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

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

“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 nonabating 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 2

Webster‟s New World Medical Dictionary. See for a description of the project.


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; TirkkonenCondit 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 eyetracking, 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




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

Funded by the Austrian Science Fund (FWF), project No. P20908-G03 (2008–11).


Susanne Göpferich, Arnt Lykke Jakobsen & Inger M. Mees

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. BayerHohenwarter‟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.



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 HjortPedersen. 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,


Susanne Göpferich, Arnt Lykke Jakobsen & Inger M. Mees

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 selfcorrections 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 selfcorrections 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 selfcorrections 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



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


Susanne Göpferich, Arnt Lykke Jakobsen & Inger M. Mees

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 logicalsemantic 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



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: Eyetracking 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 (Copenhagen Studies in Language 27). Copenhagen: Samfundslitteratur. 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,

TransComp is funded by the Austrian Science Fund (FWF) as project No. P20908G03 (September 2008–August 2011).


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

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)


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.


Susanne Göpferich

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


Figure 1. Hönig‘s model of an ideal translation process (Hönig 1991: 79; terminology adapted to Hönig 1995: 51)

Towards a model of translation competence (acquisition)


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


Susanne Göpferich

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) (2) (3) (4) As a linguistic reflex stimulated by the first contact between the projected st [sc. source text] and semantic associations in the uncontrolled workspace. As an automatic transfer from the uncontrolled workspace after a macrostrategy has been worked out. As a product of a microstrategy applied in the controlled workspace which has been approved by monitoring. 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)

Towards a model of translation competence (acquisition)


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 subcompetences 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


Susanne Göpferich 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 substrategies 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).


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

Towards a model of translation competence (acquisition)


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 selfconcept, 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).


Susanne Göpferich

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 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 communicative competence in at least 2 languages working conditions (e.g. time pressure) domain competence

tools and research competence

strategic competence motivation

psychomotor competence

translation routine activation competence

translation norms translation assignment

translator’s self concept/ professional ethos topics covered and methods employed in theoretical & practical translation training psycho-physical disposition

Figure 3. Göpferich‘s translation competence model

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

For a ―psycholinguistic model of the translation process‖, see Kiraly (1995); for a critical review of Kiraly‘s model, see Göpferich (2008: 137 ff.).

Towards a model of translation competence (acquisition)






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. Targetlanguage 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. 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. 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. 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 targetlanguage 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). 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


Susanne Göpferich

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).

Towards a model of translation competence (acquisition)


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 subcompetencies, 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).


Susanne Göpferich

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 longterm 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

Towards a model of translation competence (acquisition)


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.

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.


Susanne Göpferich

Table 1. Translation scheme
Beginning of 1 semester Beginning of 2nd semester Beginning of 3rd semester Beginning of 4th semester Beginning of 5th semester Beginning of 6th semester End of 6th semester


Group A (6 students) Text A1, Text A2, Text A3 Text A4, Text A5 Text B1 (1 semester‘s lag) Text B2 (2 semesters‘ lag) Text B3 (3 semesters‘ lag) Text B4 (3 semesters‘ lag) 7 Text B5 (4 semesters‘ lag) Text A1 (6 semesters‘ lag)

Group B (6 students) Text B1, Text B2, Text B3 Text B4, Text B5 Text A1 (1 semester‘s lag) Text A2 (2 semesters‘ lag) Text A3 (3 semesters‘ lag) Text A4 (3 semesters‘ lag) Text A5 (4 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



‗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. 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

Towards a model of translation competence (acquisition)


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 timeconsuming 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. For three of our subjects in the first wave, a Tobii eye-tracker (with the screenrecording 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.



Susanne Göpferich

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

For the availability of the materials used in the experiments, see Section 6.

Towards a model of translation competence (acquisition)


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 domainspecific knowledge, we make sure that the subjects‘ domain competence can also be regarded as a more or less controlled variable.


Susanne Göpferich

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 selfconcepts 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. Verbalreport 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

Towards a model of translation competence (acquisition)


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 subcompetences 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


Susanne Göpferich

decisions, whereas students only gradually learn to develop a macrostrategy, 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 targetlanguage 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

Towards a model of translation competence (acquisition)


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 surfaceoriented, 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

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.


Susanne Göpferich

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.

Towards a model of translation competence (acquisition)


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.


Susanne Göpferich

— & 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. (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 PostEditing Processes. Kent (Ohio), London: Kent State Univ. Press.

Towards a model of translation competence (acquisition)


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 <> (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


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 nonliteralness, 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


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


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 processoriented 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 formoriented 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.


Gerrit Bayer-Hohenwarter

Recent theories of creativity (e.g. Kußmaul 2000a) draw heavily on prototype semantics (e.g. Rosch 1977) and more specifically, on scenesand-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 scenesand-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

Translational creativity


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


Gerrit Bayer-Hohenwarter

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

Translational creativity


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, concretisation, modification and abstraction can be regarded as non-routine. Among the many researchers who see creativity as a type of problemsolving 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


Gerrit Bayer-Hohenwarter

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. remetaphorisation 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.

Translational creativity


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.


Gerrit Bayer-Hohenwarter

Table 1. Translations of the ST unit “dog jumps out of his skin”2
TT TT1 TT2 TT3 TT4 TT5 TT6 TT7 TT8 Subject JTH JZE ERE LPE MLE STO THI BKR Wave Target text t1 t1 t1 t1 t1 t1 t1 t3 wird er alles tun [um Dir zu zeigen wie sehr er dich mag.] wird er beinahe aus seiner Haut springen wird er an dir hoch springen wird er sich fast überschlagen wird er beinahe einen Luftsprung machen wird er fast aus seiner Haut herausspringen tut er alles um dir zu zeigen wird er voll Übermut und Freude um dich herumspringen wird er sich fast überschlagen wird er sich wie verrückt gebärden dann wird er alles tun zerreißt er sich fast wird er sich fast überschlagen bringt er sich beinahe um zeigt er Ihnen mit einem Freudentanz [wie sehr er Sie mag] springt er an Ihnen hoch und demonstriert Ihnen seine Zuneigung einfach umwerfend Rating Procedures pass fail pass pass pass fail pass pass + Abstraction; demetaphorisation Reproduction Concretisation; demetaphorisation Modification; remetaphorisation Modification; re-metaphorisation Reproduction Abstraction; de-metaphorisation Concretisation; de-metaphorisation explicitation Modification; re-metaphorisation Abstraction; de-metaphorisation Abstraction; de-metaphorisation Modification; re-metaphorisation Modification; re-metaphorisation Modification; re-metaphorisation Modification; re-metaphorisation Concretisation; de-metaphorisation enrichment



t3 t8 t8 t8 t8

pass pass pass pass pass pass pass +





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.

