UnQovering Stereotyping Biases via Underspecified Questions
This work addresses the issue of stereotyping biases in AI systems for researchers and practitioners in natural language processing, though it is incremental as it builds on prior studies of biases in language embeddings.
The authors tackled the problem of stereotyping biases in question answering models by introducing UNQOVER, a framework to probe and quantify biases through underspecified questions, revealing that all tested transformer-based models exhibit notable biases, with larger models often showing higher bias.
While language embeddings have been shown to have stereotyping biases, how these biases affect downstream question answering (QA) models remains unexplored. We present UNQOVER, a general framework to probe and quantify biases through underspecified questions. We show that a naive use of model scores can lead to incorrect bias estimates due to two forms of reasoning errors: positional dependence and question independence. We design a formalism that isolates the aforementioned errors. As case studies, we use this metric to analyze four important classes of stereotypes: gender, nationality, ethnicity, and religion. We probe five transformer-based QA models trained on two QA datasets, along with their underlying language models. Our broad study reveals that (1) all these models, with and without fine-tuning, have notable stereotyping biases in these classes; (2) larger models often have higher bias; and (3) the effect of fine-tuning on bias varies strongly with the dataset and the model size.