Modeling Semantic Relations Expressed by Prepositions

Transactions of the Association for Computational Linguistics, 1 (2013) 231–242. Action Editor: Noah Smith. Submitted 11/2012; Revised 2/2013; Published 5/2013. c⃝2013 Association for Computational Linguistics. Modeling Semantic Relations Expressed by Prepositions Vivek Srikumar and Dan Roth University of Illinois, Urbana-Champaign Urbana, IL. 61801. {vsrikum2, danr}@illinois.edu Abstract This paper introduces the problem of predict- ing semantic relations expressed by preposi- tions and develops statistical learning models for predicting the relations, their arguments and the semantic types of the arguments. We define an inventory of 32 relations, build- ing on the word sense disambiguation task for prepositions and collapsing related senses across prepositions. Given a preposition in a sentence, our computational task to jointly model the preposition relation and its argu- ments along with their semantic types, as a way to support the relation prediction. The an- notated data, however, only provides labels for the relation label, and not the arguments and types. We address this by presenting two mod- els for preposition relation labeling. Our gen- eralization of latent structure SVM gives close to 90% accuracy on relation labeling. Further, by jointly predicting the relation, arguments, and their types along with preposition sense, we show that we can not only improve the re- lation accuracy, but also significantly improve sense prediction accuracy. 1 Introduction This paper addresses the problem of predicting se- mantic relations conveyed by prepositions in text. Prepositions express many semantic relations be- tween their governor and object. Predicting these can help advancing text understanding tasks like question answering and textual entailment. Consider the sentence: (1) The book of Prof. Alexander on primary school methods is a valuable teaching resource. Here, the preposition on indicates that the book and primary school methods are connected by the relation Topic and of indicates the Creator- Creation relation between Prof. Alexander and the book. Predicting these relations can help answer questions about the subject of the book and also rec- ognize the entailment of sentences like Prof. Alexan- der has written about primary school methods. Being highly polysemous, the same preposition can indicate different kinds of relations, depending on its governor and object. Furthermore, several prepositions can indicate the same semantic relation. For example, consider the sentence: (2) Poor care led to her death from pneumonia. The preposition from in this sentence expresses the relation Cause(death, pneumonia). In a differ- ent context, it can denote other relations, as in the phrases copied from the film (Source) and recog- nized from the start (Temporal). On the other hand, the relation Cause can be expressed by sev- eral prepositions; for example, the following phrases express a Cause relation: died of pneumonia and tired after the surgery. We characterize semantic relations expressed by transitive prepositions and develop accurate models for predicting the relations, identifying their argu- ments and recognizing the semantic types of the ar- guments. Building on the word sense disambigua- tion task for prepositions, we collapse semantically related senses across prepositions to derive our re- lation inventory. These relations act as predicates in …

Content truncated to 500 words