NAME README - [documentation] Introduction to WordNet::Similarity SYNOPSIS There are a number of documentation files covering different aspects of WordNet::Similarity available: * intro.pod Introduction to WordNet::Similarity (this document) * install.pod How to install * utils.pod How to use the utility programs in /utls * modules.pod How the measure modules are designed * developers.pod How to write your own measure of semantic relatedness * config.pod How to set configuration options for the measures You can use pod2html, pod2latex, pod2man, or pod2text to get this documentation in a different format. See the man pages for pod2html, etc. These translators should come with Perl, but you can also download them from . DESCRIPTION This package consists of Perl modules along with supporting Perl programs that implement the semantic relatedness measures described by Leacock & Chodorow (1998), Jiang & Conrath (1997), Resnik (1995), Lin (1998), Hirst & St-Onge (1998), Wu & Palmer (1994), the extended gloss overlap measure by Banerjee and Pedersen (2002), and two measures based on context vectors by Patwardhan (2003). The details of the vector measure are described in the Master's thesis work of Patwardhan (2003) at the University of Minnesota Duluth, and the vector_pairs measure is derived from that. The Perl modules are designed as objects with methods that take as input two word senses. The semantic relatedness of these word senses is returned by these methods. A quantitative measure of the degree to which two word senses are related has wide ranging applications in numerous areas, such as word sense disambiguation, information retrieval, etc. For example, in order to determine which sense of a given word is being used in a particular context, the sense having the highest relatedness with its context word senses is most likely to be the sense being used. Similarly, in information retrieval, retrieving documents containing highly related concepts are more likely to have higher precision and recall values. A command line interface to these modules is also present in the package. The simple, user-friendly interface returns the relatedness measure of two given words. A number of switches and options have been provided to modify the output and enhance it with trace information and other useful output. Details of the usage are provided in other sections of this README. Supporting utilities for generating information content files from various corpora are also available in the package. The information content files are required by three of the measures for computing the relatedness of concepts. The following sections describe the organization of this software package and how to use it. A few typical examples are given to help clearly understand the usage of the modules and the supporting utilities. SEMANTIC RELATEDNESS We observe that humans find it extremely easy to say if two words are related and if one word is more related to a given word than another. For example, if we come across two words -- 'car' and 'bicycle', we know they are related as both are means of transport. Also, we easily observe that 'bicycle' is more related to 'car' than 'fork' is. But is there some way to assign a quantitative value to this relatedness? Some ideas have been put forth by researchers to quantify the concept of relatedness of words, with encouraging results. A number of different measures of relatedness have been implemented in this software package. These include a simple edge counting approach and a random method for measuring relatedness. The measures rely heavily on the vast store of knowledge available in the online electronic dictionary -- WordNet. So, we use a Perl interface for WordNet called WordNet::QueryData to make it easier for us to access WordNet. The modules in this package REQUIRE that the WordNet::QueryData module be installed on the system before these modules are installed. CONTENTS The package contains the semantic relatedness modules, some support Perl utilities and some sample configuration files, data files and programs. Modules All the modules that will be installed in the Perl system directory are present in the '/lib' directory tree of the package. These include the semantic relatedness modules -- jcn.pm, res.pm, lin.pm, lch.pm, hso.pm, lesk.pm, vector.pm, vector_pairs.pm, wup.pm, path.pm and random.pm -- present in the /lib/WordNet/Similarity subdirectory and the supporting modules get_wn_info.pm and string_compare.pm. There also exists a WordNet/Similarity.pm module that currently serves as a base class for all the relatedness modules and contains Perl documentation. All these modules, once installed in the Perl system directory, can be directly used by Perl programs. Supporting Utilities The '/utils' subdirectory of the package contains supporting Perl programs. 