Towards Understanding and Mitigating Social Biases in Language Models
This work addresses fairness issues in language models used in sensitive real-world applications like healthcare and legal systems, representing an incremental improvement in bias mitigation techniques.
The paper tackled the problem of social biases in large-scale pretrained language models by defining sources of representational biases, proposing new benchmarks and metrics to measure them, and developing steps to mitigate these biases during text generation, with empirical results and human evaluation showing effectiveness in reducing bias while maintaining high-fidelity text generation.
As machine learning methods are deployed in real-world settings such as healthcare, legal systems, and social science, it is crucial to recognize how they shape social biases and stereotypes in these sensitive decision-making processes. Among such real-world deployments are large-scale pretrained language models (LMs) that can be potentially dangerous in manifesting undesirable representational biases - harmful biases resulting from stereotyping that propagate negative generalizations involving gender, race, religion, and other social constructs. As a step towards improving the fairness of LMs, we carefully define several sources of representational biases before proposing new benchmarks and metrics to measure them. With these tools, we propose steps towards mitigating social biases during text generation. Our empirical results and human evaluation demonstrate effectiveness in mitigating bias while retaining crucial contextual information for high-fidelity text generation, thereby pushing forward the performance-fairness Pareto frontier.