Neural machine translation and a queer perspective on gender bias

The issue of gender in translation has garnered significant attention since the establishment of (human aided) machine translation as previously existing gender biases in natural language are now being reproduced by translation machines. Although machine translation has a long history, recent advancements, such as neural machine translation (NMT), have revolutionized the field. NMT systems rely on training algorithms and large corpora which are influenced by human choices that can be negatively biased regarding gender, race, etc. In the context of written French, écriture inclusive strategies seek to establish gender-inclusive alternatives to promote gender equality in language use. However, the debate on gender and inclusive language still predominantly focuses on binary gender representations. This study explores how Google Translate and DeepL handle écriture inclusive strategies when they are translated into German. Three main aspects will be directly addressed in this section. First, we will take a look at the common translation practices offered by the machines regarding sentences in écriture inclusive; second, we will examine the target term strategies that differ from the source language’s; and third, we will examine the instances where translations incorporate genders beyond the binary. We therefore aim to investigate how machine translation systems, specifically Google Translate and DeepL, perform in these cases. In this article, we argue that the absence of ethical frameworks for AI and data training has resulted in the reinforcement of gender biases and representational harms within machine translation systems.