How Well Do Large Language Models Understand African American Language? Causes and Implications

We focus on studying large language models (LLMs) and their ability to successfully interpret African American Language (AAL) in ways that do not distort the communicative intent of the speakers or writers. We discuss research that quantifies the gap between how well LLMs interpret AAL and how well they interpret White Mainstream English as well as research identifying the causes of this difference. This difference has an impact both on users of applications developed using LLMs and on application performance when interpreting AAL. Given this impact, researchers have begun developing approaches to mitigate these differences. We close by discussing the role of culture in speaker interactions and the difficulties in encoding culture in LLMs.