Automated Spell Checking in Nepali Texts Using LSTM and BiLSTM

Abstract

Spell checking remains a fundamental requirement in natural language processing (NLP) applications ranging from word processors to chat bots. For Nepali texts, checking the spelling is crucial for Nepali grammar. This study includes using LSTM and BiLSTM models for the task of spell checking for Nepali texts. The dataset used consists of articles from online Nepali newspapers spanning various topics. Both of the models are trained to identify and correct spelling errors within a supervised learning framework. This study evaluates their performance using standard metrics including accuracy and BLUE score. The results demonstrate that the LSTM model achieved a character-level accuracy of 65.40% and a BLEU score of 0.4871 while the BiLSTM model has 75.12% character-level accuracy and a 0.5537 BLEU score respectively. The BiLSTM model outperforms the LSTM model in terms of accuracy and performance as well.