Extraction of Entailed Semantic Relations Through Syntax-Based Comma Resolution

Proceedings of ACL-08: HLT, pages 1030–1038, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics Extraction of Entailed Semantic Relations Through Syntax-based Comma Resolution Vivek Srikumar 1 Roi Reichart2 Mark Sammons1 Ari Rappoport2 Dan Roth1 1University of Illinois at Urbana-Champaign {vsrikum2|mssammon|danr}@uiuc.edu 2Institute of Computer Science, Hebrew University of Jerusalem {roiri|arir}@cs.huji.ac.il Abstract This paper studies textual inference by inves- tigating comma structures, which are highly frequent elements whose major role in the ex- traction of semantic relations has not been hitherto recognized. We introduce the prob- lem of comma resolution, defined as under- standing the role of commas and extracting the relations they imply. We show the importance of the problem using examples from Tex- tual Entailment tasks, and present A Sentence Transformation Rule Learner (ASTRL), a ma- chine learning algorithm that uses a syntac- tic analysis of the sentence to learn sentence transformation rules that can then be used to extract relations. We have manually annotated a corpus identifying comma structures and re- lations they entail and experimented with both gold standard parses and parses created by a leading statistical parser, obtaining F-scores of 80.2% and 70.4% respectively. 1 Introduction Recognizing relations expressed in text sentences is a major topic in NLP, fundamental in applications such as Textual Entailment (or Inference), Question Answering and Text Mining. In this paper we ad- dress this issue from a novel perspective, that of un- derstanding the role of the commas in a sentence, which we argue is a key component in sentence comprehension. Consider for example the following three sentences: 1. Authorities have arrested John Smith, a retired police officer. 2. Authorities have arrested John Smith, his friend and his brother. 3. Authorities have arrested John Smith, a retired police officer announced this morning. Sentence (1) states that John Smith is a retired police officer. The comma and surrounding sen- tence structure represent the relation ‘IsA’. In (2), the comma and surrounding structure signifies a list, so the sentence states that three people were ar- rested: (i) John Smith, (ii) his friend, and (iii) his brother. In (3), a retired police officer announced that John Smith has been arrested. Here, the comma and surrounding sentence structure indicate clause boundaries. In all three sentences, the comma and the sur- rounding sentence structure signify relations essen- tial to comprehending the meaning of the sentence, in a way that is not easily captured using lexical- or even shallow parse-level information. As a hu- man reader, we understand them easily, but auto- mated systems for Information Retrieval, Question Answering, and Textual Entailment are likely to en- counter problems when comparing structures like these, which are lexically similar, but whose mean- ings are so different. In this paper we present an algorithm for comma resolution, a task that we define to consist of (1) dis- ambiguating comma type and (2) determining the relations entailed from the sentence given the com- mas’ interpretation. Specifically, in (1) we assign each comma to one of five possible types, and in …

Content truncated to 500 words