In this paper we present a semantic based partial parsing system for spontaneous dialogue processing. The processing is done following three main steps: at the first a pre-processing module normalizes spoken utterances from errors and extragrammaticalities. The second step consists of parsing the input utterances. Parsing is done on Semantic Tree Unification Grammar STUG. We propose this formalism as an extended version of traditional semantic grammar. The key features of this formalism are linguistic and cognitive motivation, robustness and efficiency in processing spoken language extragrammaticalities. The partial aspect of the parser is achieved via a selective strategy allowing the system to detect and process relevant islands in the utterance.Finally, the parsed utterances are associated with a semantic interpretation by a frame module.