Introducing a fusion model of language content attention mechanisms and structural embeddings to achieve automatic scoring of English writing

This paper proposes a novel framework for automatic scoring of English writing that combines a Fusion Language Content Attention Model (FLCAM) with an innovative fusion strategy to improve scoring accuracy and interpretability. The FLCAM captures the intricate interplay between semantic meaning and structural organization by integrating language content attention mechanisms with structural embeddings. Using a pre-trained language model, the system transforms tokens into semantic representations while structural embeddings encode grammatical and organizational relationships within the text. Through a multimodal encoder and a graphical propagation layer, the model dynamically learns contextual dependencies, enabling comprehensive analysis of writing quality across multiple linguistic dimensions. The proposed fusion strategy aligns content attention and structural representations through a balanced optimization process. It leverages a joint objective function to weight both semantic and structural contributions, while an adaptive feedback mechanism refines attention and embedding parameters iteratively. This process ensures model adaptability across diverse writing styles and proficiency levels. Experimental results demonstrate that the fusion-based framework achieves superior performance compared to existing systems, offering higher precision, consistency, and robustness. By effectively combining content comprehension with structural evaluation, this approach provides a more holistic and transparent method for assessing English writing, representing a significant step forward in automated evaluation technology.