sranjans : Semantic Textual Similarity using Maximal Weighted Bipartite Graph Matching

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 579–585, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics sranjans : Semantic Textual Similarity using Maximal Weighted Bipartite Graph Matching Sumit Bhagwani, Shrutiranjan Satapathy, Harish Karnick Computer Science and Engineering IIT Kanpur, Kanpur - 208016, India {sumitb,sranjans,hk}@cse.iitk.ac.in Abstract The paper aims to come up with a sys- tem that examines the degree of semantic equivalence between two sentences. At the core of the paper is the attempt to grade the similarity of two sentences by find- ing the maximal weighted bipartite match between the tokens of the two sentences. The tokens include single words, or multi- words in case of Named Entitites, adjec- tivally and numerically modified words. Two token similarity measures are used for the task - WordNet based similarity, and a statistical word similarity measure which overcomes the shortcomings of WordNet based similarity. As part of three systems created for the task, we explore a simple bag of words tokenization scheme, a more careful tokenization scheme which cap- tures named entities, times, dates, mone- tary entities etc., and finally try to capture context around tokens using grammatical dependencies. 1 Introduction Semantic Textual Similarity (STS) measures the degree of semantic equivalence between texts. The goal of this task is to create a unified framework for the evaluation of semantic textual similarity modules and to characterize their im- pact on NLP applications. The task is part of the Semantic Evaluation 2012 Workshop (Agirre et al., 2012). STS is related to both Textual Entailment and Paraphrase, but differs in a number of ways and it is more directly applicable to a number of NLP tasks. Also, STS is a graded similarity notion - this graded bidirectional nature of STS is useful for NLP tasks such as MT evaluation, informa- tion extraction, question answering, and summa- rization. We propose a lexical similarity approach to grade the similarity of two sentences, where a maximal weighted bipartite match is found be- tween the tokens of the two sentences. The ap- proach is robust enough to apply across different datasets. The results on the STS test datasets are encouraging to say the least. The tokens are sin- gle word tokens in case of the first system, while in the second system, named and monetary en- tities, percentages, dates and times are handled too. A token-token similarity measure is integral to the approach and we use both a statistical sim- ilarity measure and a WordNet based word sim- ilarity measure for the same. In the final run of the task, apart from capturing the aforemen- tioned entities, we heuristically extract adjecti- vally and numerically modified words. Also, the last run naively attempts to capture the context around the tokens using grammatical dependen- cies, which in turn is used to measure context similarity. Section 2 discusses the previous work done in this area. Section 3 describes the datasets, the baseline system and the evaluation measures used by the task organizers. Section 4, 5 and 6 …

Content truncated to 500 words