A Parallel Parser for Spoken Natural Language

This paper describes SYNAPSIS, a parser for performing real-time understanding of spoken utterances in a parallel computational environment. Understanding continuous speech allowing reasonably free syntax poses two main oroblems, namely the risk of erroneous interpretations and the largeness of the search space owing to the high uncertainty of the input. The parser is characterized by an approach whose major novel features are 1) a drastic reduction of idle time, thanks to the asynchronous inter agent communications, and 2) high modularity, thanks to the distribution of homogeneous pieces of knowledge rather than distribution of different parsing tasks. These features make the parser most apt for implementation on homogeneous Transputer-based distributed architectures.