MLPAS: Encoder-only Essay Scoring with Multi-level Disentanglement
Abstract: The application of language models in essay scoring has gained significant attention in recent years, typically evaluating a single model across multiple prompts. However, in a multi-prompt setup, it is crucial to understand the varying aspects of different prompts. In such settings, there are notable variations even in a trait with the same name across prompts, often overlooked in existing research. We propose introducing multi-level disentanglement into a Transformer encoder-only framework for essay scoring, preserving fine-grained semantic differences across such traits. Our method not only improves the quality of essay scoring, but also reduces memory usage and latency. Experimental results demonstrate that our framework achieves the highest agreement with human essay ratings over four SOTA approaches. The codes will become available upon acceptance.
Paper Type: Long
Research Area: NLP Applications
Research Area Keywords: essay scoring, educational applications
Languages Studied: English
Submission Number: 60
Loading