Optimizing Arabic Text Readability Measurement Using Morphological Word Embeddings

Analyzing the readability of a document means assessing its clarity and the simplicity of its language to make it comprehensible to a wide audience. This is a central concern in linguistic research. The most widespread method for representing a document is to use discriminating feature vectors. Currently, vector representation techniques, such as Word2vec for non-contextual approaches and BERT for contextual ones, are widely adopted in this field. However, these methods are not always able to capture all the crucial morphological information needed for many natural language processing applications. This study proposes a novel approach that exploits the morphological features of words to propose a vector representation of Arabic documents. We analyzed the impact of this new vector representation on text readability prediction and compared it with the performance of other vector representations of documents, both contextual and non-contextual. The results obtained with this new representation outperformed those of other approaches, highlighting its effectiveness and relevance in predicting text readability.