Semantic Role Labeling Using Complete Syntactic Analysis

Proceedings of the 9th Conference on Computational Natural Language Learning (CoNLL), pages 221–224, Ann Arbor, June 2005. c⃝2005 Association for Computational Linguistics Semantic Role Labeling Using Complete Syntactic Analysis Mihai Surdeanu Technical University of Catalunya surdeanu@lsi.upc.edu Jordi Turmo Technical University of Catalunya turmo@lsi.upc.edu Abstract In this paper we introduce a semantic role labeling system constructed on top of the full syntactic analysis of text. The la- beling problem is modeled using a rich set of lexical, syntactic, and semantic at- tributes and learned using one-versus-all AdaBoost classifiers. Our results indicate that even a simple ap- proach that assumes that each semantic ar- gument maps into exactly one syntactic phrase obtains encouraging performance, surpassing the best system that uses par- tial syntax by almost 6%. 1 Introduction Most current semantic role labeling (SRL) ap- proaches can be classified in one of two classes: approaches that take advantage of complete syntac- tic analysis of text, pioneered by (Gildea and Juraf- sky, 2002), and approaches that use partial syntac- tic analysis, championed by the previous CoNLL shared task evaluations (Carreras and M`arquez, 2004). However, to the authors’ knowledge, a clear anal- ysis of the benefits of using full syntactic analysis versus partial analysis is not yet available. On one hand, the additional information provided by com- plete syntax should intuitively be useful. But, on the other hand, the state-of-the-art of full parsing is known to be less robust and perform worse than the tools used for partial syntactic analysis, which would decrease the quality of the information pro- vided. The work presented in this paper contributes to this analysis by introducing a model that is en- tirely based on the full syntactic analysis of text, generated by a real-world parser. 2 System Description 2.1 Mapping Arguments to Syntactic Constituents Our approach maps each argument label to one syn- tactic constituent, using a strategy similar to (Sur- deanu et al., 2003). Using a bottom-up approach, we map each argument to the first phrase that has the exact same boundaries and climb as high as possible in the syntactic tree across unary production chains. Unfortunately, this one-to-one mapping between semantic arguments and syntactic constituents is not always possible. One semantic argument may be mapped to many syntactic constituents due to: (a) intrinsic differences between the syntactic and se- mantic representations, and (b) incorrect syntactic structure. Figure 1 illustrates each one of these sit- uations: Figure 1 (a) shows a sentence where each semantic argument correctly maps to one syntac- tic constituent; Figure 1 (b) illustrates the situation where one semantic argument correctly maps to two syntactic constituents; and Figure 1 (c) shows a one- to-many mapping caused by an incorrect syntactic structure: argument A0 maps to two phrases, the ter- minal “by” and the noun phrase “Robert Goldberg”, due to the incorrect attachment of the last preposi- tional phrase, “at the University of California”. Using the above observations, we separate one- 221 rising consumer prices VBG NN NNS NP P A1 developed by Robert Goldberg …

Content truncated to 500 words