arXiv:2605.25686v1 Announce Type: new Abstract: The recent shift from dedicated NMT systems to general-purpose LLMs has reshaped machine translation, with LLMs reported to produce more fluent, less literal output than their predecessors. We test whether this shift extends to the deliteralization hypothesis, the long-standing claim from translation studies that translations become progressively less literal as they are drafted and revised. Using the WMT24++ dataset, we compare the literality of human translations and post-editions to that of two NMT systems and six LLMs across 54 language pairs
The proliferation of general-purpose LLMs has recently enabled a comparison of their translation characteristics against traditional NMT systems and human translators.
Understanding the 'literalness' of LLM translations impacts the nuanced interpretation of cross-lingual communication and the future of translation workflows.
LLMs are producing more fluent and less literal translations, suggesting a potential shift in how we evaluate and integrate machine translation outputs.
- · AI-powered translation services
- · Global businesses using LLM translation
- · LLM developers
- · Traditional NMT system providers
- · Translators specializing in literal fidelity
Increased adoption of LLMs for complex or nuanced translation tasks where fluency is prioritized.
Demand for new evaluation metrics and post-editing workflows tailored to less literal LLM outputs.
Potential for LLMs to reshape cross-cultural communication by standardizing towards a less literal, more 'human-like' translation style.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL