Managing Educational Natural Language Processing Projects With Hybrid Artificial Intelligence: A Case Study in Arabic Text Simplification

We present a case study in the management of an educational research project through the design of an Arabic text simplification system. The push towards AI implementations within education has produced a parallel demand for personalized, accessible content to accommodate language learners. This research project builds towards addressing these gaps, by designing and developing a lexical simplification framework that combines linguistic resources with the latest advances in deep learning models. It takes the reader through the project lifecycle while also noting the overall decisions, methodological innovations and management strategies made throughout the project lifecycle. The system produces identifications of complex words and simplified synonym options based on an innovative combination of the Arabic WordNet, lexicons manually composed, word embeddings and the pre-trained AraBERT model. The outcomes are assessed using a combination of text readability classification and semantic similarity, to produce quantifiable improvements in readability while sufficiently retaining meaning.