Split and Rephrase

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 606–616 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Split and Rephrase Shashi Narayan† Claire Gardent‡ Shay B. Cohen† Anastasia Shimorina‡ † School of Informatics, University of Edinburgh, 10 Crichton Street, Edinburgh EH8 9AB, UK ‡ CNRS, LORIA, UMR 7503, Vandoeuvre-l`es-Nancy, F-54500, France shashi.narayan@ed.ac.uk claire.gardent@loria.fr scohen@inf.ed.ac.uk anastasia.shimorina@loria.fr Abstract We propose a new sentence simplifica- tion task (Split-and-Rephrase) where the aim is to split a complex sentence into a meaning preserving sequence of shorter sentences. Like sentence simplification, splitting-and-rephrasing has the potential of benefiting both natural language pro- cessing and societal applications. Because shorter sentences are generally better pro- cessed by NLP systems, it could be used as a preprocessing step which facilitates and improves the performance of parsers, semantic role labelers and machine trans- lation systems. It should also be of use for people with reading disabilities be- cause it allows the conversion of longer sentences into shorter ones. This paper makes two contributions towards this new task. First, we create and make available a benchmark consisting of 1,066,115 tu- ples mapping a single complex sentence to a sequence of sentences expressing the same meaning.1 Second, we propose five models (vanilla sequence-to-sequence to semantically-motivated models) to under- stand the difficulty of the proposed task. 1 Introduction Several sentence rewriting operations have been extensively discussed in the literature: sen- tence compression, multi-sentence fusion, sen- tence paraphrasing and sentence simplification. Sentence compression rewrites an input sen- tence into a shorter paraphrase (Knight and Marcu, 2000; Cohn and Lapata, 2008; Filippova and 1The Split-and-Rephrase dataset is available here: https://github.com/shashiongithub/ Split-and-Rephrase. Strube, 2008; Pitler, 2010; Filippova et al., 2015; Toutanova et al., 2016). Sentence fusion consists of combining two or more sentences with over- lapping information content, preserving common information and deleting irrelevant details (McK- eown et al., 2010; Filippova, 2010; Thadani and McKeown, 2013). Sentence paraphrasing aims to rewrite a sentence while preserving its mean- ing (Dras, 1999; Barzilay and McKeown, 2001; Bannard and Callison-Burch, 2005; Wubben et al., 2010; Mallinson et al., 2017). Finally, sentence (or text) simplification aims to produce a text that is easier to understand (Siddharthan et al., 2004; Zhu et al., 2010; Woodsend and Lapata, 2011; Wubben et al., 2012; Narayan and Gardent, 2014; Xu et al., 2015; Narayan and Gardent, 2016; Zhang and La- pata, 2017). Because the vocabulary used, the length of the sentences and the syntactic structures occurring in a text are all factors known to af- fect readability, simplification systems mostly fo- cus on modelling three main text rewriting opera- tions: simplifying paraphrasing, sentence splitting and deletion. We propose a new sentence simplification task, which we dub Split-and-Rephrase, where the goal is to split a complex input sentence into shorter sentences while preserving meaning. In that task, the emphasis is on sentence splitting and rephras- ing. There is no deletion and no lexical or phrasal simplification but the systems must learn to split complex sentences into shorter ones and …

Content truncated to 500 words