Use of Prosody in Syntactic Disambiguation: An Analysis-by-Synthesis Approach

USE OF PROSODY IN SYNTACTIC DISAMBIGUATION: AN ANALYSIS-BY-SYNTHESIS APPROACH C. W. Wightman, N. M. Veilleuz, M. Ostendorf Boston University 44 Cumrnington St. Boston, MA 02215 ABSTRACT of seven types of structural ambiguity. In addition, it ap- ~ ~ ~ ~ f i ~ ~ t ~ have shown that prosody is used by human ]is- peared that the relative size and location of prosodic phrase teners to disambiguate spoken language and, in particular, that boundaries provided the principal prosodic clue for resolv- the relative size and location of prosodic phrase boundaries pro- ing ambiguities. Thus, it seems likely that automatically vides a cue for resolving syntactic ambiguity. Therefore, auto- detected prosodic phrase breaks could be used by speech matically detected prosodic phrase boundaries can provide infor- understanding systems to reduce syntactic ambiguity. mation useful in speech understanding for choosing among several candidate parses. Here, we propose a scoring algorithm to rank candidate parses based on an analysis-by-synthesis method which compares the observed prosodic phrase structure with the pre- dicted structure for each candidate parse. In experiments with a small corpus of ambiguous sentences spoken by FM radio an- nouncers, we have achieved disambiguation performance close to the performance of human subjects in perceptual experiments. INTRODUCTION Spoken language processing is a difficult problem, in part because of the many ambiguities inherent in natural lan- guage. Syntactic ambiguity arises when a given expression can be described by more than one syntactic structure, and represents an important problem in natural language pro- cessing. In particular, attachment ambiguities occur fre- quently in language, e.g., "Show the fares for [the cheapest flights on the screen]." "Show [the fares for the cheapest flights] on the screen." Several factors may be involved in resolving such ambigu- ities, including semantics, discourse and syntactic bias. In spoken language, prosody, or the suprasegmental informa- tion in an utterance, is an important additional cue. In previous work [lo], a hierarchical set of break indices was proposed as a representation of prosodic phrase bound- aries and automatically detected break indices were used in a parser to provide constraints on rules that would prevent prosodically inconsistent parses. Here, we propose a scoring algorithm to rank candidate parses based on an analysis-by- synthesis method which involves: (1) using an algorithm to predict prosodic break locations for each candidate syntac- tic structure; (2) automatically detecting prosodic breaks in the spoken utterance; and (3) ranking the parses accord- ing to a similarity score between predicted and observed prosodic structure. Using the database of ambiguous sen- tences from [13], this approach achieves performance close to that of the human subjects. The following section describes the speech corpus and prosodic break index representation. We then examine syn- thesis algorithms for predicting the phrase structure of a sentence and evaluate them in relation to the hand-labeled speech corpus. Next, we describe an automatic method of labeling the prosodic phrase structure and a measure of the similarity between the predicted and detected prosodic structures. We then present experimental results, based …

Content truncated to 500 words