Simplifying Coreference Chains for Dyslexic Children

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1142–1151 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1142 Simplifying Coreference Chains for Dyslexic Children Rodrigo Wilkens, Amalia Todirascu LiLPa – University of Strasbourg rswilkens@gmail.com, todiras@unistra.fr Abstract This work aims to generate adapted content for dyslexic children for French, in the context of the ALECTOR project. Thus, we developed a system to adapt texts at the discourse level. This system modifies coreference chains, which are markers of text cohesion, by using rules. These rules were designed following a careful study of coreference chains in both original and simplified text versions. Moreover, aiming to evaluate the reliability of the proposed transformation rules, we analysed several coreference properties as well as the concurrent simplification operations in the aligned texts. This information is coded together with a coref- erence resolution system and a text rewritten tool. The proposed system comprises a coreference module specialised in written text and seven text transformation operations. Firstly, we evaluate the text simplification results by manual validation of three judges. The identified errors were grouped into five classes that combined can explain 93% of the errors. The second evaluation step con- sisted of measuring the simplification perception by 23 judges, which allow us to measure the simplification impact of the proposed rules. Keywords: Automatic Text Simplification (ATS), Dyslexic Children, Coreference Resolution, French 1. Introduction Automatic text simplification (ATS) aims to adapt content to a specific audience. Thus, it may be used to support learning activities (foreign or native language acquisition) and to generate appropriate content for people with lan- guage disabilities. Several ATS projects target specific au- diences (e.g. autism (Yaneva and Evans, 2015) and dyslexia (Rello et al., 2013)), and they have been applied to sev- eral languages, such as English (Barbu et al., 2013; Wood- send and Lapata, 2011), Spanish (Saggion et al., 2011), Portuguese (Alu´ısio et al., 2008; Wilkens et al., 2014), Japanese (Inui et al., 2003), Dutch (Ruiter et al., 2012; Bult´e et al., 2018), Italian (Dell’Orletta et al., 2011; Barlac- chi and Tonelli, 2013), and French (Seretan, 2012; Brouw- ers et al., 2014). It might address a specific language level (e.g. lexical or syntactic simplification), or it might be ap- plied at different levels at the same time. The lexical simpli- fication proposes the replacement of complex terms by sim- pler ones (Rello et al., 2013; Franc¸ois et al., 2016) (Billami et al., 2018), while the syntactic simplification intends to transforms complex syntactic structures into simpler ones (Seretan, 2012; Brouwers et al., 2014). However, these transformations might break the discourse structure by vio- lating cohesion or coherence constraints. Coherence and cohesion are text properties which help the text understanding (Hobbs, 1979; Schnedecker, 1997b), (Charolles, 2006). In addition, several cohesion devices, such as lexical chains (Hirst and St-Onge, 1995) or corefer- ence chains (Schnedecker, 1997b), are also strongly related to text readability and complexity (Pitler and Nenkova, 2008). Nevertheless, few existing systems (e.g. Sid- dharthan …

Content truncated to 500 words