Damien Hansen

Thesis summary:

Although machine translation has traditionally been perceived in the literary domain as a sort of avant-garde technology on a par with Oulipian techniques, the recent arrival of neural translation tools seems to have given rise to an entirely new and deeply polarized debate. In order to take a critical and objective look on the matter, we therefore attempt to build a system specifically trained on literature. While our results show that it is indeed possible to train such a system for this particular domain, despite the still evident limitations of these models, our experience on prose fiction also demonstrates that they can be tailored to the individual style of a translator. Our conclusions obviously raise newfound questions regarding the ergonomic aspects of translation technology, the ethical concerns that it raises or the societal issues that have to be addressed. But, more importantly, they indicate that it is possible to reimagine such tools and rethink their use in a way that better serves the human-machine interaction, the creative process and the quality of translated texts.