Shallow Semantic Parsing using Support Vector Machines

Shallow Semantic Parsing using Support Vector Machines∗ Sameer Pradhan, Wayne Ward, Kadri Hacioglu, James H. Martin Center for Spoken Language Research, University of Colorado, Boulder, CO 80303 {spradhan,whw,hacioglu,martin}@cslr.colorado.edu Dan Jurafsky Department of Linguistics Stanford University Stanford, CA 94305 jurafsky@stanford.edu Abstract In this paper, we propose a machine learning al- gorithm for shallow semantic parsing, extend- ing the work of Gildea and Jurafsky (2002), Surdeanu et al. (2003) and others. Our al- gorithm is based on Support Vector Machines which we show give an improvement in perfor- mance over earlier classifiers. We show perfor- mance improvements through a number of new features and measure their ability to general- ize to a new test set drawn from the AQUAINT corpus. 1 Introduction Automatic, accurate and wide-coverage techniques that can annotate naturally occurring text with semantic argu- ment structure can play a key role in NLP applications such as Information Extraction, Question Answering and Summarization. Shallow semantic parsing – the process of assigning a simple WHO did WHAT to WHOM, WHEN, WHERE, WHY, HOW, etc. structure to sentences in text, is the process of producing such a markup. When pre- sented with a sentence, a parser should, for each predicate in the sentence, identify and label the predicate’s seman- tic arguments. This process entails identifying groups of words in a sentence that represent these semantic argu- ments and assigning specific labels to them. In recent work, a number of researchers have cast this problem as a tagging problem and have applied vari- ous supervised machine learning techniques to it (Gildea and Jurafsky (2000, 2002); Blaheta and Charniak (2000); Gildea and Palmer (2002); Surdeanu et al. (2003); Gildea and Hockenmaier (2003); Chen and Rambow (2003); Fleischman and Hovy (2003); Hacioglu and Ward (2003); Thompson et al. (2003); Pradhan et al. (2003)). In this ∗This research was partially supported by the ARDA AQUAINT program via contract OCG4423B and by the NSF via grant IS-9978025 paper, we report on a series of experiments exploring this approach. For the initial experiments, we adopted the approach described by Gildea and Jurafsky (2002) (G&J) and eval- uated a series of modifications to improve its perfor- mance. In the experiments reported here, we first re- placed their statistical classification algorithm with one that uses Support Vector Machines and then added to the existing feature set. We evaluate results using both hand- corrected TreeBank syntactic parses, and actual parses from the Charniak parser. 2 Semantic Annotation and Corpora We will be reporting on results using PropBank1 (Kings- bury et al., 2002), a 300k-word corpus in which predi- cate argument relations are marked for part of the verbs in the Wall Street Journal (WSJ) part of the Penn Tree- Bank (Marcus et al., 1994). The arguments of a verb are labeled ARG0 to ARG5, where ARG0 is the PROTO- AGENT (usually the subject of a transitive verb) ARG1 is the PROTO-PATIENT (usually its direct object), etc. PropBank attempts to treat semantically related verbs consistently. In addition to these CORE ARGUMENTS, additional ADJUNCTIVE …

Content truncated to 500 words