Text Readability for Arabic as a Foreign Language What performance to expect from simple predictors?

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.