Gender bias in Dutch and Turkish word embeddings

Abstract

In this thesis, research is done on the bias of gender in natural language processing (NLP). This is done speci cally on Dutch and Turkish instead of English to show if the bias is extended to other languages. The original paper by Bolukbasi et al. [6] used a corpus of news articles. In this thesis, Wikipedia is used as a corpus. The contribution of this research is on nding e ects in languages other than English, which is limited in the literature. The languages of choice are Dutch, a language that is close to English, and Turkish, a language from a di erent language family. The contribution of this research is to show that gender bias exists in NLP, independent of the language. It does a ect the size and direction of the bias.