Fewer Errors, but More Stereotypes? The Effect of Model Size on Gender Bias

Abstract

The size of pretrained models is increasing, and so is their performance on a variety of NLP tasks. However, as their memorization capacity grows, they might pick up more social biases. In this work, we examine the connection between model size and its gender bias (specifically, occupational gender bias). We measure bias in three masked language model families (RoBERTa, DeBERTa, and T5) in two setups: directly using prompt based method, and using a downstream task (Winogender). We find on the one hand that larger models receive higher bias scores on the former task, but when evaluated on the latter, they make fewer gender errors. To examine these potentially conflicting results, we carefully investigate the behavior of the different models on Winogender. We find that while larger models outperform smaller ones, the probability that their mistakes are caused by gender bias is higher. Moreover, we find that the proportion of stereotypical errors compared to anti-stereotypical ones grows with the model size. Our findings highlight the potential risks that can arise from increasing model size. 1

A Predictive Factor Analysis of Social Biases and Task-Performance in Pretrained Masked Language Models

“…However, it is interesting to observe that social biases do not necessarily increase with this extra capacity of the MLMs. In the case of gender-related biases, Tal et al (2022) showed that even if the gender bias scores measured on Winogender (Rudinger et al, 2018) are smaller for the larger MLMs, they make more stereotypical errors with respect to gender. However, whether this observation generalises to all types of social biases remains an open question.…”

JASMINE: Arabic GPT Models for Few-Shot Learning

“…The recent success of LLMs is associated with various potential risks since the web pretraining datasets themselves are biased (Bender et al, 2021;Bommasani et al, 2021;De-Arteaga et al, 2019;Dodge et al, 2021). ; Tal et al (2022) show that the risk of biases gets higher with the increase of the model size, causing biases to resurface during the downstream tasks such as NLI (Poliak et al, 2018;Sharma et al, 2021), coreference resolution (Rudinger et al, 2018;Zhao et al, 2018), and MT (Stanovsky et al, 2019). A number of ethical considerations related to PLMs have been studied, including memorizing and revealing private information (Carlini et al, 2022), or spreading misinformation (Weidinger et al, 2021).…”

Fairness in Language Models Beyond English: Gaps and Challenges

“…Finally, while the interplay and tradeoff between privacy, efficiency, and fairness in tabular data has received extensive examination (Hooker et al, 2020;Lyu et al, 2020) comparatively fewer studies have been conducted in NLP (Tal et al, 2022;Ahn et al, 2022;Hessenthaler et al, 2022).…”