (= the language nonspecific access to lexemes in the bilingual
lexicon; see Bonin 2003 and Dijkstra 2006 for an overview; see also below for
further discussion), but I would like to illustrate it here with a personal
vignette. Some months ago a student asked me if I knew a particular word in
Bulgarian, my mother tongue. The word was какалашка, spelled as
kakalashka. This is an item that is considered rare and either indicated in
dictionaries of contemporary Bulgarian as dialectal or excluded (e.g., from the
most recent works of „middle‟ size, i.e., approximately 60,000 entries). The
word in question means “corn cob”. Upon hearing the word my automatic
response was that I knew it (as I felt it sounded familiar), but afterwards I
found myself unable on the spot to work out what it was supposed to mean.
What came to mind was that it might refer to an insect or insect-like creature.
This discrepancy between the familiarity of the form and the failure to work
out what it meant embarrassed me. Some hours later on the way home I again
meditated on this word, and this time I worked out why I thought I knew it in
my native language (while actually being misled by the word form). The word
in question sounds like two words in two other languages I know. It is similar
in shape to der Kakerlak (“cockroach” in German) and to the Russian
букашка (spelled bukashka), which means “insect”. It is on the basis of this
trilingual lexical (word-form) relationship that I thought that I knew the word
and afterwards with a degree of effort came up with the meaning of “insect”
while the true meaning of the word in Bulgarian, which has nothing to do with
insects, remained inaccessible. This outcome is possible if, and only if, in
accessing my lexicon I perform a fuzzy match – both in terms of form and
meaning – which is not language-specific, even if the form in each language
looks so different in juxtaposition.
7. Types of cognates with respect to meaning identity/difference
Barnickel (1992: 60-61) differentiates in a systematic way between the

following types of true/false cognates in English-German with respect to
meaning relationships:
1. words with the same meanings (true cognates) – mayonnaise, museum,
fauna;
2. words with additional meanings in German (partial false cognates from
English into German) – neu-new, Seite-side, Tat-deed;
3. words with additional meanings in English (e.g., Roman words) (partial
false cognates from German into English) – Aktion-action, Balance-
balance;
4. words with additional different meanings in both languages (partial false
cognates in both directions) – Apfel-apple, Auge-eye;
5. words used in different style registers – forcieren-to force (in German
used in more formal speech registers);
6. words with no senses in common (false cognates par excellence) –
Akkord-accord, dezent-decent.
The taxonomy in question identifies what sort of information we need in

order to successfully recognize the differentia specifica in the meanings of the
cognates in a certain language pair:
(a) differences in the lexical meaning of the word, e.g., differences in the
polysemous structure of a word;
(b) differences in the meanings of a word that can best be illustrated in terms
of the collocations they may enter into;
(c) differences in usage, e.g., in terms of style register(s), esp. where the
difference may lead to mistakes related to formal vs. informal or polite-
impolite language use.
As we will see below, the first two ways of differentiating between false
cognates seem to be sufficient in the majority of cases, i.e., enough for the
practical purposes of translation, at least in the direction from L2 to L1. It is
on the basis of these that a computerized dictionary of false cognates for a
particular pair of languages can be developed.
8. The structure of an entry for a cognate – an example
The main problem in dealing with partial cognates is how to assess, code and
make the differences in their meanings accessible for the user in a convenient
way. In this respect, the first thing to remember is that the cognate
relationships in a pair of languages are, as a rule, not symmetrically reversible,
i.e., a set of cognates prepared first for use in the direction English-German
cannot automatically be used in the opposite direction. This means that each
set of cognates must be treated in a direction-specific way and later presented
in two separate dictionaries. This circumstance means that it is necessary to
take into consideration the direction of translation to a much higher degree
than is the case in the currently available dictionaries of “false friends”, e.g.,
Barnickel (1992) or Schwarz (1993). Actually, the same applies to the
treatment of cognates in regular bilingual dictionaries because the compilers
assume that the future users share their L1 and do not include information they
assume is familiar in the L1. This means that, all other things being equal, a
German will benefit most from an English-German-English dictionary
compiled by German lexicographers while British users will feel more at
home with a dictionary prepared by British professionals in the same field of
expertise.
The points made above can be illustrated by means of an entry for the
partial cognate verb to think/denken in English-German and German-English
dictionaries, viz. the Langenscheidt Handwörterbuch English (2001),
reproduced in (5); Oxford Duden German Dictionary (1999), reproduced in
(6); and in Barnickel (1992), reproduced in (7):
(5) to think (itr.) 1. denken (of an acc.); ~ ahead vorausdenken, a. vorsichtig sein; ~
aloud laut denken;
2. (about, over) nachdenken (über acc.), sich überlegen;
3. ~ of (a) sich besinnen auf, sich erinnern an; (b) et. bedenken; (c)
sich et. denken; (d) Plan etc. ersinnen, ausdenken; (e) halten von;
4. meinen, denken;
5. gedenken, vorhaben, beabsichtigen;
to think (tr.) 6. et. denken; ~ away et. wegdenken; ~ out (a) sich et. ausdenken;
(b) Am. a. ~ through Problem zu Ende denken; ~ s.th. over sich et.
überlegen or durch den Kopf gehen lassen; ~ up Plan etc. aushecken,
sich ausdenken, sich et. einfallen lassen;
7. sich et. denken oder vorstellen;

