The 'Translator Choice' site (McCarthy 2003, 2004a) was developed as a means of delivering what is, from the point of view of the user, the simplest of exercises: the presentation of successive segments of a passage in the source language each accompanied by a set of 5 student-generated translations of varying accuracy displayed in random order. Students are required to either select the most accurate translation (Figure 1a) or rank the translations in order of accuracy (Figures 2a, 2b). A scoring system is applied based on the degree of discrepancy between what the student rates as accurate and the relative accuracy of each variant as determined by the instructor during the creation of the activity. Each activity can be administered in Practice or Test mode. In the Practice mode of the Select One activity, students are free to explore a bank of feedback providing a detailed linguistic explanation of the deficiencies of each alternative translation (Figure 3).
Because of the time involved, it is simply not possible to provide each student with the ideal service of a full written or oral explanation of each problematic point in each translation they submit. Neither is it possible to know how much explanation is actually required by different individuals. By allowing the development of a site such as 'Translator Choice', contemporary ICT has nevertheless brought us a lot closer to that ideal. Full details of the rationale, design and development of the site have been provided in earlier articles (McCarthy 2003, 2004b), so will not be repeated here. However, some key aspects to mention are:
Translation was closely associated with the older generation's endeavours to learn a foreign language. First steps in the study of French in a comprehensive state secondary school in Australia in the late 1950s, for example, were supported by grammar-translation textbooks which were the largely unquestioned standard in an era when few of the country's foreign language teachers had ever met a French native, let alone conversed with one or travelled to France. It was an environment in which terms such as audio-lingual, audio-visual, functional, communicative, or special purposes had yet to be associated with notions of language teaching or incorporated into syllabuses or teaching approaches. For French, as for Latin, proof of mastery of vocabulary and grammar in the junior years was demonstrated by the ability to translate carefully-engineered phrases and sentences from English into the foreign language. At higher levels of secondary education these sentences had been strung together to form cohesive passages.
The language component of French studies at university commonly took the form of 'proses' (translation into French of passages of authentic English, usually drawn from novels), dictations, 'explication de texte', conversation classes and an introduction to phonetics. The translation of passages from French into English ('unseens') was generally considered by students to be the softer option, as it was a relatively passive activity, requiring recognition rather than production of the foreign-language words, syntax and constructions, and an exercise in which first-language skills played a much greater role in determining the quality of the end result. Graduates of this now superseded methodology may not have been able function comfortably in everyday conversation in France, but they did emerge with a reasonable grasp what was required to produce a grammatically and stylistically accurate equivalent of an L1 passage in L2, at least for the purposes of passing a university language examination. Any subsequent contact with authentic French, such as an extended stay in France, provided linguistic exposure and cultural perceptions that could only improve the output.
University students of French in present-day Australia have typically not had the benefit of five or six years of foreign language study at secondary school. But they do have the opportunity to interact with French exchange students on their own campus, and some of them may have already spent time touring or living in France or a French-speaking country. They have generally not had the benefit of long-term training in comparing and contrasting two language systems. And in any case, the activity of translation now occupies quite a different position in foreign-language courses.
It may not be anathema, as it was in the eyes of some of the more purist teachers of the structuro-global audio-visual era of the mid-seventies, but most contemporary beginner courses strive to develop communicative skills primarily by moving from comprehension of authentic or semi-authentic audio or video dialogues to independent re-use of a designated set of structures by the learners. Students and teachers commonly resort to translation at different points of the process, but it is rarely presented as a routine, formal, assessable activity. However, in a world where foreign language study is undertaken for pragmatic rather than purely academic reasons, students are eager to be able to demonstrate to potential employers that they possess a high degree of proficiency in translating in addition to their general communication skills in the foreign language.
If formal translation activities are introduced only at the intermediate or advanced level after as little as two semesters of foreign language studies, students are likely to discover that their deficiencies in that area are just as great as the deficiencies in real-life spoken communication were in the foreign language skills of students of an earlier generation.
Because its texts, contexts, objectives, agents and audiences can be so diverse, it is highly improbable that the word 'translation' will ever be defined in anything but the broadest terms. Literary, artistic, journalistic, machine, technical and summary translation are each governed by a different set of parameters. They have in common the rendering of meaning in another language. The end result of the deconstruction, transfer and reconstitution of meaning can in each case be judged on its merits and defects, which are often as much an impression in the eye of the beholder as demonstrable and absolute linguistic truths. Several parameters combine to make university translation assignments a unique genre. They include the reason for undertaking the translation, its subjection to formal assessment, and the availability to the translator of one or more parallel translations.
