Corpus-based Sentence Deletion and Split Decisions for Spanish Text Simplification

Abstract

This study addresses the automaticsimplification of texts in Spanish in order to make themmore accessible to people with cognitive disabilities.A corpus analysis of original and manually simplifiednews articles was undertaken in order to identifyand quantify relevant operations to be implementedin a text simplification system. The articles werefurther compared at sentence and text level bymeans of automatic feature extraction and variousmachine learning classification algorithms, using threedifferent groups of features (POS frequencies, syntacticinformation, and text complexity measures) with theaim of identifying features that help separate originaldocuments from their simple equivalents. Finally, itwas investigated whether these features can be usedto decide upon simplification operations to be carriedout at the sentence level (split, delete, and reduce).Automatic classification of original sentences into thoseto be kept and those to be eliminated outperformed theclassification that was previously conducted on the samecorpus. Kept sentences were further classified into thoseto be split or significantly reduced in length and thoseto be left largely unchanged, with the overall F-measureup to 0.92. Both experiments were conducted andcompared on two different sets of features: all featuresand the best subset returned by an attribute selectionalgorithm.