Automatic Discovery of Manner Relations and its Applications

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 315–324, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Automatic Discovery of Manner Relations and its Applications Eduardo Blanco and Dan Moldovan Human Language Technology Research Institute The University of Texas at Dallas Richardson, TX 75080 USA {eduardo,moldovan}@hlt.utdallas.edu Abstract This paper presents a method for the auto- matic discovery of MANNER relations from text. An extended definition of MANNER is proposed, including restrictions on the sorts of concepts that can be part of its domain and range. The connections with other relations and the lexico-syntactic patterns that encode MANNER are analyzed. A new feature set spe- cialized on MANNER detection is depicted and justified. Experimental results show improve- ment over previous attempts to extract MAN- NER. Combinations of MANNER with other semantic relations are also discussed. 1 Introduction Extracting semantic relations from text is an impor- tant step towards understanding the meaning of text. Many applications that use no semantics, or only shallow semantics, could benefit by having available more text semantics. Recently, there is a growing in- terest in text semantics (M`arquez et al., 2008; Davi- dov and Rappoport, 2008). An important semantic relation for many appli- cations is the MANNER relation. Broadly speaking, MANNER encodes the mode, style, way or fashion in which something is done or happened. For ex- ample, quick delivery encodes a MANNER relation, since quick is the manner in which the delivery hap- pened. An application of MANNER detection is Question Answering, where many how questions refer to this particular relation. Consider for example the ques- tion How did the President communicate his mes- sage?, and the text Through his spokesman, Obama sent a strong message [...]. To answer such ques- tions, it is useful to identify first the MANNER rela- tions in text. MANNER occurs frequently in text and it is expressed by a wide variety of lexico-syntactic patterns. For example, PropBank annotates 8,037 ARGM-MNR relations (10.7%) out of 74,980 adjunct-like arguments (ARGMs). There are verbs that state a particular way of doing something, e.g., to limp implicitly states a particular way of walk- ing. Adverbial phrases and prepositional phrases are the most productive patterns, e.g., The nation’s industrial sector is now growing very slowly if at all and He started the company on his own. Con- sider the following example: The company said Mr. Stronach will personally direct the restructur- ing assisted by Manfred Gingl, [...]1. There are two MANNER relations in this sentence: the under- lined chunks of text encode the way in which Mr. Stronach will direct the restructuring. 2 Previous Work The extraction of semantic relations in general has caught the attention of several researchers. Ap- proaches to detect semantic relations usually focus on particular lexical and syntactic patterns. There are both unsupervised (Davidov et al., 2007; Turney, 2006) and supervised approaches. The SemEval- 2007 Task 04 (Girju et al., 2007) aimed at relations between nominals. Work has been done on detect- ing …

Content truncated to 500 words