Predicting Overt Display of Power in Written Dialogs

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 518–522, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Predicting Overt Display of Power in Written Dialogs Vinodkumar Prabhakaran Computer Science Dept. Columbia University New York, NY 10027, USA vinod@cs.columbia.edu Owen Rambow CCLS Columbia University New York, NY 10027, USA rambow@ccls.columbia.edu Mona Diab CCLS Columbia University New York, NY 10027, USA mdiab@ccls.columbia.edu Abstract We analyze overt displays of power (ODPs) in written dialogs. We present an email cor- pus with utterances annotated for ODP and present a supervised learning system to predict it. We obtain a best cross validation F-measure of 65.8 using gold dialog act features and 55.6 without using them. 1 Introduction Analyzing written dialogs (such as email exchanges) to extract social power relations has generated great interest recently. This paper introduces a new task within the general field of finding power relations in written dialogs. In written dialog, an utterance can represent an overt display of power (ODP) on the part of the utterer if it constrains the addressee’s actions beyond the constraints that the underlying dialog act on its own imposes. For example, a re- quest for action is the first part of an adjacency pair and thus requires a response from the addressee, but declining the request is a valid response. However, the utterer may formulate her request for action in a way that attempts to remove the option of declining it (“Come to my office now!”). In so doing, she re- stricts her addressee’s options for responding more severely than a simple request for action would. Our new task is to classify utterances in written dialog as to whether they are ODPs or not. Such a classifi- cation can be interesting in and of itself, and it can also be used to study social relations among dialog participants. After reviewing related work (Section 2), we de- fine “overt display of power” (Section 3) and then present manual annotations for ODP in a small sub- set of Enron email corpus. In Section 5, we present a supervised learning system using word and part-of- speech features along with features indicating dialog acts. 2 Related Work Many studies in sociolinguistics have shown that power relations are manifested in language use (e.g., (O’Barr, 1982)). Locher (2004) recognizes “restriction of an interactant’s action-environment” (Wartenberg, 1990) as a key element by which ex- ercise of power in interactions can be identified. Through ODP we capture this action-restriction at an utterance level. In the computational field, sev- eral studies have used Social Network Analysis (e.g., (Diesner and Carley, 2005)) for extracting so- cial relations from online communication. Only re- cently have researchers started using NLP to analyze the content of messages to deduce social relations (e.g., (Diehl et al., 2007)). Bramsen et al. (2011) use knowledge of the actual organizational structure to create two sets of messages: messages sent from a superior to a subordinate, and vice versa. Their task is to …

Content truncated to 500 words