Subject–Verb Agreement Error Handling Using RNN Architectures

The domain of natural language processing has reached to a level where various tools are available for automatically identifying and correcting grammatical errors in a wide range of languages text. Though these tools are productive, it is still difficult to develop an effective and accurate grammatical error handling system. In this paper, we examine the efficacy of several neural network architectures with a particular focus on subject–verb agreement, a rule of English grammar. This work investigates the capabilities and performance of deep learning models using an annotated dataset designed specifically for English language subject–verb agreement issues. The strengths and weaknesses of well-known models, including Support Vector Machine (SVM), Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), and Transformers (BERT), are assessed through testing. The study sheds light on how the performances of these models differ from one another in detecting subject–verb agreement errors in English language sentences.