Better Hit the Nail on the Head than Beat around the Bush: Removing Protected Attributes with a Single Projection

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 8395 - 8416 December 7-11, 2022 ©2022 Association for Computational Linguistics Better Hit the Nail on the Head than Beat around the Bush: Removing Protected Attributes with a Single Projection Pantea Haghighatkhah♢ Antske Fokkens♣♢ Pia Sommerauer♣ Bettina Speckmann♢ Kevin Verbeek♢ ♢TU Eindhoven, Department of Mathematics and Computer Science ♣ Vrije Universiteit Amsterdam, Computational Linguistics Text Mining Lab Abstract Bias elimination and recent probing studies attempt to remove specific information from embedding spaces. Here it is important to remove as much of the target information as possible, while preserving any other informa- tion present. INLP is a popular recent method which removes specific information through iterative nullspace projections. Multiple itera- tions, however, increase the risk that informa- tion other than the target is negatively affected. We introduce two methods that find a single tar- geted projection: Mean Projection (MP, more efficient) and Tukey Median Projection (TMP, with theoretical guarantees). Our comparison between MP and INLP shows that (1) one MP projection removes linear separability based on the target and (2) MP has less impact on the overall space. Further analysis shows that ap- plying random projections after MP leads to the same overall effects on the embedding space as the multiple projections of INLP. Applying one targeted (MP) projection hence is methodolog- ically cleaner than applying multiple (INLP) projections that introduce random effects. 1 Introduction Word embedding spaces can contain rich informa- tion that is valuable for a wide range of NLP tasks. High quality word embeddings should capture the semantic attributes associated with a word’s mean- ing. Though we can establish that points repre- senting words with similar semantic attributes tend to be close to each other, it is challenging to rea- son which attributes are captured to what extent and, in particular, which dimensions capture these attributes (Sommerauer and Fokkens, 2018). But even though we do not (yet) know exactly which attributes are represented in which way, we nev- ertheless may want to ensure that one particular attribute is no longer present in the embedding. For example, because this attribute is a protected at- tribute such as gender which can lead to harmful bias, or because we want to test how the attribute was affecting a system in a probing setup. There are various approaches to remove a par- ticular attribute from an embedding. Clearly it is important to remove as much of the target at- tribute as possible while preserving other informa- tion present. Arguably linear projections are one of the least intrusive methods to transform an embed- ding space. Iterative nullspace projection (Ravfo- gel et al., 2020, INLP) is a recent popular method which follows this paradigm and uses multiple lin- ear projections. INLP achieves linear guarding of the protected attribute, that is, a linear classifier can no longer separate instances that have the attribute from instances that do not. However, it remains unclear to what extent INLP affects the remainder of the embedding …

Content truncated to 500 words