FCICU at SemEval-2017 Task 1: Sense-Based Language Independent Semantic Textual Similarity Approach

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 125–129, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics FCICU at SemEval-2017 Task 1: Sense-Based Language Independent Semantic Textual Similarity Approach Basma Hassan1 Samir AbdelRahman2 Reem Bahgat2 Ibrahim Farag2 1Faculty of Computers and Information Fayoum University, Fayoum, Egypt bhassan@fayoum.edu.eg 2 Faculty of Computers and Information Cairo University, Giza, Egypt {s.abdelrahman, r.bahgat, i.farag}@fci-cu.edu.eg Abstract This paper describes FCICU team systems that participated in SemEval-2017 Seman- tic Textual Similarity task (Task1) for monolingual and cross-lingual sentence pairs. A sense-based language independent textual similarity approach is presented, in which a proposed alignment similarity method coupled with new usage of a se- mantic network (BabelNet) is used. Addi- tionally, a previously proposed integration between sense-based and surface-based semantic textual similarity approach is ap- plied together with our proposed approach. For all the tracks in Task1, Run1 is a string kernel with alignments metric and Run2 is a sense-based alignment similarity meth- od. The first run is ranked 10th, and the second is ranked 12th in the primary track, with correlation 0.619 and 0.617 respec- tively. 1 Introduction Semantic Textual Similarity (STS) is the task of measuring the similarity between two short texts semantically. STS is very important because a wide range of Natural Language Processing (NLP) applications rely heavily on such task. This paper describes our participation in the STS task (Task1) at SemEval 2017 in all the six monolingual and cross-lingual tracks (Cer et al., 2017). The STS task seeks to calculate a graded similarity score from 0 to 5 between two sentenc- es according to their meaning, i.e. semantically. The monolingual tracks are Arabic, English, and Spanish sentence-pairs (track1, track3, and track5 respectively), while the cross-lingual tracks are Arabic, Spanish, and Turkish sentences paired with English sentences (track2, track4a-4b, and track6 respectively). An additional Primary track is provided that presents the mean score of the re- sults of all the other tracks. The similarity between two natural language sentences can be inferred from the quanti- ty/quality of aligned constituents in both sentenc- es. Such alignments provide valuable information regarding how and to what extent the two sen- tences are related or semantically similar, where semantically equivalent text pairs are likely to have a successful alignment between their words. Our proposed sense-based approach employs this aspect to calculate the similarity between sen- tence-pairs regardless of their language. This is achieved through a proposed word-sense aligner that relies mainly on a new usage of the semantic network BabelNet. BabelNet utilization compen- sates the need of a machine translation module that is most commonly used to transfer cross- lingual STS to monolingual. Besides, the pro- posed sense-based similarity score is combined with a surface-based similarity score. The paper is organized as follows. Section 2 explains our main multilingual sense-based align- er. Section 3 describes our system that participated in all tracks. Section 4 shows the experiments conducted and analyzes the results achieved. Sec- tion 5 concludes the paper and mentions …

Content truncated to 500 words