8. halten für;
9. überlegen, nachdenken über;
10. denken, vermuten.
denken tr. and itr. I. think; (nachsinnen) reflect; logisch: reason; (vermuten)
think, imagine;
sich etwas ~ (vorstellen) imagine;
~ an think of, (sich errinern an; nicht vergessen) remember, (in Sinn
haben) have in mind, think of.
(6) to think 1. v.t. A. (consider) meinen; we ~ [that] he will come wir denken
od. glauben, dass er kommt; B. (coll.: remember); C. (intend), D.
(imagine) sich vorstellen;
2. v.i. A. [nach]denken; B. (have intention);
3. ~ about v.t. A. (consider) nachdenken über (+ Acc.); B. (consider
practicality of) sich (Dat.) überlegen;
4. ~ ahead v.i. vorausdenken;
5. ~ back to v.t. sich zurückerinnern an (+ Acc.);
6. ~ of v.t. A. (consider) denken an (+ Acc.); B. (be aware of in the
mind) denken an C. (consider the possibility of) denken an (+ Akk.);
[…] F. (remember) sich erinnern an;
7. ~ out v.t. A. (consider carefully) durchdenken; B. (devise) sich
*Dat.) ausdenken <Plan, Verfahren>;
8. ~ over v.t. sich (Dat.) überlegen; überdenken;
9. ~ through v.t. [gründlich] durchdenken <Problem, Angelegen-
heit>;
10. ~ up v.t. (coll.) sich (Dat.) ausdenken <Plan>.
denken 1. itr. think (auf, über);

