The Use of Prosody in Syntactic Disambiguation

THE USE OF PROSODY IN SYNTACTIC DISAMBIGUATION Patti Price (SRI), Mari Ostendorf (BU), Stefanie Shattuck-Hufnagel (MIT), Cynthia Fong (BU) SRI International, Menlo Park, CA 94025 Boston University, Boston, MA 02215 MIT, Cambridge, MA 02139 ABSTRACT Prosodic structure and syntactic structure are not identical; neither are they unrelated. Knowing when and how the two eorrespoud could yield better quality speech synthesis, could aid in the disambiguation of com- peting syntactic hypotheses in speech understanding, and could lead to a more comprehensive view of human speech processing. In a set of exper- iments involving 35 pairs of phonetically similar sentence.s representing seven types of structural contrasts, the perceptual evidence shows that some, but not all, of the pairs can be disambiguated on the basis of pro- sodie differences. The phonological evidence relates the disambiguation primarily to boundary phenomena, although prominences sometimes play a role. Finally, phonetic analyses describing the attributes of these phonological markers indicate the importance of both absolute and rela- tive measures. INTRODUCTION The syntax of spoken utterances is frequently ambiguous. Yet listeners usually arrive at something close to the intended mean- ing. Information listeners might use in disambiguation includes knowledge of the world, shared context, and a source of non-syn- tactic information that is under-represented in written communi- cation: the prosody of the utterance. By 'prosody' we mean suprasegmental information in speech, such as phrasing and stress, which can alter perceived sentence meaning without changing the segmental identity of the components. Since prosody plays an important role in speech communica- tion, a clear understanding of the mapping between prosodic and syntactic structure would reveal significant aspects of the cogni- tive processes of speech production and perception. In addition, it would provide guidelines for the synthesis of more natural- sounding speech. Further, any contribution that prosody can make to the resolution of structural ambiguities will be particu- larly helpful in spoken-language understanding, where lexical and structural ambiguities of written forms are compounded by difficulties in finding word boundaries and in identifying words reliably in automatic speech recognition. Here, we study the mapping between prosody and syntax by minimizing the contri- bution of other possible cues to the resolution of ambiguity. This study forms the foundation for further work on modeling prosody by assessing a set of syntactic environments in which prosody alone might be used to disambiguate sentences, and by analyzing the correspondence between the phonological and phonetic attributes of the prosodic structure of utterances and their per- ceived meanings. We begin by discussing previous work on the relationship between prosody and syntax. We then describe the recording of the corpus, and present results for the experimental studies which consider: (1) the accuracy and confidence of listeners in disam- biguating different types of syntactic structures, (2) the phono- logical analysis of prosodic cues associated with the different structures, and their relation to the disambiguation results, and (3) a phonetic analysis of the phonological markers. Finally, we discuss the implications of these results, and raise some unre- solved questions that suggest …

Content truncated to 500 words