Research project
Metaphor in Machine Translation: Reactions, Responses, Repercussions
This project investigates which metaphors are problematic in machine translation, when, why, and for whom.
- Duration
- 2025 - 2029
- Contact
- Lettie Dorst
- Funding
- NWO Vidi
One defining characteristic of human communication that remains problematic for Artificial Intelligence and Natural Language Processing is metaphors. Metaphors are pervasive in all forms of language use, from newspapers announcing ‘waves’ of migrants and ‘soaring’ prices to casual conversations in which we say we feel ‘down’ or ‘stuck’. They are notoriously hard to translate, even for humans, because of their linguistic and cultural embeddedness. And we do not currently know what happens to their forms, functions and effects when they are translated by machines lacking in world knowledge and cultural sensitivity. Yet people increasingly rely on machine translation websites and apps like Google Translate and SayHi for all kinds of written and spoken discourse, from tourists translating a menu or talking to the hotel receptionist to large multinationals and NGOs incorporating the technology into their standard workflow.
While research shows considerable improvements in the quality of machine translation for many language combinations and text types, there is a growing concern that most machine-translated content is no longer checked or corrected by human beings. Even the best systems make mistakes and we do not know which errors lead to critical misunderstandings. The current project therefore undertakes the first comprehensive study of how machine translation may be reshaping our metaphors.
The Machine-Translated Metaphor (MTM) project answers the question: Which metaphors are problematic in machine translation, when, why, and for whom? To understand their broader scientific and societal relevance, the project studies how machine-translated metaphors work simultaneously as embedded in language, cognition and culture. The team will pioneer an interdisciplinary mixed-methods approach to investigate the influence of machine translation on the forms, functions and effects of metaphors from the perspectives of the Machine, the Professional User, and the Non-expert Reader. By aligning these perspectives, the project offers the first empirical, systematic and comprehensive investigation of both the reactions and responses to, as well as the repercussions of machine-translated metaphor.
The MTM-project asks the following questions:
- How do machine translation architectures (Transformer-Based Neural Machine Translation and Large Language Models) translate different types of metaphor (e.g. creative, extended) in journalistic and literary texts?
- How do professional users (translators and journalists) react and respond to machine-translated metaphors in literary and journalistic texts and what are the implications for professional practice?
- How do non-expert readers react and respond to machine-translated metaphors in literary and journalistic texts and what are the implications for our understanding and appreciation of such texts?
- Is the use of machine translation reshaping our metaphors and does this have repercussions for language use, human-machine relations, and society in general? That is: is machine translation changing the forms and functions of our metaphors (language)?; can and should we change the way machines translate metaphors (human-machine relations)?; and do machine-translated metaphors affect our cultural and ideological values (society)?
The MTM-project synthesizes the answers to these questions to provide new and urgently needed insights into how machines may be reshaping our metaphors in a way that is both fascinatingly complex and deceivingly inconspicuous.