UOW-SHEF: SimpLex – Lexical Simplicity Ranking based on Contextual and Psycholinguistic Features

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 477–481, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics UOW-SHEF: SimpLex – Lexical Simplicity Ranking based on Contextual and Psycholinguistic Features Sujay Kumar Jauhar Research Group in Computational Linguistics University of Wolverhampton Stafford Street, Wolverhampton WV1 1SB, UK Sujay.KumarJauhar@wlv.ac.uk Lucia Specia Department of Computer Science University of Sheffield Regent Court, 211 Portobello Sheffield, S1 4DP, UK L.Specia@dcs.shef.ac.uk Abstract This paper describes SimpLex,1 a Lexical Simplification system that participated in the English Lexical Simplification shared task at SemEval-2012. It operates on the basis of a linear weighted ranking function composed of context sensitive and psycholinguistic fea- tures. The system outperforms a very strong baseline, and ranked first on the shared task. 1 Introduction Lexical Simplification revolves around replacing words by their simplest synonym in a context aware fashion. It is similar in many respects to the task of Lexical Substitution (McCarthy and Navigli, 2007) in that it involves elements of selectional preference on the basis of a central predefined criterion (sim- plicity in the current case), as well as sensitivity to context. Lexical Simplification envisages principally a hu- man target audience, and can greatly benefit chil- dren, second language learners, people with low lit- eracy levels or cognitive disabilities, and in general facilitate the dissemination of knowledge to wider audiences. We experimented with a number of features that we posited might be inherently linked with tex- tual simplicity and selected the three that seemed the most promising on an evaluation with the trial dataset. These include contextual and psycholin- guistic components. When combined using an SVM 1Developed by co-organizers of the shared task ranker to build a model, such a model provides re- sults that offer a statistically significant improve- ment over a very strong context-independent base- line. The system ranked first overall on the Lexical Simplification task. 2 Related Work Lexical Simplification has received considerably less interest in the NLP community as compared with Syntactic Simplification. However, there are a number of notable works related to the topic. In particular Yatskar et al. (2010) leverage the relations between Simple Wikipedia and English Wikipedia to extract simplification pairs. Biran et al. (2011) extend this base methodology to apply lexi- cal simplification to input sentences. De Belder and Moens (2010), in contrast, provide a more general architecture for the task, with scope for possible ex- tension to other languages. These studies and others have envisaged a range of different target user groups including children (De Belder and Moens, 2010), people with low liter- acy levels (Aluisio et al., 2008) and aphasic readers (Carroll et al., 1998). The current work differs from previous research in that it envisages a stand-alone lexical simpli- fication system based on linguistically motivated and cognitive principles within the framework of a shared task. Its core methodology remains open to integration into a larger Text Simplification system. 3 Task Setup The English Lexical Simplification shared task at SemEval-2012 (Specia et al., 2012) required sys- 477 tems to …

Content truncated to 500 words