RESPONSIBLE WOMEN AND ANALYTICAL MEN Developing Swedish Gendered Lexica for Detection of Gender Bias in Job Advertisements

Abstract

In this research, we examine gender bias in Swedish job advertisements with the use of gendered lexica. Gendered lexica can be employed to ascertain whether job advertisements are written with words that are associated with more masculine or feminine traits. The main purpose of this study is to investigate translating English gendered lexica using different machine translation methods: Google Translate, frequency-based and word-embedding-based. In the absence of a gold standard, we evaluated the translations by conducting quantitative and qualitative experiments. The embedding-based translation was evaluated as the most consistent method for the development of gendered lexica. Further testing of the embedding-based lexicon showed that Swedish job advertisements seem to be written with more feminine coded words, regardless of the gender of the majority of the workers in the advertised occupation. Advertisements for technical universities, specifically, tend to be written with more masculine coded words, while advertisements for universities that offer a wider range of education contain more feminine coded words.