2. tr. think;
3. refl. A. (sich vorstellen) think; B. sich (Dat.) etw. by etw. ~
(beabsichtigen) mean sth by sth.
(7) denken = think What do you think about it? | to think badly about sb | I think the
same. | What will people think! | She doesn‟t think anything of it.
denken Ich denke mit gemischten Gefühlen daran. I have mixed feelings
about it. | Denk d‟ran! Don‟t forget! | So war das nicht gedacht. That
wasn‟t what I had in mind. | Ich habe mir das so gedacht. That‟s what
I had in mind.| Ich darf gar nicht dran denken. It doesn‟t bear
thinking about. | Ich denk‟ nicht daran! No way (I‟m going to do
that)! | Wo denkst du hin! What an idea! | Der Mensch denkt und
Gott lenkt. Man proposes, God disposes.
think (nach can / could) verstehen: I can‟t think why he did it. | (infml) to
think big gross planen | es sich gut überlegen: I should think twice
before accepting this offer.
In entries (5) and (6) for to think/denken the phraseology component (which is
quite developed) has been skipped, because the point is to display the different
ways of presenting the structure of the polysemous entries.
When comparing the British and German lexicographic traditions of
dictionary entry construction, one can see similarities as well as considerable
differences. Both Langenscheidt and Oxford Duden devote much less attention
to the elucidation of the meaning of denken than to to think (as it has a less
developed and more straightforward meaning structure in German compared
to its English cognate), though Langenscheidt seems to be more thorough.
Here the guiding principle for grouping the information for to think in German
is the thematic and sense structure expressed by the word, while for the
Oxford Duden the guiding principle in organizing the entry is the way in
which the verb forms are constructed. The two alternative choices result in
different strategies for information search and retrieval in the two dictionaries
– either from sense to form (Langenscheidt) or from form to meaning (Oxford
Duden), especially when looking for a translation match for the phrasal verbs.
Each of the strategies has advantages and disadvantages. The challenge
remains to discover which of the two is more helpful to a bilingual or a
translator during online dictionary use. To my knowledge, there are no
investigations concerning the usability of information structure of a lexical
entry in a dictionary. The requirements that dictionary developers are
currently concerned with are lexicological and lexicographic concerns and
conventions. Nevertheless, it is obvious that optimizing the conditions for
finding information in a dictionary entry is comparable to other types of
information search and retrieval by operators of a human-computer system,
e.g., websites, etc. It would suffice to open the two dictionaries cited above in
order to show that in the case of words with many senses and developed
phraseology the search for the appropriate meaning in a particular context may
become quite a demanding task in itself. And this turns out to be the case even
more so with the advent of computerized dictionaries when the search is done
on the computer monitor and the information displayed is dense and can
occupy more than a single page on the screen.
Entry (7) offers us an additional opportunity to compare what is
currently available in an entry for a dictionary of false cognates with the
entries in bilingual dictionaries. One notices considerable differences in the
treatment of to think/denken in Barnickel (1992) compared to the bilingual
dictionaries, which are likely to be related to the presupposition that users of a
dictionary of cognates will consult it only in cases when they feel that they
may be misled by the similarity of meanings in the two languages and/or
inability to find a correct translation for a phraseological unit containing the
word that is a cognate. This entry may also help us become aware that the very
format of presenting what is similar and what is different in the meaning
structure of a cognate in English-German-English offers a considerable
challenge from the point of view of direction specificity.
9. Structure of the bilingual lexicon, lexical access, and word recognition
It was pointed out above that in two different languages the chance of
complete overlap in the form of words is rare – both with respect to true
cognates (Section 6) and chance homonyms (Section 5). This is confirmed by
our everyday experience – in hearing just a single word, we can detect that it
is not from our own language or that it is pronounced by a foreigner. This said,
it comes as a surprise that our own mental lexicons while performing their job
of lexical access and word recognition seem to neglect the differences both in
form and meaning that are established during higher-level language
processing. Psycholinguistic research into monolingual and bilingual lexical
access and recognition (for an overview see Bonin 2003; Dijkstra 2005, 2006)
has shown that L1 words that resemble words in the foreign language (with
respect to spelling and/or pronunciation) are easier to recognize than words
that are less similar in the corresponding language (and this seems to apply
cross-linguistically, too, at least if the same alphabet is in use in two or more
languages; cf. Schwartz & Kroll 2006: 975). This is called the neighborhood
effect. This effect interacts with another effect – that of frequency of
occurrence – in such a way that low-frequency words like mail from large
neighborhoods, e.g., rail, bail, tail, wail, sail, hail, are recognized faster than
low-frequency words from small neighborhoods. However, low-frequency
words like bog with high-frequency neighbors, e.g., dog, log, are recognized
more slowly due to the competition. Finally, high-frequency words are not
affected by neighborhood size. This is the sort of dynamics whose impact we
cannot measure and evaluate without carrying out strictly controlled
psycholinguistic experiments. Such experiments are able to display aspects of
the very fast automatic processes enacted in the mental lexicon when
performing lexical access and recognition of words. After taking into account
the details of subjects‟ access to word forms, we can appreciate in a more
informed way why the nominal availability of so many homographs and
homophones in English (cf. Hobbs 2006) does not pose an obstacle to fast and
effective disambiguation already at the level of lexical access, i.e., before
taking into account the impact of sentence processing, context and situation in
which a particular word occurs. This behavior during lexical access by
approximation prepares us to carry out the selection and recognition of word
forms and also provides us with an explanation as to why, when it comes to
cognates, we do not need complete form overlap in order to perceive them as
“the same” if they share meaning – the neighborhood effect helps boost the
performance of the mental lexicon during access and recognition of cognates.
The effects not only of lexical form but also of semantic neighborhood
in accessing the mental lexicon appear to be rather complex. The effects of
semantic neighborhood size, e.g., in terms of semantic features, associations,
number of related meanings of different words and/or senses of a polysemous
word, is also found to have an impact both during monolingual and bilingual
lexical access. Printed words with rich semantic representations like to think
are recognized faster than words with poor semantic representations like to
bicker. For further detail and discussion of semantic influences on word
recognition, see e.g., Rastle (2007: 81-82). This again, looks counterintuitive,
as we would be inclined to believe that during word recognition one would
need more time to deal with more information than less information, but the
way we access the mental lexicon differs from the way we obtain information
from a dictionary. However, this does not mean that we cannot exploit the
specificity of lexical access and use it to optimize the way we organize and
access information from dictionaries, e.g., through prompting with
appropriately calibrated minimum information and achieving broadly
conceived semantic impact that helps users of a dictionary in performing their
tasks, e.g., writing a translation.
Psycholinguistic studies of priming are close to our concerns with
respect to prompting the translator (Section 10 below). Priming has been
conceptualized as one of the key tools used to study a variety of word and
memory processes, for example, word recognition and the structure and
mechanism of the mental lexicon. In priming experiments, subjects are usually
presented with pairs of items displaced in time. The first item, or “prime”,
serves to establish some type of context and the second item, or “target”, is
usually manipulated so that it either fits or does not fit into the prime context.
Numerous studies have shown that processing of the target can be greatly
influenced by the nature of the relationship between the prime and target
stimuli. It is well established, too, that there are different priming effects
owing to different types of priming. For instance, we have priming from
visual, phonological, semantic and syntactic phenomena. Especially pertinent
in our case are studies involving tasks where the prime and the target are in
two different languages, e.g., think and denken. In this context the study of
cognates becomes a window into the bilingual lexicon and the way of
representing and activating the lexical entries that share aspects of meaning
and/or form. And here, again, it is important to acknowledge that priming is a
mechanism that reflects the way the mental lexicon works while in developing
a dictionary in which we intend to exploit priming-like strategies of reminding
the translator, we have to reinterpret anew what the purpose would be of
presenting the words in L2 and L1 in such a way.
The point of the present section has been twofold. On the one hand,
there is a large amount of experimental research and accumulated knowledge
in psycholinguistics concerning the nature and the mechanisms of the
(bilingual) mental lexicon. This work has contributed to our understanding of
how the mental lexicon operates, as well as to what extent it differs in its
principles of organization from the dictionaries we have developed and use for
different purposes. On the other hand, up to the present time the knowledge
acquired within psycholinguistics has not been taken into account for the
purposes of optimizing the structure of computerized dictionaries in line with
the potential provided by the modern human-computer interaction
technologies. If this is the case, we need to develop user-friendly dictionaries
that provide information about the lexical and phraseological entries in a
format (or a set of formats) that conforms to the way the mental lexicon
functions, and thus optimize its performance when carrying out language
comprehension and translation tasks. One way of achieving such optimization
is by the concept of prompting (see below).
10. Prompting – the very idea
Many years ago Haas (1962: 48) pointed out that “[t]he perfect dictionary is
one in which you can find the thing you are looking for preferably in the very
first place you look.” The idea looks fascinating but almost immediately one
becomes aware that what one would expect to find first in a lexical entry may
differ according to the circumstances, including variable word forms, variety
of senses, context, situation and many others. The possibility, correspondingly,
to realize this idea is rather like a mission impossible, unless, for instance, the
computer in a human-computer interaction system, can find out before us what
we need and offer it as the first suggestion in any cognitive task involving
information search and retrieval in general and in looking for the meaning of a
word in a specific situation and context in particular.
The idea of prompting can be seen as a way to operationalize Haas‟s
idea and make the mission possible, at least for a specific purpose. It is based
on the following intuition. Even if we know what a word in a source language
means, from time to time its translation equivalent in target language may not
be available when we need it. Sometimes we can translate a word
immediately; sometimes, however, we have to search in our memory for a
longer time (for a whole set of different reasons) in order to find the correct
translation equivalent. During this time we “look around” and, suddenly, we
have an „aha‟ experience: the word we looked for has popped up. Prompting
serves the function of alleviating the search for an appropriate word in target
language under the assumption that translators already know the word in the
source language and its meaning, but have the problem of finding a word in
target language that matches its meaning. One possible strategy in developing
a prompting dictionary is to follow the logic of abridgement of lexical entries
used when developing pocket format dictionaries from ones that are larger and
more detailed in their treatment of lexical entries. This strategy, however,
cannot fit our requirements and orientation because of the radically different
purposes pocket dictionaries and prompting dictionaries serve. The former
helps us find the meaning of the word we are looking for as a very first
approximation, i.e., it provides the most frequent usage of the English word,
while a prompting dictionary offers help in cases where this association is
difficult to establish or in order to prevent a translator from producing a
misleading association. Its aim is to help translators do their job without going
through an exhaustive and time-consuming search in conventional dictionaries
and databases, especially in cases where the available tools usually included in
a translator‟s workbench cannot help them to translate online as quickly as
desirable without distraction and extended searches (as is the case with e.g.,
computerized bilingual vocabularies and thesauri, translation memories and
machine translation assistants). It is important to acknowledge that the prompt
is conceptualized not on the basis of a certain model and/or theory in
linguistics or psycholinguistics per se, but in relation to the function it is
supposed to serve in a real world situation, namely translating from one
natural language into another. What and how we prompt is based on our
knowledge not only of the processes of lexical access in the human mind but
also of how they can be supported for the purposes of translation by means of
appropriate technology. In developing the prompting function a great deal of
knowledge and research accumulated in psycholinguistics on the nature of the
bilingual lexicon in general and of lexical access and priming in particular can
help us optimize this function for effective use.
11. Technologies that provide a shortcut to meaning: genies, popup

