What can Artificial Intelligence (still) not handle in translation?

On the use of technology in teaching translation

Abstract

The remarkably rapid development of digital technologies—especially tools based on neural machine translation (NMT) and large language models (LLMs)—has long been a source of concern within the translation community. This concern is well-founded: discussions on both public and closed translators’ forums in social media indicate that the number of “from-scratch” translation assignments is steadily declining in favor of post-editing machine translations, which are sometimes commissioned under the label of “proofreading,” without disclosing that the first “translator” was an automatic system. In fact, only sworn translators, game localizers, and conference interpreters have not experienced drastic market changes. Consequently, more and more voices are calling for a rebranding of the profession and strongly discouraging young people from pursuing a career in translation or studying it at university. Translators emphasize that, from the client’s perspective, time and low cost are often more important than the quality of the final product. In the case of highly specialized texts, such “proofreading” frequently means working from scratch—only for half the pay. However, technology is continuously improving, and the widespread implementation of ChatGPT-5 and other LLM-based tools may significantly transform both the profession of translation and the translation services market.

Ignoring reality is hardly a wise approach; instead, artificial intelligence can be applied sensibly in the teaching process. Students eagerly use automatic translators (DeepL, Google Translate, ChatGPT), achieving fairly satisfactory results when working with popular languages and relatively general or standard texts. Of course, instructors may prohibit the use of such tools—especially in the early stages of learning, when students are just beginning to acquire basic translation competences. Nevertheless, it is worth rethinking the course structure, individual assignments, and the selection of teaching materials so as to illustrate both the advantages of digital technologies (in the broad sense) and their weaknesses (or weaker points).

In this paper, I will discuss several examples of exercises conducted with Russian-language students at the Institute of Applied Linguistics, University of Warsaw, as part of a written translation course (2nd and 3rd year of BA). In my opinion, a carefully designed sequence of exercises (raw MT without pre-editing, MT with pre-editing, and MT with post-editing) can provide students with a convincing argument—either for or against the use of digital translation tools. It is essential that the decision be made by the student themselves—and that they are able to justify it convincingly.