Learning to Live with Machine Translation

Abstract:Rapid advancements in technologies of text and image generation have increasingly put the perceived autonomy of human creativity under threat. Even before ChatGPT and other large-language models sent such anxieties into overdrive, literary critics were arguing for a hermeneutics of automatic writing and revisiting long-held assumptions about artistic originality. Few, however, gave much thought to these model's quirky cousins—a family branch that once ruled over the utopian dreams invested in AI: machine translation (MT). This essay reflects on why translation has been lost in all the recent talk about these models and offers a necessary corrective. It considers what a critical response to MT might look like when reframed around an understanding of current technologies and a vision of MT as potential collaborator rather than human replacement. First, it offers an overview of current neural-based MT and the theories of translation that underwrite it. It then uses literary texts as a limit case for surveying the technology's most visible gaps, providing a deep, qualitative analysis of Japanese literary texts machine translated into English. Finally, it takes a speculative turn and considers what "good enough" machine translation of a large corpus of world literature might be good for in a future of ubiquitous and ever more accessible MT. The results hint at more immediate ways that MT invites inquiry into the present conditions of world literature, but also to a future where the entanglement of human translation and agency with the material agency of the technology bring forth potentials in both.