Word embeddings, which create a numerical representation of text data, are an important part of many information retrieval systems. These methods find application in tasks such as search, classification, and sentiment analysis. However, a number of studies have observed the presence of bias in embedding methods: famously, the word woman is closely associated with nurse, while man is associated with doctor. In this section, we discuss techniques to debias word embeddings, including both hard-debiasing, which removes the part of the vector representation that lies along a particular (e.g., gender) subspace, and soft-debiasing, which reduces the effect of a particular subspace.