functions and eye-tracking aids
Technologies that provide shortcuts to meaning are already available on the

market and/or have been developed to the level of being capable of being used
in commercial dictionaries. For example, two monolingual dictionaries of
English include in their computerized versions applications of prompting-like
functions that are implemented (1) as a Genie (in OUP 2005) and (2) as a
popup function (in Longman 2005). The function of the Oxford Genie is
explained as follows in the Help section of the CD-version of this dictionary
(similar both in form and way of use is the popup function of Longman 2005):
This tool lets you look up words in the Oxford Advanced Learner‟s Dictionary, 7th
edition, while you are working with other programs, such as Internet Explorer and
Microsoft Word. This is the Genie window:
In Microsoft Windows when the cursor is pointing at a word in Internet Explorer,

the Genie will look it up straight away. It can do this in Microsoft Word too. […]
You can also look a word up by typing it into the search box and pressing the Enter
key, or by clicking on the magnifying glass symbol. (OUP, 2005)
From an inspection of the way the Genie works, it becomes obvious,

however, that the entries are not abbreviated as would be appropriate for the
purposes of reminding/prompting, i.e., they cannot serve a prompting-like
function. On the other hand, the way the Genie is implemented – activation
and popup as soon as the cursor points at a word on the computer screen for a
time longer than X (e.g., 1 sec) – offers a direct analogy to gaze-based
applications like iDict (see below). The difference between mouse-driven and
gaze-based applications is that the former is based on the intentional
manipulation by the user of the dictionary and the system, while in the latter
we have a human-computer interaction system where the system itself
monitors the behavior (eye movements) of the user.
Another technology capable of implementing the function of prompting
is the computerized tool iDict. It was originally developed within the EU FP5
IST project “Interacting with Eyes: Gaze Assisted Access to Information in
Multiple Languages (iEye)” with the intention of alleviating difficulties of
comprehension during reading. It creates a rectangular text mask (i.e. each
word is represented by a small rectangle that encloses this word) of the
document loaded before a user starts reading. Later, during the reading
process it uses gaze data from a Tobii 1750, or other eye trackers, to detect the
word being read (word-in-focus). iDict recognizes a word as “problematic”
when the reader looks at the word for longer than a certain threshold period of
time. It then shows as a prompt a translation word slightly above the
problematic word in a smaller and different-color font. If the reader is
dissatisfied with the translation, he or she can look at the “extended translation
area” panel on the right for full information from a dictionary entry. For
further details on iDict, see Hyrskykari (2006). This tool comes quite close to
a prompting dictionary with the caveat that it is not a tool for translation but a
comprehension aid and what is the best prompt for comprehension may not
necessarily also be effective for the purposes of translation. There is a second
difference that should be mentioned. Only one characteristic of the words in
the L2 was used as a criterion for the purpose of prompting-for-
comprehension, namely frequency.
The iDict was further developed for the purposes of the project EYE-to-
IT into a GWM/Translog tool and now possesses a functionality that provides
the potential for implementing a prompting dictionary. It can detect from the
gaze data certain parameters of eye tracking behavior that are considered to be
indicative of difficulty in reading and comprehending a text. It can be related
to a dictionary, and it can present the information in it in the three formats that
are discussed in section 12 below.5
12. A dictionary format for prompting
If we assume that the problem prompting should be capable of solving is to

find the correct translation equivalent for a word in the source language, i.e.,
that translators know the word in the source language but experience difficulty
in finding its equivalent in the target language, the most straightforward seems
to be the strategy to prompt them in three formats. The first shortcut is to
prompt them with one or two translations of the most deviating senses of the
partial cognate in L2 compared to L1. What is “most deviating” is a matter of
expert judgment (an example is provided in (8)a below) or of extended
experimental work, especially when we have as cognates words with a
polysemic structure in a pair of languages. In the majority of cases, however,
the senses that are asymmetric in the direction from the source to target
language will be restricted to one or two, thus making this maximally abridged
way of prompting feasible in practical terms. If this is unsuccessful, the
second strategy is to prompt the translator with the equivalent of all the senses
in which an entry may deviate in the source language as compared with the
target language. This is especially appropriate in cases where the entry in the
source language has a widely developed polysemous structure, as is the case
with the verb to think in English (examples 5 and 6 above). What we need to
achieve is an appropriate abridgement (or pruning) of these very long and
dense entries in the dictionaries mentioned above.
If we compare the treatment of to think in (5) and in (8b) below, we can
see that what is deleted in the latter compared to the former is all the
information related to grammatical categories (except transitivity), thematic
structure and phrasal combinations (including phrasal verbs). All this is
known to the translators, at least in the great majority of cases. What they are
looking for is a prompting to find the right translation equivalent. And this is
5
The latest version of GWM tool is available at Oleg Špakov‟s personal web page
http://www.cs.uta.fi/~oleg/gwm.html.
what should be made available for translating into German (on the basis of the
bilingual dictionaries) with the exception of the true cognate – in this case
denken. What is also made available is the basic polysemic structure of to
think (with the exception of the overlapping part). However, it is provided
only implicitly in the form of listing ten different senses of the verb given in
their translation equivalents in German. If translators have problems finding a
translation equivalent of a phrase that includes as a reference word to think,
we will prompt them in a separate part/space on the monitor with translation
equivalents of a set of phraseological units, as illustrated in (8)c below.
Bearing in mind that, e.g., in Longman (2005), the entry for to think (as
a verb) is presented as having 43 different senses (including phraseology plus
six phrasal verbs), the abridgment achieved in (8)b can be considered quite
significant (it remains to be verified with appropriate usability experiments
since it should not be developed ad hoc but on a principled basis). As I have
shown how the procedure of pruning can be carried out in a complicated case
with a word with a rich asymmetric polysemic structure, I assume that in the
great majority of other possible cases involving partial cognates this procedure
would be easier and more straightforward to enact.
Thus we arrive at the following structure for a prompting entry in a
dictionary of cognates. The first prompt is conceived as a shortcut to appear
above the problematic word as soon as a problem is detected. The potential
target-language equivalent (or at most two or three) in question should be the
one that is most difficult/problematic to think of during translation. The
second prompt would be made available in a separate window (as in the OUP
Genie) and would involve meaning-oriented prompting that provides all the
translation equivalents of a certain word (for meaning structure we have
chosen Langenscheidt 2001; cf. 8a and 8b below). On a third level, we may
prompt translators with the phraseological component of the standard lexical
entry in a dictionary (for phraseology in 8c preference was given to Oxford
Duden, 1999). All the information that is offered in the three components is
provided without abbreviations, tildes, etc., in an easily readable screen format
(unlike the strategy in the hardcopy dictionaries, which, in order to save space,
abound in problems as far as readability is concerned).
It is proposed that the prompting entry for to think in English looks as

