Transformer-LSTM models for automatic scoring and feedback in English writing assessment

Writing assessment is one of the most important stages in the educational process, but it is also the most resource-demanding one. To address the challenges of scalability and inconsistency, this study proposes a Transformer-LSTM model for automated scoring and feedback generation, enhancing accuracy and reliability in assessment. Integrating the contextual reading abilities of transformers with the sequential analysis strength of LSTMs, the model analyzes significant metrics of writing quality, including grammar, coherence, and structure, while providing individualized, actionable feedback. Using annotated datasets and evaluation metrics like RMSE and feedback relevance, it was established that the model performs well overall and that improvements in grammar and coherence seemed to be the most significant contributors to writing ability. It was also demonstrated that feedback relevance enhances these outcomes, thus confirming its valuable role in promoting structural and grammatical accuracy. Understanding that most existing systems do not encourage significant human feedback, this work demonstrates a scalable approach with potential alignment with human evaluation standards. Finally, this study shows hybrid models’ promise for automated writing assessment as promising scalable, equitable, impact-based tools for global enhancement of educational outcomes.

Text 121575.pdf - Published Version Available under License Creative Commons Attribution. Download (2MB) Official URL or Download Paper: https://ieeexplore.ieee.org/document/10969843/

Actions (login required)

View Item