on measuring social biases in sentence encoders

• DOI: 10.48550/arxiv.1903.10561
• Publication Date: 2019-01-01
• Authors: Chandler May; Shikha Bordia; Samuel R. Bowman; Alex Wang; Rachel Rudinger
• Keywords: FOS: Computer and information sciences; Computer Science - Computers and Society; Computer Science - Computation and Language; Computers and Society (cs.CY); Computation and Language (cs.CL); 01 natural sciences; 0105 earth and related environmental sciences

NAACL 2019<br/>The Word Embedding Association Test shows that GloVe and word2vec word embeddings exhibit human-like implicit biases based on gender, race, and other social constructs (Caliskan et al., 2017). Meanwhile, research on learning reusable text representations has begun to explore sentence-level texts, with some sentence encoders seeing enthusiastic adoption. Accordingly, we extend the Word Embedding Association Test to measure bias in sentence encoders. We then test several sentence encoders, including state-of-the-art methods such as ELMo and BERT, for the social biases studied in prior work and two important biases that are difficult or impossible to test at the word level. We observe mixed results including suspicious patterns of sensitivity that suggest the test's assumptions may not hold in general. We conclude by proposing directions for future work on measuring bias in sentence encoders.<br/>