Gender-neutral language detection in instructional texts

Abstract

Gender-neutral language plays a large part in the linguistic inclusion of women and binary and non-binary trans individuals. Its implementation can depend on different factors like language and guidelines from specific organizations. In this work, the gender-neutral language of instructional texts is analyzed. Five different classifiers are compared using a gold dataset, that consists of annotated revisions from wikiHow.com. The best performing classifier used a static list of gendered words to find gendered terms in the revision. Once a gendered term is found, the revision is classified as gendered. Using this classifier, the remaining 11 million revisions are classified and analyzed. The analysis suggests, that even though 3/4 articles gender-neutral, there was no concerted effort of the editors to change the ones that are gendered to be gender-neutral.