Translational creativity


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


Gerrit Bayer-Hohenwarter

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.

Translational creativity


Table 2. Creativity assessment of the translations of “dog jumps out of his skin”
Dog jumps out of his skin Primary procedure(s) Re-produce Secondary procedure(s) Target texts Creativity indicators Creativity rating

Group A: wird er beinahe/fast aus seiner Haut heraus/springen fahrt er fast aus der Haut Group B: wird er alles tun wird er sich wie verrückt gebärden

TT2_t1 (x) TT6_t1 (x)



De-metaphorise De-metaphorise


TT1_t1 TT7_t1 TT12_t8 TT10_t8

Group C: wird er sich fast überschlagen Group C: wird er beinahe einen Luftsprung machen zerreißt er sich fast Group D: wird er an dir hoch springen Group D: wird er voll Übermut und Freude um dich herumspringen springt er an Ihnen hoch und demonstriert Ihnen seine Zuneigung einfach umwerfend zeigt er Ihnen mit einem Freudentanz



TT4_t1 TT9_t3 TT11_t8 TT14 TT5_t1

Acceptability Comprehension P-flexibility (1) Acceptability Comprehension P-flexibility (1) Uniqueness Acceptability Comprehension P-flexibility (1) Acceptability Comprehension P-flexibility (1) Uniqueness Acceptability Comprehension P-flexibility (1) Uniqueness Acceptability Comprehension P-flexibility (2) Uniqueness Outstanding quality Acceptability Comprehension P-flexibility (2) Uniqueness Acceptability Comprehension P-flexibility (2) Uniqueness Outstanding quality





TT13_t8 Concretise De-metaphorise TT3_t1



De-metaphorise Explicitate




De-metaphorise Enrich



Concretise Modify





Gerrit Bayer-Hohenwarter

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 ITT1 ITT2 ITT3 ITT4 ITT5 TT fahrt er fast aus der Haut fährt er fast aus der Haut Rating Procedures fail fail Reproduction Reproduction Modification Modification Modification Modification; re-metaphorisation

macht er fast ‟nen Kopfstand fail reißt er sich ein Bein aus zerreißt er sich zerreißt er sich fast fail fail pass

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

Translational creativity


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


Gerrit Bayer-Hohenwarter

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:

Translational creativity


Table 4. Creativity assessment for intermediate translations of “dog jumps out of his skin” (HOB_t8)
HOB t8 ITT1 ITT2 Target text fahrt er fast aus der Haut fährt er fast aus der Haut macht er fast ‟nen Kopfstand reißt er sich ein Bein aus Conditions of production main phase: written down after a long pause main phase: after 3 minutes of producing other TT and re-reading ITT1 but without delay main phase: generated spontaneously after ITT2; dismissed without further comment main phase: HOB produces English primary association fall over backward and uses it as a search term in online dictionary. ITT4 is one search result. main phase: solution taken from online dictionary main phase: generated and self-dictated immediately after comment Creativity x (Reproduction) x (Reproduction)



Modification; re-metaphorisation Spontaneity x (Modification) Change of perspective x (Modification) Modification; re-metaphorisation Spontaneity Generativity (4) Procedures (1) Spontaneity (2) Others (1)


zerreißt er sich zerreißt er sich fast

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 intermediate 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


Gerrit Bayer-Hohenwarter

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

Translational creativity


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.


Gerrit Bayer-Hohenwarter

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. <> [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 übersetzungswissenschaftlichen 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

Translational creativity


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 Buchgesellschaft. 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 Crosscultural 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


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


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 nonliteralness. 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 nonliteral 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


Kristian T.H. Jensen

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 ARI Flesch-Kincaid Coleman-Liau Gunning Fog SMOG Flesch Reading Ease LIX Formula 4.71*characters/words+0.5*words/sentences-21.43 11.8*syllables/words+0.39*words/sentences-15.59 5.89*characters/words-0.3*sentences/(100*words)-15.8 0.4*(words/sentences+100*((words >= 3 syllables)/words)) square root of (((words >= 3 syllables)/sentence)*30) + 3 206.835-84.6*syllables/words-1.015*words/sentences 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,



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. 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.

Indicators of text complexity


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) ( and LIX4 ( 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.



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. LIX is a Swedish abbreviation for läsbarhetsindex (i.e. readability index).


Kristian T.H. Jensen

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 (2535), 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.

Indicators of text complexity


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


Kristian T.H. Jensen

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 nonliteral 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

Merriam-Webster: sesquipedalian: (1) having many syllables; (2) given to or characterised by the use of long words.

Indicators of text complexity


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


Kristian T.H. Jensen

Table 2. K2-K10 words in texts A, B and C
Text A Guilty, imprisoned, medicine, murder, nurse, patients, sentences, strangely, trial, victims, weak, burden, motive Text B Ahead, alarming, analysts, below, climbed, cough, customers, fuel, nurses, professionals, racing, salary, struggles, suffered, warned, supermarkets, inflation, sector, escalating, insistence, soar, soaring, healthcare Text C China, Chinese, criminal, embarrass, emphasizing, extensive, humanity, international, maintains, negative, oil, sought, suffering, wake, withdrawal, bulk, flaring, gesture, rattle, halt, 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 K2K10 frequency band and the word counts does not, it is probable that a


The frequency bands are based on the British National Corpus.

Indicators of text complexity


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

Back-translation: wave goodbye to (something).


Kristian T.H. Jensen

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
Text A

Only the awareness of other hospital staff put a stop to him and to the killings

Text B Families have to cough up an extra £1,300

Text C Spielberg shows Beijing red card

Prices are racing ahead of salary increases

Spielberg pulled out of the Olympics His withdrawal comes in the wake of fighting Khartoum bears the bulk of the responsibility

Indicators of text complexity


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.


Kristian T.H. Jensen

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
Text A (None) Text B Hit with increase (mp) Prices/bills soar (mp) Prices (…) have climbed (mp) Cut interest rates (mp) Struggles to keep inflation … under control (mp) Escalating prices (mp) Government (mt) Text C Beijing (mt) Rattle the Chinese government (mp) China (mt) Sudan (mt) Fighting flaring up (mp) Has sought to halt (mp) Negative fallout (mp) Close ties Close links Government (mt) Khartoum (mt)


Mp.: metaphor. Mt.: metonym.

Indicators of text complexity


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 Indicator ARI Flesch-Kincaid Coleman-Liau Gunning Fog SMOG Flesch-Reading Ease Score LIX Word frequency Idiom density Metonymy & metaphor density Least Inbetween Text B Text B Text B Text B Text B Text B Text B Text B Text B Text B Most

Text A Text A Text A Text A Text A Text A Text A Text A Text A Text A

Text C Text C Text C Text C Text C Text C Text C Text C Text C 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.


