Accurate Function Parsing

Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 620–627, Vancouver, October 2005. c⃝2005 Association for Computational Linguistics Accurate Function Parsing Paola Merlo Department of Linguistics University of Geneva 1204 Geneva Switzerland merlo@lettres.unige.ch Gabriele Musillo Depts of Linguistics and Computer Science University of Geneva 1204 Geneva Switzerland musillo4@etu.unige.ch Abstract In this paper, we extend an existing parser to produce richer output annotated with function labels. We obtain state-of-the-art results both in function labelling and in parsing, by automatically relabelling the Penn Treebank trees. In particular, we ob- tain the best published results on seman- tic function labels. This suggests that cur- rent statistical parsing methods are suffi- ciently general to produce accurate shal- low semantic annotation. 1 Introduction With recent advances in speech recognition, parsing, and information extraction, some domain-specific interactive systems are now of practical use for tasks such as question-answering, flight booking, or restaurant reservation (Stallard, 2000). One of the challenges ahead lies in moving from hand-crafted programs of limited scope to robust systems inde- pendent of a given domain. While this ambitious goal will remain in the future for some time to come, recent efforts to develop language processing sys- tems producing richer semantic outputs will likely be the cornerstone of many successful developments in natural language understanding. In this paper, we present a parser that outputs la- bels indicating the syntactic or semantic function of a constituent in the tree, such as NP-SBJ or PP-TMP shown in bold face in the tree in Figure 1. These labels indicate that the NP is the subject of the sen- tence and that the PP conveys temporal information. (Labels in parentheses will be explained later in the paper.) Output annotated with such informative la- bels underlies all domain-independent question an- S ������ HHHHH H NP-SBJ  PPP the authority VP  � �� @@ @ PPPPPPPP P VBD dropped PP-TMP �� H H IN(TMP) at NP NN midnight NP-TMP NNP(TMP) Tuesday PP-DIR �� H H TO(DIR) to NP QP  PPP $ 2.8 trillion Figure 1: A sample syntactic structure with function labels. swering (Jijkoun et al., 2004) or shallow semantic interpretation systems (Collins and Miller, 1998; Ge and Mooney, 2005). We test the hypothesis that a current statistical parser can output such richer in- formation without any degradation of the parser’s accuracy on the original parsing task. Briefly, our method consists in augmenting a state-of-the-art sta- tistical parser (Henderson, 2003), whose architec- ture and properties make it particularly adaptive to new tasks. We achieve state-of-the-art results both for parsing and function labelling. Statistical parsers trained on the Penn Treebank (PTB) (Marcus et al., 1993) produce trees annotated with bare phrase structure labels (Collins, 1999; Charniak, 2000). The trees of the Penn Treebank, however, are also decorated with function labels. Figure 1 shows the simplified tree representation with function labels for a sample sentence from the Penn Treebank corpus (section 00) The Gov- ernment’s borrowing authority dropped at midnight Tuesday to …

Content truncated to 500 words