Multilingual models invert sentiment polarity 28.7% of the time on Bengali text and show asymmetric affective weighting plus a 57% rise in error on formal dialect compared with colloquial Bengali.
InPro- ceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 2238–
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Cross-Lingual Sentiment Misalignment: Auditing Multilingual Language Models for Inversion Risk, Dialectal Representation, and Affective Stability
Multilingual models invert sentiment polarity 28.7% of the time on Bengali text and show asymmetric affective weighting plus a 57% rise in error on formal dialect compared with colloquial Bengali.