Probabilistic Bias Mitigation in Word Embeddings
This addresses bias mitigation in NLP for fairness, but it appears incremental as it builds on prior methods to improve robustness.
The paper tackles the problem of bias in word embeddings by proposing a probabilistic framework and a novel method that effectively reduces bias according to three measures while maintaining embedding quality on semantic tasks.
It has been shown that word embeddings derived from large corpora tend to incorporate biases present in their training data. Various methods for mitigating these biases have been proposed, but recent work has demonstrated that these methods hide but fail to truly remove the biases, which can still be observed in word nearest-neighbor statistics. In this work we propose a probabilistic view of word embedding bias. We leverage this framework to present a novel method for mitigating bias which relies on probabilistic observations to yield a more robust bias mitigation algorithm. We demonstrate that this method effectively reduces bias according to three separate measures of bias while maintaining embedding quality across various popular benchmark semantic tasks