TRACT Conference: The Impact of AI on Literary Translation: Assessing Changes in Translation Theory, Practice and Creativity, 20-21 October, Sorbonne Nouvelle University, Paris

The impact of AI on literary translation: assessing changes in translation theory, practice and creativity

In early 2020, we discussed the possibility of organizing a TRACT seminar series on machine translation (MT) of literary texts. Since then, this topic has been the subject of an ever-increasing number of conferences, articles and monographs. It is probably the spectacular “progress” of MT tools now available to the general public — in particular DeepL and Google Translate, taking advantage of recent advances in neural machine translation (NMT) — that has made it inevitable for the literary translation community to take this phenomenon into account.

            Indeed, these tools, especially because of their ability to process an impressive quantity of texts almost instantaneously, reinforce the idea that translation, i.e. going from one language to another, is quite a straightforward operation, the manifestation of a one-to-one relationship between two languages. This reflects a simplistic conception of language, seen as a code, which translators would simply have to decode and then re-encode, following transformation rules or algorithms.

            And this is precisely how the first translation machines were imagined and designed, before being supplanted by statistical translation, and then by so-called “neural” machine translation. However, the blatant failure of the first attempts at machine translation led to the total and brutal suppression of the budget allocated to this research in 1966 in the United States following the conclusions of the ALPAC report. On the other hand, the still perceptible imperfections of MT, based only on the statistical processing of huge parallel corpora, never seemed likely to call into question the role of human translators (otherwise called “bio-translators”). Until recently, only specialized or pragmatic translators often resorted to computer-assisted translation or CAT. However, with the rapid advent of CAT, even literary translators fear that their autonomy, their authorial status, their agency might be threatened. The creative dimension of their work, which translators have been claiming for so many years, is at risk of being forgotten and replaced by the ancillary activity of post-editing. Man at the service of the machine, so to speak.

            It is easy to see what advantages unscrupulous publishers could gain from this new situation. This is particularly true for so-called “genre literature” (fantasy, romance, sci-fi, etc.) that tends to follow repetitive and set patterns. The neural machine translation of fantasy or romance books, for example, would save a lot of time and therefore money, which would certainly change the practices of the publishing world.

            Faced with this situation, it seems that literary translation practitioners and theorists can no longer remain on the sidelines. “L’observatoire de la traduction automatique” [The Machine Translation Study Centre] set up in 2019 by ATLAS, the Association for the Promotion of Literary Translation, is a concrete example of this in France. It is not a question of adopting a defensive position, but of taking full account of the paradigm shift in translation that the emergence of NMT implies. In any case, it will not disappear and is even likely, according to some A.I. specialists, to make progress that could, in the long run, supplant bio-translators.

            That is why, beyond the fears aroused by NMT among translation professionals, and beyond the criticism of the quality of the translations it produces, we wish to question the shifts that NMT induces in our ways of considering translation. In other words, what NMT does to the concept of translation and, consequently, to translation theory — how our experience of translation, modified by the presence of the machine, necessarily affects the way we think about translation. Is the machine capable of capturing the singularity of an author's style, of what the author does with and to language? Can NMT find a strategy capable of restoring the complexity of the translation process, in one way or another? This leads us to a renewed questioning of what it means to “understand” a text, and more generally to “read” a text, especially if we consider with G. Steiner that “to understand is to translate”. Can we say that the machine reads the text in order to translate it the way the bio-translator does? Translating implies the implementation of an extremely refined form of thought. And this brings us back to the question posed by Alan Turing, one of the fathers of artificial intelligence (AI), back in 1950: "Can machines think?”

           How does the human translator understand the source text? Is reading the text to be translated different from reading for pleasure? How does the translator arrive at the target text, through hesitations, backtracking, dictionary consultations, etc.? Can research on the cognitive processes at work in human translators shed useful light on these questions?

Our seminar proposes to investigate the topic in three directions (which necessarily intersect at certain points):

• We would like to introduce literary translators, Translation Studies specialists, researchers, students to the new tools coming from AI, CAT, NMT, enlightening them on how they work, the role of computational linguistics, cognitive science, neurosciences, their history, the perspectives of progress, their limits etc.
• How does NMT measure up to literary texts; what challenges does literature — especially poetry — with its equivocation, ambiguities, enigmatic meanings, points of untranslatability pose to NMT? Conversely, what part can NMT, CAT tools, play in the renewal of literary creativity?
• Does NMT effect a paradigm shift for translation? To what extent do omnipresent machines allow us to gain awareness of the fact that certain processes that used to be performed by expert translations only have now become automatic? What is the place of bio-translators? Do they become liberated or alienated by the machine (which cannot function without human-produced data)? In what way can the translators' lived experience of these changes help to map out a new paradigm, which includes but also exceeds the pragmatic dimension of this work?

As part of the above, the following questions might be addressed:

• Could the new MT tools really replace human translators in the long run?
• Consequences of and new directions in teaching translation at universities in the age of NLP
• Can corpus translatology or CAT improve the quality of literary translations or retranslations?
• To what extent are the practices of pragmatic translators transferable to literary translators?
• Does the machine make the bio-translator an augmented or a diminished translator? What role for the machine, what role for the human?
• How do NNT and CAT change the translator's relationship to the literary text, his or her reading of the text, and thus his or her engagement with the text?
• Human/machine interaction in literary translation: is collaboration possible, desirable, or harmful?
• Aren't literary translators in danger of being strongly encouraged by publishers to become specialized editors (development of post-editing)? Won't the machine reduce them to an ancillary function that they have been trying to free themselves from for decades?
• Can't the machine become the ally of literary creativity, through the randomness it introduces into the translation, or through the formal constraints that can be instilled in it (rhymes and feet in the translation of poetry, for example)?
• Isn't genre literature, which often responds to fairly formatted forms of writing (fantasy, romance, etc.) an ideal target for the development of NMT in the literary field?
• What happens to the “translation project" — dear to Antoine Berman — if we entrust the text to a machine?
• Does corpus stylistics allow us better to study and compare the translation strategies implemented by human translators? Is it relevant for comparing machine translation and bio-translation?
• Does readers’ reception of literary texts differ depending on the modalities of their translation?

Deadline for abstracts: 6 June 2022

For more information, click here