Understanding Gender-coded Wording in Job Postings with Word-vectors and BERT

img
Biased gender-coded words still exist in job advertisements today. These words can strongly influence candidates' perception of the job, discourage diverse candidates from applying and even reduce their sense of belonging to the occupation. Gaucher et al (2011)\cite{gaucher} provides two lists of known gender-coded words in job advertisements. One for masculine and one for feminine coded words. However, these lists are likely incomplete and may miss more subtle or sentence-level gender coding. In this paper I propose that by using word vectors and BERT, we can discover additional gender-coded words and detect gender bias at the grammatical/sentence level.