Exploring the Impact of Training Data Distribution and Subword Tokenization on Gender Bias in Machine Translation

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F23%3A10475930" target="_blank" >RIV/00216208:11320/23:10475930 - isvavai.cz</a>

Výsledek na webu

<a href="http://www.afnlp.org/conferences/ijcnlp2023/proceedings/main-long/cdrom/pdf/2023.ijcnlp-long.57.pdf" target="_blank" >http://www.afnlp.org/conferences/ijcnlp2023/proceedings/main-long/cdrom/pdf/2023.ijcnlp-long.57.pdf</a>

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Exploring the Impact of Training Data Distribution and Subword Tokenization on Gender Bias in Machine Translation

Popis výsledku v původním jazyce

We study the effect of tokenization on gender bias in machine translation, an aspect that has been largely overlooked in previous works. Specifically, we focus on the interactions between the frequency of gendered profession names in training data, their representation in the subword tokenizer&apos;s vocabulary, and gender bias. We observe that female and non-stereotypical gender inflections of profession names (e.g., Spanish &quot;doctora&quot; for &quot;female doctor&quot;) tend to be split into multiple subword tokens. Our results indicate that the imbalance of gender forms in the model&apos;s training corpus is a major factor contributing to gender bias and has a greater impact than subword splitting. We show that analyzing subword splits provides good estimates of gender-form imbalance in the training data and can be used even when the corpus is not publicly available. We also demonstrate that fine-tuning just the token embedding layer can decrease the gap in gender prediction accuracy between female and male forms without impairing the translation quality.

Název v anglickém jazyce

Exploring the Impact of Training Data Distribution and Subword Tokenization on Gender Bias in Machine Translation

Popis výsledku anglicky

We study the effect of tokenization on gender bias in machine translation, an aspect that has been largely overlooked in previous works. Specifically, we focus on the interactions between the frequency of gendered profession names in training data, their representation in the subword tokenizer&apos;s vocabulary, and gender bias. We observe that female and non-stereotypical gender inflections of profession names (e.g., Spanish &quot;doctora&quot; for &quot;female doctor&quot;) tend to be split into multiple subword tokens. Our results indicate that the imbalance of gender forms in the model&apos;s training corpus is a major factor contributing to gender bias and has a greater impact than subword splitting. We show that analyzing subword splits provides good estimates of gender-form imbalance in the training data and can be used even when the corpus is not publicly available. We also demonstrate that fine-tuning just the token embedding layer can decrease the gap in gender prediction accuracy between female and male forms without impairing the translation quality.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

<a href="/cs/project/GA23-06912S" target="_blank" >GA23-06912S: Identifikace a prevence nechtěné genderové zaujatosti v neuronových jazykových modelech</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2023

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)

ISBN

979-8-89176-014-1

ISSN

—

e-ISSN

—

Počet stran výsledku

12

Strana od-do

885-896

Název nakladatele

Association for Computational Linguistics

Místo vydání

Stroudsburg, PA, USA

Místo konání akce

Nusa Dua, Bali, Indonesia

Datum konání akce

1. 11. 2023

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—