Introduction

Although the first artificial neurons were invented in 1943 by two mathematicians, W. McCulloch and W. Pitts, A. Turing is often considered as the precursor of artificial intelligence, even if he did not use the term in his 1950 paper ‘Computing Machinery and Intelligence’. Instead, the term was introduced in the summer of 1956 in a conference at Dartmouth College. The ambition of researchers at the time was to design a machine capable of matching or even reproducing human cognitive abilities. Since then, AI has alternately aroused interest or, on the contrary, been the victim of a certain apathy, due to the limitations affecting its effectiveness. However, it was re-established after 1996, when Deep Blue, an IBM expert system, triumphed over G. Kasparov, who was the world chess champion at the time: this feat rekindled interest in AI. The advent of the Internet and the introduction of big data in the 2000s were a huge breakthrough that has led to spectacular advances in this field to date (PIEKOSZEWSKI-CUQ, [s. d.], 12‑18). In France, a national strategy for AI was set up in 2018. In 2019, UNESCO held the ‘Beijing Consensus’ on AI and education, and ChatGPT launched a free, open-access version in 2020 (BLUTEAU, 2024, 14). Since then, there has been a steady increase in the discussions and the scientific disseminations about the usage of AI (BILAL & DEVILLERS, 2018; BILBARD & LABOURET, 2024; BONIFACE & PELPEL, 2024; Institut de France, 2024; KILIC & PAYET, 2024; OUDEYER & ROY, 2024; SAUQUET & VIELAJUS, 2014; SOLIGNAC, 2024). AI is everywhere; we could even think that it is excessively present sometimes. Because of its practical effects and its undeniable utility, it imposed itself in our daily lives at high speed (in around two decades), up to the point of transforming them. Therefore, we must live with it and use it. However, it is not always easy to measure its pertinence, its challenges and the underlying risks. Its usage raises some technical, intellectual, scientific and ethical questions, which led to the creation of two AI European regulations in 2024 (BLUTEAU, 2024, 14‑15). We must also acknowledge the emotional dimension that the use of AI raises, whether it is for the fascination, the appeal or the fear that it provokes, concerning its real or potential effects. A fortiori, within the field of education, the issue becomes more pertinent. Professors have to train their students. They have to help them develop their own intelligence, memory, knowledge, and skills, while, increasingly, AI can deliver a comparable assignment, or even one of a higher quality, which encourages the students not to do it themselves or to put in less effort. That is why, for several years now, scientific productions, academic works and dissemination articles regarding the use of AI in the teaching context have multiplied.

Similar questions arise in the field of translation, since tools like Reverso, Reverso Context, Wordreference, Pons or Systran are developing increasingly and more efficiently. Amongst them, aligned corpora like Linguee or TradooIT can be distinguished, along with Neural Machine Translation tools (NMT) such as DeepL and Google Translate that are trained based on text samples from Linguee, for example, or even ChatGPT, as generative AI. Their practicality and pertinence are checked by the users, professionals or not, and by researchers interested in these matters (BREYEL-STEINER & GRASS, 2021; HANSEN et al., 2022; LOFFLER-LAURIAN, 1996; NAOUAL & LAHBIB, 2017; UTUSHKINA, 2023). Regarding translation, the presence of such software modifies the translating activity – in writing, but more and more orally- making us believe that they could maybe replace humans in some scenarios. Therefore, the translating and interpreting professions are changing rapidly, threatening a part of the (future) professionals of the field or making them improve their capacity to train themselves and to constantly update their skills, in view of these evolving tools. This issue is of the utmost concern for most bio-translators (BREYEL-STEINER & GRASS, 2021; DIRAND & ROSSI, 2019). All of them feel directly involved.

With that in mind, teaching translation raises a problem: how to teach others how to translate, with tools evolving with such rapidity, and the role of the human translator changing constantly? Academic papers about students Computer Assisted Tools (CAT) use and the translating teaching practices concerning these tools are both necessary and useful, as demonstrated by recent publications and scientific events (CLOISEAU, 2022; RUCART, 2024). One of the questions that arises is the following: “What are the current practices and expectations of students, as regards of the CAT?”

To answer this question, and in line with G. Cloiseau’s analysis (2022), the idea of carrying out a survey at the Faculty of Arts and Humanities (FLSH) of the University of Limoges (France) arose. The objective was to assess the teaching practices in the light of the expectations and needs of the students of this institution. In other words, how do the current translation teaching practices match the expectations of the students regarding the AI, knowing that a large majority of them use these evolving tools whether they are authorized or not, and that the professors have not been trained to use it, considering it or often presenting it to the students as an unloyal help or an obstacle to learning translating practices?

