Predicting Intonational Boundaries Automatically from Text: The ATIS Domain

Predicting Intonational Boundaries Automatically from Text: The ATIS Domain Michelle Q. Wang Churchill College Cambridge University Cambridge UK Julia Hirschberg AT&T Bell Laboratories 600 Mountain Avenue Murray Hill NJ 07974 February 11, 1991 Abstract Relating the intonational characteristics of an utter- ance to other features inferable from its text is impor- tant both for speech recognition and for speech syn- thesis. This work investigates techniques for predicting the location of intonational phrase boundaries in natural speech, through analyzing a utterances from the DARPA Air Travel Information Service database. For statistical modeling, we employ Classification and Regression Tree (CART) techniques. We achieve success rates of just over 90%. 1 Introduction Intuitively, intonational phrasing divides an utterance into meaningful 'chunks' of information [3]. Variation in phrasing can change the meaning hearers assign to tokens of a given sentence. For example, 'Bill doesn't drink because he% un- happy' is likely to be interpreted one way when uttered as a single phrase (i.e., Bill drinks, but not because he's un- happy) and another when uttered with a boundary between drink and because (the cause of Bill's failure to drink is his unhappiness). While phrase boundaries are perceptual categories, they are associated with certain acoustic features. Generally, phrases may be identified by one of more of the following features: pauses (which may be filled or not), changes in am- plitude and in the pitch contour, and lengthening of the final syllable in the phrase (sometimes accompanied by glottaliza- tion of that syUable and perhaps preceding syllables). Major phrase boundaries tend to be associated with longer pauses, more pronounced contour excursions, and greater amounts of final lengthening than minor boundaries. 2 Inferring Phrasing from Text How the intonational phrasing of an utterance is related to as- pects of the text uttered is potentially an important source of information for speech recognition, to constrain the set of al- lowable hypotheses by identifying boundary locations in both the recognized text and the acoustic signal or to moderate durational information at likely boundary locations. How- ever, to date, syntactically-based prediction of intonational boundaries has met with limited success. While considerable work has been done on the relationship between some par- ticular syntactic configurations and intonational boundaries [12, 2, 6, 9], the prediction of boundaries in unrestricted and spontaneous speech rarely been attempted [1]. 1 Predicting boundaries solely from information available automaticaLly from text analysis presents a further challenge, which must also be addressed if predictions are to be useful in real spoken language systems. To address these issues, we experimented with the pre- diction of intonational boundaries from text analysis, using 298 utterances from 26 speakers in the Air Travel Informa- tion Service (ATIS) database for training and testing. ~ To prepare data for analysis, we labeled the speech prosodically by hand, noting location and type of intonational boundaries and presence or absence of pitch accents, using both the wave- form and pitchtracks of each utterance. Although major and minor boundaries were distinguished in the labeling process, in the …

Content truncated to 500 words