Kristian T.H. Jensen

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.

Indicators of text complexity


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


Kristian T.H. Jensen

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 [accessed 19 March 2009 from]. 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 ], an adaptation of Heatley & Nation‟s (1994) Range. [accessed 19 March 2009 from]. 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.

Indicators of text complexity


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] 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.

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

Number of characters with spaces: 837 Length of headline in characters with spaces: 41


Kristian T.H. Jensen

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

Number of characters with spaces: 846 Length of headline in characters with spaces: 44

(Text C) Spielberg shows Beijing red card over Darfur Source: The Daily Telegraph on 13 February 2008 1 2 3 4 5 6 7 8 9 10 11 In a gesture sure to rattle the Chinese Government, Steven Spielberg pulled out of the Beijing Olympics to protest against China‟s backing for Sudan‟s policy in Darfur. His withdrawal comes in the wake of fighting flaring up again in Darfur and is set to embarrass China, which has sought to halt the negative fallout from having close ties to the Sudanese government. China, which has extensive investments in the Sudanese oil industry, maintains close links with the Government, which includes one minister charged with crimes against humanity by the International Criminal Court in The Hague. Although emphasizing that Khartoum bears the bulk of the responsibility for these ongoing atrocities, Spielberg maintains that the international community, and particularly China, should do more to end the suffering.

Number of characters with spaces: 856 Length of headline in characters with spaces: 44

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


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 finalyear 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), eyetracking (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


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 postprocess 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


Nataša Pavlović

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

Collaborative translation protocols


(“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 250word 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


Nataša Pavlović

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 decisionmaking 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.]

Collaborative translation protocols


Figure 1 – CNA based on 60 translations of the same ST sentence


Nataša Pavlović

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

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.

Collaborative translation protocols 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


Nataša Pavlović

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 thinkaloud 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.


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.

Collaborative translation protocols 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]


Nataša Pavlović

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!

Collaborative translation protocols 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

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).


Nataša Pavlović

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; TirkkonenCondit 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

Collaborative translation protocols


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.


Nataša Pavlović

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:

Collaborative translation protocols


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


Nataša Pavlović

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:

Collaborative translation protocols


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-domainspecific 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


Nataša Pavlović

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

It is perhaps worth noting that screen recordings take up large amounts of disk space.

Collaborative translation protocols


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


Nataša Pavlović

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.

Collaborative translation protocols


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, Interdependence 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. Visited February 2007. Gile, D. 2004. Integrated problem and decision reporting as a translator training tool. JoSTrans 2: 2-20. 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 (Copenhagen Studies in Language 24). Copenhagen: Samfundslitteratur. Hansen, G. (ed.) 2002. Empirical Translation Studies: Process and Product (Copenhagen Studies in Language 27). Copenhagen: Samfundslitteratur. Hansen, G. 2006. Retrospection methods in translator training and translation research. JoSTrans 5. 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


Nataša Pavlović

Process of Translation: Outlooks on Empirical Research. Amsterdam and 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 Process of Translation: Outlooks on Empirical Research. Amsterdam and Philadelphia: John Benjamins. 71-82. 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. Tirkkonen-Condit, and R. Jääskeläinen (eds). Tapping and Mapping the Process of Translation: Outlooks on Empirical Research. Amsterdam and Philadelphia: John Benjamins. 97-110. 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 Philadelphia: Multilingual Matters. 159-176. 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 übersetzungswissenschaftlichen 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 DanishEnglish 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:

Collaborative translation protocols


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): 7595. Séguinot, C. 2000. Management issues in the translation process. In S. Tirkkonen-Condit, and R. Jääskeläinen (eds). Tapping and Mapping the Process of Translation: Outlooks on Empirical Research. Amsterdam and 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 Process of Translation: Outlooks on Empirical Research. Amsterdam and Philadelphia: John Benjamins. Translog. Tool for analyzing text production processes. 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


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 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:


See 3.3 for categories of explicitation and implicitation.



Dorrit Faber & Mette Hjort-Pedersen

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 „adoptionslovgivningen‟ 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 twosided 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.

Inference processes in legal 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.


Dorrit Faber & Mette Hjort-Pedersen

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


“Disambiguation” did not apply to this particular translation task.

Inference processes in legal translation


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

114 T2: I don’t even know what they mean

Dorrit Faber & Mette Hjort-Pedersen

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:

Inference processes in legal translation


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/ Focus point Number of verbalised mental explicitations by 4 groups of students 0 3 2 3 3 3 3 Verbalised mental explicitation Professional 1 – + + + + – + Verbalised mental explicitation Professional 2

RA Retsplejeloven RA RA RA EN EN EN
Administration of Justice Act Byretten The City Court afgørelsen the decision kærendes processkrift the appellant’s statement of case findes is found som sket as was the case fremsættelse (af afvisningspåstand) submission (of motion of dismissal) indsigelse (mod værnetinget objection (to venue) (sagens ) forberedelse (for byretten) preparation (of the case)(before the City Court) det tiltrædes is accepted

– – – – – – –


2 4

– +

– –




0 out of 10 = 0%

27 out of 403 = 67.5% 7 out of 10 = 70%

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

4 groups x 10 focus points.



Dorrit Faber & Mette Hjort-Pedersen

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 Unit/Focus point Professional translator 2 men- linguistic men- linguistic men- linguistic tal expl impl tal expl impl tal expl impl 0 A(4) A A 4 groups of students Professional translator 1

RA Retspleje-


loven Administration of Justice Act Byretten The City Court the decision kærendes processkrift the appellant’s statement of case findes is found as was the case afvisningspåstand) submission of a motion of dismissal) indsigelse (mod værnetinget) objection (to venue) sagens forberedelse (for byretten ) preparation (of the case)(before the City Court) det tiltrædes is accepted

3 2 3

A(1) S(1) S(1)

+ + +



RA afgørelsen RA


3 3 3

A(1) R(2) A(1) R(1) S(1) A(1) R(1)

+ + R R



EN som sket

EN fremsættelse (af


2 4

S(1) S(4)






26.20/12.04  45.95%



Overall time/ Focus point time  % of overall time

56.06/41.18 73.46% 49.09/38.02  77.45% 47.06/29.35  62.37% 34.55/12.01  34.76%

15.42/6.13  39.75%

Inference processes in legal translation


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.



Dorrit Faber & Mette Hjort-Pedersen

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.

Inference processes in legal translation


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


Dorrit Faber & Mette Hjort-Pedersen

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 facethreatening 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

Inference processes in legal translation


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


Dorrit Faber & Mette Hjort-Pedersen

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

Inference processes in legal translation


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.
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 Communication. 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.