follows:
(8) a. to think vorstellen, vorhaben, überlegen;
b. to think(tr/itr)1. vorausdenken, vorsichtig sein;

2. nachdenken, sich überlegen;
3. sich besinnen auf, sich erinnern an; (b) bedenken; (c) sich
denken; (d) ersinnen, ausdenken; (e) halten von;
4. meinen;
5. gedenken, vorhaben, beabsichtigen;
6. wegdenken; (a) sich ausdenken; (b) sich überlegen;
aushecken, sich ausdenken, sich einfallen lassen;
7. sich vorstellen;
8. halten für;
9. überlegen, nachdenken über;
10. vermuten.
c. to think (now that I) come to think of it = dabei fällt mir ein;

think much (highly) of = viel halten von;
think nothing of = wenig halten von / nichts dabei finden;
he thinks himself very fine = er meint er sei etwas Besonderes;
it is not thought proper = es gilt als unschicklich;
I thought as much/so = das habe ich mir schon gedacht;
I should think not! = auf keinen Fall!
You are a model of tact, I don‟t think! = du bist mir vielleicht
ein Ausbund von Taktgefühl!
I need time to think = ich muss es mir erst überlegen;
I‟ve been thinking = ich habe nachgedacht;
think on one‟s feet (coll.) = sich ausdem
Stegreif etwas überlegen;
It doesn‟t bear thinking about = man
darf gar nicht daran denken;
but I can‟t think of everything at once! = aber ich habe
schliesslich auch nur einen Kopf!
to think of it! =stell dir das bloss vor!
(now I come) to think of it, … = wenn
ich es mir recht überlege, …
not for a minute would she think of helping anybody else =
ihr würde es nicht im Traum einfallen, anderen zu helfen;
I couldn‟t think of such a thing = das würde mir nicht im
Traum einfallen;
we‟ll think of something = wir werden uns etwas einfallen
lassen;
what will they think of next? = was werden sie sich wohl noch
alles einfallen lassen?
I just can‟t think of her name = ich komme einfach nicht auf
ihren Namen;
Think little/nothing of somebody = wenig/nichts von
jemandem halten;
think not much of somebody/something = nicht viel von
jemandem/etwas halten;
I will think it over = ich lasse es mir durch den Kopf gehen.
The role and the effectiveness of the meaning-oriented simplification in the

computer-assisted entry presentation suggested here remains to be proven in
practice, but in principle such a dictionary with 2,500-3,000 entries, especially
targeting partial cognates, would not appear to be difficult to develop once the
strategy of prompting and the formats are established and verified in
appropriate psycholinguistic experiments. Some work has already been carried
out in the course of the EYE-to-IT project. In terms of basic research, some
experiments have been performed aiming at verifying the psycholinguistic and
neurolinguistic correlates of prompting in comparison to standard procedures
and research on priming (cf. Stamenov et al., forthcoming, and Section 10
above). In terms of usability research, a platform for integrating the
technologies for eye tracking, keystroke logging, commercial dictionary
software and formats for prompting has been developed. 5 The construction of
an actual dictionary for prompting purposes would require further experiments
amounting to a separate research program. Let me point out again that such a
prompting dictionary format, i.e., a format that comes closer to the way the
mind of a user works during lexical access and recognition, could, in
principle, be developed for any bilingual or monolingual dictionary and not
just for the sake of prompting cognates.
13. Conclusion
In this article several points have been made. Firstly, cognates are a significant
class of words in the bilingual mental lexicon of all individuals who master
more than one language. The percentage of such words may nowadays easily
reach 20 % to 25 % of the items the individual in question knows actively or
passively for languages with a common origin and/or that have had extensive
language contacts. Secondly, the main source of cognates nowadays is the
borrowing of words from one language into another. In the majority of cases
the borrowed words have more senses in the source language than in the target
language. This means that with more borrowing the cognates will not only
tend to proliferate but they will also as a rule be partial cognates that offer
specific challenges. It is for this reason that the development and maintenance
of specialized dictionaries of cognates are justified, especially when it comes
to the optimization of the work of professional translators. The available
dictionaries of false cognates, however, suffer drawbacks in two directions –
they are not developed to fit the direction of translation, and they follow the
traditional way of structuring the entries of a dictionary that is not optimal for
finding a shortcut to the meaning in translation. The shortcut proposed here is
conceptualized in terms of prompting translators and presupposes that all the
information related to the word in a text and its possible translation is known
to the users and they merely need to be reminded of the correct translation
match. Contemporary technology offers a range of possibilities to implement
such a prompting dictionary into a human-computer interaction system, e.g.,
one based on eye-tracking technology. Some of the possibilities of developing
a dictionary of cognates and of detecting with eye-tracking technology when
translators need information in order to prompt them have been explored in
the course of EYE-to-IT (cf. Gerganov et. al. 2008; Stamenov et al.
forthcoming) but much remains to be done in developing user-friendly
computerized dictionaries for different purposes with various formats and
ways of accessing and manipulating the information included in them.
Acknowledgment
The research reported in the present article was supported by a grant under the
auspices of the 6th Framework Programme of the EC, FET-517590,
“Development of Human-Computer Monitoring and Feedback Systems for the
Purposes of Studying Cognition and Translation (EYE-to-IT)”.
References
Balkandgieva, B. 1999. English-Bulgarian Law Dictionary. Sofia: Sofi-R.

