Prosody as a cue to syntactic structure: The role of prosodic information in disambiguating phonetically ambiguous sentences
Patti J. Price, M. Ostendorf, Stefanie Shattuck‐Hufnagel, Cynthia Fong · The Journal of the Acoustical Society of America · 1989
To go beyond dictation tasks, speech recognition needs to be integrated with an understanding component. Natural language understanding research, however, has typically considered text input, and has not had to face the massive phonetic ambiguity arising from speech input: Even with no phonemic errors, word boundary ambiguities can yield thousands of parses for a substantial portion of moderate length sentences. This study assesses the potential role of prosody in aiding speech understanding. Using phonetically similar sentence pairs with and without a major structural break, seven types of structural ambiguity were investigated. Professional speakers read ten sentences of each type after reading diambiguating contexts. Naive listeners selected the appropriate context for each sentence. For recording and listening, only one member of each pair was used in each session; sentences were otherwise randomized. The results indicate substantial differences across structural types: e.g., appositions and tags were nearly always correctly identified, whereas main-main versus main-subordinate pairs were at the chance level. This suggests that prosody can disambiguate some, but not all, structural ambiguities. [Work supported by NSF.]