Instance Based Lexical Entailment for Ontology Population

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 248–256, Prague, June 2007. c⃝2007 Association for Computational Linguistics Instance Based Lexical Entailment for Ontology Population Claudio Giuliano and Alfio Gliozzo FBK-irst, Istituto per la Ricerca Scientifica e Tecnologica I-38050, Trento, ITALY {giuliano,gliozzo}@itc.it Abstract In this paper we propose an instance based method for lexical entailment and apply it to automatic ontology population from text. The approach is fully unsupervised and based on kernel methods. We demonstrate the effectiveness of our technique largely surpassing both the random and most fre- quent baselines and outperforming current state-of-the-art unsupervised approaches on a benchmark ontology available in the liter- ature. 1 Introduction Textual entailment is formally defined as a relation- ship between a coherent text T and a language ex- pression, the hypothesis H. T is said to entail H, denoted by T → H, if the meaning of H can be in- ferred from the meaning of T (Dagan et al., 2005; Dagan and Glickman., 2004). Even though this no- tion has been recently proposed in the computational linguistics literature, it has already attracted a great attention due to the very high generality of its set- tings and to the indubitable usefulness of its (poten- tial) applications. In this paper, we concentrate on the problem of lexical entailment, a textual entailment subtask in which the system is asked to decide whether the sub- stitution of a particular word w with the word e in a coherent text Hw = HlwHr generates a sentence He = HleHr such that Hw → He, where Hl and Hr denote the left and the right context of w, re- spectively. For example, given the word ‘weapon’ a system may substitute it with the synonym ‘arm’, in order to identify relevant texts that denote the sought concept using the latter term. A particular case of lexical entailment is recognizing synonymy, where both Hw → He and He → Hw hold. In the literature, slight variations of this problem are also referred to as sense matching (Dagan et al., 2006), lexical reference (Glickman et al., 2006a) and lexical substitution (Glickman et al., 2006b). They have been applied to a wide variety of tasks, such as semantic matching, subtitle generation and Word Sense Disambiguation (WSD). Modeling lex- ical entailment is also a prerequisite to approach the SemEval-2007 lexical substitution task1, consisting of finding alternative words that can occur in given context. In this paper, we propose to apply an approach for lexical entailment to the ontology population task. The basic idea is that if a word entails another one in a given context then the former is an instance or a subclass of the latter. This approach is intuitively appealing because lexical entailment is intrinsically an unsupervised task, therefore it does not require lexical resources, seed examples or manually anno- tated data sets. Unsupervised approaches are partic- ularly suited for ontology population, whose goal is to find instances of concepts …

Content truncated to 500 words