Intonation, Discourse Structure and Information Status in Spontaneous Speech Luciana Lucente 1 , Julia Hirschberg 2 , Plínio A. Barbosa 1 1 State University of Campinas, 2 Columbia University lucente.luciana@gmail.com; julia@cs.columbia.edu; pabarbosa.unicampbr@gmail.com The relation between intonational focus and information status, especially to with given and new information (Terken & Hirschberg, 1994; Gravano & Hirschberg, 2006) has been accepted for some time. Recent studies on Brazilian Portuguese (henceforth BP) intonation explore this relationship in VoCE spontaneous speech corpus. This research is being developed in the framework of the DaTo (Dynamic Tones of Brazilian Portuguese Intonation) intonation annotation system (Lucente & Barbosa, 2008; 2010). The VoCE corpus is a corpus essentially composed of spontaneous speech data from radio broadcasts and podcasts from the internet. This corpus has been being collected since 2006. It contains approximately 10 hours of conversation among announcers and guests. The DaTo system attempts not only to present a system for intonational annotation, but also to describe functional aspects of intonational contours in spontaneous speech, e. g., their pragmatic function. As research about information status and intonation in the DaTo framework follows a pragmatic/cognitive approach, in this paper we attempt to analyze whether given/new information status is related to (i) a specific DaTo intonational contour; (ii) a position in a discourse segment purpose (DSP) (according to Grosz & Sidner (1986)); and (iii) a grammatical function. The choice of these three parameters is based on the hypothesis that speakers use intonational features and discourse structure to differentiate given information from new information. The first step before data analysis was the classification nouns as given or new following Prince (1981) classification: DnHn – discourse new, hearer new; DnHg – discourse new, hearer given; DgHn – discourse given, hearer new; DgHg – discourse given, hearer given. Such classification assumes a cognitive basis, for which the information status is defined according to the information achievement in both speaker and hearer minds (Chafe, 1974). The next step was to relate the information status classification with the DaTo intonational contours (LH, >LH, vLH, HL, LHL and HLH) associated with each given/new token. The given/new classification was also related to grammatical function (subject/object) and surface position (initial, medial, final) in the DSPs, into which the entire corpus was segmented. The annotation referring to intonational contours, information status and DSP segmentation was done in separate tiers using Praat software package. A segmentation of the material in vowel-to-vowel units (Barbosa, 2007) and orthographic words was also carried out. Our preliminary results have not yet shown significant correlations between information status and intonational contours. It may be that information status in BP is related to focus emphasis and not to contour type. A perception test was carried out to evaluate the degree of emphasis on given vs. new information. This study showed tendency for given information be evaluated as more prominent than new information. The increase in prominence on given information may be due to two reasons: (i) the tendency of given information to occur in subject position, and for new information to occur in object position, and (ii) a speaker’s strategy to (re)activate the information in the listener’s consciousness.