Probabilistic Soft Logic for Semantic Textual Similarity

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 1210–1219, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Probabilistic Soft Logic for Semantic Textual Similarity Islam Beltagy§ Katrin Erk† Raymond Mooney§ §Department of Computer Science †Department of Linguistics The University of Texas at Austin Austin, Texas 78712 §{beltagy,mooney}@cs.utexas.edu †katrin.erk@mail.utexas.edu Abstract Probabilistic Soft Logic (PSL) is a re- cently developed framework for proba- bilistic logic. We use PSL to combine logical and distributional representations of natural-language meaning, where distri- butional information is represented in the form of weighted inference rules. We ap- ply this framework to the task of Seman- tic Textual Similarity (STS) (i.e. judg- ing the semantic similarity of natural- language sentences), and show that PSL gives improved results compared to a pre- vious approach based on Markov Logic Networks (MLNs) and a purely distribu- tional approach. 1 Introduction When will people say that two sentences are sim- ilar? This question is at the heart of the Semantic Textual Similarity task (STS)(Agirre et al., 2012). Certainly, if two sentences contain many of the same words, or many similar words, that is a good indication of sentence similarity. But that can be misleading. A better characterization would be to say that if two sentences use the same or similar words in the same or similar relations, then those two sentences will be judged similar.1 Interest- ingly, this characterization echoes the principle of compositionality, which states that the meaning of a phrase is uniquely determined by the meaning of its parts and the rules that connect those parts. Beltagy et al. (2013) proposed a hybrid ap- proach to sentence similarity: They use a very 1Mitchell and Lapata (2008) give an amusing example of two sentences that consist of all the same words, but are very different in their meaning: (a) It was not the sales manager who hit the bottle that day, but the office worker with the serious drinking problem. (b) That day the office manager, who was drinking, hit the problem sales worker with a bottle, but it was not serious. deep representation of sentence meaning, ex- pressed in first-order logic, to capture sentence structure, but combine it with distributional sim- ilarity ratings at the word and phrase level. Sen- tence similarity is then modelled as mutual entail- ment in a probabilistic logic. This approach is in- teresting in that it uses a very deep and precise representation of meaning, which can then be re- laxed in a controlled fashion using distributional similarity. But the approach faces large hurdles in practice, stemming from efficiency issues with the Markov Logic Networks (MLN) (Richardson and Domingos, 2006) that they use for performing probabilistic logical inference. In this paper, we use the same combined logic- based and distributional framework as Beltagy et al., (2013) but replace Markov Logic Networks with Probabilistic Soft Logic (PSL) (Kimmig et al., 2012; Bach et al., 2013). PSL is a proba- bilistic logic framework designed to have efficient inference. Inference …

Content truncated to 500 words