Too many cooks spoil the model : are bilingual models for Slovene better than a large multilingual model?
- Author
- Pranaydeep Singh (UGent) , Aaron Maladry (UGent) and Els Lefever (UGent)
- Organization
- Abstract
- This paper investigates whether adding data of typologically closer languages improves the performance of transformer-based models for three different downstream tasks, namely Part-of-Speech tagging, Named Entity Recognition, and Sentiment Analysis, compared to a monolingual and plain multilingual language model. For the presented pilot study, we performed experiments for the use case of Slovene, a low(er)-resourced language belonging to the Slavic language family. The experiments were carried out in a controlled setting, where a monolingual model for Slovene was compared to combined language models containing Slovene, trained with the same amount of Slovene data. The experimental results show that adding typologically closer languages indeed improves the performance of the Slovene language model, and even succeeds in outperforming the large multilingual XLM-RoBERTa model for NER and PoS-tagging. We also reveal that, contrary to intuition, distantly or unrelated languages also combine admirably with Slovene, often out-performing XLM-R as well. All the bilingual models used in the experiments are publicly available at https://github.com/pranaydeeps/BLAIR
- Keywords
- lt3
Downloads
-
EACL2023.pdf
- full text (Published version)
- |
- open access
- |
- |
- 375.08 KB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-01H0Q97T5GP8RFQRD1W73FHMAH
- MLA
- Singh, Pranaydeep, et al. “Too Many Cooks Spoil the Model : Are Bilingual Models for Slovene Better than a Large Multilingual Model?” Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023, edited by Jakub Piskorski et al., Association for Computational Linguistics, 2023, pp. 32–39.
- APA
- Singh, P., Maladry, A., & Lefever, E. (2023). Too many cooks spoil the model : are bilingual models for Slovene better than a large multilingual model? In J. Piskorski, M. Marcińczuk, P. Nakov, M. Ogrodniczuk, S. Pollak, P. Přibáň, … R. Yangarber (Eds.), Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023 (pp. 32–39). Association for Computational Linguistics.
- Chicago author-date
- Singh, Pranaydeep, Aaron Maladry, and Els Lefever. 2023. “Too Many Cooks Spoil the Model : Are Bilingual Models for Slovene Better than a Large Multilingual Model?” In Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023, edited by Jakub Piskorski, Michał Marcińczuk, Preslav Nakov, Maciej Ogrodniczuk, Senja Pollak, Pavel Přibáň, Piotr Rybak, Josef Steinberger, and Roman Yangarber, 32–39. Association for Computational Linguistics.
- Chicago author-date (all authors)
- Singh, Pranaydeep, Aaron Maladry, and Els Lefever. 2023. “Too Many Cooks Spoil the Model : Are Bilingual Models for Slovene Better than a Large Multilingual Model?” In Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023, ed by. Jakub Piskorski, Michał Marcińczuk, Preslav Nakov, Maciej Ogrodniczuk, Senja Pollak, Pavel Přibáň, Piotr Rybak, Josef Steinberger, and Roman Yangarber, 32–39. Association for Computational Linguistics.
- Vancouver
- 1.Singh P, Maladry A, Lefever E. Too many cooks spoil the model : are bilingual models for Slovene better than a large multilingual model? In: Piskorski J, Marcińczuk M, Nakov P, Ogrodniczuk M, Pollak S, Přibáň P, et al., editors. Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023. Association for Computational Linguistics; 2023. p. 32–9.
- IEEE
- [1]P. Singh, A. Maladry, and E. Lefever, “Too many cooks spoil the model : are bilingual models for Slovene better than a large multilingual model?,” in Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023, Dubrovnik, Croatia, 2023, pp. 32–39.
@inproceedings{01H0Q97T5GP8RFQRD1W73FHMAH,
abstract = {{This paper investigates whether adding data of typologically closer languages improves the performance of transformer-based models for three different downstream tasks, namely Part-of-Speech tagging, Named Entity Recognition, and Sentiment Analysis, compared to a monolingual and plain multilingual language model. For the presented pilot study, we performed experiments for the use case of Slovene, a low(er)-resourced language belonging to the Slavic language family. The experiments were carried out in a controlled setting, where a monolingual model for Slovene was compared to combined language models containing Slovene, trained with the same amount of Slovene data. The experimental results show that adding typologically closer languages indeed improves the performance of the Slovene language model, and even succeeds in outperforming the large multilingual XLM-RoBERTa model for NER and PoS-tagging. We also reveal that, contrary to intuition, distantly or unrelated languages also combine admirably with Slovene, often out-performing XLM-R as well. All the bilingual models used in the experiments are publicly available at https://github.com/pranaydeeps/BLAIR}},
author = {{Singh, Pranaydeep and Maladry, Aaron and Lefever, Els}},
booktitle = {{Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023}},
editor = {{Piskorski, Jakub and Marcińczuk, Michał and Nakov, Preslav and Ogrodniczuk, Maciej and Pollak, Senja and Přibáň, Pavel and Rybak, Piotr and Steinberger, Josef and Yangarber, Roman}},
isbn = {{9781959429579}},
keywords = {{lt3}},
language = {{eng}},
location = {{Dubrovnik, Croatia}},
pages = {{32--39}},
publisher = {{Association for Computational Linguistics}},
title = {{Too many cooks spoil the model : are bilingual models for Slovene better than a large multilingual model?}},
url = {{https://aclanthology.org/2023.bsnlp-1.5}},
year = {{2023}},
}