An Unsupervised Dynamic Bayesian Network Approach to Measuring Speech Style Accommodation

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 787–797, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics An Unsupervised Dynamic Bayesian Network Approach to Measuring Speech Style Accommodation Mahaveer Jain1, John McDonough1, Gahgene Gweon2, Bhiksha Raj1, Carolyn Penstein Ros´e1,2 1. Language Technologies Institute; 2. Human Computer Interaction Institute Carnegie Mellon University Pittsburgh, PA 15213 {mmahavee,johnmcd,ggweon,bhiksha,cprose}@cs.cmu.edu Abstract Speech style accommodation refers to shifts in style that are used to achieve strate- gic goals within interactions. Models of stylistic shift that focus on specific fea- tures are limited in terms of the contexts to which they can be applied if the goal of the analysis is to model socially motivated speech style accommodation. In this pa- per, we present an unsupervised Dynamic Bayesian Model that allows us to model stylistic style accommodation in a way that is agnostic to which specific speech style features will shift in a way that resem- bles socially motivated stylistic variation. This greatly expands the applicability of the model across contexts. Our hypothesis is that stylistic shifts that occur as a result of social processes are likely to display some consistency over time, and if we leverage this insight in our model,we will achieve a model that better captures inherent struc- ture within speech. 1 Introduction Sociolinguistic research on speech style and its resulting social interpretation has frequently fo- cused on the ways in which shifts in style are used to achieve strategic goals within interac- tions, for example the ways in which speakers may adapt their speaking style to suppress differ- ences and accentuate similarities between them- selves and their interlocutors in order to build solidarity (Coupland, 2007; Eckert & Rickford, 2001; Sanders, 1987). We refer to this stylis- tic convergence as speech style accommodation. In the language technologies community, one tar- geted practical benefit of such modeling has been the achievement of more natural interactions with speech dialogue systems (Levitan et al., 2011). Monitoring social processes from speech or language data has other practical benefits as well, such as enabling monitoring how beneficial an in- teraction is for group learning (Ward & Litman, 2007; Gweon, 2011), how equal participation is within a group (DiMicco et al., 2004), or how conducive an environment is for fostering a sense of belonging and identification with a community (Wang et al., 2011). Typical work on computational models of speech style accommodation have focused on spe- cific aspects of style that may be accommodated, such as the frequency or timing of pauses or backchannels (i.e., words that show attention like ’Un huh’ or ’ok’), pitch, or speaking rate (Ed- lund et al., 2009; Levitan & Hirschberg, 2011). In this paper, we present an unsupervised Dynamic Bayesian Model that allows us to model speech style accommodation in a way that does not re- quire us to specify which linguistic features we are targeting. We explore a space of models de- fined by two independent factors, namely the di- rect …

Content truncated to 500 words