Text readability for Arabic as a foreign language

In this study, we evaluate the informativeness of lexical, morphological and semantic features in determining the readability of texts geared towards learners of Arabic as a foreign language. We have gathered low-complexity features with the purpose of establishing a baseline for future research in readability assessment, using freely available natural language processing (NLP) and machine learning (ML) tools on a publicly accessible corpus. We tested common classification algorithms, as well as random forests-an ensemble learning method-and report on their results using several evaluation measures for comparability with similar work. Our results suggest that a small set of easily computed features can be indicative of the reading level of a text. Moreover, our findings will serve as a common ground, for ourselves and others, to evaluate and compare the performance of more elaborate techniques and feature sets.