SimPA: A Sentence-Level Simplification Corpus for the Public Administration Domain

SimPA: A Sentence-Level Simplification Corpus for the Public Administration Domain Carolina Scarton, Gustavo Henrique Paetzold, Lucia Specia Department of Computer Science, University of Sheffield Regent Court, 211 Portobello, Sheffield, S1 4DP, UK {c.scarton, g.h.paetzold, l.specia}@sheffield.ac.uk Abstract We present a sentence-level simplification corpus with content from the Public Administration (PA) domain. The corpus contains 1, 100 original sentences with manual simplifications collected through a two-stage process. Firstly, annotators were asked to simplify only words and phrases (lexical simplification). Each sentence was simplified by three annotators. Secondly, one lexically simplified version of each original sentence was further simplified at the syntactic level. In its current version there are 3, 300 lexically simplified sentences plus 1, 100 syntactically simplified sentences. The corpus will be used for evaluation of text simplification approaches in the scope of the EU H2020 SIMPATICO project – which focuses on accessibility of e-services in the PA domain – and beyond. The main advantage of this corpus is that lexical and syntactic simplifications can be analysed and used in isolation. The lexically simplified corpus is also multi-reference (three different simplifications per original sentence). This is an ongoing effort and our final aim is to collect manual simplifications for the entire set of original sentences, with over 10K sentences. Keywords: text simplification, simplification corpora, public administration 1. Introduction Text simplification (TS) is the task of reducing lexical and/or structural complexity of texts (Siddharthan, 2004). It is common to divide this task in two: lexical simpli- fication (LS) and syntactic simplification (SS). LS deals with the identification and replacement of difficult words or phrases, while SS focuses on making complex syntac- tic structures simpler, e.g. by changing passive into active voice or splitting a sentence with coordination in two sen- tences. LS has been widely explored in recent years. Work in- clude simple word frequency-based approaches (Carroll et al., 1998; Carroll et al., 1999; Biran et al., 2011), unsuper- vised approaches that use word embeddings (Glavaˇs and ˇStajner, 2015; Paetzold and Specia, 2016c), and supervised approaches (Horn et al., 2014; Paetzold and Specia, 2017). For SS, some approaches apply hand-crafted rules (Sid- dharthan, 2011; Candido Jr. et al., 2009; Bott et al., 2012; Brouwers et al., 2014; Barlacchi and Tonelli, 2013; Scarton et al., 2017), while others use parallel data to learn simpli- fication operations (Woodsend and Lapata, 2011; Paetzold and Specia, 2013; Siddharthan and Angrosh, 2014; Zhu et al., 2010; Coster and Kauchak, 2011; Wubben et al., 2012; Narayan and Gardent, 2014; Xu et al., 2016; Zhang and La- pata, 2017; Nisioi et al., 2017). Corpus-based approaches tend to learn both LS and SS transformations jointly. For English, two main parallel corpora exist: Simple Wikipedia (Zhu et al., 2010) and Newsela (Newsela, 2016), the latter with simplifications performed by professionals. However, such corpora do not distinguish among different types of simplification (i.e. lexical from syntactic transfor- mations). Moreover, they cover general domain texts, and therefore may not be sufficient to model operations for spe- cific domains. In terms of specific corpora …

Content truncated to 500 words