Dorrit Faber & Mette Hjort-Pedersen

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: 189209. 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 postediting 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 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 logicalsemantic 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.


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


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

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).


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 contextdependent, 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

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).

Unique items in translation


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 logicalsemantic 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


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.


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 ellers MA students (19) BA students (34) Total (53) 12 8 20 alternatives 0 1 (tidligere = „earlier‟) 1 0 0 0 ST misinterpretations

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,

Unique items in translation


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
TT explicitations ellers alternatives MA students (15) 0 5 0 ST misinterpretations

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


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.


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.

Unique items in translation


Table 3. Product data. TT explicitations
TT explicitations ellers Adversative connectors MA students (30) 2 0 0 ST misinterpretations

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 (TirkkonenCondit 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 adversativeconcessive 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


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
TT explicitations ellers Adversative connectors MA students (30) 0 0 0 ST misinterpretations

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 adversativeconcessive 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

Unique items in translation


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 thinkaloud 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.


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.

Unique items in translation


Table 5a. The first experiment. Product data. TT explicitations
TT explicitations ellers Adversative connectors Professionals (5) MA students (7) Total (12) 0 0 0 0 0 0 0 0 0 ST misinterpretations

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


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 logicalsemantic 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 misinterpretations Explicitations Comments Explicitations 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

Unique items in translation


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 meanhearted, 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 adversativeconcessive 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


Louise Denver

relation, categorised it as adversative or concessive. In this way, the data related to example (5) suggest some confusion as to the adversativeconcessive 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.).

Unique items in translation


(1) The first experiment Table 6a contains the empirical data for example (6).
Table 6a. The first experiment. Product data. TT explicitations.
TT explicitations ST misinterpretations ellers Professionals (5) MA students (7) Total (12) 0 2 2 Idiom 3 1 4 0 1 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


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

Unique items in translation


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


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

From the second experiment (see Tables 5b and 6b) only the figures related to phase I, the translation phase, have been reported.

Unique items in translation


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 decisionmaking. 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,

The figure includes all the explicitations made in the TTs for examples (5) and (6) from the second experiment.


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. 223235. 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.

Unique items in translation


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. Jakobsen, A. L. 1999. Logging target text production with Translog. In G. Hansen (ed.). Probing the Process in Translation: Methods and Results (Copenhagen Studies in Language 24.). Copenhagen: Samfundslitteratur. 920. 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.

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.”


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 selfconfidence 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


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 secondlanguage 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


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 minimarket, 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.

Tracking self-corrections with Translog


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 preestablished 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 selfcorrections to prepositions that we would predict to find in HebrewEnglish translation – were included in Table 2 (Refining the Target Text).

In each example from the logs, the original translation appears in Roman and the selfcorrection in italics, separated by a slash: more and more of our time / more and more time.


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 selfcorrections 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 EnglishHebrew 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 culturebound expressions.

Tracking self-corrections with Translog


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


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.

Tracking self-corrections with Translog


Table 1. Correcting the Target Text

Type of self-correction Insertion of an omitted element Substitution of one preposition for another Meaning Grammar


Examples from the logs Instances in spite the fact / in spite of the fact 69 bestseller / idiotic bestseller appear in / appear on 45 starting with / starting from dramatic rise / dramatic decrease 44 hoards / hordes excessive aggressive behavior / 7 excessively aggressive behavior clothes that was so / clothes that were so 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 selfcorrections. 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.


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 spellcheck feature, presumably they would have used the spell checker after completing their translation.
Table 2. Refining the Target Text

Type of self-correction Replacing a word or phrase with a synonymous one Punctuation Articles

Tense and aspect

False starts and retyping the identical word or phrase Singular to plural and vice versa Adding or deleting an extra preposition


Examples from the logs Instances registration / enrollment 685 cause irreversible damage to our blood vessels / cause our blood vessels irreversible damage Once we were / Once, we were 158 Universities / universities in the country clubs / in country clubs 80 exclusive neighborhoods / the exclusive neighborhoods we replaced / did we replace / have we 59 replaced is / will athletic cl / athletic styles 57 The State of Education Today / The State of Education Today an ad hoc solution / ad hoc solutions 37 country clubs / country club in the Stone Age or the Middle Ages / 16 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 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.

Tracking self-corrections with Translog


Table 3. Self-corrections to Lexicalizable Strings, False Cognates, and Culture-bound Terms

Type of self-correction Lexicalizable strings False cognates Culture-bound terms Total

Instances 16 15 0 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 Insertion of an adjective before the noun Tense and aspect Commas Capitalization Addition or deletion of an extra preposition Indefinite article Changing “life are” to “life is” Total

Instances 62 59 57 43 16 11 3 251

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).


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 culturebound 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.

Tracking self-corrections with Translog


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


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 selfcorrections 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 selfcorrections 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 selfcorrections, 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.

For a detailed treatment see Göpferich (2008: 254).







Tracking self-corrections with Translog


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 / Philadelphia: John Benjamins. 243-250. 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. 7789. 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. nonprofessional 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.


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 / Philadelphia: Multilingual Matters. 159-176. 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 / Philadelphia: John Benjamins. 65-82. 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.

Tracking self-corrections with Translog


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),


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: 180181). 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.


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


Ricardo Muñoz Martín

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 (“otherdrafts”). 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 Verbal comprehension Working Memory Processing Speed Structure (English) Reading (English) 109 116 125 95 92 Bruno 100 100 114 85 86 Carla 106 129 84 83 71 Diana 112 104 120 95 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.


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 selfrevisions ST Publ. Session words 465 11/11/06 11/16/06 470 11/11/06 11/23/06 544 11/11/06 12/07/06 490 09/14/06 01/11/07 453 10/26/06 549 507 11/02/06 09/14/06 Other-revisions Draft Words Session

Code T1 T2 T3 T4 R1 R2 R3

Titles Politics this week Without prejudice South Africa rethinks Powering up (1st part) All creatures great and small Eyeing up the collaboration Powering up (2nd part)

514 586 556

11/16/06 12/15/06 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


Ricardo Muñoz Martín

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.


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 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Categ.

