Comparative Study of Sentence Completion Techniques for Low-Resource Languages: A Focus on Persian language

- Automated sentence completion and word prediction play a vital role in advancing Natural Language Processing (NLP), especially for low-resource languages. This paper offers a comparative overview of multiple deep learning approaches, including Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), transformer-based models such as BERT and ensemble and hybrid models. Drawing on insights from prior studies in diverse languages—Urdu, Arabic, Persian, Punjabi, Hindi, and English—the paper evaluates model performance, highlights data scarcity challenges, and discusses strategies for enhancing model accuracy. The comparative findings indicate that transformer-based models, particularly BERT and its variants, typically outperform traditional recurrent models and are more effective for complex sentence prediction tasks involving large datasets. However, LSTM demonstrates superior performance when applied to small and simple datasets due to its ability to capture sequential dependencies efficiently.