An Analysis of Social Biases Present in BERT Variants Across Multiple Languages

TL;DR

This paper investigates social biases in monolingual BERT models across English, Greek, and Persian, analyzing gender, religious, and ethnic biases using a novel template-based measurement approach that accounts for linguistic diversity.

Contribution

It introduces a language-agnostic bias measurement method and provides a cross-linguistic analysis of biases in BERT models, highlighting cultural and linguistic differences.

Findings

Bias measurement varies significantly across languages.

Cultural and linguistic factors influence bias expression.

Higher biases in non-English models may relate to training data content.

Abstract

Although large pre-trained language models have achieved great success in many NLP tasks, it has been shown that they reflect human biases from their pre-training corpora. This bias may lead to undesirable outcomes when these models are applied in real-world settings. In this paper, we investigate the bias present in monolingual BERT models across a diverse set of languages (English, Greek, and Persian). While recent research has mostly focused on gender-related biases, we analyze religious and ethnic biases as well and propose a template-based method to measure any kind of bias, based on sentence pseudo-likelihood, that can handle morphologically complex languages with gender-based adjective declensions. We analyze each monolingual model via this method and visualize cultural similarities and differences across different dimensions of bias. Ultimately, we conclude that current methods of probing for bias are highly language-dependent, necessitating cultural insights regarding the unique ways bias is expressed in each language and culture (e.g. through coded language, synecdoche, and other similar linguistic concepts). We also hypothesize that higher measured social biases in the non-English BERT models correlate with user-generated content in their training.