Evaluating the cultural alignment of multilingual LLMs in typical Japanese workplace scenarios
Zhiwei Gao, Nobuyuki Shimizu, Sumio Fujita, Shaowen Peng, Shoko Wakamiya, Eiji Aramaki · 2026
Read Evaluating the cultural alignment of multilingual LLMs in typical Japanese workplace scenarios by Zhiwei Gao, Nobuyuki Shimizu, Sumio Fujita, Shaowen Peng, Shoko Wakamiya, Eiji Aramaki (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/plos-pone-0338524.txt.
Description
Gao and colleagues, in a 2026 study, test how well large language models fit the culture of Japanese work life. They look at five models, including one built in Japan, and ask native Japanese raters to judge replies to common work scenes. Their scoring draws on a well known frame of six cultural traits.
Native raters gave the top foreign models scores as high as, or higher than, the Japanese model. A closer look split skill into three layers: polite word forms, social values, and fitting action in context. On that split, the Japanese model leaned hard on surface politeness yet lagged on values and action. The authors argue that judging culture needs more than one score, since a single mark can hide weak spots in how a model acts.
Catalog record
| CREATOR | Zhiwei Gao, Nobuyuki Shimizu, Sumio Fujita, Shaowen Peng, Shoko Wakamiya, Eiji Aramaki |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42560 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |