AI in Identity Construction and Sociolinguistics

Project Overview

In both peer-to-peer interactions and peer-to-educator interactions, ideologies of race and sexuality present barriers in the form of language-based discriminatory practices towards marginalized and minoritized learners. Joining emergent perspectives in sociolinguistics about language as identity construction, and employing recent advances in machine learning AI, our study explicates the centrality of intersectionality in how social evaluations of individuals and their language are formed. Our findings will further understandings of discrimination in institutional settings like the classroom, where insights on perceptions of intersectional identities can inform more inclusive pedagogical practices. The current study seeks to add nuance to our understanding of Asian American identity construction as it intersects with queerness. While Asians make up a significant demographic portion of the United States, little is known about how the group uses language in identity construction. Even less is known about the language of queer individuals within this group.

Previous work in the classroom has demonstrated that social evaluations of a talker’s language function as a proxy for social evaluations of the individual. From this perspective, we hypothesize that minoritized and marginalized individuals will be more negatively evaluated regardless of the language they use. More specifically, we expect that a single audio stimulus be evaluated more negatively by listeners who are told either that the talker is a queer South Asian woman or a queer East Asian woman versus a straight white woman. The effects of these negative evaluations disempowers learners and reinforces systematic barriers to education by reproducing dominant ideologies of white hegemony and the heteropatriarchy.

The data will be analyzed using cutting edge machine learning methodologies. To our knowledge, this will be the first study in sociolinguistics which makes use of such approaches. While many social scientists have welcomed machine learning technologies, linguists have generally been slow to adopt such resources. By incorporating AI analytic methods, we aim to underscore the insights that these approaches have to offer in both interpreting sociolinguistic data and thus in improving our understanding of language and meaning-making more broadly.

We will collect and analyze perceptual data from an online survey. Each participant will be primed about a talker’s race (South Asian, East Asian, or white) and sexuality (queer or straight) before listening to their voice and rating them on various scales (ie., perceived intelligence and perceived assertiveness). Different listeners will hear the same audio recording but will be primed with different information about the talker’s sexuality and race. Listeners will be given one of two audio recorded stimuli, with only one phonetic feature differing between each recording. This will allow us to probe the link between social evaluations of the talker and the language they use.

In undertaking this study, we underscore a perspective that analyzing language with a central focus on the intersections of various social categories like gender, race, and sexuality, is imperative in developing a nuanced understanding of related ideologies. In contributing to an understanding of these ideologies as they pertain to Asian American and queer identities, we hope to make clear the roots of language-based discrimination in various settings, including the classroom.

Team