Gender bias in legal corpora and debiasing it

Abstract Word embeddings have become important building blocks that are used profoundly in natural language processing (NLP). Despite their several advantages, word can unintentionally accommodate some gender- and ethnicity-based biases present within the corpora they trained on. Therefore, ethical concerns been raised since extensively high-level algorithms. Studying such debiasing them recently an research endeavor. Various studies conducted to measure extent of bias capture eradicate them. Concurrently, as another subfield has started gain traction recently, applications NLP field law increase develop rapidly. As a direct utmost effect on people’s lives, issues for legal domain certainly important. However, best our knowledge, not yet studied context corpora. In this article, we approach gender problem from scope text domain. embedding models composed by documents legislation different countries utilized eliminate documents. Several methods employed reveal degree observe its variations over countries. Moreover, method neutralize unwanted bias. The preservation semantic coherence debiased vector space also demonstrated using tasks. Finally, overall results implications discussed