You Can Have Your Data and Balance It Too: Towards Balanced and Efficient Multilingual Models

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F23%3A10476179" target="_blank" >RIV/00216208:11320/23:10476179 - isvavai.cz</a>

Výsledek na webu

<a href="https://aclanthology.org/2023.sigtyp-1.1.pdf" target="_blank" >https://aclanthology.org/2023.sigtyp-1.1.pdf</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.18653/v1/2023.sigtyp-1.1" target="_blank" >10.18653/v1/2023.sigtyp-1.1</a>

Jazyk výsledku

angličtina

Název v původním jazyce

You Can Have Your Data and Balance It Too: Towards Balanced and Efficient Multilingual Models

Popis výsledku v původním jazyce

Multilingual models have been widely used for the cross-lingual transfer to low-resource languages. However, the performance on these languages is hindered by their under-representation in the pretraining data. To alleviate this problem, we propose a novel multilingual training technique based on teacher-student knowledge distillation. In this setting, we utilize monolingual teacher models optimized for their language. We use those teachers along with balanced (sub-sampled) data to distill the teachers&apos; knowledge into a single multilingual student. Our method outperforms standard training methods in low-resource languages and retains performance on high-resource languages while using the same amount of data. If applied widely, our approach can increase the representation of low-resource languages in NLP systems.

Název v anglickém jazyce

You Can Have Your Data and Balance It Too: Towards Balanced and Efficient Multilingual Models

Popis výsledku anglicky

Multilingual models have been widely used for the cross-lingual transfer to low-resource languages. However, the performance on these languages is hindered by their under-representation in the pretraining data. To alleviate this problem, we propose a novel multilingual training technique based on teacher-student knowledge distillation. In this setting, we utilize monolingual teacher models optimized for their language. We use those teachers along with balanced (sub-sampled) data to distill the teachers&apos; knowledge into a single multilingual student. Our method outperforms standard training methods in low-resource languages and retains performance on high-resource languages while using the same amount of data. If applied widely, our approach can increase the representation of low-resource languages in NLP systems.

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

—

Návaznosti

S - Specificky vyzkum na vysokych skolach

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 5th Workshop on Research in Computational Linguistic Typology and Multilingual NLP

ISBN

978-1-959429-56-2

ISSN

—

e-ISSN

—

Počet stran výsledku

11

Strana od-do

1-11

Název nakladatele

Association for Computational Linguistics

Místo vydání

Stroudsburg, PA, USA

Místo konání akce

Dubrovnik, Croatia

Datum konání akce

2. 5. 2023

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—