Gender bias in MT from Basque into Spanish: the case of gender stereotypical adjectives and occupations

Abstract

With the rise of artificial intelligence, the use of machine translation (MT) has become commonplace. However, concern has been voiced with regard to MT output, from the perspective of both the quality and a range of biases that are evident in texts translated using this technology. In this study, we focus on the phenomenon of gender bias, specifically in the case of translation from languages that have no explicit grammatical gender, such as Basque, to languages that do, such as Spanish.

In the study, we collected samples from three corpora created from texts drawn from different fields (literature, science, and journalism) and examined the translations proposed by an MT system (Elia) with regard to use of stereotypical masculine and feminine adjectives and occupations. Our findings suggest that when no explicit gender is given in Basque, the MT system primarily selects the masculine option in Spanish. Nevertheless, in certain occupations, we observed that the use of certain translation methods can contribute to producing less stereotypical target texts.