Prosodic Similarities of Dialog Act Boundaries Across Speaking Styles

Spontaneous speech differs in a multitude of ways from read, formal, or laboratory speech (Maclay & Osgood 1959, Goldman-Eisler 1968, Levelt 1983, Biber 1988, Howell & Kadi-Hanifi 1991, Eskenazi 1993, Shriberg 1994, Swerts et al. 1996, Bruce 1995, Hirschberg 1995, Laan 1997). Although the labels “spontaneous” and “read” each reflect an underlying multidimensional space of different genres or “styles” (Eskenazi 1993), generally speaking, spontaneous speech exhibits greater segmental and suprasegmental variability than does read speech (Lieberman et al. 1985, Llisterri & Poch 1991, Wajskop et al. 1992, Kohler 1996, Greenberg 1999, Maekawa 2003, Tseng 2005, Benzeghiba et al. 2007). It is no surprise, then, that automatic speech processing techniques typically have more difficulty with spontaneous than with read speech (Weintraub et al. 1996, Greenberg 1999, McAllister et al. 1998, Riley et al. 1999, Ostendorf 2000, Binnenpoorte et al. 2004, Benzeghiba et al. 2007). For example, an early study in large-vocabulary speech recognition found degraded performance for spontaneous speech even when recording conditions, speaker, and word sequences were held constant (Weintraub et al. 1996). In the study, speakers read transcripts of what they had said in previous spontaneous conversations; the read versions were significantly easier to recognize than the spontaneous originals. In this paper we focus on comparisons of prosody across speaking styles, specifically for the task of dialog act segmentation. Dialog act segmentation aims to segment the