To conduct this survey, two student groups of the University of Limoges were selected in autumn 2024. One from second year of MA in LEA (Applied Languages) in Intercultural Management (IM) and the other one from second year of MA in Cultural Transfers and Translation at the FLSH. Guided by the Research Engineer in Data Production, Processing and Analysis Assistant of the faculty, D. Santos Araujo, and the professor-researcher and Project Management tutor V. Lagarde, six MA students prepared and distributed a survey about the use of machine translation in the translation courses at the university. Composed of 23 questions, this survey was presented as a poll made with Lime Survey, which focused mainly on education level (Bachelor’s or Master’s degree), the academic program training (LLCER – Bachelor in Languages, Literature, Foreign and Regional Civilisations), LEA, TCT, Intercultural Management (IM) or ITC – Identity and Cultural Transfers), the languages involved, or even the type of document to translate (literary, technical, audio-visual, etc.), aiming to identify the student’s practices both in the context of the courses and elsewhere.

This work was carried out between September and December 2024: the entire second year class of the Cultural Transfers and Trilingual Translation (Spanish-English-French) MA, consisting of three students, initiated a literature review and elaborated the poll. Then, three Applied Foreign Languages MA second year students (Intercultural Management option) managed it: after loading it into Lime Survey, the translation professors, as well as all students in language programs taking translation courses, were informed about this project. Then, the LEA MA students shared the poll by going to the classes or sending messages to the target via mailing list. They followed up and sent out reminders, and started a first frequency table, thanks to the statistics proposed by Lime Survey. This was an exploratory study – not a confirmatory one. It was a continuation of the work already carried out by the LEA MA students in 2023 (who were by then in first year), which could also be further pursued in the future.

Following this initial presentation of the issues raised and the procedure implemented, here are the results of the survey, which will open a discussion that will show the diversity of situations depending on the context and will suggest some improvements for the translation professors at the FLSH or other similar institutions. 1

1.Results

1.2.Participants’ profiles

A total of 604 FLSH students, from L1 (first year of bachelor’s degree) to M2 (last year of MA) inclusive, were targeted by the survey - i.e., all those taking translation courses at the faculty. 2 There were 174 valid answers. The response rate was therefore 30 % on average, although it varied according to level and major. The variability in response rates can be explained by ethical and quantitative reasons. First of all, the fact that 70 % of the population did not respond or gave an incomplete answer, may be explained by a high volume of solicitations, as many questionnaires and surveys are regularly sent out within the university. Additionally, a low response rate may be due to a lack of interest in the subject or a failure to understand certain questions or their relevance. Also, fear of how the answers might be interpreted, or fear of being recognised when the data is cross-referenced (which could be the case in small classes), partly explains the incomplete or absent answers in some cases. Furthermore, there are other hazards to be considered before analysing the results of the survey. First, the reliability or honesty of some responses must be evaluated. Given that these tools are seen, particularly by some professors, as fraudulent, it is likely that some students are reluctant to admit to using them, or at least to admit how often they use them. Secondly, some students probably censored themselves regarding their (dis)satisfaction with certain courses in connection with AI, which is highly present in their lives but still rarely studied and taught at the University of Limoges. Last, some of the questions in the survey may have been understood differently depending on the individual, such as those regarding to the frequency of use of CAT tools: this parameter depends on each person's judgement and is therefore debatable.

If we take a closer look at the profile of the participants, we can see that there are differences in the response rate of students depending on their education level: in overall, the majority of participants were from Bachelor's degree programmes (47 juniors L1, 71 from second year L2, and 25 seniors L3), while Master's programmes were fewer in number (18 from first year -M1- and 13 from second year -M2-). However, the responses from L1 and L3 were less representative. Less than 20 % of responses were obtained for these two classes, whereas more than half of the L2 and M1 students responded, as did 45 % of the M2 students. Concerning the profile of participants by major, the questionnaires were mainly completed within the English LLCER Bachelor's degree (98 complete responses) but the results are particularly convincing for the LEA Intercultural Management and the Cultural Transfers and Translation MAs because, in both cases, the participants represent respectively 62 % and 50 % of their class. Even if the number of students concerned remains, overall, relatively limited (21 students for the LEA-IM MA and 5 for the TCT MA).

