A Monolingual Tree-based Translation Model for Sentence Simplification

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 1353–1361, Beijing, August 2010 A Monolingual Tree-based Translation Model for Sentence Simplification∗ Zhemin Zhu1 Department of Computer Science Technische Universit¨at Darmstadt Delphine Bernhard2 LIMSI-CNRS Iryna Gurevych1 Department of Computer Science Technische Universit¨at Darmstadt 1http://www.ukp.tu-darmstadt.de 2delphine.bernhard@limsi.fr Abstract In this paper, we consider sentence sim- plification as a special form of translation with the complex sentence as the source and the simple sentence as the target. We propose a Tree-based Simplification Model (TSM), which, to our knowledge, is the first statistical simplification model covering splitting, dropping, reordering and substitution integrally. We also de- scribe an efficient method to train our model with a large-scale parallel dataset obtained from the Wikipedia and Simple Wikipedia. The evaluation shows that our model achieves better readability scores than a set of baseline systems. 1 Introduction Sentence simplification transforms long and dif- ficult sentences into shorter and more readable ones. This helps humans read texts more easily and faster. Reading assistance is thus an impor- tant application of sentence simplification, espe- cially for people with reading disabilities (Carroll et al., 1999; Inui et al., 2003), low-literacy read- ers (Watanabe et al., 2009), or non-native speakers (Siddharthan, 2002). Not only human readers but also NLP ap- plications can benefit from sentence simplifica- tion. The original motivation for sentence sim- plification is using it as a preprocessor to facili- tate parsing or translation tasks (Chandrasekar et al., 1996). Complex sentences are considered as stumbling blocks for such systems. More recently, sentence simplification has also been shown help- ful for summarization (Knight and Marcu, 2000), ∗ This work has been supported by the Emmy Noether Program of the German Research Foundation (DFG) under the grant No. GU 798/3-1, and by the Volkswagen Founda- tion as part of the Lichtenberg-Professorship Program under the grant No. I/82806. sentence fusion (Filippova and Strube, 2008b), se- mantic role labeling (Vickrey and Koller, 2008), question generation (Heilman and Smith, 2009), paraphrase generation (Zhao et al., 2009) and biomedical information extraction (Jonnalagadda and Gonzalez, 2009). At sentence level, reading difficulty stems ei- ther from lexical or syntactic complexity. Sen- tence simplification can therefore be classified into two types: lexical simplification and syntac- tic simplification (Carroll et al., 1999). These two types of simplification can be further implemented by a set of simplification operations. Splitting, dropping, reordering, and substitution are widely accepted as important simplification operations. The splitting operation splits a long sentence into several shorter sentences to decrease the complex- ity of the long sentence. The dropping operation further removes unimportant parts of a sentence to make it more concise. The reordering operation interchanges the order of the split sentences (Sid- dharthan, 2006) or parts in a sentence (Watanabe et al., 2009). Finally, the substitution operation re- places difficult phrases or words with their simpler synonyms. In most cases, different simplification opera- tions happen simultaneously. It is therefore nec- essary to consider the simplification process as a combination of different operations and treat them …

Content truncated to 500 words