The paradox of multilingual emotion detection

Publication type
C1
Publication status
Published
Author
De Bruyne, L.
Editor
Jeremy Barnes, Orphée De Clercq and Roman Klinger
Series
Proceedings of the 13th Workshop on Computational Approaches to Subjectivity, Sentiment, and Social Media Analysis
Pagination
458-466
Publisher
Association for Computational Linguistics (ACL) (Toronto, Canada)
Conference
13th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis, collocated with ACL 2023 (WASSA 2023) (Toronto, Canada)
Download
(.pdf)
View in Biblio
(externe link)

Abstract

The dominance of English is a well-known issue in NLP research. In this position paper, I turn to state-of-the-art psychological insights to explain why this problem is especially persistent in research on automatic emotion detection, and why the seemingly promising approach of using multilingual models to include lower-resourced languages might not be the desired solution. Instead, I campaign for the use of models that acknowledge linguistic and cultural differences in emotion conceptualization and verbalization. Moreover, I see much potential in NLP to better understand emotions and emotional language use across different languages.