The Integration of Unification-Based Syntax/Semantics and Memory-Based Pragmatics for Real-Time Understanding of Noisy Continuous Speech Input

Hideto Tomabechi, Masaru Tomita

Real-time understanding of speech input is difficult especially because the input is often noisy and elliptic. Multiple morphophonemic and lexical hypotheses generated for a single input sentence cannot be resolved by local semantics alone. We have developed a system in which unilication-based parsing of speech input is integrated with thematic memory-based spreading activation that supplies extra-sentential knowledge that can help to disambiguate noisy and elliptic real-time speaker-independent continuous speech input.