Optimizing Grammar Error Correction Performance with Attention-Based Neural Networks

The task of correcting grammatical errors, known as Grammatical Error Correction (GEC) is addressed in this research. Various errors in text such as punctuation, spelling, grammatical, and choice of words errors are detected and corrected. Communication is crucial in everyone’s daily lives, and language is key to share information in verbal and written communication. Among all languages, English has an important position due to its global use in business, education, and entertainment. Yet, mastering its grammar can be tough for learners. With rapid commercial use of Grammarly, ChatGPT and QuillBot there is a new focus on GEC domain. In this research, the seq2seq neural networks-based on attention are applied in Grammar error correction, using a special type of LSTM model with FastText embedding at the vectorization. The attention mechanism which has been so far popular in Language translation, Computer vision and sentiment analysis has now been deployed in English language grammatical error correction. Validation of the implemented model is done on Lang-8 Corpus. Results show that attention-based model optimizes Grammatical error correction with good GLUE score and less loss.