La traduction automatique ou le supplice du Nutella® - Commission européenne Aller au contenu principal
EMT Blog
  • Article d’actualité
  • 15 décembre 2020
  • 15 min de lecture

La traduction automatique ou le supplice du Nutella®

Jean-Yves Bassole, dr.Phil. dr.Litt., Directeur de l’Institut de Traducteurs, d’Interprètes et de Relations Internationales (ITIRI), Faculté des Langues, Université de Strasbourg, France.

Robotrad poster de la conférence

Les enseignants de traduction sont de plus en plus souvent confrontés à une réalité menaçante, qui est susceptible de nuire à leurs cours : les outils de traduction automatique. Certes, cette matière est enseignée en tant que telle et les cursus de traduction prévoient en général des cours de post-édition. Pourtant, la tentation est souvent trop grande : certains étudiants ne peuvent s’empêcher d’avoir recours à divers outils de traduction automatique mis gracieusement à leur disposition par les entreprises qui les ont élaborés. Qui renoncerait sans broncher à un pot de Nutella ? Disons-le sans ambages, nous pouvons tous comprendre cette tentation – qui n’y a jamais succombé ?  

Ce qui nous gêne le plus en réalité, ce n’est pas que la tentation se concrétise – nul ne saurait critiquer ses étudiants d’utiliser tous les instruments mis à leur disposition. N’est-ce pas finalement ce que nous leur conseillons de faire le jour où ils seront nos collègues dans le métier de traducteur ? Non, ce qui nous gêne, ce n’est pas qu’ils le fassent, mais bien qu’ils le fassent trop tôt, sans avoir préalablement suivi les cours appropriés, qui sont de nature à leur faire prendre conscience des atouts et des faiblesses des instruments offerts.

Une question se pose alors, côté enseignants : si les étudiants n’ont pas la patience d’attendre le bon moment pour utiliser les outils susmentionnés, ne serait-il pas avisé de repenser nos cursus en intégrant plus tôt la prise de contact avec ces instruments ? Les réponses varient selon les pays, les cultures et les établissements. La réponse à cette question est également liée à la structuration même des études de traduction : certains systèmes universitaires estiment qu’on peut commencer à étudier la traduction dès la première année de licence, d’autres considèrent qu’il faut d’abord avoir acquis de solides bases linguistiques et culturelles avant d’aborder la pratique de la traduction.

Tel est le cas de l’Institut de Traducteurs, d’Interprètes et de Relations Internationales (ITIRI), Faculté des Langues, Université de Strasbourg : les études de traduction débutent en première année de master, le ticket d’entrée étant constitué, d’une part, de trois années d’études universitaires (en langues, lettres, droit ou tout autre domaine) et, de l’autre, de la réussite à des examens d’admission visant à s’assurer du niveau linguistique et du bagage culturel du candidat.

On le voit, notre conception est celle d’un minimum requis pour passer au niveau supérieur. Dans ces conditions, il ne faut pas s’étonner de nous voir maintenir les cours de post-édition en seconde année de master de traduction professionnelle, de traduction littéraire ou de traduction audio-visuelle et accessibilité. Il nous semble évident que les outils de traduction automatique ne peuvent être d’une réelle utilité que lorsque l’étudiant est déjà suffisamment formé à la traduction.

Face à ce choix de principe des formateurs et à la tentation étudiante évoquée ci-dessus, la seule solution viable consiste à informer les étudiants de première année de master des dangers inhérents aux outils de traduction automatique. C’est dans cet esprit que je présenterai une communication au prochain congrès organisé par l’ITIRI, intitulé ROBOTRAD, qui est programmé pour l’automne 2021. Je m’y attacherai à exposer ce que je tente de démontrer à nos étudiants, à savoir qu’en règle générale nul ne devient virtuose sans avoir étudié le solfège.

