Correcting Arabic Soft Spelling Mistakes using BiLSTM-based Machine Learning

Soft spelling mistakes are a class of that is widespread among native Arabic speakers and foreign learners alike. Some these typographical in nature. They occur due to orthographic variations some letters the complex rules dictate their correct usage. Many people forgo rules, given identical phonetic sounds, they often confuse such letters. In this paper, we investigate how use machine learning there no su?icient datasets train correction models. errors detection an active field natural language processing. We generate training using proposed transformed input approach stochastic error injec-tion approach. These approaches applied two acclaimed represent Classical Modern Standard Arabic. treat problem as character-level, one-to-one sequence transcription problem. This include omissions deletions possible with adopted simple transformations. permits bidirectional long short-term memory (BiLSTM) models more effective compared other alternatives encoder-decoder Based on investigating multiple alternatives, recommend configuration has BiLSTM layers, trained injection rate 40%. The best model corrects 96.4%of injected achieves low character 1.28% real test set soft mistakes.