Using Sociolinguistic Variables to Reveal Changing Attitudes Towards Sexuality and Gender

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9918–9938 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 9918 Using Sociolinguistic Variables to Reveal Changing Attitudes Towards Sexuality and Gender Sky CH-Wang† Department of Computer Science Columbia University skywang@cs.columbia.edu David Jurgens School of Information University of Michigan jurgens@umich.edu Abstract Individuals signal aspects of their identity and beliefs through linguistic choices. Studying these choices in aggregate allows us to exam- ine large-scale attitude shifts within a popula- tion. Here, we develop computational meth- ods to study word choice within a sociolin- guistic lexical variable—alternate words used to express the same concept—in order to test for change in the United States towards sexu- ality and gender. We examine two variables: i) referents to significant others, such as the word “partner” and ii) referents to an indefi- nite person, both of which could optionally be marked with gender. The linguistic choices in each variable allow us to study increased rates of acceptances of gay marriage and gender equality, respectively. In longitudinal analyses across Twitter and Reddit over 87M messages, we demonstrate that attitudes are changing but that these changes are driven by specific demo- graphics within the United States. Further, in a quasi-causal analysis, we show that passages of Marriage Equality Acts in different states are drivers of linguistic change. 1 Introduction A person’s identity and attitudes are reflected in the language they use (Norton, 1997; Huffaker and Calvert, 2005; De Fina, 2012). In particular, the linguistic choice for a concept can reveal the in- dividual’s stance or attitudes (Jaffe et al., 2009); for example, the use of “illegals” or “undocu- mented” in reference to immigrants reveals the speaker’s attitude on immigration (Lakoff and Fer- guson, 2006). These alternations in word choice are known as lexical variables in sociolinguistics. Examining the relative frequencies of a variable’s words can reveal the underlying attitudes within a †Work performed while the author was an undergraduate research assistant at the University of Michigan. population that drive the linguistic choice. Here, we examine changes in attitude towards sexuality and gender in the United States through two lexi- cal variables. Sociolinguistics has long focused on variation in language with respect to identity and attitudes (Labov, 1963; Eckert and Rickford, 2001; Trudg- ill, 2002). Recent computational studies have built upon this line of research (Nguyen et al., 2016), showing not only that this variation occurs in so- cial media (Eisenstein et al., 2014; Hovy et al., 2015), but also that the large scale of social media enables the study of broader societal trends (Abit- bol et al., 2018; Grieve et al., 2018). Our work ex- pands this line of research by examining longitudi- nal changes in linguistic variation to show chang- ing societal attitudes. Here, we test for change in attitudes about sex- uality and gender by computationally measuring variation for two lexical variables associated with these attitudes from a massive longitudinal study of 73M Twitter posts and 14M Reddit comments across nearly ten years. The …

Content truncated to 500 words