Javanese part-of-speech tagging using cross-lingual transfer learning

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F25%3AG87LC4VZ" target="_blank" >RIV/00216208:11320/25:G87LC4VZ - isvavai.cz</a>

Výsledek na webu

<a href="https://www.scopus.com/inward/record.uri?eid=2-s2.0-85200057183&doi=10.11591%2fijai.v13.i3.pp3498-3509&partnerID=40&md5=3bc107ded6fef1573c58cdb8f371ff2c" target="_blank" >https://www.scopus.com/inward/record.uri?eid=2-s2.0-85200057183&doi=10.11591%2fijai.v13.i3.pp3498-3509&partnerID=40&md5=3bc107ded6fef1573c58cdb8f371ff2c</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.11591/ijai.v13.i3.pp3498-3509" target="_blank" >10.11591/ijai.v13.i3.pp3498-3509</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Javanese part-of-speech tagging using cross-lingual transfer learning

Popis výsledku v původním jazyce

Large datasets that are publicly available for part-of-speech (POS) tagging do not always exist for some languages. One of those languages is Javanese, a local language in Indonesia, which is considered as a low-resource language. This research aims to examine the effectiveness of cross-lingual transfer learning for Javanese POS tagging by fine-tuning the state-of-the-art transformer-based models (such as IndoBERT, mBERT, and XLM-RoBERTa) using different kinds of source languages that have a higher resource (such as Indonesian, English, Uyghur, Latin, and Hungarian languages), and then fine-tuning it again using the Javanese language as the target language. We found that the models using cross-lingual transfer learning can increase the accuracy of the models with-out using cross-lingual transfer learning by 14.3%–15.3% over long short-time memory (LSTM)-based models, and by 0.21%–3.95% over transformer-based models. Our results show that the most accurate Javanese POS tagger model is XLM-RoBERTa that is fine-tuned in two stages (the first one using Indonesian language as the source language, and the second one using Javanese language as the target language), capable of achieving an accuracy of 87.65%. © 2024, Institute of Advanced Engineering and Science. All rights reserved.

Název v anglickém jazyce

Javanese part-of-speech tagging using cross-lingual transfer learning

Popis výsledku anglicky

Large datasets that are publicly available for part-of-speech (POS) tagging do not always exist for some languages. One of those languages is Javanese, a local language in Indonesia, which is considered as a low-resource language. This research aims to examine the effectiveness of cross-lingual transfer learning for Javanese POS tagging by fine-tuning the state-of-the-art transformer-based models (such as IndoBERT, mBERT, and XLM-RoBERTa) using different kinds of source languages that have a higher resource (such as Indonesian, English, Uyghur, Latin, and Hungarian languages), and then fine-tuning it again using the Javanese language as the target language. We found that the models using cross-lingual transfer learning can increase the accuracy of the models with-out using cross-lingual transfer learning by 14.3%–15.3% over long short-time memory (LSTM)-based models, and by 0.21%–3.95% over transformer-based models. Our results show that the most accurate Javanese POS tagger model is XLM-RoBERTa that is fine-tuned in two stages (the first one using Indonesian language as the source language, and the second one using Javanese language as the target language), capable of achieving an accuracy of 87.65%. © 2024, Institute of Advanced Engineering and Science. All rights reserved.

Druh

J_SC - Článek v periodiku v databázi SCOPUS

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

—

Rok uplatnění

2024

Kód důvěrnosti údajů

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

Název periodika

IAES International Journal of Artificial Intelligence

ISSN

2089-4872

e-ISSN

—

Svazek periodika

13

Číslo periodika v rámci svazku

3

Stát vydavatele periodika

US - Spojené státy americké

Počet stran výsledku

12

Strana od-do

3498-3509

Kód UT WoS článku

—

EID výsledku v databázi Scopus

2-s2.0-85200057183