Quantifying Intimacy in Language

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 5307–5326, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 5307 Quantifying Intimacy in Language Jiaxin Pei School of Information University of Michigan pedropei@umich.edu David Jurgens School of Information University of Michigan jurgens@umich.edu Abstract Intimacy is a fundamental aspect of how we relate to others in social settings. Language encodes the social information of intimacy through both topics and other more subtle cues (such as linguistic hedging and swear- ing). Here, we introduce a new computa- tional framework for studying expressions of the intimacy in language with an accompany- ing dataset and deep learning model for ac- curately predicting the intimacy level of ques- tions (Pearson’s r=0.87). Through analyzing a dataset of 80.5M questions across social me- dia, books, and films, we show that individu- als employ interpersonal pragmatic moves in their language to align their intimacy with so- cial settings. Then, in three studies, we further demonstrate how individuals modulate their in- timacy to match social norms around gender, social distance, and audience, each validating key findings from studies in social psychology. Our work demonstrates that intimacy is a per- vasive and impactful social dimension of lan- guage. 1 Introduction Intimacy is a vital ingredient in the hierarchy of hu- man needs (Maslow, 1981; Erikson, 1993; Sullivan, 2013), playing key roles in development (Harlow and Zimmermann, 1959) and well-being (Sneed et al., 2012). Language provides multiple means of conveying intimacy in a social context as individ- uals make decisions on the topic of conversation, phrasing, and markers relating the speaker to the world. Much like how social status and power are constructed and expressed, individuals negotiate in- timacy in language to fulfill fundamental and strate- gic needs, while respecting social norms about the appropriate intimacy (Chaikin and Derlega, 1974; Korobov and Thorne, 2006). In this paper, we aim to quantify the intimacy expressed in language and demonstrate how this intimacy is constructed and employed across diverse social settings. While sociolinguistics and social psychology have long pointed to how people shape their lan- guage to convey social information (Labov, 1972; Brown and Levinson, 1978; Clark and Schunk, 1980; Weber, 2008; Locher and Graham, 2010), only recently, have computational models focused on making this information explicit (Choi et al., 2012; Danescu-Niculescu-Mizil et al., 2013; Bak et al., 2014). In particular, works on social status and power have shown how individuals use lexical cues and linguistic strategies like accommodation to express their perceived status in relation to oth- ers (Danescu-Niculescu-Mizil et al., 2013; Prab- hakaran et al., 2014). Much like status in society, intimacy is a natural concept describing how an in- dividual relates to their audience in their perceived interdependence, warmth, and willingness to per- sonally share (Perlman and Fehr, 1987). Our work provides the first model of intimacy in language and tests its implications. In this paper, we examine the intimacy of ques- tions. As requests for information, questions pro- vide a natural mechanism for studying how people …

Content truncated to 500 words