A Learning Approach to Shallow Parsing

A Learning Approach to Shallow Parsing* Marcia Mufioz t Vasin Punyakanok* Dan Roth* Day Zimak Department of Computer Science University of Illinois at Urbana-Champaign Urbana, IL 61801 USA Abstract A SNoW based learning approach to shallow pars- ing tasks is presented and studied experimentally. The approach learns to identify syntactic patterns by combining simple predictors to produce a coherent inference. Two instantiations of this approach are studied and experimental results for Noun-Phrases (NP) and Subject-Verb (SV) phrases that compare favorably with the best published results are pre- sented. In doing that, we compare two ways of mod- eling the problem of learning to recognize patterns and suggest that shallow parsing patterns are bet- ter learned using open/close predictors than using inside/outside predictors. 1 Introduction Shallow parsing is studied as an alternative to full-sentence parsers. Rather than producing a complete analysis of sentences, the alternative is to perform only partial analysis of the syntactic structures in a text (Harris, 1957; Abney, 1991; Greffenstette, 1993). Shallow parsing informa- tion such as NPs and other syntactic sequences have been found useful in many large-scale lan- guage processing applications including infor- mation extraction and text summarization. A lot of the work on shallow parsing over the past years has concentrated on manual construction of rules. The observation that shallow syntactic information can be extracted using local infor- mation - by examining the pattern itself, its nearby context and the local part-of-speech in- formation - has motivated the use of learning methods to recognize these patterns (Church, 1988; Ramshaw and Marcus, 1995; Argamon et al., 1998; Cardie and Pierce, 1998). * Research supported by NSF grants IIS-9801638 and SBR-9873450. t Research supported by NSF grant CCR-9502540. This paper presents a general learning ap- proach for identifying syntactic patterns, based on the SNoW learning architecture (Roth, 1998; Roth, 1999). The SNoW learning architecture is a sparse network of linear ftmctions over a pre- defined or incrementally learned feature space. SNoW is specifically tailored for learning in do- mains in which the potential number of infor- mation sources (features) taking part in deci- sions is very large - of which NLP is a princi- pal example. Preliminary versions of it have al- ready been used successfully on several tasks in natural language processing (Roth, 1998; Gold- ing and Roth, 1999; Roth and Zelenko, 1998). In particular, SNoW's sparse architecture sup- ports well chaining and combining predictors to produce a coherent inference. This property of the architecture is the base for the learning ap- proach studied here in the context of shallow parsing. Shallow parsing tasks often involve the iden- tification of syntactic phrases or of words that participate in a syntactic relationship. Com- putationally, each decision of this sort involves multiple predictions that interact in some way. For example, in identifying a phrase, one can identify the beginning and end of the phrase while also making sure they are coherent. Our computational paradigm suggests using a SNoW based predictor as a building block that learns to perform …

Content truncated to 500 words