Certains de mes collègues enseignant la traduction donnent parfois à leurs étudiants un texte à traduire accompagné de plusieurs ‘traductions’ proposées par des outils de traduction automatique. Cette opération hautement pédagogique, que j’assimilerais tout compte fait à une sorte de déminage, se fait généralement sur des textes relevant de ce qu’il est convenu d’appeler la ‘traduction pragmatique’. Pour ma part, j’ai souhaité faire l’expérience avec des textes tirés d’auteurs du 20e siècle mais avec une préférence pour des passages qui relèvent plus de la langue parlée.

C’est ainsi que j’ai pu sélectionner une palette de difficultés, de vocabulaire, de syntaxe ou de morphologie, sans oublier un ou deux traits d’esprit. Ces phrases ou groupes de phrases ont été soumis à cinq moteurs de traduction automatique pour traduction vers l’anglais. La méthodologie retenue a consisté à soumettre, un même jour, pratiquement à la même heure, une phrase aux cinq moteurs. Cette même procédure a été répétée deux fois, après un délai minimal d’un jour, l’objectif étant de confirmer ou non la stabilité des réponses. La même procédure a été mise en œuvre à partir de deux autres postes de travail.

Sans entrer ici dans le détail des constatations effectuées, trois grandes lignes se sont immédiatement dégagées :

  • dans de nombreux cas, la traduction proposée ne présente aucun caractère de stabilité : la même phrase peut être traduite de manière différente par le même moteur de traduction à 24 ou 48 heures d’intervalle ;
  • les traductions proposées d’un poste de travail à un autre ne présentent pas non plus un caractère de stabilité ;
  • les expressions relevant du langage parlé ou de l’humour ne sont pas souvent identifiées.

A titre d’illustration, voici les traductions obtenues pour le titre de ce billet, que j’aurais humblement traduit « Machine Translation or the Nutella Ordeal » :

The machine translation or the torment of Nutella.

Machine translation or the torment of Nutella.

Machine translation or Nutella supplementation.

Machine translation or Nutella torture.

S’agissant des traductions proposées, on constate autant d’améliorations sensibles que de détériorations étonnantes.

Très vite, les étudiants arrivent à la conclusion qu’ils ne peuvent se fier à ces traductions. Ils découvrent, à travers la multiplicité des solutions, le manque de stabilité des réponses et leur variabilité d’un poste de travail à un autre, la nécessité de se méfier, de douter et de considérer que ces outils, mis gratuitement à leur disposition sur l’Internet, ne sont pas de nature à les aider mais peuvent plutôt constituer un piège qui va leur faire perdre plus de temps qu’il ne leur en fera gagner.

Ceci une fois compris, les étudiants parviennent souvent à identifier des tours fautifs qui se répètent ; parfois même, ils arrivent à proposer une explication à certains types de déficiences. Il ne fait alors aucun doute qu’ils sont prêts à poursuivre leur formation sans risquer de se laisser séduire par les sirènes de la traduction automatique ; ils seront également prêts, une fois ce premier cycle achevé, à aborder sereinement les cours de post-édition.

 

Google Translate

Automatic translation or the torment of Nutella®

Translation teachers are more and more often confronted with a threatening reality which is likely to affect their lessons: machine translation tools. Of course, this subject is taught as such and translation courses generally provide for post-editing courses. Yet the temptation is often too great: some students cannot help but use various machine translation tools made available to them free of charge by the companies that have developed. Who would give up a jar of Nutella without flinching? Let's put it bluntly, we can all understand this temptation - who has never succumbed to it?

What really bothers us the most is not that the temptation materializes - no one can criticize their students for using all the tools at their disposal. Isn't that ultimately what we will be asking them to do the day they become our colleagues in the profession of translator? No, what bothers us is not that they do it, but that they do it too early, without having previously taken the appropriate courses, which are likely to make them aware of the strengths and weaknesses of the instruments offered.

A question then arises, on the teachers' side: if students do not have the patience to wait for the right moment to use the aforementioned tools, would it not be wise to rethink our courses by integrating earlier contact with these? instruments? Answers vary by country, culture and institution. The answer to this question is also linked to the very structure of translation studies: some university systems consider that one can start studying translation from the first year of the license, others consider that it is necessary first to have acquired solid linguistic and cultural foundations before starting the practice of translation.