{×}{«1},_•••••el_ programa_gener{«5}del_ordenador_genera_ [04.20]port•ador•es_ del_•VIG{«1}H{«18}con_SIDA•••_) de_Sudáfrica,_pero_está_pla{«4}n_plan•eando_••• •••sol•icitan_un_seguro••,••_[29.01]{«2}_pero_que_•••mueren_ [11.14]vot•ó_•legalizar•_l•os{«13}la_legalizaci•ón_de_los_ _de_Pusan_•Nationa_Un{«3}l_University{-26}_la_•Universidad_de {×} _asciende_u {«1} a_unos •• _tres_millones_de_rands_ • _( {«3} _ • ([05.12]9 •••• {«1} •• 9 •• {«1} 392 •••• . • 87 • 0 ••• € • ) personas_que_qui{«3}[04.69]quieren_co•ntrat{«7}seleccionar._ m•ísm{«3}ismos•_ NO_CU••{«5}no_cubre_[08.31]{»2} ••ass••o{«4}Ass•ociaton{-2}i proteje{«2}ger_sus_ ••••absorbe••_•un_for{«1}••t•ón_ d•e_mae{«1}teriales_más_ {+2}{×}{«3}os_paneles_soalres_{«6}••lares_ [05.30],_crear_una_corriente_ec{«1}••l•éctrica del_prime_{«1}r





1W 1M 1K 1+ 1S

••••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.


Ricardo Muñoz Martín

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).

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.


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


Ricardo Muñoz Martín

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.


Table 4. Average interventions and MPs in other-revisions, translations, and selfrevisions by text
Other-revs Int MPs 35 4 37,25 6,75 39 5 Trans Int MPs 137 27 153,75 24 218 37 167,25 33 Self-revs Int MPs 29,3 13 15,3 12 22,3 23 19,5 14

Text 1 2 3 4

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 otherrevisions 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


Ricardo Muñoz Martín

related MPs in other-revision and self-revision tasks were similar, which seems to support this explanation.
Table 5. Average interventions and MPs in other-revisions, translations, and selfrevisions, divided into typos (T) and other phenomena (NT=ILC+L+SL+T)
Other- revs Translations Text T NT T NT 7.5 28.5 81.2 65.5 1 2 8 29.25 80 78.5 8.6 32.6 111.3 123.3 3 4

Missed phenomena
Self revs Other- revs Translations Self-revs T NT T NT T NT typos NT


9.5 8.75 0.5 4 3.3 3.3 78.2 3.2 3.25

1 0 0

3 6.7 6.8

4.5 4.75 4.7 6

31.5 24.5 39.4 37

3.5 4.7 0.25 4.7 2.25 7 0 6.2


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 selfrevising 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.
Table 6. Average interventions and MPs in other-revisions, translations, and selfrevisions by subject.
Other-revs Int MPs 25.00 3.33 39.33 5.33 30.33 4.33 53.67 8.00 Trans Int MPs 123.75 35.25 147.50 20.50 165.25 34.75 239.50 30.50 Self-revs Int MPs 21.25 14.50 24.50 17.50 21.75 16.50 18.75 12.25

Subject Andrés Bruno Carla Diana

Curiously, subjects seemed to miss more phenomena when they performed more interventions in a particular text. This is particularly noticeable in

Typos & Co.


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.
Table 7. Average interventions and MPs in other-revisions, translations, and selfrevisions, divided into typos (T) and other phenomena (NT=ILC+L+SL+T), by subject
Other- revs T Andrés Bruno Carla Diana 5.00 8.00 4.33 14.67 NT 21.33 31.33 26.00 39.00

Translations T NT 78.25 51.50 71.75 83.75 85.00 91.75 133.50 113.25 Self-revs T 4.25 3.50 5.25 4.25 NT 8.25 5.50 3.75 2.25 T 0.67 0.33 0.33 0.00

Missed phenomena
Other-revs NT Translations T NT 25.75 32.00 44.50 33.50 Self-revs T 1.25 1.50 2.25 1.50 NT 7.75 6.75 4.50 5.25

2.67 13.25 5.00 0.75 4.33 4.25 8.33 3.75

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


Ricardo Muñoz Martín

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 otherrevisions, 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.


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


Ricardo Muñoz Martín

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.


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


Ricardo Muñoz Martín

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 F SQ 65 39 41 35 27 14 7 6 Bruno F SQ 34 29 43 29 23 27 4 5 Carla F SQ 62 52 23 12 25 21 2 6 Diana F SQ 35 72 64 79 48 50 14 8

1W 1M 1K 1S

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.


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_


Ricardo Muñoz Martín

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.


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


Ricardo Muñoz Martín

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.


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. 7599. 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.

The authors would like to thank Igor A.L. Silva for a careful reading of this paper and for the invaluable suggestions provided.


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


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. 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


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


Fabio Alves and Tânia Liparini Campos

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

Translation technology in time


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


Fabio Alves and Tânia Liparini Campos

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

Translation technology in time


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


Fabio Alves and Tânia Liparini Campos

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.

Translation technology in time


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.

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.


Fabio Alves and Tânia Liparini Campos

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

Translation technology in time


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

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.


Fabio Alves and Tânia Liparini Campos

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

Translation technology in time


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)
ESG/T1 GSG/T2 Average ESG/T3 GSG/T4 Average ESG/T5 GSG/T6 Average ESG/T7 GSG/T8 Average OP SISO SESO DISO DESO RP SISR SESR DISR DESR Total 65% 54% 9% 0% 2% 35% 32% 3% 0% 0% 100% 74% 64% 7% 1% 2% 26% 23% 1% 1% 1% 100% 70% 59% 8% 1% 2% 30% 27% 2% 1% 0% 100% 73% 35% 9% 27% 2% 27% 19% 3% 4% 1% 100% 85% 27% 6% 49% 3% 15% 13% 1% 1% 0% 100% 79% 31% 8% 38% 2% 21% 16% 2% 3% 0% 100% 78% 62% 14% 0% 2% 22% 18% 3% 1% 0% 100% 75% 67% 5% 1% 2% 25% 23% 2% 0% 0% 100% 76% 64% 9% 1% 2% 24% 20% 3% 1% 0% 100% 79% 37% 12% 28% 2% 21% 16% 3% 2% 0% 100% 80% 48% 9% 22% 1% 20% 18% 0% 1% 1% 100% 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.


Fabio Alves and Tânia Liparini Campos

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

Translation technology in time


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.


Fabio Alves and Tânia Liparini Campos

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
ESG/T1 GSG/T2 Average ESG/T3 GSG/T4 Average ESG/T5 GSG/T6 Average ESG/T7 GSG/T8 Average OP SISO SESO DISO DESO RP SISR SESR DISR DESR Total 1% 1% 0% 0% 0% 99% 93% 5% 0% 1% 100% 0% 0% 0% 0% 0% 100% 96% 4% 0% 0% 100% 0% 0% 0% 0% 0% 100% 95% 5% 0% 0% 100% 0% 0% 0% 0% 0% 100% 95% 2% 3% 0% 100% 0% 0% 0% 0% 0% 100% 95% 1% 4% 0% 100% 0% 0% 0% 0% 0% 100% 95% 2% 3% 0% 100% 10% 10% 0% 0% 0% 90% 88% 1% 0% 1% 100% 0% 0% 0% 0% 0% 100% 94% 6% 0% 0% 100% 6% 6% 0% 0% 0% 94% 91% 3% 0% 0% 100% 9% 5% 0% 4% 0% 91% 91% 0% 0% 0% 100% 0% 0% 0% 0% 0% 100% 96% 2% 0% 2% 100% 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 endrevision 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.


