Author Commitment and Social Power: Automatic Belief Tagging to Infer the Social Context of Interactions

Proceedings of NAACL-HLT 2018, pages 1057–1068 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Author Commitment and Social Power: Automatic Belief Tagging to Infer the Social Context of Interactions Vinodkumar Prabhakaran Stanford University vinod@cs.stanford.edu Premkumar Ganeshkumar Agolo, Inc. prem@agolo.com Owen Rambow Elemental Cognition, Inc. owenr@elementalcognition.com Abstract Understanding how social power structures af- fect the way we interact with one another is of great interest to social scientists who want to answer fundamental questions about hu- man behavior, as well as to computer scien- tists who want to build automatic methods to infer the social contexts of interactions. In this paper, we employ advancements in extra- propositional semantics extraction within NLP to study how author commitment reflects the social context of an interactions. Specifi- cally, we investigate whether the level of com- mitment expressed by individuals in an orga- nizational interaction reflects the hierarchical power structures they are part of. We find that subordinates use significantly more instances of non-commitment than superiors. More im- portantly, we also find that subordinates at- tribute propositions to other agents more of- ten than superiors do — an aspect that has not been studied before. Finally, we show that en- riching lexical features with commitment la- bels captures important distinctions in social meanings. 1 Introduction Social power is a difficult concept to define, but is often manifested in how we interact with one an- other. Understanding these manifestations is im- portant not only to answer fundamental questions in social sciences about power and social inter- actions, but also to build computational models that can automatically infer social power struc- tures from interactions. The availability and ac- cess to large digital repositories of naturally occur- ring social interactions and the advancements in natural language processing techniques in recent years have enabled researchers to perform large scale studies on linguistic correlates of power, such as words and phrases (Bramsen et al., 2011; Gilbert, 2012), linguistic coordination (Danescu- Niculescu-Mizil et al., 2012), agenda control (Tay- lor et al., 2012), and dialog structure (Prabhakaran and Rambow, 2014). Another area of research that has recently gar- nered interest within the NLP community is the modeling of author commitment in text. Ini- tial studies in this area were done in process- ing hedges, uncertainty and lack of commit- ment, specifically focused on scientific text (Mer- cer et al., 2004; Di Marco et al., 2006; Farkas et al., 2010). More recently, researchers have also looked into capturing author commitment in non- scientific text, e.g., levels of factuality in newswire (Saur´ı and Pustejovsky, 2009), types of commit- ment of beliefs in a variety of genres including conversational text (Diab et al., 2009; Prabhakaran et al., 2015). These approaches are motivated from an information extraction perspective, for instance in aiding tasks such as knowledge base popula- tion.1 However, it has not been studied whether such sophisticated author commitment analysis can go beyond what is expressed in language and reveal the underlying social contexts in which lan- guage is exchanged. …

Content truncated to 500 words