Title: The laws of "LOL": Computational approaches to sociolinguistic variation in online discussions

Ian Stewart

PhD Candidate in Human-Centered Computing

School of Interactive Computing

Georgia Institute of Technology

-----------------------------------------

Date: Monday August 10, 2020

Time: 4:00 - 7:00 PM EST

Location (virtual): https://primetime.bluejeans.com/a2m/live-event/rbjdtbqr

Committee:

-----------------------------------------

Jacob Eisenstein (advisor), School of Interactive Computing, Georgia Institute of Technology

Diyi Yang (advisor), School of Interactive Computing, Georgia Institute of Technology

Munmun De Choudhury, School of Interactive Computing, Georgia Institute of Technology

Mark Riedl, School of Interactive Computing, Georgia Institute of Technology

David Jurgens, School of Information Science, University of Michigan

Tim Baldwin, School of Computing and Information Systems, University of Melbourne

Abstract:

-----------------------------------------

When speaking or writing, a person often chooses one form of language over another based on social constraints, including expectations in a conversation, participation in a global change, or expression of underlying attitudes. Sociolinguistic variation (e.g. choosing <going> versus <goin'>) can reveal consistent social differences among groups such as dialects and consistent social motivations in discussions. While traditional sociolinguistics investigates variation in spoken communication, computational sociolinguistics explores variation in online communication on social media. The structured nature of online discussions and the diversity of language patterns allow computational sociolinguists to test highly specific hypotheses about communication, such as the effect of different configurations of listener "audience." Understanding the communication decisions of people on discussion platforms helps answer long-standing sociolinguistic questions that are otherwise hard to tackle, and helps social media platforms anticipate their members' sometimes complicated patterns of participation in conversations.

To that end, this thesis explores open questions in sociolinguistic research by leveraging insight from online discussions. I use a combination of natural language processing and statistical analysis to investigate sociolinguistic patterns in social media. I leverage the "birds-eye" view of social media to focus on three major questions in sociolinguistics research relating to authors' participation in online discussions. First, I test the role of conversation expectations in the context of content bans and crisis events, and I show that authors vary their language to adjust to audience expectations in line with community standards and shared knowledge. Next, I investigate language change in online discussions and show that language structure, more than social context, explains word adoption. Lastly, I investigate the expression of social attitudes among multilingual speakers, and I find that such attitudes can explain language choice when the attitudes have a clear social meaning based on the discussion context. This thesis demonstrates the rich opportunities that social media provides to addressing sociolinguistic questions and provides insight into how people adapt to the communication affordances in online platforms.