A Rule-Based Grammar Approach to Detect Noun Number Errors

Abstract: English is the world's most dominant language, making automated grammar checking a vital task in natural language processing. Among various grammatical mistakes, noun number errors represent a critical issue that requires accurate detection and correction. This paper presents a rule-based, feature-driven approach designed to detect and resolve noun number errors in English sentences. To evaluate performance, the method was tested on two datasets: a custom dataset of 500 sentences containing noun number errors and the NUS Corpus of Learner English (NUCLE). Additionally, the model's accuracy on the custom dataset was comparative to the widely used Grammarly application. Results: The proposed rule-based model achieved an accuracy of approximately 78% on the custom 500-sentence dataset. On the same custom dataset, the proposed method outperformed Grammarly. But, on the broader NUCLE dataset, the model achieved an accuracy of 38%. The results indicate that while a pure rule-based model effectively addresses noun number error patterns, its performance drops on complex corpus data. Future efficiency can be improved by transitioning to a hybrid model that combines rule-based methods with deep learning techniques, such as bidirectional LSTM dependency parsing.