Machine translation (MT) is one of the Ur-scenes of artificial intelligence: AI itself emerged from the attempt to solve the problem of translation. In his 1949 memorandum advocating for the development of translating machines, Warren Weaver, a mathematician and pioneer in MT, speculated about a “real but as yet undiscovered universal language,” an “open basement” shared by all linguistic edifices. The premise at the origins of MT was that language could be understood through common structures or “invariant properties,” with semantics emerging from statistical-mathematical regularities shared across tongues. From the outset, translation was construed as resting on the possibility of equivalence between languages—and by extension, between cultures—even as literary translation studies have long demonstrated that such equivalence is always partial and each time negotiated by the translator.

Today’s large language models (LLMs) seem to fulfill what was unimaginable in Weaver’s time: the “unimpeded” access to texts in any language, including minority languages with limited training data. Recently, researchers have argued that the LLMs' vectors representing the relation of between words become associated across languages through “language-agnostic representations” (Tiyajamorn et al. 2021), a move that recalls Weaver’s idea of universal concepts underpinning linguistic variation. While AI's statistical universalism erases the normative construction of datasets and models, the dominance of English, and the ideological power wielded by those who design AI systems, it also marginalizes the work of the translator as a mediator and negotiator of the relation between languages and between cultures.

While the proposal foregrounds the contemporary moment of LLMs, the panel situates this discussion within longer histories of machine translation, comparative approaches across political and cultural contexts, and the broader question of how translation mediates between language, cultures, and knowledge. The panel invites diverse perspectives from literary studies, translation theory, philosophy, and media theory. The aim is to foster a wide-ranging conversation about the stakes of translation in the age of AI.

Possible topics include:

• The critical potential of literary translation in response to the new ontologies of AI
• The ethics and politics of translation in the age of LLMs
• The critical discussion of LLMs' ontologies
• (Comparative) histories of machine translation in the Western and Soviet blocs
• The future of translation studies in the context of the growing use of LLMs