Recognizing Social Constructs from Textual Conversation

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1293–1298, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Recognizing Social Constructs from Textual Conversation Somak Aditya and Chitta Baral and Nguyen H. Vo and Joohyung Lee and Jieping Ye, Zaw Naung and Barry Lumpkin and Jenny Hastings Dept. of Computer Science, Arizona State University Richard Scherl Dept. of Computer Science, Monmouth University Dawn M. Sweet Dept. of Psychology, Iowa State University Daniela Inclezan Dept. of Computer Science, Miami University Abstract In this paper we present our work on rec- ognizing high level social constructs such as Leadership and Status from textual conversa- tion using an approach that makes use of the background knowledge about social hierarchy and integrates statistical methods and sym- bolic logic based methods. We use a stratified approach in which we first detect lower level language constructs such as politeness, com- mand and agreement that help us to infer inter- mediate constructs such as deference, close- ness and authority that are observed between the parties engaged in conversation. These in- termediate constructs in turn are used to de- termine the social constructs Leadership and Status. We have implemented this system successfully in both English and Korean lan- guages and achieved considerable accuracy. 1 Introduction and Related Works The traditional information extraction paradigm has seen success in extracting simplistic behaviors or emotions from text. However, to detect high-level social constructs such as leadership or status, we re- quire robustly defined notions about language con- structs that cannot always be directly inferred from text. Hence, in this paper we focus on extracting information from text that requires additional back- ground knowledge and inference. There are a few works in this direction, such as (Tari et al., 2010), however our focus in this paper is to extract infor- mation pertaining to different social constructs from textual conversation. The earlier research in ana- lyzing conversations includes developing annotated chat corpuses (Shaikh et al., 2010), and developing a socio-cultural phenomena model from discourse with a small-scale implementation (Strzalkowski et al., 2010). Other researchers have focused on auto- matically annotating social behavior in conversation using statistical approaches (Mayfield et al., 2013). The discourse structure of a conversation is modeled as a Hidden Markov Model in (Stolcke, 2000) to de- termine dialogue acts such as Statement, Question and Agreement. In (Prabhakaran et al., 2012) an- notated email threads are presented for facilitating detection of social relations. Among recent works, (Gilbert, 2012) uses lin- guistic cues to discover workplace hierarchy from emails. The use of phrases to detect language use such as “commands” is motivating. However, due to the lack of logical explanation and robust defi- nition, the effectiveness of this method decreases in the semi-formally moderated Wikipedia community, which has interplay of several different LUs such as command, politeness and informal language. In (Danescu-Niculescu-Mizil et al., 2012), the authors explain how reflection of linguistic styles can shed light on power differentials; though, …

Content truncated to 500 words