Purpose. The primary reason for undertaking the translation of a given passage is that it has been set as an assessable task in a foreign language course. The translator is conscious of the fact that the assignment is part of a broader learning process. The passage was chosen by the teacher rather than the student, and for reasons best known to the teacher: e.g. to give the student exposure to a particular style or register of language, or because it provides the student with an appropriate linguistic challenge. And the translation is being done for an audience of one: the teacher-assessor. This does not mean that the student is not endeavouring to produce a translation that would convey the meaning of the original document fully and accurately to a target-language audience, it is just that that audience remains a hypothetical one.
Assessment. It is not unusual for a translation to be subject to evaluation, particularly if it displays conspicuous flaws. Such assessment may take the form of anything from the complete layperson's 'You can tell that was translated!' or 'I wonder who translated that!' or 'What's that supposed to mean?' in relation to the set of instructions accompanying a household appliance produced in a foreign country, to formal editorial amendment of a document submitted for publication. But in the case of the student assignment, regardless of how vague or detailed the accompanying commentary may be, the piece of work will be awarded a numeric grade either as a measure of its quality in relation to an absolute standard or as a means of ranking it in relation to the efforts of other students. Issues surrounding the formulation and implementation of a consistent, formal, objective evaluation process will be discussed below.
Linguistic impediment. Any genuine foreign language student will by definition have only a partial grasp of one of the two languages involved in their translation exercise. In fact, one of the principal objectives of an L2-to-L1 translation assignment is often to determine whether the student has fully grasped the meaning of the original. In less controlled activities such as listening comprehension or conversation, where grasp of the big picture, or simply getting by and keeping communication going, are of primary importance, a student's failure to fully understand individual words and phrases, or to make appropriate connections between different elements of the discourse, can easily remain concealed - with potentially serious consequences.
Comparison. At some point or other, students completing a translation assignment will normally see one or more versions of the translation other than their own, either through informal collaboration with their peers or in the fair copy provided by the instructor. This provides them with a unique opportunity to put their own work in perspective and to learn to discriminate between an acceptable variant an inaccurate or incomplete translation.
The purpose of the teacher in producing a fair copy of a translation passage set as an assignment is not the same as that of a translator employed to produce a translation of a novel or newspaper article or academic article for publication, even if both are the work of 'professionals' and the text generated might be the same in each case. The functions of the fair copy and the published translation are also different. Instead of being a freestanding text whose readership will in all probability neither need nor be in a position to compare it to the source-language text, the fair copy stands as a benchmark against which students can measure their efforts. The process of comparison reveals two things. The first is that the student's translation may be different from the fair copy at many points and yet not have elicited any comment. The implicit message to students is that just because something they have written is different from the fair copy doesn't mean it is unacceptable. This message may not always be detected, but it accounts for why the term 'fair copy' is used rather than 'answer'. The second is to provide an alternative translation for anything that might have been signalled as problematic, deficient or erroneous in the student's work. This can be a sensitive matter. It is very easy at this point for students to lose sight of the fact that their text is being evaluated not just as a piece of English, but on the extent to which it is a faithful rendering of the meaning of the source text. Offence at poorly reasoned or inadequately explained criticism is usually expressed by remarks in the vein of 'English is my mother tongue, I speak and understand it every bit as well as you, so what right do you have to say that what I've written is wrong?'
A key premise in establishing the 5 variants in any 'Translator Choice' activity is that there are quantifiable, objectively explainable differences that produce qualitative differences in a translation. The grading of student translations is based on such differences. Use of the term 'mistakes' or 'errors' in relation to translation is a sensitive issue, but, if marks are being deducted because of the assessor's preference of one linguistic form over another, then, regardless of the terminology, as far as the student who has lost the marks is concerned, they have made a mistake. Curiosity and professional responsibility compel the foreign-language teacher to explore how their position fits with that of mainstream professional translators.
