Semantic change of female-denoting nouns in diachronic German corpora

Abstract

Computational studies of language change using word embedding models have focused on lexical semantic shifts. Current work in Natural Language Processing has revealed gender bias in word embeddings trained on internet data. However, little is known about what distributional methods can reveal about possible gender bias in diachronic corpora. This paper addresses this gap by examining gender-specific nouns denoting humans, e.g. Magd ‘maidservant’, Knecht ‘farmhand’ in a historical corpus of German (1350-1899). We show that word embedding models provide evidence for gender bias and negative evaluations associated with female-denoting terms in diachronic German corpora.