This is the case of the Institute of Translators, Interpreters and International Relations (ITIRI), Faculty of Languages, University of Strasbourg: translation studies begin in the first year of the master, the entry ticket being made up of, '' on the one hand, three years of university studies (in languages, humanities, law or any other field) and, on the other, success in admission exams aimed at ensuring linguistic level and background cultural background of the candidate.

As we can see, our conception is that of a minimum required to move to the next level. Under these conditions, we should not be surprised to see us maintain post-editing courses in the second year of the master's degree in professional translation, literary translation or audio-visual translation and accessibility. It seems obvious to us that machine translation tools can only be of real use when the student is already sufficiently trained in translation.

Faced with this choice in principle of the trainers and the student temptation mentioned above, the only viable solution consists in informing the first year master's students of the dangers inherent in automatic translation tools. It is in this spirit that I will present a paper at the next congress organized by ITIRI, entitled ROBOTRAD, which is scheduled for the fall of 2021. I will endeavor to expose what I am trying to demonstrate to our students, to know that as a general rule no one becomes a virtuoso without having studied music theory.

Some of my translation colleagues sometimes give their students a text to translate along with several "translations" provided by machine translation tools. This highly educational operation, which I would consider altogether to a kind of demining, is generally carried out on texts relating to what has come to be called pragmatic translation ’. For my part, I wanted to experiment with texts taken from 20th century authors but with a preference for passages that are more related to the spoken language.

This is how I was able to select a palette of difficulties, vocabulary, syntax or morphology, without forgetting one or two wit. These sentences or groups of sentences were submitted to five machine translation engines for translation into English. The methodology adopted consisted in submitting a sentence to the five engines on the same day, practically at the same time. This same procedure was repeated twice, after a minimum of one day, the objective being to confirm or not the stability of the responses. The same procedure was implemented from two other workstations.

Without going into the details of the findings here, three main lines immediately emerged:

  • in many cases, the proposed translation does not present any stability character: the same sentence can be translated in a different way by the same translation engine 24 or 48 hours apart;
  • the translations offered from one workstation to another are not stable either;
  • expressions relating to spoken language or humor are often not identified.

By way of illustration, here are the translations obtained for the title of this post, which I humbly translated « Machine Translation or the Nutella Ordeal »:

The machine translation or the torment of Nutella.

Machine translation or the torment of Nutella.

Machine translation or utella supplementation.

Machine translation or Nutella torture. 

With regard to the proposed translations, there are as many appreciable improvements as there are astonishing deteriorations.

Very quickly, the students come to the conclusion that they cannot trust these translations. They discover, through the multiplicity of solutions, the lack of stability of responses and their variability from one workstation to another, the need to be wary, to doubt and to consider that these tools, made available free of charge on Internet, are not likely to help them but can rather constitute a trap which will make them waste more time than it will save them.

This once understood, the students often manage to identify wrong turns which are repeated; sometimes they even manage to offer an explanation for certain types of deficiencies. There is no doubt then that they are ready to continue their training without risking being seduced by the sirens of machine translation; they will also be ready, once this first cycle is completed, to approach post-publishing courses with confidence.

DeepL (free version)

Machine translation or the torture of Nutella®

Translation teachers are increasingly confronted with a threatening reality, which is likely to affect their lessons: machine translation tools. Admittedly, this subject is taught as such and translation curricula usually include post-editing courses. However, the temptation is often too great: some students cannot help but resort to various machine translation tools made available to them free of charge by the companies that developed them. Who would give up a jar of Nutella without flinching? To put it bluntly, we can all understand this temptation - who has never succumbed to it? 

What really bothers us most is not that the temptation actually materialises - no one can criticise their students for using all the instruments at their disposal. Isn't that what we advise them to do when they become our colleagues in the translation profession? No, what bothers us is not that they do it, but that they do it too early, without having first attended the appropriate courses, which are likely to make them aware of the strengths and weaknesses of the instruments offered.

