Abstract
Large language models (LLMs) are increasingly used for translation,
yet their ability to convey pragmatic meaning remains underexplored. A qualitative
analysis grounded in speech act theory and politeness theory was applied to refusals,
requests and compliment responses from high-context languages (Uzbek, Japanese)
rendered into English. Translations tend to preserve propositional content but flatten
indirectness, address forms and face-saving strategies, drifting toward low-context
Anglo-American directness. These patterns are interpreted as pragmatic transfer and
cultural bias rooted in training-data imbalance; pragmatics-aware evaluation and
prompt design are proposed.
References

This work is licensed under a Creative Commons Attribution 4.0 International License.
