Unlocking Latent Discourse Translation in LLMs Through Quality-Aware Decoding

ACL ARR 2025 May Submission4008 Authors

Abstract: Large language models have emerged as strong contenders in machine translation. Yet, they often fall behind specialized neural machine translation systems in addressing discourse phenomena, such as pronoun resolution and lexical cohesion at the document level. In this study, we thoroughly investigate the discourse phenomena performance of LLMs in document-level translation. We demonstrate that discourse knowledge is encoded within LLMs and propose the use of quality-aware decoding (QAD) to effectively extract this knowledge, showcasing its superiority over other decoding methods through comprehensive analysis. Furthermore, we illustrate that QAD enhances the semantic richness of translations and aligns them more closely with human preferences.

Paper Type: Long

Research Area: Machine Translation

Research Area Keywords: automatic evaluation, human evaluation

Contribution Types: NLP engineering experiment

Languages Studied: English, Portuguese, German, French, Russian, Korean, Arabic

Submission Number: 4008