In all the other majors, the response rate was considerably less than half. Only 14 % of LEA students responded (34 responses), 26 % of Spanish majors (11 responses), 39 % of English majors, and 28 % of MA students in the Identity and Cultural Transfers programme (5 students, who are also Anglicists). To complete the description of the participants’ profile, it should be pointed out that the majority of Bachelor's degree participants were pursuing traditional studies, without retaking a year or re-enrolling at a later age. Most of them are 18 in L1, 19 in L2 and 20 in L3. On the other hand, students’ admission to the MA programmes is often delayed, since by the time they reach M1 most students are over 22 years old, which shows that other studies or experiences of at least one year, after they have obtained their Bachelor's degree, often precede the preparation of this new degree. This is why MA students are, on average, older than expected at this education level. Furthermore, 87 % of the students surveyed are native speakers of French, but they are likely to use other languages, including some that are not taught at university (this is the case for 43 % of them). Beyond the university context, students mainly use Latin languages, but also Asian languages, Germanic languages, Arabic, and, to a lesser extent, African dialects, which reflects a certain diversity of practices outside university.

1.2.Findings

Regarding Machine Translation (MT) software, the survey questions focused on three different contexts of CAT tools use: firstly, whilst in class, secondly, at home as part of the class, and finally, in a non-academic context.

In the first case, i.e., during classes, 82 % of participants said they used these tools, although some thought they did so “rarely”. However, in this case, use is directed towards documents that correspond to the content of their studies, as shown in Graph 2.

In nearly all majors, CAT tools are used primarily to translate literary documents, except in the LEA MA, where students use them primarily to translate non-specialised documents (71 %), followed by technical documents (67 %) and then literary documents (57 %). On the other hand, these tools, which are still less effective orally, are used little, or not at all, for translating audiovisual documents. In the ITC MA, CAT tools are used neither for this type of source nor for non-specialised documents.

In terms of their use during classes, the CAT tools are used both for the “thème” (i.e., from French as a source language (“SL”) into another language) and for the “version” (i.e., from another language into French as a target language (“TL”)) in all majors.

Of the 142 students (82 % of all respondents, as indicated above) who admit to using CAT tools in class, the level of satisfaction is high, since more than half of them have a positive or even very positive view of these tools: this is the case for 57 % of Bachelor's students and 81 % of MA students. The satisfaction rate is much higher for MA students, as shown in graph 3.

The second context studied concerns the use of MT software by students outside the classroom, but related to their assignments: 89 % of respondents said that they used it in this context, which seems to show that the practice is slightly more significant outside the university than within it (as already shown, 82 % used it in class). This suggests that either professors want to check students' translation skills in class and so forbid them from using CAT tools when they are in the classroom, or that students are more encouraged, feel more entitled or take more freedom to use them when away from their professors. There is a difference according to the major, as shown in the fourth graph.

While just over half of translation specialists (TCT MA) admit to using CAT tools, students in the other Masters programmes all use them, albeit with varying frequency. The LEA and ITC Masters are indeed language-oriented, but learning to translate is not a priority, since the students concerned are not oriented towards this activity. Moreover, this use increases with the level of study, as can be seen in Graph 5.

While more than 60 % of students say they use these tools at home as part of their academic activities, regardless of the education level, the percentage is higher in MA (M1 and M2) than in Bachelor's degree (L1, L2 and L3). As already shown, since 55 % of M1 and 45 % of M2 students responded to the survey, the results obtained from MA are more reliable. Whereas the responses from L3 students, for example, are probably less representative, since only 19 % responded.

Regarding the type of documents, these tools are primarily used for translating literary documents assignments, and then for technical documents. And this is the case in all majors, except for the LEA-MI MA, where technical documents come first, followed by other sources, whether literary or not, which is logical given the professional orientation of these students and the content of their courses.

Once again, audiovisual documents are the least frequently translated using CAT tools in all majors. This is probably due to technical reasons. If you want to use a translation software, it is sometimes necessary to make a transcription of the speech before proceeding with the translation. CAT tools that translate speech directly are still rare and are rarely used in university settings.

This is confirmed by the answers to the questions about the type of translation – oral or written: most students in all majors use CAT tools mainly for written rather than oral translation. Of course, depending on the results, a small number of Undergraduate students use these tools for oral translation, but it is possible that the question asked on this subject was misunderstood, as interpreting is rarely taught at this level, and even less from home, in preparation for lessons. Therefore, it makes little sense for Bachelor's students to use software that automatically translates speech (unless these younger students are aware of oral translation tools that are accessible outside the university and unknown to their elders).