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.

Translation technology in time


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. 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.

Drafting phase


Fabio Alves and Tânia Liparini Campos

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

Translation technology in time


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 Processoriented 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. Jakobsen, A. L. 1999. Logging target text production with Translog. In G. Hansen (ed.) Probing the Process in Translation: Methods and Results (Copenhagen Studies in Language 24). Copenhagen: Samfundslitteratur. 920. Jakobsen, A. L. 2002. Translation drafting by professional translators and by translation students. In G. Hansen (ed.) Empirical Translation Studies:


Fabio Alves and Tânia Liparini Campos

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.

Translation technology in time



Table 4: Occurrences of types of support in the drafting phase in T1/T2
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 31 39 40 40 35 31 28 38 72 85 32 74 SISO 20 33 33 39 35 20 28 34 61 68 29 56 SESO 10 4 6 1 0 8 0 3 8 15 1 10 DISO 0 0 0 0 0 0 0 0 0 0 2 3 DESO 1 2 1 0 0 3 0 1 3 2 0 5 RP 23 9 35 58 7 12 15 9 33 16 18 19 SISR 19 9 33 56 7 8 14 8 25 15 18 15 SESR 4 0 2 1 0 3 1 1 1 1 0 1 DISR 0 0 0 0 0 1 0 0 4 0 0 3 DESR 0 0 0 1 0 0 0 0 3 0 0 0 Total 54 48 75 98 42 43 43 47 105 101 50 93

Table 5: Percentages of types of support in the drafting phase in T1/T2
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP 57% 81% 53% 41% 83% 72% 64% 65% 81% 69% 84% 64% 80% 74% 69% SISO 37% 69% 44% 40% 83% 46% 53% 65% 73% 58% 67% 58% 60% 64% 58% SESO 18% 8% 8% 1% 0% 19% 9% 0% 6% 8% 15% 2% 12% 7% 8% DISO 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 4% 3% 1% 1% DESO 2% 4% 1% 0% 0% 7% 2% 0% 2% 3% 2% 0% 5% 2% 2% RP 43% 19% 47% 59% 17% 28% 35% 35% 19% 31% 16% 36% 20% 26% 30% SISR 35% 19% 44% 57% 17% 19% 32% 33% 17% 23% 15% 36% 16% 23% 27% SESR 8% 0% 3% 1% 0% 7% 3% 2% 2% 1% 1% 0% 1% 1% 2% DISR 0% 0% 0% 0% 0% 2% 0% 0% 0% 4% 0% 0% 3% 1% 1% DESR 0% 0% 0% 1% 0% 0% 0% 0% 0% 3% 0% 0% 0% 1% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%


Fabio Alves and Tânia Liparini Campos

Table 6: Occurrences of types of support in the drafting phase in T3/T4
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 48 30 17 48 51 63 53 26 55 84 48 112 SISO 36 17 12 25 10 11 13 8 20 27 7 55 SESO 3 8 2 4 1 15 4 0 8 6 6 5 DISO 9 4 3 17 38 35 35 17 26 46 35 44 DESO 0 1 0 2 2 2 1 1 1 5 0 8 RP 11 8 10 29 10 27 6 1 9 21 19 22 SISR 7 6 7 24 9 13 3 1 8 18 18 19 SESR 1 1 1 2 0 6 0 0 1 1 1 0 DISR 3 1 2 1 1 8 3 0 0 2 0 3 DESR 0 0 0 2 0 0 0 0 0 0 0 0 Total 59 38 27 77 61 90 59 27 44 105 67 134

Table 7: Percentages of types of support in the drafting phase in T3/T4
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP 81% 80% 63% 62% 84% 70% 73% 90% 96% 86% 80% 71% 84% 85% 79% SISO 61% 46% 45% 32% 16% 12% 35% 22% 30% 31% 26% 10% 41% 27% 31% SESO 5% 21% 7% 5% 2% 17% 9% 7% 0% 12% 6% 9% 4% 6% 8% DISO 15% 10% 11% 22% 63% 39% 27% 59% 63% 41% 44% 52% 33% 49% 38% DESO 0% 3% 0% 3% 3% 2% 2% 2% 3% 2% 4% 0% 6% 3% 2% RP 19% 20% 37% 38% 16% 30% 27% 10% 4% 14% 20% 29% 16% 16% 22% SISR 12% 16% 26% 31% 14% 14% 19% 5% 4% 12% 17% 27% 14% 13% 17% SESR 2% 2% 4% 3% 0% 7% 3% 0% 0% 2% 1% 2% 0% 1% 2% DISR 5% 2% 7% 1% 2% 9% 4% 5% 0% 0% 2% 0% 2% 2% 3% DESR 0% 0% 0% 3% 0% 0% 1% 0% 0% 0% 0% 0% 0% 0% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%

Translation technology in time


Table 8: Occurrences of types of support in the drafting phase in T5/T6
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 13 24 26 41 24 19 19 16 55 58 35 65 SISO 11 16 25 34 23 9 18 15 54 49 35 50 SESO 1 8 0 5 1 10 1 0 0 8 0 9 DISO 0 0 0 2 0 0 0 1 0 0 0 2 DESO 1 0 1 0 0 0 0 0 1 1 0 4 RP 2 3 8 16 6 12 17 6 15 12 18 5 SISR 2 3 8 15 5 5 15 4 15 12 18 4 SESR 0 0 0 0 1 5 2 2 0 0 0 1 DISR 0 0 0 1 0 2 0 0 0 0 0 0 DESR 0 0 0 0 0 0 0 0 0 0 0 0 Total 15 27 34 57 30 31 36 22 70 70 53 70

Table 9: Percentages of types of support in the drafting phase in T5/T6
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP 87% 89% 76% 72% 80% 61% 78% 53% 73% 79% 83% 66% 93% 75% 76% SISO 73% 59% 73% 60% 77% 29% 62% 50% 68% 77% 70% 66% 71% 67% 64% SESO 7% 30% 0% 9% 3% 32% 14% 3% 0% 0% 11% 0% 13% 5% 9% DISO 0% 0% 0% 3% 0% 0% 0% 0% 5% 0% 0% 0% 3% 1% 1% DESO 7% 0% 3% 0% 0% 0% 2% 0% 0% 2% 2% 0% 6% 2% 2% RP 13% 11% 24% 28% 20% 39% 22% 47% 27% 21% 17% 34% 7% 25% 24% SISR 13% 11% 24% 26% 17% 16% 18% 42% 18% 21% 17% 34% 6% 23% 20% SESR 0% 0% 0% 0% 3% 16% 3% 5% 9% 0% 0% 0% 1% 2% 3% DISR 0% 0% 0% 2% 0% 7% 1% 0% 0% 0% 0% 0% 0% 0% 1% DESR 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%


