Stepmothers are mean and academics are pretentious: What do pretrained language models learn about you?

TL;DR

This paper explores the stereotypes encoded in pretrained language models, introduces a new dataset, and examines how stereotypes and associated emotions change through fine-tuning on different text sources.

Contribution

It presents the first dataset of social stereotypes for language models and a method to analyze stereotypes and emotions in an unsupervised manner.

Findings

Models encode varying stereotypes of social groups.

Stereotypes and emotions can shift rapidly during fine-tuning.

Fine-tuning on different sources affects model attitudes and emotional associations.

Abstract

In this paper, we investigate what types of stereotypical information are captured by pretrained language models. We present the first dataset comprising stereotypical attributes of a range of social groups and propose a method to elicit stereotypes encoded by pretrained language models in an unsupervised fashion. Moreover, we link the emergent stereotypes to their manifestation as basic emotions as a means to study their emotional effects in a more generalized manner. To demonstrate how our methods can be used to analyze emotion and stereotype shifts due to linguistic experience, we use fine-tuning on news sources as a case study. Our experiments expose how attitudes towards different social groups vary across models and how quickly emotions and stereotypes can shift at the fine-tuning stage.