FeasPar - A Feature Structure Parser Learning to Parse Spoken Language

FeasPar - A Feature Structure Parser Learning to Parse Spoken Language Finn Dag Bu¢ and Alex Waibel Interactive Systems Laboratories University of Karlsruhe, Germany and Carnegie Mellon University, USA {finndag[waibel}~ira. uka. de Abstract We describe and experimentally evalu- ate a system, FeasPar, that learns pars- ing spontaneous speech. To train and run FeasPar (Feature Structure Parser), only limited handmodeled knowledge is required. The FeasPar architecture consists of neu- ral networks and a search. The networks spilt the incoming sentence into chunks, which are labeled with feature values and chunk relations. Then, the search finds the most probable and consistent feature structure. FeasPar is trained, tested and evaluated with the Spontaneous Schednling Task, and compared with a handmodeled LR- parser. The handmodeling effort for Fea- sPar is 2 weeks. The handmodeling ef- fort for the LR-parser was 4 months. FeasPar performed better than the LR- parser in all six comparisons that are made. 1 Introduction When building a speech parsing component for small domains, an important goal is to get good performance. If low hand labor is involved, then it's even better. Unification based formalisms, e.g.(Gazdar et al., 1985; Kaplan and Bresnan, 1982; Pollard and Sag, ]987), have been very successful for analyz- ing written language, because they have provided parses with rich and detailed linguistic informa- tion. However, these approaches have two major drawbacks: first, they require hand-designed sym- bolic knowledge like lexica and grammar rules, and second, this knowledge is too rigid, causing problems with ungranlmaticality and other devi- ations from linguistic rules. These deviations are manageable and low in number, when analyzing written language, but not for spoken language. The latter also contains spontaneous effects and speech recognition errors. (On the other hand, the good thing is that spoken language tend to contain less complex structures than written language.) Several methods have been suggested compensate for these speech related problems: e.g. score and penalties, probabilistic rules, and skipping words (Dowding et al., 1993; Seneff, 1992; Lavie and Tomita, 1993; Issar and Ward, 1993). A small community have experimented with ei- ther purely statistical approaches(Brown et al., 1990; Schiitze, 1993) or connectionist based ap- proaches (Berg, 1991; Miikkulainen and Dyer, 1991; Jain, 1991; Wermter and Weber, 1994). The main problem when using statistical ap- proaches for spoken language processing, is the large amounts of data required to train these mod- els. All connectionist approaches to our knowl- edge, have suffered from one or more of the fol- lowing problems: One, parses contains none or too few linguistic attributes to be used in translation or understanding, and/or it is not shown how to use their parse formalism in a total NLP system. Two, no clear and quantitative statement about overall performance is made. Three, the approach has not been evaluated with real world data, but with highly regular sentences. Four, millions of training sentences are required. In this paper, we present a parser that produces complex feature structures, as known from e.g. GPSG(Gazdar et al., 1985). This parser requires only …

Content truncated to 500 words