Assessing Conformance of Manually Simplified Corpora with User Requirements: the Case of Autistic Readers

Abstract

In the state of the art, there are scarce resources available to support development and evaluation of automatic text simplification (TS) systems for specific target populations. These comprise parallel corpora consisting of texts in their original form and in a form that is more accessible for different categories of target reader, including neurotypical second language learners and young readers. In this paper, we investigate the potential to exploit resources developed for such readers to support the development of a text simplification system for use by people with autistic spectrum disorders (ASD). We analysed four corpora in terms of nineteen linguistic features which pose obstacles to reading comprehension for people with ASD. The results indicate that the Britannica TS parallel corpus (aimed at young readers) and the Weekly Reader TS parallel corpus (aimed at second language learners) may be suitable for training a TS system to assist people with ASD. Two sets of classification experiments intended to discriminate between original and simplified texts according to the nineteen features lent further support for those findings.