Modeling Socio-Cultural Phenomena in Discourse

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 1038–1046, Beijing, August 2010 Modeling Socio-Cultural Phenomena in Discourse Tomek Strzalkowski1,2, George Aaron Broadwell1, Jennifer Stromer-Galley1, Samira Shaikh1, Sarah Taylor3 and Nick Webb1 1ILS Institute, University at Albany, SUNY 2IPI, Polish Academy of Sciences 3Lockheed Martin Corporation tomek@albany.edu Abstract In this paper, we describe a novel ap- proach to computational modeling and understanding of social and cul- tural phenomena in multi-party dia- logues. We developed a two-tier ap- proach in which we first detect and classify certain social language uses, including topic control, disagreement, and involvement, that serve as first order models from which presence the higher level social constructs such as leadership, may be inferred. 1. Introduction We investigate the language dynamics in small group interactions across various set- tings. Our focus in this paper is on English online chat conversations; however, the mod- els we are developing are more universal and applicable to other conversational situations: informal face-to-face interactions, formal meetings, moderated discussions, as well as interactions conducted in languages other than English, e.g., Urdu and Mandarin. Multi-party online conversations are particu- larly interesting because they become a per- vasive form of communication within virtual communities, ubiquitous across all age groups. In particular, a great amount of communica- tion online occurs in virtual chat-rooms, typi- cally conducted using a highly informal text dialect. At the same time, the reduced-cue environment of online interaction necessitates more explicit linguistic devices to convey social and cultural nuances than is typical in face-to-face or even voice conversations. Our objective is to develop computational models of how certain social phenomena such as leadership, power, and conflict are signaled and reflected in language through the choice of lexical, syntactic, semantic and conversa- tional forms by discourse participants. In this paper we report the results of an initial phase of our work during which we constructed a prototype system called DSARMD-1 (De- tecting Social Actions and Roles in Multi- party Dialogue). Given a representative seg- ment of multiparty task-oriented dialogue, DSARMD-1 automatically classifies all dis- course participants by the degree to which they deploy selected social language uses, such as topic control, task control, involve- ment, and disagreement. These are the mid-level social phenomena, which are de- ployed by discourse participants in order to achieve or assert higher-level social con- structs, including leadership. In this work we adopted a two-tier empirical approach where social language uses are modeled through observable linguistic features that can be automatically extracted from dialogue. The high-level social constructs are then inferred from a combination of language uses attrib- uted to each discourse participant; for exam- ple, a high degree of influence and a high de- gree of involvement by the same person may indicate a leadership role. In this paper we limit our discussion to the first tier only: how to effectively model and classify social lan- guage uses in multi-party dialogue. 2. Related Research Issues related to linguistic manifestation of social phenomena have not been systemati- cally …

Content truncated to 500 words