A Proposal for Function-Based Phonological Intonation Categories
We present a preliminary scheme for using functional semantic categories generally agreed in the literature to influence prosody directly as phonological intonation categories. We propose that these categories, including focus (kontrast), theme-rheme and speaker-hearer orientation, are weighted features affecting the prosodic output of an utterance (in this case the F0 contour) at either the syllable, word or prosodic phrase level. The influence of each can be seen as having a weighted effect on the prosody of the utterance, according to the other relevant features of the speech stream, including phrase length, lexical stress, lexical choice and physiological factors. Rather than presupposing the phonetic effects of each feature (e.g. by specifying ToBI elements), we wish to automatically extract them by building a computational model including these features which can predict the F0 of a given utterance from a corpus of suitably annotated conversational speech. Exploratory results of our attempts to implement this scheme are given using the SPOT corpus (Schafer et al 2000) of spontaneous restricted game-oriented dialogues within a discussion of the usefulness of this approach to intonational phonology.