Soft Cardinality + ML: Learning Adaptive Similarity Functions for Cross-lingual Textual Entailment

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 684–688, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics Soft Cardinality + ML: Learning Adaptive Similarity Functions for Cross-lingual Textual Entailment Sergio Jimenez Universidad Nacional de Colombia, Bogota, Ciudad Universitaria edificio 453, oficina 220 sgjimenezv@unal.edu.co Claudia Becerra Universidad Nacional de Colombia, Bogota cjbecerrac@unal.edu.co Alexander Gelbukh CIC-IPN Av. Juan Dios Bátiz, Av. Mendizábal, Col. Nueva Industrial Vallejo, CP 07738, DF, México gelbukh@gelbukh.com Abstract This paper presents a novel approach for building adaptive similarity functions based on cardinality us- ing machine learning. Unlike current approaches that build feature sets using similarity scores, we have developed these feature sets with the cardinal- ities of the commonalities and differences between pairs of objects being compared. This approach al- lows the machine-learning algorithm to obtain an asymmetric similarity function suitable for direc- tional judgments. Besides using the classic set cardi- nality, we used soft cardinality to allow flexibility in the comparison between words. Our approach used only the information from the surface of the text, a stop-word remover and a stemmer to address the cross-lingual textual entailment task 8 at SEMEVAL 2012. We have the third best result among the 29 systems submitted by 10 teams. Additionally, this paper presents better results compared with the best official score. 1 Introduction Adaptive similarity functions are those functions that, be- yond using the information of two objects being com- pared, use information from a broader set of objects (Bilenko and Mooney, 2003). Therefore, the same sim- ilarity function may return different results for the same pair of objects, depending on the context of where the objects are. Adaptability is intended to improve the per- formance of the similarity function in relation to the task in question associated with the entire set of objects. For example, adaptiveness improves relevance of documents retrieved for a query in an information retrieval task for a particular document collection. In text applications there are mainly three methods to provide adaptiveness to similarity functions: term weighting, adjustment or learning the parameters of the similarity function, and machine learning. Term weight- ing is a common practice that assigns a degree of im- portance to each occurrence of a term in a text collec- tion (Salton and Buckley, 1988; Lan et al., 2005). Sec- ondly, if a similarity function has parameters, these can be adjusted or learned to adapt to a particular data set. Depending on the size of the search space defined by these parameters, they can be adjusted either manually or using a technique of AI. For instance, Jimenez et al. manually adjusted a single parameter in the gener- alized measure of Monge-Elkan (1996) (Jimenez et al., 2009) and Ristrad and Yanilios (1998) learned the costs of editing operations between particular characters for the Levenshtein distance (1966) using HMMs. Thirdly, the machine-learning approach aims to learn a similar- ity function based on a vector representation of texts us- ing a subset of texts for training and a learning func- tion (Bilenko …

Content truncated to 500 words