Collecting a Large-Scale Gender Bias Dataset for Coreference Resolution and Machine Translation
This addresses gender bias in NLP systems, which is harmful in commercial deployments, but is incremental as it builds on prior work with synthetic datasets.
The authors tackled the problem of gender bias in coreference resolution and machine translation models by creating a large-scale dataset of 108K real-world English sentences, finding that all tested models over-relied on gender stereotypes and showing that finetuning with their dataset mitigates bias.
Recent works have found evidence of gender bias in models of machine translation and coreference resolution using mostly synthetic diagnostic datasets. While these quantify bias in a controlled experiment, they often do so on a small scale and consist mostly of artificial, out-of-distribution sentences. In this work, we find grammatical patterns indicating stereotypical and non-stereotypical gender-role assignments (e.g., female nurses versus male dancers) in corpora from three domains, resulting in a first large-scale gender bias dataset of 108K diverse real-world English sentences. We manually verify the quality of our corpus and use it to evaluate gender bias in various coreference resolution and machine translation models. We find that all tested models tend to over-rely on gender stereotypes when presented with natural inputs, which may be especially harmful when deployed in commercial systems. Finally, we show that our dataset lends itself to finetuning a coreference resolution model, finding it mitigates bias on a held out set. Our dataset and models are publicly available at www.github.com/SLAB-NLP/BUG. We hope they will spur future research into gender bias evaluation mitigation techniques in realistic settings.