Detection of English Grammatical Errors and Correction using Graph Dual Encoder Decoder with Pyramid Attention Network

Abstract

In English, grammatical errors pose a significant challenge, prompting the exploration of diverse detection and correction methods. Existing approaches, however, often fall short of delivering satisfactory results and achieving high accuracy. An innovative solution, the Optimized Graph Dual Encoder Decoder with Pyramid Attention (OGDED-PA), is introduced to overcome these limitations. The model utilizes the C4_200M synthetic dataset for input data, followed by preprocessing and applying hybrid Squared Root of Term Frequency Variants with Mean Semi-absolute Deviation Factors for morphological feature extraction. Bidirectional long short-term memory with conditional random field segmentation is employed, and OGDED-PA, integrating a dual encoder-decoder architecture and pyramid attention mechanism, is then applied. This model aims to enhance accuracy in identifying and correcting grammar, syntax, punctuation, and spelling errors by capturing intricate linguistic patterns. The graph-based representation leverages Improved Border Collie Optimization (IBCO) to optimize the weight parameter, allowing the model to analyze syntactic and semantic relationships and address a broad spectrum of grammatical errors. The proposed method is implemented using the Python platform. Compared to existing methods, the proposed approach achieves 99.3% accuracy, 98.7% precision and 98.6% F0.5.