Aveling (2002:4) comments that 'surprisingly, the word 'mistakes' is not one that often enters into the contemporary theorisation of translation - outside language learning classrooms' and he remarks on the dearth of references to 'mistakes' or 'errors' in many contemporary books on translation. The discourse appears to favour terms such as equivalence, adequacy, accuracy, appropriateness, completeness, correspondence, exactness, fidelity, identity, nuance, phrasing, precision or usage, accompanied by some form of negation or qualification. This may be because, in many cases, experienced translators are writing on the subject for the benefit of their peers. A certain level of proficiency is assumed, and observations relate to the limitations imposed on the full transferability of meaning by differences in linguistic systems, cultural frameworks, audiences and artistic expression. Neither do errors appear to be a major preoccupation of courses in translation studies. In the course descriptions doctoral program in Multilingual communication at Barcelona's Universitat Pompeu Fabra, for example, where 8 of the 24 core subjects relate explicitly to translation, the term 'errors' features only once - in a methodological course on description, criticism and comparison of literary translation.
It is interesting that Aveling made a special case of language learning classrooms. They are, however, not alone in the category. The list of exceptions could be expanded to include journalists and translator accreditation authorities.
As commentators on society and current affairs, journalists not uncommonly report incidents of glaring errors. The following are two recent, high-profile cases:
'Microsoft has also managed to upset women and entire countries. A Spanish-language version of Windows XP, destined for Latin American markets, asked users to select their gender between "not specified," "male" or "bitch," because of an unfortunate error in translation.' (Best, 2004).
Pirated Potter book lacks magic
A pirate version of the latest Harry Potter blockbuster has hit the streets of Venezuela, but the illegal book lacks the magic of the original and the translator apologises on virtually every page.
The rogue Spanish edition of JK Rowling's Harry Potter and the Order of the Phoenix is peppered with notes from a translator who had to leave phrases in English with a Spanish explanation "(Sorry, I didn't understand what this means)."
A similar note on the same page tells readers: "(I didn't really understand what this phrase meant, so I paraphrased)".
The translation may leave a lot to be desired but the books were selling like hot cakes in downtown Caracas, where street vendors were hawking it for around $US25.
The book first appeared in English in June but an authorised translation into Spanish is not due to go on sale in Latin America for several months.
"Whoever translated this wanted to beat the market ... but I don't think it quite catches the spirit of Ms Rowling's original," said Jose Farres of Editorial Oceano, which holds the rights for the book in Latin America. (Australian Broadcasting Corporation, 2003)
Translator and interpreter accreditation bodies have a reputation for applying rigorous assessment criteria. The certification exam for the American Translators Association (ATA) (2002) 'is challenging with an overall pass rate below 20%'. The California-based provider of translation services 'Transcend' (2000) claims that, for California Court Interpreter certification "the 96% failure rate is evidence of the difficulty in mastering these skills'. And, although less forthcoming statistically than its US counterparts, the statement by the National Accreditation Authority for Translators and Interpreters of Australia (NAATI) regarding pass rates in its Overseas Testing Guidelines (2004:11) that 'NAATI accreditation is the benchmark for entry into the profession of interpreting and translation in Australia. Consequently, the standards are high.' no doubt alludes to the application of a similarly stringent marking scheme.
The main objective of 'Translator Choice' is to provide students with a means of discovering the range of factors that can influence the quality of a translation. It is achieved by allowing them to explore various options and study the feedback generated by the instructor for each variant. The priority given to the provision of feedback places the creator of a 'Translator Choice' activity in a very different position from the ATA examiner, for example, who is operating in an environment where 'Before taking the exam, candidates must sign a statement acknowledging that they know they will be informed only of the final result of the exam, without comments from graders, and that the exam will remain the property of ATA.' (ATA 2002). It would be absurd to claim this level of confidentiality conceals some form of unjustifiable conduct on the part of the examiners. On the contrary, the ATA is one of the very rare bodies that spells out and illustrates its assessment criteria in great detail (see below), and there is every reason to assume that these criteria are applied fairly and professionally. But their objective is to assess, not to instruct.
With 'Translator Choice', the teacher's accountability is transparent: the student can read, and potentially challenge, the feedback provided for each translation deficiency. The likelihood of students absorbing this information in a detached manner is no doubt increased by the fact that they are never in the position of having to defend a translation that is the result of their personal effort or that has cost them money to submit.
