ИСТИНА |
Войти в систему Регистрация |
|
ИПМех РАН |
||
The talk addresses the issue of bias in data used for language modelling. Gender bias is a widely recognized phenomenon, which manifests itself in different forms of preferring one gender over the other. This study investigates how gender stereotypes interact with acceptability. We take scenarios that are stereotypically attributed to males or females in previous studies and test whether the change of protagonist’s gender to the opposite value affects the contrast in acceptability between target conditions.