The Untranslatable for Artificial Intelligence

Abstract

This article explores the epistemological boundaries of artificial intelligence in the field of translation, moving beyond the technical challenges of neural networks. Drawing on the philosophy of language of Ludwig Wittgenstein, particularly the concepts of “language games” and “family resemblance,” the author considers translation as a tool for conveying meaning rather than a simple exchange of lexical and grammatical units. An analysis of errors systematically reproduced by various translation programs (Google, DeepL, Yandex Translate, ChatGPT) when working with Russian-language texts has revealed key limitations of AI: the inability to distinguish subjective authorial intentions, difficulties in adapting to new contexts of existing concepts, as well as the disregard for the pragmatic level of communication and text types. The paper substantiates the thesis that, despite the ability to analyze context within the text (text-centeredness), AI remains insensitive to the intersubjective nature of language, where meaning is formed in the dialogue between the speaker and the listener. It is concluded that AI is only capable of a “paraphrase of content” but cannot convey the subjective enunciation of the author, the speaker’s intentions, or the listener’s evaluative judgments, which constitutes the space of the untranslatable for artificial intelligence.