A systematic pipeline for diagnosing and reducing gender-stereotype bias in Japanese PLMs for sentiment analysis
Yang Liu, Ziyang Li · 2026
Read A systematic pipeline for diagnosing and reducing gender-stereotype bias in Japanese PLMs for sentiment analysis by Yang Liu, Ziyang Li (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13272434.txt.
Description
Yang Liu and Ziyang Li present in 2026 a three-stage method to find and cut gender bias in Japanese language models used to judge sentiment. Most bias work has focused on English texts, so Japanese models have gone unchecked. Their method tests for bias first, then retrains the model, then tests again with the same tools.
The test sets set stereotype-matching lines against stereotype-breaking and neutral lines in the same sentiment task. Two new scores measure how far the model leans toward stereotypes and how its sentiment differs for male, female, and neutral wording. Retraining on gender-swapped text then aims to repair the lean, giving builders an audit they can rerun.
Catalog record
| CREATOR | Yang Liu, Ziyang Li |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |