Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Manna, Chiara, Mohebbi, Hosein, Alishahi, Afra, Blain, Frédéric, Vanmassenhove, Eva
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2603.17952
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917351860994048
author Manna, Chiara
Mohebbi, Hosein
Alishahi, Afra
Blain, Frédéric
Vanmassenhove, Eva
author_facet Manna, Chiara
Mohebbi, Hosein
Alishahi, Afra
Blain, Frédéric
Vanmassenhove, Eva
contents While Large Language Models achieve state-of-the-art results across a wide range of NLP tasks, they remain prone to systematic biases. Among these, gender bias is particularly salient in MT, due to systematic differences across languages in whether and how gender is marked. As a result, translation often requires disambiguating implicit source signals into explicit gender-marked forms. In this context, standard benchmarks may capture broad disparities but fail to reflect the full complexity of gender bias in modern MT. In this paper, we extend recent frameworks on bias evaluation by: (i) introducing a novel measure coined "Prior Bias", capturing a model's default gender assumptions, and (ii) applying the framework to decoder-only MT models. Our results show that, despite their scale and state-of-the-art status, decoder-only models do not generally outperform encoder-decoder architectures on gender-specific metrics; however, post-training (e.g., instruction tuning) not only improves contextual awareness but also reduces the masculine Prior Bias.
format Preprint
id arxiv_https___arxiv_org_abs_2603_17952
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Gender Disambiguation in Machine Translation: Diagnostic Evaluation in Decoder-Only Architectures
Manna, Chiara
Mohebbi, Hosein
Alishahi, Afra
Blain, Frédéric
Vanmassenhove, Eva
Computation and Language
While Large Language Models achieve state-of-the-art results across a wide range of NLP tasks, they remain prone to systematic biases. Among these, gender bias is particularly salient in MT, due to systematic differences across languages in whether and how gender is marked. As a result, translation often requires disambiguating implicit source signals into explicit gender-marked forms. In this context, standard benchmarks may capture broad disparities but fail to reflect the full complexity of gender bias in modern MT. In this paper, we extend recent frameworks on bias evaluation by: (i) introducing a novel measure coined "Prior Bias", capturing a model's default gender assumptions, and (ii) applying the framework to decoder-only MT models. Our results show that, despite their scale and state-of-the-art status, decoder-only models do not generally outperform encoder-decoder architectures on gender-specific metrics; however, post-training (e.g., instruction tuning) not only improves contextual awareness but also reduces the masculine Prior Bias.
title Gender Disambiguation in Machine Translation: Diagnostic Evaluation in Decoder-Only Architectures
topic Computation and Language
url https://arxiv.org/abs/2603.17952