LLMs & Misgendering - Data

This study will investigate the extent to which Large Language Models (LLM) are able to process non-binary language across English, French and German. This will be examined with two primary intentions, the first of which is to distinguish various strengths and weaknesses of different LLMs within different languages, specifically, which models are more or less likely to misgender non-binary referents, and in which linguistic contexts. Secondly, selected model outputs will be studied as instances of non-human errors, in order to explore how the (in)ability of LLMs to comprehend and generate non-binary language contributes to the broader discussion of whether these models can truly understand language in a human-like way.