Finally, a last group of questions focused on the use of these tools completely outside the university context, i.e., neither during classes nor for their assignments. In this case, a distinction was made between languages in which the students were proficient, even though they were not taught at university, and languages in which they were not proficient at all. In the first case, only 141 students replied that they were proficient in languages other than those taught to them at university. The majority of these used CAT tools for these languages, particularly in ITC MA (100 %), as the seventh graph shows.

On the eighth graph it appears that the rate of positive responses for the use of these tools in the case of languages in which they are not proficient, is over 85 % in all the Bachelor's degrees, over 90 % in the LEA-IM MA and up to 100 % for the other two Master's degrees, ITC and TCT.

This suggests that the software compensates partially for the lack of knowledge of the language and the lack of translation skills.

1.3.Expectations and requirements

It is possible to establish a difference between current practices at the FLSH and the need of the students. Students say that most of Bachelor's degree professors (L1, L2, and L3) do not talk about CAT tools in their lessons (between 68 % and 69 % for each level). On the other hand, only 56 % of M1 students consider that their professors do not mention these tools. The situation is different for the M2 students since almost 77 % of professors regularly mention these tools in class, even if it is not part of the syllabus. In other words, at the lower levels, professors talk little or nothing about CAT tools, but the situation changes as studies progress. In fact, in the current syllabuses of the FLSH, language courses generally do not include specific courses on CAT tools, except in the TCT MA, and only during the first year. However, the results by major show that a small number of undergraduate students (LEA and LLCER) seem to have benefited from classes on the subject:

Based on the results, it is possible to interpret graph 9 in two ways for the first two columns: either undergraduate students who replied that they had taken courses on CAT tools were mistaken because they had not fully understood the question or because they made a mistake while filling in the questionnaire; or their answer is correct, which would mean that a small percentage of L1 and L2 students had access to specific lectures on the subject, even if this was not part of the syllabus. In any case, the absence of classes about CAT tools, which in general concerns most majors and students, contrasts with the expressed interest of the participants to be able to take such classes, as shown in the following graph:

If we consider this interest by education level, at least 48 % (in L3) and 49 % (in L1) of students would like to receive training on CAT tools. However, these two percentages are the least reliable, since only 17 % of L1 and 19 % of L3 students responded to the survey. On the other hand, in L2, where the results are more conclusive, 66 % of students already expressed an interest to be trained in CAT tools; 85 % in M1, and 100 % in M2, of students would like this.

The L2, M1, and M2 students, whose respondents are the most representative of their class - since the response rate is 54 %, 55 % and 45 % respectively - are strongly interested in receiving training in CAT tools. And this interest, which seems to be general, seems to increase as the education level does: if we exclude the L1 and L3 students’ responses, which are less representative, the older and more advanced the students are in their studies (since we have seen that date of birth and level of study are correlated in the Bachelor's degrees), and the more their language and translation level improves, the more they look forward to being trained in the use of CAT tools. We can even see that the interest for a specific course increases in a similar way: 14 % of requests from L2, 17 % from M1 and 31 % from M2.

A last correlation can be established between the interest on learning about CAT tools and the training already received, as Graph 11 shows very clearly:

While more than half of the students (56 %) who are not informed on the subject would like to receive classes on these tools, specifically or not, this percentage rises when students receive more training. It goes up to 72 % when students hear about them regularly in their classes, and up to 80 % for those who have specific classes on the subject (i.e., mainly for M1 TCT students, which are the only ones whose syllabus includes a volume of hours dedicated to this). If we focus on the desire of these students to benefit from specific courses, this number also increases when students receive more detailed training on CAT tools: it is 8 % for students who never hear about them, 17 % for those who receive regular information on the subject, and 40 % for those who already have specific classes on the subject, which reiterates the fact that the more the students are trained and informed about CAT tools at university, the more they want to receive training. Conversely: the less the students hear about these tools, the less they perceive a need for training.

2.Discussion

The results of this survey highlight the general use of CAT tools by FLSH language-oriented major’s students in a variety of circumstances. However, there is a difference between students’ use of machine translation tools as part of their classes, for their assignments, or for non-academic purposes. The less proficient students are in a language, the more they use these tools. And this use, which is mainly for written documents and is widely used, seems generally satisfactory to them. However, this contrasts with the low priority given to these tools by professors, who are few to teach specific subjects about them, and who do not always mention them in their not-specific classes. In fact, only the TCT MA first year students have dedicated courses on machine translation tools. And it is precisely the students of this Master’s specialising in translation who are the most interested in courses of this type. But it is surprising to see that most students (almost half in L1 but 100 % in M2) are interested in courses, specific or not, on the subject.

