Gender Bias in BERT -- Measuring and Analysing Biases through Sentiment Rating in a Realistic Downstream Classification Task
This addresses the problem of harmful biases in widely used pretrained language models for real-life applications, with findings emphasizing the need for responsible usage, though it is incremental in refining bias measurement and analysis.
The paper tackled gender bias in BERT models by introducing a novel bias measure based on sentiment rating differences between female and male sample versions, and analyzed biases in a realistic IMDB movie classifier across 63 models, finding significant biases in almost all conditions.
Pretrained language models are publicly available and constantly finetuned for various real-life applications. As they become capable of grasping complex contextual information, harmful biases are likely increasingly intertwined with those models. This paper analyses gender bias in BERT models with two main contributions: First, a novel bias measure is introduced, defining biases as the difference in sentiment valuation of female and male sample versions. Second, we comprehensively analyse BERT's biases on the example of a realistic IMDB movie classifier. By systematically varying elements of the training pipeline, we can conclude regarding their impact on the final model bias. Seven different public BERT models in nine training conditions, i.e. 63 models in total, are compared. Almost all conditions yield significant gender biases. Results indicate that reflected biases stem from public BERT models rather than task-specific data, emphasising the weight of responsible usage.