BOLD: Dataset and Metrics for Measuring Biases in Open-Ended Language Generation
This addresses the issue of embedded prejudices in language models for users in downstream applications like conversation bots, though it is incremental as it builds on existing bias measurement efforts.
The authors tackled the problem of social biases in open-ended language generation by introducing the BOLD dataset with 23,679 prompts across five domains and new automated metrics, finding that three popular models exhibit larger biases than human-written Wikipedia text.
Recent advances in deep learning techniques have enabled machines to generate cohesive open-ended text when prompted with a sequence of words as context. While these models now empower many downstream applications from conversation bots to automatic storytelling, they have been shown to generate texts that exhibit social biases. To systematically study and benchmark social biases in open-ended language generation, we introduce the Bias in Open-Ended Language Generation Dataset (BOLD), a large-scale dataset that consists of 23,679 English text generation prompts for bias benchmarking across five domains: profession, gender, race, religion, and political ideology. We also propose new automated metrics for toxicity, psycholinguistic norms, and text gender polarity to measure social biases in open-ended text generation from multiple angles. An examination of text generated from three popular language models reveals that the majority of these models exhibit a larger social bias than human-written Wikipedia text across all domains. With these results we highlight the need to benchmark biases in open-ended language generation and caution users of language generation models on downstream tasks to be cognizant of these embedded prejudices.