The Utility of Filled Pauses, Interjections, and Parentheticals in Parsing Conversational Language
Don Engel · 2001
It has been hypothesized [5,7] that the information contained by the placement and content of filled pauses, interjections, and parentheticals is useful in the parsing of conversational language. To investigate the impact of the inclusion of such information on the performance of a parser, we have built wrappers to work between the "Switchboard " corpus [3] and an existing parser [1]. These wrappers each removed one or more types of linguistic elements from the corpus. Training and testing on each wrapped corpus, we achieve different success rates. Our data indicates that some forms of misspeaking have little effect, but others impede the functionality of the parser.