Hello, Who is Calling?: Can Words Reveal the Social Nature of Conversations?

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 112–119, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Hello, Who is Calling?: Can Words Reveal the Social Nature of Conversations? Anthony Stark, Izhak Shafran and Jeffrey Kaye Center for Spoken Language Understanding, OHSU, Portland USA. {starkan,shafrani,kaye}@ohsu.edu Abstract This study aims to infer the social nature of conversations from their content automat- ically. To place this work in context, our moti- vation stems from the need to understand how social disengagement affects cognitive decline or depression among older adults. For this purpose, we collected a comprehensive and naturalistic corpus comprising of all the in- coming and outgoing telephone calls from 10 subjects over the duration of a year. As a first step, we learned a binary classifier to fil- ter out business related conversation, achiev- ing an accuracy of about 85%. This clas- sification task provides a convenient tool to probe the nature of telephone conversations. We evaluated the utility of openings and clos- ing in differentiating personal calls, and find that empirical results on a large corpus do not support the hypotheses by Schegloff and Sacks that personal conversations are marked by unique closing structures. For classifying different types of social relationships such as family vs other, we investigated features re- lated to language use (entropy), hand-crafted dictionary (LIWC) and topics learned using unsupervised latent Dirichlet models (LDA). Our results show that the posteriors over top- ics from LDA provide consistently higher ac- curacy (60-81%) compared to LIWC or lan- guage use features in distinguishing different types of conversations. 1 Introduction In recent years, there has been a growing interest in analyzing text in informal interactions such as in Internet chat, newsgroups and twitter. The em- phasis of most such research has been in estimating network structure (Kwak et al., 2010) and detecting trending topics (Ritter et al., 2010), sentiments (Pak and Paroubek, 2010) and first stories (Petrovi´c et al., 2010). The focus has been on aggregating informa- tion from large number of users to analyze popula- tion level statistics. The study reported in this paper, in contrast, fo- cuses on understanding the social interactions of an individual over long periods of time. Our motiva- tion stems from the need to understand the factors of social engagement that ameliorate the rate of cogni- tive decline and depression in older adults. Since the early work of Glass (1997) and colleagues, several studies on large cohorts over extended duration have confirmed that older adults with few social relation- ships are at an increased risk of suffering depression and dementia. The limited information available in currently used coarse measures, often based on self- reports, have hindered epidemiologists from probing the nature of this association further. While social engagement is typically multi- faceted, older adults, who are often less mobile, rely on telephone conversations to maintain their social relationships. This is reflected in a recent survey by Pew Research Center which reported that among …

Content truncated to 500 words