Large language models (LLMs) 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 the seminar, I will present our recent work where we thoroughly investigate the discourse phenomena performance of LLMs for 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 approaches through comprehensive analysis. Furthermore, we illustrate that QAD enhances the semantic richness of translations and aligns them more closely with human preferences.
