An Automatic Modern Standard Arabic Text Simplification System: A Corpus-Based Approach

This thesis brings together an overview of Text Readability (TR) about Text Simplification (TS) with an application of both to Modern Standard Arabic (MSA). It will present our findings on using automatic TR and TS tools to teach MSA, along with challenges, limitations, and recommendations about enhancing the TR and TS models. Reading is one of the most vital tasks that provide language input for communication and comprehension skills. It is proved that the use of long sentences, connected sentences, embedded phrases, passive voices, non- standard word orders, and infrequent words can increase the text difficulty for people with low literacy levels, as well as second language learners. The thesis compares the use of sentence embeddings of different types (fastText, mBERT, XLM-R and Arabic-BERT), as well as traditional language features such as POS tags, dependency trees, readability scores and frequency lists for language learners. The accuracy of the 3-way CEFR (The Common European Framework of Reference for Languages Proficiency Levels) classification is F-1 of 0.80 and 0.75 for Arabic-Bert and XLM-R classification, respectively and 0.71 Spearman correlation for the regression task. At the same time, the binary difficulty classifier reaches F-1 0.94 and F-1 0.98 for the sentence-pair semantic similarity classifier. TS is an NLP task aiming to reduce the linguistic complexity of the text while maintaining its meaning and original information (Siddharthan, 2002; Camacho Collados, 2013; Saggion, 2017). The simplification study experimented using two approaches: (i) a classification approach and (ii) a generative approach. It then evaluated the effectiveness of these methods using the BERTScore (Zhang et al., 2020) evaluation metric. The simple sentences produced by the mT5 model achieved P 0.72, R 0.68 and F-1 0.70 via BERTScore while combining Arabic- BERT and fastText achieved P 0.97, R 0.97 and F-1 0.97. To reiterate, this research demonstrated the effectiveness of the implementation of a corpus-based method combined with extracting extensive linguistic features via the latest NLP techniques. It provided insights which can be of use in various Arabic corpus studies and NLP tasks such as translation for educational purposes.