Modeling Latent Biographic Attributes in Conversational Genres

Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, pages 710–718, Suntec, Singapore, 2-7 August 2009. c⃝2009 ACL and AFNLP Modeling Latent Biographic Attributes in Conversational Genres Nikesh Garera and David Yarowsky Department of Computer Science, Johns Hopkins University Human Language Technology Center of Excellence Baltimore MD, USA {ngarera,yarowsky}@cs.jhu.edu Abstract This paper presents and evaluates several original techniques for the latent classifi- cation of biographic attributes such as gen- der, age and native language, in diverse genres (conversation transcripts, email) and languages (Arabic, English). First, we present a novel partner-sensitive model for extracting biographic attributes in con- versations, given the differences in lexi- cal usage and discourse style such as ob- served between same-gender and mixed- gender conversations. Then, we explore a rich variety of novel sociolinguistic and discourse-based features, including mean utterance length, passive/active usage, per- centage domination of the conversation, speaking rate and filler word usage. Cu- mulatively up to 20% error reduction is achieved relative to the standard Boulis and Ostendorf (2005) algorithm for classi- fying individual conversations on Switch- board, and accuracy for gender detection on the Switchboard corpus (aggregate) and Gulf Arabic corpus exceeds 95%. 1 Introduction Speaker attributes such as gender, age, dialect, na- tive language and educational level may be (a) stated overtly in metadata, (b) derivable indirectly from metadata such as a speaker’s phone number or userid, or (c) derivable from acoustic proper- ties of the speaker, including pitch and f0 contours (Bocklet et al., 2008). In contrast, the goal of this paper is to model and classify such speaker attributes from only the latent information found in textual transcripts. In particular, we are inter- ested in modeling and classifying biographic at- tributes such as gender and age based on lexi- cal and discourse factors including lexical choice, mean utterance length, patterns of participation in the conversation and filler word usage. Fur- thermore, a speaker’s lexical choice and discourse style may differ substantially depending on the gender/age/etc. of the speaker’s interlocutor, and hence improvements may be achived via dyadic modeling or stacked classifiers. There has been substantial work in the sociolin- guistics literature investigating discourse style dif- ferences due to speaker properties such as gender (Coates, 1997; Eckert, McConnell-Ginet, 2003). Analyzing such differences is not only interesting from the sociolinguistic and psycholinguistic point of view of language understanding, but also from an engineering perspective, given the goal of pre- dicting latent author/speaker attributes in various practical applications such as user authenticaion, call routing, user and population profiling on so- cial networking websites such as facebook, and gender/age conditioned language models for ma- chine translation and speech recogntition. While most of the prior work in sociolinguistics has been approached from a non-computational perspec- tive, Koppel et al. (2002) employed the use of a linear model for gender classification with manu- ally assigned weights for a set of linguistically in- teresting words as features, focusing on a small de- velopment corpus. Another computational study for gender classification …

Content truncated to 500 words