Fabio Alves and Tânia Liparini Campos

Table 10: Occurrences of types of support in the drafting phase in T7/T8
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 34 23 26 28 37 35 20 16 29 39 25 60 SISO 15 14 11 16 17 8 7 12 21 20 13 44 SESO 3 3 5 2 0 16 5 0 3 6 3 3 DISO 15 6 8 10 20 10 8 4 5 12 9 9 DESO 1 0 2 0 0 1 0 0 0 1 0 4 RP 10 3 3 6 16 18 2 4 8 6 15 12 SISR 10 2 2 6 14 8 2 4 8 5 14 9 SESR 0 1 0 0 1 5 0 0 0 1 0 0 DISR 0 0 1 0 1 5 0 0 0 0 1 1 DESR 0 0 0 0 0 0 0 0 0 0 0 2 Total 44 26 29 34 53 53 22 20 37 45 40 72

Table 11: Percentages of types of support in the drafting phase in T7/T8
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP 77% 88% 90% 82% 70% 66% 79% 91% 80% 78% 87% 62% 83% 80% 80% SISO 34% 53% 38% 47% 32% 15% 37% 32% 60% 57% 45% 32% 61% 48% 42% SESO 7% 12% 17% 6% 0% 30% 12% 23% 0% 8% 13% 7% 4% 9% 11% DISO 34% 23% 28% 29% 38% 19% 28% 36% 20% 13% 27% 23% 13% 22% 25% DESO 2% 0% 7% 0% 0% 2% 2% 0% 0% 0% 2% 0% 5% 1% 2% RP 23% 12% 10% 18% 30% 34% 21% 9% 20% 22% 13% 38% 17% 20% 20% SISR 23% 8% 7% 18% 26% 15% 16% 9% 20% 22% 11% 35% 13% 18% 17% SESR 0% 4% 0% 0% 2% 9,50% 3% 0% 0% 0% 2% 0% 0% 0% 1% DISR 0% 0% 3% 0% 2% 9,50% 2% 0% 0% 0% 0% 3% 1% 1% 2% DESR 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 3% 1% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%

Translation technology in time


Table 12: Occurrences of types of support in the revision phase in T1/T2
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 0 0 0 0 2 0 0 0 0 0 0 0 SISO 0 0 0 0 2 0 0 0 0 0 0 0 SESO 0 0 0 0 0 0 0 0 0 0 0 0 DISO 0 0 0 0 0 0 0 0 0 0 0 0 DESO 0 0 0 0 0 0 0 0 0 0 0 0 RP 21 1 14 25 54 19 59 5 0 20 11 28 SISR 19 1 14 25 47 16 59 5 0 17 11 26 SESR 2 0 0 0 7 2 0 0 0 3 0 2 DISR 0 0 0 0 0 0 0 0 0 0 0 0 DESR 0 0 0 0 0 1 0 0 0 0 0 0 Total 21 1 14 25 56 19 59 5 0 20 11 28

Table 13: Percentages of types of support in the revision phase in T1/T2
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP 0% 0% 0% 0% 4% 0% 1% 0% 0% x 0% 0% 0% 0% 0% SISO 0% 0% 0% 0% 4% 0% 1% 0% 0% x 0% 0% 0% 0% 0% SESO 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% DISO 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% DESO 0% 0% 0% 0% 0% 0% 0% 0% 0% X 0% 0% 0% 0% 0% RP 100% 100% 100% 100% 96% 100% 99% 100% 100% x 100% 100% 100% 100% 100% SISR 90% 100% 100% 100% 84% 84% 93% 100% 100% x 85% 100% 93% 96% 95% SESR 10% 0% 0% 0% 12% 11% 5% 0% 0% x 15% 0% 7% 4% 5% DISR 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% DESR 0% 0% 0% 0% 0% 5% 1% 0% 0% x 0% 0% 0% 0% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% x 100% 100% 100% 100% 100%


Fabio Alves and Tânia Liparini Campos

Table 14: Occurrences of types of support in the revision phase in T3/T4
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 0 0 0 0 0 0 0 0 0 0 0 0 SISO 0 0 0 0 0 0 0 0 0 0 0 0 SESO 0 0 0 0 0 0 0 0 0 0 0 0 DISO 0 0 0 0 0 0 0 0 0 0 0 0 DESO 0 0 0 0 0 0 0 0 0 0 0 0 RP 5 1 10 15 29 23 49 11 0 27 22 33 SISR 5 1 10 15 28 17 48 11 0 27 17 33 SESR 0 0 0 0 1 2 0 0 0 0 1 0 DISR 0 0 0 0 0 4 1 0 0 0 4 0 DESR 0 0 0 0 0 0 0 0 0 0 0 0 Total 5 1 10 15 29 23 49 11 0 27 22 33

Table 15: Percentages of types of support in the revision phase in T3/T4
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% SISO 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% SESO 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% DISO 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% DESO 0% 0% 0% 0% 0% 0% 0% 0% 0% X 0% 0% 0% 0% 0% RP 100% 100% 100% 100% 100% 100% 100% 100% 100% x 100% 100% 100% 100% 100% SISR 100% 100% 100% 100% 97% 74% 95% 98% 100% x 100% 77% 100% 95% 95% SESR 0% 0% 0% 0% 3% 9% 2% 0% 0% x 0% 5% 0% 1% 2% DISR 0% 0% 0% 0% 0% 17% 3% 2% 0% x 0% 18% 0% 4% 3% DESR 0% 0% 0% 0% 0% 0% 0% 0% 0% x 0% 0% 0% 0% 0% Total 100% 100% 100% 100% 100% 100% 100% 100% 100% x 100% 100% 100% 100% 100%

Translation technology in time


Table 16: Occurrences of types of support in the revision phase in T5/T6
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 0 0 0 0 0 1 0 0 0 0 0 0 SISO 0 0 0 0 0 1 0 0 0 0 0 0 SESO 0 0 0 0 0 0 0 0 0 0 0 0 DISO 0 0 0 0 0 0 0 0 0 0 0 0 DESO 0 0 0 0 0 0 0 0 0 0 0 0 RP 13 0 4 24 31 1 5 0 0 1 0 17 SISR 13 0 4 22 29 1 5 0 0 1 0 14 SESR 0 0 0 2 1 0 0 0 0 0 0 3 DISR 0 0 0 0 0 0 0 0 0 0 0 0 DESR 0 0 0 0 1 0 0 0 0 0 0 0 Total 13 0 4 24 31 2 5 0 0 1 0 17