As shown by the answers to the open-ended questions added at the end of the Lime Survey poll, this difference between students' practice and expectations can certainly be partly explained by a generational gap, by a lack of training on the part of the professors themselves, and by the different perception that professors and students have of AI. On the one hand, professors, who are generally older, do not have the same experience with these tools, since they discovered at a late age. By contrast, today's students are very quickly led to use them in their daily lives, even beyond the specific issue of CAT tools: this is proved by the generative AI, ChatGPT.

Most university professors have not received any training in these tools while they were at university. National directives do mention that students should be taught the knowledge and skills to use the latest digital tools, but professors rarely have the opportunity to train themselves specifically in AI. Students are aware of this, and do not hesitate to mention it in their answers to open-ended questions: they think that if their professors don't mention these tools, or don't mention them much, it's because they don't master them sufficiently themselves.

The third difference between professors and students lies in the perception of these tools: while 75 % of professors consider the use of AI to be fraudulent, only 65 % of students agree (BLUTEAU, 2024, 15). The fear and fascination caused by ChatGPT, seen as “ami des élèves, ennemi des professeurs” [“friend of students, enemy of professors”] (MAINE, 2024, 18) seems to be applicable to Machine Translation software. There are three reasons that explain this: firstly, students use these tools more spontaneously than their professors; secondly, professors, and therefore students, are not sufficiently trained in these tools, and the students' lack of understanding of professors' expectations: professors aim to pass on knowledge and translation skills, not to obtain a pre-made translation. Students are looking for speed, efficiency and translation quality, even if it's provided by CAT tools: they don't always clearly understand that using such a tool can inhibit their own thinking and progress as a translator. As a result, some of the open-ended responses to the survey reveal that professors advise students against or forbid them to use these tools, which they see as an obstacle to learning and training. As we've seen, other professors don't talk about it at all, probably because they are not entirely comfortable with a subject that is beyond them - since generative AI and CAT tools are progressing very quickly - which some students perceive as a “taboo”, and which makes them point out the distinction between their own perception - that CAT tools are a “tool” as the acronym indicates - and that of their professors, who see them more as a fraudulent instrument.

Conclusion

In conclusion, the divergence revealed by this research between students' practices and their expectations and needs raises several observations. First of all, it would not be useful to systematically forbid students from using these tools if they have access to a connected device, either in class or at home, since they are used to using them. Most professional translators use these tools themselves, so students obviously need to be trained to use them correctly. Nevertheless, while learning, these tools can act as a restraint, or even an obstacle, since they are able to provide translations, replacing an increasingly substantial part of the work expected of the student. However, human translators still have to carry out post-editing work, and to do so, they need to have sufficient translation skills themselves - skills they can only acquire by regularly making the effort to train themselves to translate without automatic tools. There are several recommendations that could be made: firstly, for professors to be trained more in CAT tools themselves. Secondly, they should talk to their students about them, either as part of their regular classes or in specialized ones. Thirdly, they should guide and specify the use of CAT tools, authorizing them or not, depending on the assignment, the student's level of translation competence, etc. For example, when a student is a beginner in translation, he or she might be asked to work in the classroom without a computer, so he or she can understand the translation process and learn to use his or her own knowledge of vocabulary, grammar, etc. in both languages. The student must be able to practice translation itself, mastering the whole process from understanding the source text to its transformation onto the target text, depending on the target audience and expectations. Gradually, as students become more competent and able to judge the work delivered by a CAT tool, they can be entrusted with more post-editing assignments. While it is true that CAT tools can be used as a tool as well as a fraudulent instrument, it is important to know how to optimize their use, so that they can actually be used to obtain quality translations, while respecting intellectual property rights. These recommendations need to be considered according to students' professional objectives and their major, as their interest in language and translation may vary, and the time they have available to learn how to translate will differ accordingly.

Perhaps AI will partially replace professors, via deep learning, even in the field of translation, but for the time being, students are mainly looking for classes that include new data. This can only encourage us to “développer de nouvelles méthodes” [“develop new methods”] and new pedagogical practices (BLUTEAU, 2024, 12). Students who want to be trained in these tools want to “penser avec logique, repérer les erreurs, exercer [leur] esprit critique” [“think logically, identify errors, exercise [their] critical thinking skills”] (BLUTEAU, 2024, 12). The aim is to help them learning how to translate, and to distinguish what remains a part of the bio-translator’s job and what the machine cannot do, such as “create, think, prioritize”, etc., while at the same time teaching them to make the best use of AI as an increasingly efficient and powerful tool, so that they are able to control this tool, and not the other way around (QUENET, 2024, 17), and can still taste the pleasure of translating