A question then arises on the teachers' side: if students do not have the patience to wait for the right moment to use the above-mentioned tools, would it not be wise to rethink our curricula by integrating earlier contact with these instruments? Answers vary from country to country, culture to culture and institution to institution. The answer to this question is also linked to the very structuring of translation studies: some university systems believe that one can start studying translation as early as the first year of the bachelor's degree, while others consider that one must first have acquired a solid linguistic and cultural basis before starting to study translation.

This is the case at the Institute of Translators, Interpreters and International Relations (ITIRI), Faculty of Languages, University of Strasbourg: translation studies begin in the first year of the master's degree, the entry ticket consisting, on the one hand, of three years of university studies (in languages, literature, law or any other field) and, on the other hand, the successful completion of entrance exams aimed at ensuring the candidate's linguistic level and cultural background.

As you can see, our conception is that of a minimum requirement to move to the next level. Under these conditions, it should come as no surprise that we maintain the post-edition courses in the second year of the Master's degree in Professional Translation, Literary Translation or Audiovisual Translation and accessibility. It seems obvious to us that machine translation tools can only be of real use when the student is already sufficiently trained in translation.

Faced with this choice of principle by the instructors and the student temptation mentioned above, the only viable solution is to inform first-year master's students of the dangers inherent in machine translation tools. It is in this spirit that I will be presenting a paper at the next congress organised by ITIRI, entitled ROBOTRAD, which is scheduled for autumn 2021. In it, I will attempt to explain what I am trying to demonstrate to our students, namely that as a general rule, no one becomes a virtuoso without having studied solfeggio.

Some of my colleagues who teach translation sometimes give their students a text to be translated accompanied by several 'translations' offered by machine translation tools. This highly educational operation, which I would describe as a kind of mine clearance, is generally carried out on texts that fall within the scope of what is known as 'pragmatic translation'. For my part, I wanted to experiment with texts drawn from 20th century authors, but with a preference for passages that are more relevant to the spoken language.

This enabled me to select a range of difficulties, vocabulary, syntax or morphology, not forgetting one or two witty features. These sentences or groups of sentences were submitted to five machine translation engines for translation into English. The methodology used consisted of submitting one sentence to the five engines on the same day, at practically the same time. This same procedure was repeated twice, after a minimum delay of one day, with the aim of confirming or not the stability of the responses. The same procedure was carried out from two other workstations.

Without going into the details of the findings here, three broad outlines immediately emerged:

  • in many cases, the proposed translation is not stable: the same sentence can be translated differently by the same translation engine 24 or 48 hours apart;
  • the translations offered from one workstation to another are not stable either;
  • expressions in the spoken language are not often identified.

By way of illustration, here are the translations obtained for the title of this post, which I would have humbly translated « Machine Translation or the Nutella Ordeal » :

The machine translation or the torment of Nutella.

Machine translation or the torment of Nutella.

Machine translation or Nutella supplementation.

Machine translation or Nutella torture.

As far as the proposed translations are concerned, there have been as many significant improvements as astonishing deteriorations

Very quickly, students come to the conclusion that they cannot rely on these translations. They discover, through the multiplicity of solutions, the lack of stability of the answers and their variability from one workstation to another, the need to be wary, to have doubts and to consider that these tools, made available to them free of charge on the Internet, are not likely to help them but can rather constitute a trap that will make them lose more time than they gain.

Once this is understood, students often manage to identify repeated wrong turns; sometimes they even manage to offer an explanation for certain types of disabilities. There is no doubt that they are ready to continue their training without the risk of being seduced by the sirens of machine translation; they will also be ready, once this first cycle is completed, to tackle the post-editing courses calmly.

Détails

Date de publication
15 décembre 2020
Langue
  • anglais
  • français
  • néerlandais
Catégorie EMT
  • Compétences en traduction