Simple or Complex? Assessing the readability of Basque Texts

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 334–344, Dublin, Ireland, August 23-29 2014. Simple or Complex? Assessing the readability of Basque Texts Itziar Gonzalez-Dios, Mar´ıa Jes´us Aranzabe, Arantza D´ıaz de Ilarraza, Haritz Salaberri IXA NLP Group University of the Basque Country (UPV/EHU) itziar.gonzalezd@ehu.es Abstract In this paper we present a readability assessment system for Basque, ErreXail, which is going to be the preprocessing module of a Text Simplification system. To that end we compile two corpora, one of simple texts and another one of complex texts. To analyse those texts, we imple- ment global, lexical, morphological, morpho-syntactic, syntactic and pragmatic features based on other languages and specially considered for Basque. We combine these feature types and we train our classifiers. After testing the classifiers, we detect the features that perform best and the most predictive ones. 1 Introduction Readability assessment is a research line that aims to grade the difficulty or the ease of the texts. It has been a remarkable question in the educational domain during the last century and is of great importance in Natural Language Processing (NLP) during the last decade. Classical readability formulae like Flesh formula (Flesch, 1948), Dale-Chall formula (Chall and Dale, 1995) and The Gunning FOG index (Gun- ning, 1968) take into account raw and lexical features and frequency counts. NLP techniques, on the other hand, make possible the consideration of more complex features. Recent research in NLP (Si and Callan, 2001; Petersen and Ostendorf, 2009; Feng, 2009) has demon- strated that classical readability formulae are unreliable. Moreover, those metrics are language specific. Readability assessment is also used as a preprocess or evaluation in Text Simplification (TS) systems e.g. for English (Feng et al., 2010), Portuguese (Alu´ısio et al., 2010), Italian (Dell’Orletta et al., 2011), German (Hancke et al., 2012) and Spanish (ˇStajner and Saggion, 2013). Given a text the aim of these systems is to decide whether a text is complex or not. So, in case of being difficult, the given text should be simplified. As far as we know no specific metric has been used to calculate the complexity of Basque texts. The only exception we find is a system for the auto-evaluation of essays Idazlanen Autoebaluaziorako Sistema (IAS) (Aldabe et al., 2012) which includes metrics similar to those used in readability assessment. IAS analyses Basque texts after several criteria focused on educational correction such as the clause number in a sentence, types of sentences, word types and lemma number among others. It was foreseen to use this tool in the Basque TS system (Aranzabe et al., 2012). The present work means to add to IAS the capacity of evaluating the complexity of texts by means of new linguistic features and criteria. In this paper we present ErreXail, a readability assessment system for Basque, a Pre-Indo-European agglutinative head-final pro-drop language, which displays a rich inflectional morphology and whose orthography is phonemic. ErreXail classifies the texts and decides if they should be simplified or not. This …

Content truncated to 500 words