Because the activity was created as an intuitive response to a student need rather than for the sake of applying or testing a theory, the feedback to date has been totally ad hoc: whatever errors or inadequacies were thrown up by a student translation are explained to the best of the teachers ability and as objectively as possible. This has generated as many different explanations as there have been points of divergence from an acceptable rendering. Across the 50 passages they will run to thousands. Points eliciting comment can be as diverse as spelling mistakes (e.g. 'to procede' instead of eto proceedf), unidiomatic word-for-word translation (e.g. 'day fall' rather than 'nightfall' as a translation of 'la chute du jour'), inappropriate word choice for a given context (e.g. translating 'faire demi-tour' as 'to bang a u-turn' rather than 'to turn back' or 'to turn it around' in the context of handling a boat on a flooded river), and serious mistranslation arising from misinterpreted word forms and failure to understand a grammatical construction (e.g. translating 'Cfest aussi fin qufune grive' as 'It's the end of the thrush' instead of 'It's as tasty as thrush' with reference to the culinary potential of a hunted bird).
The teacher progressively builds up a profile of the types of problems and their distribution as the corpus evolves. If the activity had been designed to be static, there would be no need to do anything more than provide a full bank of feedback, and perhaps refine it in the light of subsequent insights or in response to input from students as they work through the exercises. But if the site is to realise its potential to serve as a tool for understanding the obstacles to translation encountered by students and for improving the teaching of translation in foreign language classes, it is not enough to have a sea of data. For patterns and principles to emerge it is necessary to establish a taxonomy of errors. Thorough classification will provide a mechanism for tagging feedback items (provision has been made for this in the design), which will in turn allow an investigator to study the degree to which students are sensitive to that feature as they assess a translation. Findings can then be incorporated into awareness-raising classroom activities and practical exercises.
If it is rare to find explicit references to 'translation errors', it is even rarer to find taxonomies of those errors. One area of translation in which the development of classification systems has been inevitable is that of Machine Translation (MT). As White (2003:211) states, 'Evaluation has always been central to the consciousness of those involved in the field of Machine Translation.' He points out that such evaluation frequently determines the direction of subsequent MT developments, it helps establish the cost-effectiveness of any investment in MT, and that it is hard to do because there is no single standard: 'The most obvious standard for MT, i.e., the "right" translation, is the very thing translation itself cannot provide.' (2003:213). It is, however, unlikely that with its need to satisfy the requirements of an array of stakeholders including monolingual or bilingual end-users, managers, developers, vendors and investors, MT evaluation will provide a suitable framework for the type of error analysis required to hone the translation skills of the foreign-language learner. It is also unlikely that such categories as those provided by the Framework for the Evaluation of Machine Translation (FEMTI) of the International Society for Language Engineering (ISLE) (2004) will provide the most useful starting point. The framework evaluates machine rather than human generated output. While there may be some overlap between the type of errors generated by students and those generated by a machine responding to comprehensive models and routines established by language engineers, they are likely to be viewed in different ways, given that in FEMTI (section 1.3.1) it is stated that 'the required translation quality is generally not high' and that, while fidelity is 'important', both style and wellformedness are rated as 'not a very important factor'.
In the search for the criteria governing expert human assessment of human translation it is natural to turn to what is likely to be the strictest and most highly developed environment: the official accreditation of translators. In some cases, national accreditation authorities provide nothing in the way of assessment indicators. In others, such as NAATI (2004), broad categories are provided: 'Marks are deducted in translation tests for mistranslations, inappropriate vocabulary, incorrect punctuation, incorrect grammar, incorrect spelling, distortion of meaning, unidiomatic usage, stylistic infelicities'.
The information on Exam Error Marking (revised 2002), reached from ATA's Certification page [http://www.atanet.org/bin/view.pl/12438.html], is refreshingly clear and thorough. It lists the following categories of errors: Incomplete passage, Illegible, Misunderstanding of original text, Mistranslation into target language, Addition or omission, Terminology, word choice, Register, Too freely translated, Too literal, word-for-word translation, False cognate, Indecision?gave more than one option, Inconsistency (same term translated differently), Ambiguity, Grammar, Syntax, Punctuation, Spelling, Accents and other diacritical marks, Case (upper/lower), Word Form, Usage, Style. Categories are accompanied by clear definitions, comments on how they are viewed by examiners, and helpful examples. There is inevitably a degree of generalisation in the descriptors, as they derive from and apply to many different language pairs. But they provide a very practical starting point for the classification of human translation errors. It is anticipated that when applied to the extensive corpus of French-to-English translation data currently embedded in 'Translator Choice', they will be able to be refined in such a way as to provide the teacher with a foundation on which to construct a valuable set of complementary instructional activities.
