Extracting Social Power Relationships from Natural Language

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 773–782, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Extracting Social Power Relationships from Natural Language Philip Bramsen Louisville, KY bramsen@alum.mit.edu* Ami Patel Massachusetts Institute of Technology Cambridge, MA ampatel@mit.edu* Martha Escobar-Molano San Diego, CA mescobar@asgard.com* Rafael Alonso SET Corporation Arlington, VA ralonso@setcorp.com Abstract Sociolinguists have long argued that social context influences language use in all manner of ways, resulting in lects 1. This paper ex- plores a text classification problem we will call lect modeling, an example of what has been termed computational sociolinguistics. In particular, we use machine learning techniques to identify social power relationships between members of a social network, based purely on the content of their interpersonal communica- tion. We rely on statistical methods, as op- posed to language-specific engineering, to extract features which represent vocabulary and grammar usage indicative of social power lect. We then apply support vector machines to model the social power lects representing su- perior-subordinate communication in the En- ron email corpus. Our results validate the treatment of lect modeling as a text classifica- tion problem – albeit a hard one – and consti- tute a case for future research in computational sociolinguistics. 1 Introduction Linguists in sociolinguistics, pragmatics and re- lated fields have analyzed the influence of social context on language and have catalogued countless phenomena that are influenced by it, confirming many with qualitative and quantitative studies. In- * This work was done while these authors were at SET Corpo- ration, an SAIC Company. 1 Fields that deal with society and language have inconsistent terminology; “lect” is chosen here because “lect” has no other English definitions and the etymology of the word gives it the sense we consider most relevant. deed, social context and function influence lan- guage at every level – morphologically, lexically, syntactically, and semantically, through discourse structure, and through higher-level abstractions such as pragmatics. Considered together, the extent to which speak- ers modify their language for a social context amounts to an identifiable variation on language, which we call a lect. Lect is a backformation from words such as dialect (geographically defined lan- guage) and ethnolect (language defined by ethnic context). In this paper, we describe lect classifiers for so- cial power relationships. We refer to these lects as: • UpSpeak: Communication directed to someone with greater social authority. • DownSpeak: Communication directed to someone with less social authority. • PeerSpeak: Communication to someone of equal social authority. We call the problem of modeling these lects Social Power Modeling (SPM). The experiments reported in this paper focused primarily on modeling Up- Speak and DownSpeak. Manually constructing tools that effectively model specific linguistic phenomena suggested by sociolinguistics would be a Herculean effort. Moreover, it would be necessary to repeat the ef- fort in every language! Our approach first identi- fies statistically salient phrases of words and parts of speech – known as n-grams – in training texts generated in conditions where the social …

Content truncated to 500 words