Unlocking Latent Discourse Translation in LLMs Through Quality-Aware Decoding

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Mohammed, Wafaa, Niculae, Vlad, Zerva, Chrysoula
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916996406312960
author Mohammed, Wafaa
Niculae, Vlad
Zerva, Chrysoula
author_facet Mohammed, Wafaa
Niculae, Vlad
Zerva, Chrysoula
contents Large language models (LLMs) have emerged as strong contenders in machine translation.Yet, they still struggle to adequately handle 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 context-aware 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.
format Preprint
id arxiv_https___arxiv_org_abs_2510_06866
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Unlocking Latent Discourse Translation in LLMs Through Quality-Aware Decoding
Mohammed, Wafaa
Niculae, Vlad
Zerva, Chrysoula
Computation and Language
Large language models (LLMs) have emerged as strong contenders in machine translation.Yet, they still struggle to adequately handle 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 context-aware 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.
title Unlocking Latent Discourse Translation in LLMs Through Quality-Aware Decoding
topic Computation and Language
url https://arxiv.org/abs/2510.06866