Itzulpen Automatikoko Sistemen Analisia: Genero Alborapenaren Kasua

Abstract

The aim of this work is to analyze the gender bias in three automatic translation systems widely used by the Basque community. As these commercial tools have become more popular, it is essential to see what role they play in the recreation of stereotypes defined in society. In this analysis, the gender bias found in different professions has been studied using two languages that treat gender differently: basque (which has a neutral gender mark) and spanish (which uses marks that differentiate gender). The impact of the contextual information in the original sentence and the effect of the translation directions have been analyzed. These experiments have shown how using contextual information for making gender explicit is not enough for resolving gender bias. Highlighting these biases is seen as a priority towards higher quality translations and a fairer scenario.