Sentence Simplification with Deep Reinforcement Learning

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 584–594 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Sentence Simplification with Deep Reinforcement Learning Xingxing Zhang and Mirella Lapata Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh 10 Crichton Street, Edinburgh EH8 9AB x.zhang@ed.ac.uk,mlap@inf.ed.ac.uk Abstract Sentence simplification aims to make sen- tences easier to read and understand. Most recent approaches draw on insights from machine translation to learn simpli- fication rewrites from monolingual cor- pora of complex and simple sentences. We address the simplification problem with an encoder-decoder model coupled with a deep reinforcement learning frame- work. Our model, which we call DRESS (as shorthand for Deep REinforcement Sentence Simplification), explores the space of possible simplifications while learning to optimize a reward function that encourages outputs which are simple, flu- ent, and preserve the meaning of the in- put. Experiments on three datasets demon- strate that our model outperforms compet- itive simplification systems.1 1 Introduction The main goal of sentence simplification is to re- duce the linguistic complexity of text, while still retaining its original information and meaning. The simplification task has been the subject of several modeling efforts in recent years due to its relevance for NLP applications and individ- uals alike (Siddharthan, 2014; Shardlow, 2014). For instance, a simplification component could be used as a preprocessing step to improve the perfor- mance of parsers (Chandrasekar et al., 1996), sum- marizers (Beigman Klebanov et al., 2004), and se- mantic role labelers (Vickrey and Koller, 2008; Woodsend and Lapata, 2014). Automatic simplifi- cation would also benefit people with low-literacy skills (Watanabe et al., 2009), such as children and 1Our code and data are publicly available at https:// github.com/XingxingZhang/dress. non-native speakers as well as individuals with autism (Evans et al., 2014), aphasia (Carroll et al., 1999), or dyslexia (Rello et al., 2013). The most prevalent rewrite operations which give rise to simplified text include substituting rare words with more common words or phrases, ren- dering syntactically complex structures simpler, and deleting elements of the original text (Sid- dharthan, 2014). Earlier work focused on individ- ual aspects of the simplification problem. For ex- ample, several systems performed syntactic sim- plification only, using rules aimed at sentence splitting (Carroll et al., 1999; Chandrasekar et al., 1996; Vickrey and Koller, 2008; Siddharthan, 2004) while others turned to lexical simplification by substituting difficult words with more common WordNet synonyms or paraphrases (Devlin, 1999; Inui et al., 2003; Kaji et al., 2002). Recent approaches view the simplification pro- cess more holistically as a monolingual text- to-text generation task borrowing ideas from statistical machine translation. Simplification rewrites are learned automatically from exam- ples of complex-simple sentences extracted from online resources such as the ordinary and sim- ple English Wikipedia. For example, Zhu et al. (2010) draw inspiration from syntax-based translation and propose a model similar to Ya- mada and Knight (2001) which additionally per- forms simplification-specific rewrite operations (e.g., sentence splitting). Woodsend and Lapata (2011) formulate simplification …

Content truncated to 500 words