A Survey on Machine Translation of Low-Resource Arabic Dialects

For many years, machine translation has been one of the most studied topics for AI researchers, this initially started with simple and traditional approaches like statistical machine translation (SMT) and rule-based machine translation. The new era of deep learning and the new advancements in the AI techniques like neural approaches have set new benchmarks in machine translation accuracy. As a result, many researchers have shifted their focus from developing statistical machine translation and rule-based machine translation to work on the neural machine translation (NMT) methods due to its superior performance with limited parallel data, known as the low-resource setting. Over the past decade, NMT has experienced rapid growth and reached a mature stage. Despite being the most widely adopted solution for machine translation, its performance on low-resource language pairs remains sub-optimal compared to high-resource counterparts due to the scarcity of large parallel corpora. Consequently, there has been a significant focus on implementing NMT techniques for low-resource languages and dialects pairs, attracting research attention. This literature survey traces the origins and the development timeline for methods of MT of low-resource Arabic dialects, examines key branches, categorizes various research orientations, and discusses the research trends in this field.