DeepAI AI Chat
Log In Sign Up

Unmasking the Mask – Evaluating Social Biases in Masked Language Models

by   Masahiro Kaneko, et al.

Masked Language Models (MLMs) have shown superior performances in numerous downstream NLP tasks when used as text encoders. Unfortunately, MLMs also demonstrate significantly worrying levels of social biases. We show that the previously proposed evaluation metrics for quantifying the social biases in MLMs are problematic due to following reasons: (1) prediction accuracy of the masked tokens itself tend to be low in some MLMs, which raises questions regarding the reliability of the evaluation metrics that use the (pseudo) likelihood of the predicted tokens, and (2) the correlation between the prediction accuracy of the mask and the performance in downstream NLP tasks is not taken into consideration, and (3) high frequency words in the training data are masked more often, introducing noise due to this selection bias in the test cases. To overcome the above-mentioned disfluencies, we propose All Unmasked Likelihood (AUL), a bias evaluation measure that predicts all tokens in a test case given the MLM embedding of the unmasked input. We find that AUL accurately detects different types of biases in MLMs. We also propose AUL with attention weights (AULA) to evaluate tokens based on their importance in a sentence. However, unlike AUL and AULA, previously proposed bias evaluation measures for MLMs systematically overestimate the measured biases, and are heavily influenced by the unmasked tokens in the context.


page 1

page 2

page 3

page 4


Social Biases in Automatic Evaluation Metrics for NLG

Many studies have revealed that word embeddings, language models, and mo...

Are Some Words Worth More than Others?

Current evaluation metrics for language modeling and generation rely hea...

Debiasing isn't enough! – On the Effectiveness of Debiasing MLMs and their Social Biases in Downstream Tasks

We study the relationship between task-agnostic intrinsic and task-speci...

Evaluating Metrics for Bias in Word Embeddings

Over the last years, word and sentence embeddings have established as te...

This Prompt is Measuring <MASK>: Evaluating Bias Evaluation in Language Models

Bias research in NLP seeks to analyse models for social biases, thus hel...

Improving Tokenisation by Alternative Treatment of Spaces

Tokenisation is the first step in almost all NLP tasks, and state-of-the...

Code Repositories