Table 17: Percentages of types of support in the revision phase in T5/T6
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP 0% x 0% 0% 0% 50% 10% 0% x x 0% x 0% 0% 6% SISO 0% x 0% 0% 0% 50% 10% 0% x x 0% x 0% 0% 6% x 0% 0% 0% x x 0% x 0% 0% 0% 0% 0% 0% SESO 0% DISO 0% x 0% 0% 0% 0% 0% 0% x x 0% x 0% 0% 0% DESO 0% X 0% 0% 0% 0% 0% 0% X X 0% X 0% 0% 0% RP 100% x 100% 100% 100% 50% 90% 100% x x 100% x 100% 100% 94% SISR 100% x 100% 92% 94% 50% 88% 100% x x 100% x 82% 94% 91% x 18% 6% 3% x x 0% x 0% 0% 3% 0% 1% 0% SESR 0% DISR 0% x 0% 0% 0% 0% 0% 0% x x 0% x 0% 0% 0% x 0% 0% 0% x x 0% x 0% 0% 3% 0% 1% 0% DESR 0% Total 100% x 100% 100% 100% 100% 100% 100% x x 100% x 100% 100% 100%


Fabio Alves and Tânia Liparini Campos

Table 18: Occurrences of types of support in the revision phase in T7/T8
Subject E1 E2 E3 E4 E5 E6 G1 G2 G3 G4 G5 G6 OP 0 0 0 1 1 0 0 0 0 0 0 0 SISO 0 0 0 0 1 0 0 0 0 0 0 0 SESO 0 0 0 0 0 0 0 0 0 0 0 0 DISO 0 0 0 1 0 0 0 0 0 0 0 0 DESO 0 0 0 0 0 0 0 0 0 0 0 0 RP 0 0 0 12 9 0 29 0 0 22 0 33 SISR 0 0 0 12 9 0 29 0 0 21 0 31 SESR 0 0 0 0 0 0 0 0 0 1 0 0 DISR 0 0 0 0 0 0 0 0 0 0 0 0 DESR 0 0 0 0 0 0 0 0 0 0 0 2 Total 0 0 0 13 10 0 29 0 0 22 0 33

Table 19: Percentages of types of support in the revision phase in T7/T8
Subject E1 E2 E3 E4 E5 E6 Average G1 G2 G3 G4 G5 G6 Average Final Average OP x x x 8% 10% x 9% 0% X X 0% x 0% 0% 4% SISO x x x 0% 10% x 5% 0% x x 0% x 0% 0% 2% SESO x x x 0% 0% x 0% 0% x x 0% x 0% 0% 0% DISO x x x 8% 0% x 4% 0% x x 0% x 0% 0% 2% DESO x x x 0% 0% x 0% 0% x x 0% x 0% 0% 0% RP x x x 92% 90% x 91% 100% x x 100% x 100% 100% 96% SISR x x x 92% 90% x 91% 100% x x 95% x 94% 96% 94% SESR x x x 0% 0% x 0% 0% x x 5% x 0% 2% 1% DISR x x x 0% 0% x 0% 0% x x 0% x 0% 0% 0% DESR x x x 0% 0% x 0% 0% x x 0% x 6% 2% 1% Total x x x 100% 100% x 100% 100% x x 100% x 100% 100% 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


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


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 2, consulted 07.01.2009. The reader is also referred to the online bibliography of false friends:


Maxim I. Stamenov

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 FinnoUgric. 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;

By language union (Sprachbund) is meant a set of languages, geographically close but not necessarily genetically related, in which similar phenomena can be found.

Cognates in language, in the mind, and in a prompting dictionary


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)

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.


Maxim I. Stamenov

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

Cognates in language, in the mind, and in a prompting dictionary


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 EnglishGerman), 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 EnglishBulgarian, 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


Maxim I. Stamenov

(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

Cognates in language, in the mind, and in a prompting dictionary


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”; 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


Maxim I. Stamenov

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

Cognates in language, in the mind, and in a prompting dictionary


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


Maxim I. Stamenov

(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

Cognates in language, in the mind, and in a prompting dictionary


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.


Maxim I. Stamenov

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.

Cognates in language, in the mind, and in a prompting dictionary


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, Balancebalance; 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 politeimpolite 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.


Maxim I. Stamenov

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; ~ over sich et. überlegen or durch den Kopf gehen lassen; ~ up Plan etc. aushecken, sich ausdenken, sich et. einfallen lassen;

Cognates in language, in the mind, and in a prompting dictionary 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. 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, Angelegenheit>; 10. ~ up v.t. (coll.) sich (Dat.) ausdenken <Plan>. 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.


to think



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.

236 think

Maxim I. Stamenov (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

Cognates in language, in the mind, and in a prompting dictionary


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


Maxim I. Stamenov

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

Cognates in language, in the mind, and in a prompting dictionary


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


Maxim I. Stamenov

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

Cognates in language, in the mind, and in a prompting dictionary


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.


Maxim I. Stamenov

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

Cognates in language, in the mind, and in a prompting dictionary


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-forcomprehension, namely frequency. The iDict was further developed for the purposes of the project EYE-toIT 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


Maxim I. Stamenov

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

The latest version of GWM tool is available at Oleg Špakov‟s personal web page

Cognates in language, in the mind, and in a prompting dictionary


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).


Maxim I. Stamenov

It is proposed that the prompting entry for to think in English looks as follows:
(8) a. b. to think vorstellen, vorhaben, überlegen;

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. 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;


Cognates in language, in the mind, and in a prompting dictionary


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


Maxim I. Stamenov

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)”.

Cognates in language, in the mind, and in a prompting dictionary


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 DeutschEnglisch. 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.


Maxim I. Stamenov

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 95144-6643-8, Available at Kaldieva-Zaharieva, S. 2007. Калдиева-Захариева, Стефана. 2007. Румънскобългарски 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.

Cognates in language, in the mind, and in a prompting dictionary


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 RuhrUniversität 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 empiricalexperimental 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: 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 ( Her main research interest lies in translation process research. E-Mail: 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, screenrecording 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:


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: 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: 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 (Translationsprozessforschung: Stand – Methoden – Perspektiven, Tübingen: Narr, 2008) and the longitudinal study TransComp. E-Mail: Website: 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


semantics. With Dorrit Faber she is currently involved in a project on the phenomena of explicitation and implicitation in legal translation. E-mail: 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 ( and and is used to study writing processes worldwide. Email: Website: 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: 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


Notes on contributors

Cognates” [Meta], and “Translation as a Decision Process: Evidence from Cognates” [Babel]. E-mail: 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: 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: 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: 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


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 ( 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:

Sign up to vote on this title
UsefulNot useful