Learning Entailment Rules for Unary Templates

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 849–856 Manchester, August 2008 Learning Entailment Rules for Unary Templates Idan Szpektor Department of Computer Science Bar-Ilan University Ramat Gan, Israel szpekti@macs.biu.ac.il Ido Dagan Department of Computer Science Bar-Ilan University Ramat Gan, Israel dagan@macs.biu.ac.il Abstract Most work on unsupervised entailment rule acquisition focused on rules between templates with two variables, ignoring unary rules - entailment rules between templates with a single variable. In this pa- per we investigate two approaches for un- supervised learning of such rules and com- pare the proposed methods with a binary rule learning method. The results show that the learned unary rule-sets outperform the binary rule-set. In addition, a novel directional similarity measure for learning entailment, termed Balanced-Inclusion, is the best performing measure. 1 Introduction In many NLP applications, such as Question An- swering (QA) and Information Extraction (IE), it is crucial to recognize whether a specific target meaning is inferred from a text. For example, a QA system has to deduce that “SCO sued IBM” is inferred from “SCO won a lawsuit against IBM” to answer “Whom did SCO sue?”. This type of reasoning has been identified as a core semantic inference paradigm by the generic Textual Entail- ment framework (Giampiccolo et al., 2007). An important type of knowledge needed for such inference is entailment rules. An entailment rule specifies a directional inference relation be- tween two templates, text patterns with variables, such as ‘X win lawsuit against Y → X sue Y ’. Applying this rule by matching ‘X win lawsuit against Y ’ in the above text allows a QA system to c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. infer ‘X sue Y ’ and identify “IBM”, Y ’s instantia- tion, as the answer for the above question. Entail- ment rules capture linguistic and world-knowledge inferences and are used as an important building block within different applications, e.g. (Romano et al., 2006). One reason for the limited performance of generic semantic inference systems is the lack of broad-scale knowledge-bases of entailment rules (in analog to lexical resources such as WordNet). Supervised learning of broad coverage rule-sets is an arduous task. This sparked intensive research on unsupervised acquisition of entailment rules (and similarly paraphrases) e.g. (Lin and Pantel, 2001; Szpektor et al., 2004; Sekine, 2005). Most unsupervised entailment rule acquisition methods learn binary rules, rules between tem- plates with two variables, ignoring unary rules, rules between unary templates (templates with only one variable). However, a predicate quite of- ten appears in the text with just a single variable (e.g. intransitive verbs or passives), where infer- ence requires unary rules, e.g. ‘X take a nap → X sleep’ (further motivations in Section 3.1). In this paper we focus on unsupervised learn- ing of unary entailment rules. Two learning ap- proaches are proposed. In our main approach, rules are learned by measuring how similar the variable instantiations of two templates in a corpus …

Content truncated to 500 words