Balkandgieva, B. 2000. Bulgarian-English Law Dictionary. Sofia: Sofi-R.
Barnickel, K-D. 1992. Falsche Freunde. Ein vergleichendes Wörterbuch Deutsch-
Englisch. Heidelberg: Julius Groos.
Bennemann, H., Herting, B., Prause, T. 1994. Typische Fehler Englisch: 2500
“false friends” Englisch und Deutsch. 2nd edn. Berlin: Langenscheidt.
Bonin, P. (ed.). 2003. Mental Lexicon: Some Words to Talk About Words.
Hauppauge, NY: Nova.
Chamizo-Domínguez, P.J. 2007. Semantics and Pragmatics of False Friends.
London: Routledge.
Dijkstra, T. 2005. Bilingual visual word recognition and lexical access. In: J. F.
Kroll & A.M.B. de Groot (eds) Handbook of Bilingualism: Psycholinguistic
Approaches. Oxford: Oxford University Press. 179-201.
Dijkstra, T. 2006. The multilingual lexicon. In M. Gareth Gaskell (ed.) The Oxford
Handbook of Psycholinguistics. Oxford: Oxford University Press. 251-265.
Dijkstra, T., J. Grainger & Heuven, W.J.B. van. 1999. Recognition of cognates and
interlingual homographs: the neglected role of phonology. Journal of Memory
and Language 41: 496–518.
Friel, B.M. & Kennison, S.M. 2001. Identifying German-English cognates, false
cognates and non-cognates: Methodological issues and descriptive norms.
Bilingualism: Language and Cognition 4(3), 249-274.
Gerganov, A., Kaiser, V., Braunstein, V., Popivanov, I., Brunner, C., Neuper, C., &
Stamenov, M. 2008. Priming bilingual brain with correct and incongruent
translations of true and false cognates in English-German during a translation
task: An EEG and eye tracking study. Poster presented at the Ghent Workshop
on Bilingualism, September 18-20, 2008, Ghent, Belgium.
Gottlieb, K. H. M. 1985. Wörterbuch der „Falschen Freunde des Übersetzers‟.
Moscow: Russkij Jasyk.
Gouws, R., Prinsloo, D. J. & Schryver, G-M de. 2004. Friends will be friends – true
or false. Lexicographic approaches to the treatment of false friends. In
EURALEX 2004 Proceedings: Lexicological Issues of Lexicographic
Relevance, July 06-10, 2004, Lorient, France. 797-806.
Granger, S. & Swallow, H. 1988. False friends: A kaleidoscope of translation
difficulties. Langage et l‟Homme 23 : 108-120.
Haas, Mary. 1962. What belongs in a bilingual dictionary? In F. W. Householder &
S. Saporta (eds) Problems in Lexicography. International Journal of American
Linguistics 28: 45-50.
Haspelmath, M. 2001. The European linguistic area: Standard Average European. In
M. Haspelmath, E. König, W. Oesterreicher, & W. Raible (eds) Language
Typology and Language Universals, (Handbücher zur Sprach- und
Kommunikationswissenschaft). Berlin: de Gruyter. 1492-1510.
Hobbs, J. B. 2006. Homophones and Homographs: An American Dictionary. 4th

edn. Jefferson, NC: McFarland.
Hyrskykari, A. 2006. Eyes in Attentive Interfaces: Experience from Creating iDict, a
Gaze-Aware Reading Aid. Doctoral Dissertation, Tampere, Finland, ISBN 951-
44-6643-8, Available at http://acta.uta.fi/pdf/951-44-6643-8.pdf.
Kaldieva-Zaharieva, S. 2007. Калдиева-Захариева, Стефана. 2007. Румънско-
български речник: общи и подобни думи [Romanian-Bulgarian dictionary of
shared and similar words]. София: Академично издателство “Марин
Дринов”.
Kileva-Stamenova, R. & Dentscheva, E. 1997. Килева-Стаменова, Ренета и
Емилия Денчева. Речник “неверните приятели на преводача”[Dictionary
of “false friends” of the translator]. Plovdiv: Lettera.
Kühnel, H. 1995. Typische Fehler Französisch (2800 “falsche Freunde”
Französisch und Deutsch). 6th edn. Berlin: Langenscheidt.
Lalor, E. & Kirsner, K. 2001. The representation of “false cognates” in the bilingual
lexicon. Psychonomic Bulletin & Review 8(3): 552-559.
Langenscheidt 2001. Handwörterbuch Englisch. Berlin: Langenscheidt. 1656 pp.
Longman 2005. Longman Dictionary of Contemporary English. Fourth edn with
CD-ROM. Harlow: Longman.
Oxford Duden 1999. Oxford Duden German Dictionary. 2nd edn Oxford: Oxford
University Press. 1728 pp.
OUP 2005. Oxford Advanced Learner‟s Dictionary. 7th edn with CD-ROM. Oxford:
Oxford University Press.
Ramat, P. 2000. The notion of „standard average European‟. Paper presented at the
conference Le lingue d‟Europa, October 12-14, 2000, Bologna, Italy.
Rastle, K. 2007. Visual word recognition. In M. Gareth Gaskell (ed.) The Oxford
Handbook of Psycholinguistics. Oxford: Oxford University Press. 71-87.
Sanchez-Casas, R. & Garcia-Albea, J. E. 2005. The representation of cognate and
noncognate words in bilingual memory: can cognate status be characterized as
a special kind of morphological relation? In: J. F. Kroll & A. M. B. de Groot
(eds) Handbook of Bilingualism: Psycholinguistic Approaches. Oxford:
Oxford University Press. 226-250.
Schwartz, A. & J. F. Kroll. 2006. Language processing in bilingual speakers. In M.
J. Traxler & M. A. Gernsbacher (eds) Handbook of Psycholinguistics. 2nd edn.
Amsterdam: Elsevier. 907-999.
Schwarz, H. 1993. False Friends: Lumske ligheder. Dansk-engelsk. Frederiksberg:
Samfundslitteratur.
Sherkina, M. 2003. The cognate facilitation effect in bilingual speech processing.
Toronto Working Papers in Linguistics 21: 135–151.
Stamenov, M.I., Gerganov, A. & Popivanov, I. (forthcoming). What can prompting
cognates reveal about the bilingual lexicon and the possibilities to optimize
access to it during translation. In G. Shreve & E. Angelone (eds) Translation
and Cognition. Amsterdam: John Benjamins.
Szpila, G. 2005. False friends in dictionaries. Bilingual false cognates lexicography

