Abstract: Word embeddings (WEs) often reflect biases present in their training data, and various bias mitigation and evaluation techniques have been proposed to address this. Existing benchmarks for ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results