Efficient semantic parsing of conversational speech

This paper presents an empirical method for mapping speech input to shallow semantic representation. Semantic parsing is realized through a bottom-up type parsing paradigm where the operators are based on semantic concepts, obtained from a lexicon. A statistically trained model specializes the parser, by guiding the runtime beam-like search of possible parses. The semantic representation is a logical form equivalent to a Discourse Representation Structure (DRS). Each output of the parser is given a probability according to how similar, given a contextual word similarity measure, the parsing process for the input was to those collected during the training phase. Contextual information during parsing allows for better coverage of large domains. The non syntactic but very semantic nature of the parser would make it very tolerant to noisy (recognized) speech input. Shallow parsing using First-Order Logic (FOL) allows for fast but meaningful enough processing of the input, which makes the parser well suited for real-time Spoken Dialogs Systems (SDS).