in Poland. International Journal of Lexicography 19(1): 73-97.
Yankova, D. 2003. Towards achieving equivalence in Bulgarian-English translation
of statutory writing. Foreign Language Teaching 3: 52-63.
Yankova, D. 2005. Culture-specific features of statutory texts. In M. Grozeva (ed.),
Spaces, Gaps, Borders. Sofia: St. Kliment Ochridski University Press. 15-21.
Notes on Contributors
Fabio Alves is Professor of Translation Studies and a researcher of

LETRA, the Laboratory for Experimentation in Translation, at the Federal
University of Minas Gerais (Brazil). He holds a PhD from the Ruhr-
Universität Bochum (Germany) with a process-oriented study of cognitive
differences and similarities observed among Brazilian and Portuguese
translators. His current research focuses primarily on the empirical-
experimental investigation of the translation process as well as on the
development of expertise in translation. His publications include articles in
Meta, Journal of Translation Studies, TradTerm and Cadernos de
Tradução as well as book chapters in the Benjamins Translation Library.
E-mail: fabio-alves@ufmg.br
Gerrit Bayer-Hohenwarter has six years of experience as a technical

translator and leader of a translation team in Graz. She was employed as a
translation teacher at the University of Graz and is now working on her
PhD on translational creativity. Her thesis forms part of the longitudinal
study TransComp (http://gams.uni-graz.at/container:tc). Her main research
interest lies in translation process research.
E-Mail: gerrit.bayer-hohenwarter@uni-graz.at
Tânia Liparini Campos has an MA in translation studies from

Universidade Federal de Minas Gerais (UFMG), Brazil, where she is
currently working on her PhD on the impact of translation technology and
time pressure on cognitive effort in translation, using key-logging, screen-
recording and retrospection to collect data. Her research interests include
cognitive processes in translation, human-computer interaction and time
pressure. She is currently associated with LETRA, the Laboratory for
Experimentation in Translation at UFMG, and the SEGTRAD project,
which investigates the impact of translation technology on the cognitive
process of professional translators.
E-mail: t.liparini@ig.com.br
254 Notes on contributors
Louise Denver is Associate Professor at the Copenhagen Business School,

Department of International Language Studies and Computational
Linguistics. Her research interests include linguistics, translation studies,
especially cognitive processes in translation, and currently she is engaged
in a project on the increasing use of English as a lingua franca in the
international degree programmes at CBS.
E-mail: ld.isv@cbs.dk
Dorrit Faber is Associate Professor in the Department of International

Language Studies and Computational Linguistics at the Copenhagen
Business School, where she teaches specialised communication and
translation, in particular English-Danish and Danish-English translation
within the domains of law and business economics. Her research interests
include translation studies and translation process research focusing in
particular on the translation of legal texts. With Mette Hjort-Pedersen she is
currently involved in a project on the phenomena of explicitation and
implicitation in legal translation.
E-mail: df.isv@cbs.dk
Susanne Göpferich is Professor of Translation Studies at the University of

Graz/Austria. From 1997 to 2003 she was Professor of Technical
Communication and Documentation at the Karlsruhe University of Applied
Sciences/Germany. Her main fields of research comprise text linguistics,
specialized communication and translation, comprehensibility research,
technical writing, translation theory and didactics, as well as translation
process research, the topic of her most recent book (Translationsprozess-
forschung: Stand – Methoden – Perspektiven, Tübingen: Narr, 2008) and
the longitudinal study TransComp.
E-Mail: susanne.goepferich@uni-graz.at
Website: www.susanne-goepferich.de
Mette Hjort-Pedersen is Associate Professor in the Department of

International Language Studies and Computational Linguistics at the
Copenhagen Business School. She teaches legal translation and business
communication. Her main fields of research comprise legal communication
and translation, translation theory, translation processes and lexical
Notes on contributors 255
semantics. With Dorrit Faber she is currently involved in a project on the

phenomena of explicitation and implicitation in legal translation.
E-mail: mhp.isv@cbs.dk
Arnt Lykke Jakobsen is Professor of Translation and Translation

Technology in the Department of International Language Studies and
Computational Linguistics at Copenhagen Business School, and Director of
the Centre for Research and Innovation in Translation and Translation
Technology, which he founded in 2005. He taught English literature at
Copenhagen University from 1972 to 1985, where he developed an interest
in translation. Since 1985 he has been with the Copenhagen Business
School. In 1995 he developed the first version of the software program
Translog, which is now a key technology in the EU Eye-to-IT project
(www.translog.dk and cogs.nbu.bg/eye-to-it) and is used to study writing
processes worldwide.
Email: alj.isv@cbs.dk
Website: www.cbs.dk/critt
Kristian T. H. Jensen has an MA in translation from CBS, where he is

currently working on his PhD on cognitive effort in translation, using eye
tracking and key-logging to collect data. His research interests include
cognitive processes in translation and human-computer interaction, and he
is currently associated with the EU Eye-to-IT project. In the spring of 2009,
he spent four months at the Dublin City University School of Applied
Language and Intercultural Studies. He has worked as a freelance translator
since 2003.
E-mail: kthj.isv@cbs.dk
Brenda Malkiel spent six years directing the Hebrew-English Translation

