Published April 10, 2024 | Version v1
Conference paper Open

MultiLexBATS: Multilingual Dataset of Lexical Semantic Relations

  • 1. ROR icon University of Vienna
  • 2. ROR icon University of Coimbra
  • 3. ROR icon Universidad de Zaragoza
  • 4. ROR icon Polytechnic University of Bucharest
  • 5. ROR icon University of Prishtina
  • 6. ROR icon University of Aveiro
  • 7. ROR icon Universidade Nova de Lisboa
  • 8. ROR icon Institute for Computational Linguistics “A. Zampolli”
  • 9. ROR icon Slovak Academy of Sciences
  • 10. ROR icon University of Ljubljana
  • 11. ROR icon Athena Research and Innovation Center In Information Communication & Knowledge Technologies
  • 12. ROR icon Jerusalem College of Technology
  • 13. University of Naples - L'Orientale
  • 14. Institute of Croatian Language and Linguistics
  • 15. Mykolas Romeris University
  • 16. Université Grenoble Alpes
  • 17. Laboratoire d'Informatique de Grenoble
  • 18. Universite Grenoble Alpes UFR Informatique Mathématiques et Mathématiques Appliquées de Grenoble
  • 19. Universidade do Porto Faculdade de Letras
  • 20. Uppsala Universitet
  • 21. Universitatea Politehnica din Bucuresti

Description

Understanding the relation between the meanings of words is an important part of comprehending natural language. Prior work has either focused on analysing lexical semantic relations in word embeddings or probing pretrained language models (PLMs), with some exceptions. Given the rarity of highly multilingual benchmarks, it is unclear to what extent PLMs capture relational knowledge and are able to transfer it across languages. To start addressing this question, we propose MultiLexBATS, a multilingual parallel dataset of lexical semantic relations adapted from BATS in 15 languages including low-resource languages, such as Bambara, Lithuanian, and Albanian. As experiment on cross-lingual transfer of relational knowledge, we test the PLMs’ ability to (1) capture analogies across languages, and (2) predict translation targets. We find considerable differences across relation types and languages with a clear preference for hypernymy and antonymy as well as romance languages.

Files

2475_Paper.pdf

Files (233.9 kB)

Name Size Download all
md5:b1ecbb671008b71b99849b3b3d4add7a
233.9 kB Preview Download

Additional details