UNT-SimpRank: Systems for Lexical Simplification Ranking

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 493–496, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics UNT-SIMPRANK: Systems for Lexical Simplification Ranking Ravi Sinha University of North Texas 1155 Union Circle #311277 Denton, Texas 76203-5017 RaviSinha@my.unt.edu Abstract This paper presents three systems that took part in the lexical simplification task at SE- MEVAL 2012. Speculating on what the con- cept of simplicity might mean for a word, the systems apply different approaches to rank the given candidate lists. One of the systems performs second-best (statistically significant) and another one performs third-best out of 9 systems and 3 baselines. Notably, the third- best system is very close to the second-best, and at the same time much more resource-light in comparison. 1 Introduction Lexical simplification (described in (Specia et al., 2012)) is a newer problem that has arisen follow- ing a recent surge in interest in the related task of lexical substitution (McCarthy et al., 2007). While lexical substitution aims at making systems generate suitable paraphrases for a target word in an instance, which do not necessarily have to be simpler versions of the original, it has been speculated that one pos- sible use of the task could be lexical simplification, in particular in the realm of making educational text more readable for non-native speakers. The task of lexical simplification, which thus derives from lexical substitution, uses the same data set, and has been introduced at the 6th International Workshop on Semantic Evaluation (SEMEVAL 2012), in conjunction with the First Joint Conference on Lexical and Computational Seman- tics (*SEM 2012). Instead of asking systems to pro- vide substitutes, the task provides the systems with all substitutes and asks them to be ranked. The task provides several instances of triplets of a context C, a target word T, and a set of gold stan- dard substitutes S. The systems are supposed to rank the substitutes si ∈ S from the simplest to the most difficult, and match their predictions against the provided human annotations. The organizers de- fine simple loosely as words that can be understood by a wide variety of people, regardless of their lit- eracy and cognitive levels, age, and regional back- grounds. The task is novel in that so far most work has been done on syntactic simplification and not on lexical simplification. Carroll et. al. (Carroll et al., 1998) seem to have pioneered some methodology and eval- uation metrics in this field. Yatskar et. al. (Yatskar et al., 2010) use an unsupervised learning method and metadata from the Simple English Wikipedia. 2 Data The data (trial and test, no training) have been adopted from the original lexical substitution task (McCarthy et al., 2007). The trial set has 300 examples, each with a context, a target word, and a set of substitutions. The test set has 1710 exam- ples. The organizers provide a scorer for the task, the trial gold standard rankings, and three baselines. The data is provided in XML format, with tags iden- …

Content truncated to 500 words