Program at Beit Berl College (Israel) and currently teaches in the
Department of Translation and Interpreting Studies at Bar-Ilan University
(Israel). Her studies of translation students include “The Effect of
Translator Training on Interference and Difficulty” [Target], “What Can
Grades Teach Us?” [Perspectives], “When Idioti (Idiotic) Becomes
Fluffy”: Translation Students and the Avoidance of Target-language
256 Notes on contributors
Cognates” [Meta], and “Translation as a Decision Process: Evidence from

Cognates” [Babel].
E-mail: brendamalkiel@gmail.com
Inger M. Mees studied at Leiden and Edinburgh before moving to the

Copenhagen Business School as Associate Professor teaching English
phonetics. Together with Inge Livbjerg, she has published research on
dictionary use in translation employing think-aloud protocols. With
Beverley Collins, she has co-authored books and articles on pronunciation
training, accent varieties, language change, and historiography. She is
currently working on a project on English-medium education in the
international degree programmes at CBS.
Email: im.isv@cbs.dk
Ricardo Muñoz has been a freelance translator since 1987. He graduated

in English Studies (Univ. Valencia) and Translation and Interpreting (Univ.
Granada), and was awarded a PhD in Hispanic Linguistics at UC Berkeley.
He coordinates the research efforts of the group Expertise and Environment
in Translation (PETRA, Spanish acronym). His main research focus is on
the interface between cognitive science and empirical approaches to
translation processes. Dr Muñoz is an associate professor of translation at
the University of Las Palmas de Gran Canaria.
Email: rmunoz@dfm.ulpgc.es
Nataša Pavlović teaches translation theory and practice at the University

of Zagreb, Croatia. She has a PhD in Translation and Intercultural Studies
from Universitat Rovira i Virgili in Tarragona, Spain, where she defended a
thesis on directionality in translation processes written under the
supervision of Gyde Hansen and Anthony Pym. Her research interests
include translation processes, directionality, translator education and
research methodology. She has worked as a freelance translator since 1991.
E-mail: natasa.pavlovic@zg.t-com.hr
Maxim Stamenov is a senior research fellow at the Institute of the

Bulgarian Language, Bulgarian Academy of Sciences, Sofia, where he has
been working since 1983. In January 2006 he joined the Central and
Notes on contributors 257
Eastern European Centre for Cognitive Science at the New Bulgarian

University (NBU) as scientific coordinator on behalf of the NBU of the
FET-517590 EYE-to-IT project (cogs.nbu.bg/eye-to-it) under the auspices
of the Sixth Framework Programme of the European Commission. His
research interests are in the field of language and cognition in general, and
language and consciousness in particular, including investigation and
modeling the processes of monitoring and control of language-specific
processing for different purposes.
E-mail: mstamen@nbu.bg

CBS in Language 37

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

CBS in Language 37

Caricato da

Copyright:

Formati disponibili

Behind the Mind

Behind the Mind

INTRODUCTION: BEHIND THE MIND OF TRANSLATORS................................ 1

Notes on contributors .............................................................................. 253

of introspective data in translation). The 1990s saw the development of

the present article she discusses how flexibility can be identified in

Another paper discussing methodological concerns in our attempts

they detail the problems involved in the set-up of experiments designed to

significance of typos is the very issue explored by Ricardo Muñoz. He puts

from the spatial limitations imposed by a screen, and therefore the

Copenhagen and Graz, April 2009

Alves, F. (ed.) 2003. Triangulating translation. Perspectives in Process Oriented

1. Translation competence acquisition – providing empirical evidence

The development of models of translation competence and, even more so,

and longitudinal studies are extremely cumbersome and time-consuming.

2. Central components of a translation competence model

There is consent among translation scholars that translation competence is

[T]he training of translators involves the creation of the following two-fold

2.1 Hönig’s model of an ideal translation process

Figure 1. Hönig‘s model of an ideal translation process (Hönig 1991: 79;

the risk of getting lost in the maze of micro-strategies in their controlled

The relationship between the two central components of translation

2.2 The PACTE group’s revised translation competence model

The PACTE group‘s first translation competence model dates back to

Figure 2. The PACTE group‘s revised translation competence model

3 What has to be taken into account in think-aloud studies is that according to

By bilingual sub-competence the PACTE group means ―pragmatic,

2.3 My translation competence model

translator’s self concept/

Figure 3. Göpferich‘s translation competence model

As illustrated in Fig. 3, I differentiate between the following sub-

respective cultures. Communicative competence in the source language

translation problems in a narrower sense, because from this the memory

3. The PACTE group’s translation competence acquisition model

Since, as we have seen, the translation competence models developed so

Figure 4. The PACTE group‘s translation competence acquisition model

The model is based on the PACTE group‘s revised translation competence

4. Findings in expertise research

The PACTE group‘s description of the development of translation

5. TransComp – a longitudinal study into the development of

As mentioned in Section 1, the above-mentioned assumptions and findings

translation competence if longitudinal study is understood in the strictest

5.1 The design of TransComp

In TransComp twelve undergraduate students of translation at the

experiments in order not to risk obtaining incomplete accounts in the

they are thinking aloud, interferences should be smaller than in studies

5.2 The focus of TransComp and the ‘measurement’ of (the development

TransComp will concentrate on the following components of translation

decisions, whereas students only gradually learn to develop a macro-

10 In another study, we also plan to investigate whether translation routine activation

automatically, to the latter category. Increased automaticity in a category is

6. Availability of materials and results

Anderson, J. R. 1983. The Architecture of Cognition. Cambridge, Mass.: Harvard

— & Charness, N. 1997. Cognitive and developmental factors in expert

Neubauer, A. & Stern, E. 2007. Lernen macht intelligent: Warum Begabung

are many concepts which cannot be defined by common properties with

weaknesses in different areas, e.g. high fluency but little evaluation

Creative procedures reviewed

Recent theories of creativity (e.g. Kußmaul 2000a) draw heavily on

Kußmaul’s types of creative translation

belong to different levels of abstraction. As explained before, frame refers

Three types of creative translation procedures

said to correspond to the levels of categorisation suggested by Brown