As it is used, the site gathers data on the choices made by students, the time taken and the score achieved. This data can be studied in relation to individuals or groups, and with reference to isolated segments or the passage as a whole.
Figure 4 shows the overall statistics for a passage taken in Test mode by two groups of students, one doing it as a Select One activity, the other as Rank the Translations. It is possible to gather data for any combination of passages and groups.
Research potential: Data for whole groups provide insights into the relative difficulty of passages; comparison of the difficulty of the Select One and Rank the Translations activities; amount of time dedicated to reflecting on translation outside of class. Data for an individual student's attempts over time would make it possible to discern any improvement in their perception of accuracy.
Figure 5 is a display of data for the individual segments of a passage (in this case using Rank the Translations option), showing the total number of times it was completed by students, the average time taken, and the average score.
Research potential: Time taken and average score give an indication of whether or not students are readily identifying particular sets of problems and processing them accurately.
Figure 6 is a display of information relating to student choices in a Select One activity: it indicates the beginning of the segment in French; the beginning of each of the alternative translations in recommended order, followed by the number of times that translation was selected as most accurate by students, the total number of student attempts for the segment, the average time taken on that segment, and the average score.
Research potential: Allows for the detection of 'blind spots' in students' filtering of information. And, if students who have consistently demonstrated a high level of perceptiveness in relation to most of the translations rank the variants differently, it allows the teacher to challenge the assumptions of the original ranking and admit that they may have been the result of personal preferences rather than differences that can be objectively established on linguistic grounds.
Figure 7a and 7b show the most detailed level of data analysis for a Rank the Translations and a Select One activity respectively. In each case, once a full classification system has been developed for feedback, it will be converted to tags showing in detail the set of difficulties attached to each translation variant. Research potential: This configuration of data will make it possible to gain insights into the nature and number of errors or problems that students can make in their translations, establish some sort of ranking of their impact on overall accuracy, and determine which type of problems are most readily recognisable. In Practice activities (feedback is not accessible when activities are done as Tests), when the full bank of feedback becomes available for each passage, it will be possible to better understand when and how often students are exploring the linguistic explanations of problems they did recognise in their assessment of accuracy. Preparation of the feedback will allow language-specific refinement of the ATA classification system.
The conception and creation of any foreign-language teaching resource is an exercise in applied linguistics. It requires the teacher not only to understand the linguistic system of the foreign language, but to manage the material in such a way as to create a bridge that carries students forward in their understanding and mastery of the language. That the activity is enriching for the student is axiomatic. The process also provides the teacher with insights that can be reinvested into teaching and research. In the case of this project, those insights relate to the peculiar status and nature of translation when it is carried out in the context of a foreign-language classroom. Moreover, they point to the need to evaluate the translations produced by students against some standard, and to apply clear, consistent, and linguistically justifiable criteria in explaining any shortcomings.
From the point of view of the student user, 'Translator Choice' is a simple interactive online exercise assessing the ability to recognise and evaluate the linguistic features of human translation. Its test activities are an integral component of their semester's work, and its practice activities provide the opportunity to access expert commentary on linguistic phenomena and distinctions that might otherwise elude the student.
From the perspective of the teacher-designer, the site is multifaceted, serving as an editable and expandable manager of a valuable but otherwise unwieldy activity, and as a unique database. Analysis of the data gathered will make it possible to establish a functional taxonomy of the errors and inadequacies characteristic of the translations produced by foreign-language learners. It will also improve our understanding of such things as patterns of use, student progress, the relative difficulty of passages, and recognition or non-recognition of translation difficulties by students or teachers. This, in conjunction with informal feedback from students, will complete the teaching-learning-research triangle by improving existing teaching materials, shaping the development of new ones, and by creating a mechanism for the regulation of the contents of the site.
Brian McCarthy is Associate Professor in French at the University of Wollongong, N.S.W., Australia. His principal research interests are in foreign language teaching methodology, computer assisted language learning, software development, phonetics and translation.