Semantic Role Labelling of Prepositional Phrases

Semantic Role Labelling of Prepositional Phrases Patrick Ye1 and Timothy Baldwin1,2 1 Department of Computer Science and Software Engineering, University of Melbourne, VIC 3010, Australia 2 NICTA Victoria Laboratories, University of Melbourne, VIC 3010, Australia {jingy, tim}@cs.mu.oz.au Abstract. We propose a method for labelling prepositional phrases ac- cording to two different semantic role classifications, as contained in the Penn treebank and the CoNLL 2004 Semantic Role Labelling data set. Our results illustrate the difficulties in determining preposition seman- tics, but also demonstrate the potential for PP semantic role labelling to improve the performance of a holistic semantic role labelling system. 1 Introduction Prepositional phrases (PPs) are both common and semantically varied in open English text. Learning the semantics of prepositions is not a trivial task in gen- eral. It may seem that the semantics of a given PP can be predicted with rea- sonable reliability independent of its context. However, it is actually common for prepositions or even identical PPs to exhibit a wide range of semantic fuctions in different open English contexts. For example, consider the PP to the car: this PP will generally occur as a directional adjunct (e.g. walk to the car), but it can also occur as an object to the verb (e.g. refer to the car) or contrastive argu- ment (e.g. the default mode of transport has shifted from the train to the car); to further complicate the situation, in key to the car it functions as a complement to the N-bar key. Based on this observation, we may consider the possibility of constructing a semantic tagger specifically for PPs, which uses the surrounding context of the PP to arrive at a semantic analysis. It is this task of PP semantic role labelling that we target in this paper. A PP semantic role labeller would allow us to take a document and identify all adjunct PPs with their semantics. We would expect this to include a large portion of locative and temporal expressions, e.g., in the document, providing valuable data for tasks such as information extraction and question answering. Indeed our initial foray into PP semantic role labelling relates to an interest in geospatial and temporal analysis, and the realisation of the importance of PPs in identifying and classifying spatial and temporal references. The contributions of this paper are to propose a method for PP semantic role labelling, and evaluate its performance over both the Penn treebank (including comparative evaluation with previous work) and also the data from the CoNLL Semantic Role Labelling shared task. As part of this process, we identify the R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 779–791, 2005. c ⃝ Springer-Verlag Berlin Heidelberg 2005 780 P. Ye and T. Baldwin Fig. 1. An example of the preposition semantic roles in Penn Teebank level of complementarity of a dedicated PP semantic role labeller with a conven- tional holistic semantic role labeller, suggesting PP semantic role labelling as a potential avenue for boosting the performance of existing systems. 2 Preposition …

Content truncated to 500 words