'similarity.pl' is a command-line interface to the relatedness modules. A number of Perl programs, that generate information content files from various corpora, are provided. As part of the standard install, these are also installed into the system directories, and can be accessed from any working directory if the common system directories (/usr/bin, /usr/local/bin, etc) are in your path. Samples If you downloaded this package as a tar-gzipped file from the web, you will find a '/samples' subdirectory in the package. There is a separate README in that directory. The directory contains sample configuration files for the modules, sample programs showing usage of the modules and sample data files (e.g., relation files). REFERENCES 1 Wu Z. and Palmer M. 1994. Verb Semantics and Lexical Selection. In Proceedings of the 32nd Annual Meeting of the Association for Computational Linguistics. Las Cruces, New Mexico. 2 Resnik P. 1995. Using information content to evaluate semantic similarity. In Proceedings of the 14th International Joint Conference on Artificial Intelligence, pages 448-453. Montreal. 3 Jiang J. and Conrath D. 1997. Semantic similarity based on corpus statistics and lexical taxonomy. In Proceedings of International Conference on Research in Computational Linguistics. Taiwan. 4 Fellbaum C., editor. WordNet: An electronic lexical database. MIT Press, 1998. 5 Leacock C. and Chodorow M. 1998. Combining local context and WordNet similarity for word sense identification. In Fellbaum 1998, pp. 265-283. 6 Lin D. 1998. An information-theoretic definition of similarity. In Proceedings of the 15th International Conference on Machine Learning. Madison, WI. 7 Hirst G. and St-Onge D. 1998. Lexical Chains as representations of context for the detection and correction of malapropisms. In Fellbaum 1998, pp. 305-332. 8 Schütze H. 1998. Automatic Word Sense Discrimination. Computational Linguistics, 24(1):97-123. 9 Resnik P. 1999. Semantic Similarity in a Taxonomy: An Information- Based Measure and its Applications to Problems of Ambiguity in Natural Language. Journal of Artificial Intelligence Research, 11, 95-130. 10 Budanitsky A. and Hirst G. 2001. Semantic distance in WordNet: An experimental, application-oriented evaluation of five measures. In Workshop on WordNet and Other Lexical Resources, Second meeting of the North American Chapter of the Association for Computational Linguistics. Pittsburgh, PA. 11 Banerjee S. and Pedersen T. 2002. An Adapted Lesk Algorithm for Word Sense Disambiguation Using WordNet. In Proceeding of the Fourth International Conference on Computational Linguistics and Intelligent Text Processing (CICLING-02). Mexico City. 12 Patwardhan S., Banerjee S. and Pedersen T. 2002. Using Semantic Relatedness for Word Sense Disambiguation. In Proceedings of the Fourth International Conference on Intelligent Text Processing and Computational Linguistics. Mexico City. 13 Banerjee S. and Pedersen T. 2003. Extended Gloss Overlaps as a Measure of Semantic Relatedness. In the Proceedings of the Eighteenth International Joint Conference on Artificial Intelligence. Acapulco, Mexico. 14 Patwardhan S. and Pedersen T. 2006. Using WordNet-based Context Vectors to Estimate the Semantic Relatedness of Concepts. In the Proceedings of the EACL 2006 Workshop Making Sense of Sense - Bringing Computational Linguistics and Psycholinguistics Together. Trento, Italy. 15 Banerjee S. Adapting the Lesk algorithm for word sense disambiguation to WordNet. Master Thesis, University of Minnesota, Duluth, 2002. 16 Patwardhan S. Incorporating dictionary and corpus information into a vector measure of semantic relatedness. Master Thesis, University of Minnesota, Duluth, 2003. SEE ALSO intro.pod Mailing list: Project Home page: ACKNOWLEDGEMENTS We would like to thank the following for their support and contribution towards the development of this package. We thank Jason Rennie for his QueryData package, the WordNet guys at Princeton for WordNet, Resnik, Hirst, St-Onge, Jiang, Conrath, Lin, Wu, Palmer, Leacock, and Chodorow for their algorithms and work on the relatedness measures. We also thank Bano (Satanjeev Banerjee) for his work on the extended gloss overlap module. We are grateful to Wybo Wiersma for contributing his optimizations to the GlossFinder code. We also appreciate the many helpful suggestions and bug patches from Ben Haskell. AUTHORS Ted Pedersen, University of Minnesota Duluth tpederse at d.umn.edu Siddharth Patwardhan, University of Utah, Salt Lake City sidd at cs.utah.edu Satanjeev Banerjee, Carnegie Mellon University, Pittsburgh banerjee+ at cs.cmu.edu Jason Michelizzi COPYRIGHT Copyright (c) 2005-2008, Ted Pedersen, Siddharth Patwardhan, Satanjeev Banerjee, and Jason Michelizzi Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. Note: a copy of the GNU Free Documentation License is available on the web at and is included in this distribution as FDL.txt.