Modern Bias

Multilingual models show significant sentiment misalignment, especially for Bengali

A new research paper highlights significant cross-lingual sentiment misalignment in multilingual language models, particularly affecting low-resource languages like Bengali. The study found that a compressed model archi…