An Experiment in MT Post-Editing by a Class of Intermediate/Advanced French Majors

This paper describes an experiment whose goal was to introduce MT via postediting to a third-year university class in French-to-English translation. Even though most of the students in our French programs do not go on to become professional translators, it seemed a worthy objective to give them at least a taste of the challenges and opportunites which MT offers today, given how widespread it has become in the commercial and technical sector. It would also be valuable to know to what extent a brief hands-on encounter with MT would help them better understand the nature of translation. After providing background on the course content, level of students and type of documents chosen for the experiment, I briefly look at the question of how to “teach” MT and give reasons for the focus on post-editing. I then outline the experiment itself, including the process of choosing an MT program, the type of texts, the error analysis adopted for evaluating both human and MT output of those texts, a comparison of human and MT results and finally student reaction to the experiment. I conclude with a glance at the next phase of this project, a parallel application in English-to-French translation, where, unlike the first phase, the target language is not the students’ native tongue. 1 1 This paper is a revision of Kliffer 2001 which describes the experiment in French. I have added a discussion of post-editing, focusing on the differences between its use in commercial/technical translation and its exploitation in the classroom. 1. Background: course, level of students, type of documents The course chosen for the experiment is a level 3 French-to-English course which is one of four translation courses intended as a supplement to language courses rather than part of a translator-training program. This distinction is important to note for the experiment because the course’s general orientation means that the focus is on mainly journalistic, topical texts instead of the specialized subjects expected in a course for training professionals. The course continues the presentation of basic translation skills, such as proper dictionary usage, contextbased inferencing strategies, and English-French contrasts in lexicon, style, and advanced syntax. As the only one of our translation courses working exclusively from French to English, the native language of most of our students, it places more emphasis on precise interpretation of the source text. The students have typically had several years of French in secondary school, so that by third year in our program their French is quite advanced. Our courses immerse them in French, whether the content is literature, linguistics, culture/civilization, or language-based. Yet, the course where the experiment was carried out is the first time most are exposed extensively to journalistic texts and in spite of their extensive study of literature, they find journalistic translation quite a challenge, thanks largely to the abundant cryptic references, the cultural subtext, and the often opaque figurative language. Of course, this is not to downplay target text An Experiment in MT Post-Editing by a Class of Intermediate/Advanced French Majors EAMT 2005 Conference Proceedings 161 factors, such as the audience’s existing beliefs and encyclopedic knowledge, but several years of teaching this course have confirmed that the biggest challenge for all but the best students is to arrive at a correct understanding of the source text. 2. “Teaching” MT: why focus on post-editing ? MT is one of five ancillary topics on translation theory that the students read about as a complement to their regular “hands-on” work. Before the students attempt post-editing, they read Hutchins 1992 and Sampson 1987 order to learn about the major prototypes of MT, its chequered history, controversial claims and the reasons why many commercial and governmental organizations now consider PE indispensable for producing publishable-quality output. An experienced translator may well question the appropriateness of simulating PE in a course on general translation. As Melby & Warner 1995 insist, successful MT today is confined to texts which utilize sub-languages restricted to a particular, usually technical, domain. This is the diametric opposite of journalistic texts, aimed at a general public and free of any such constraints. Moreover, PE specialists stress efficiency, which entails that the revisor should not devote too much time to stylistic niceties and should avoid the temptation to rewrite throughly the output text, caveats going against the methodology of our translation courses, which encourage the student to aim for semantic and stylistic precision via multiple rewritings. Nevertheless, there are good reasons for having students perform PE in order to learn about MT. First, it is arguably the component of MT that ties in best with translation desiderata already emphasized within the course, notably concern for semantic and functional accuracy vis-à-vis the source text. The other components would entail either practices that contravene the course methodology (e.g. pre-editing, which suits technical writing but in most cases not journalistic texts) or abilities beyond its scope (e.g. writing MT code). On the other hand, revising MT output is feasible: if students have access to the output and the source text, they can apply their existing knowledge of both languages to make corrections, without any special technical training. PE thus makes student aware of MT’s capabilities and limitations, as well as providing additional translation practice through revising.