Automated Cross-prompt Scoring of Essay Traits

The majority of current research in Automated Essay Scoring (AES) focuses on prompt-specific scoring either the overall quality an essay or with regards to certain traits. In real-world applications obtaining labelled data for a target prompt is often expensive unfeasible, requiring AES system be able perform well when predicting scores essays from unseen prompts. As result, some recent has been dedicated cross-prompt AES. However, this line thus far only concerned holistic, scoring, no exploration into different users systems require feedback aspects their writing, trait necessary component effective system. Therefore, address need, we introduce new task named Cross-prompt Traits, which requires model trained solely non-target-prompt and predict score as number specific traits target-prompt essays. This challenges model's ability generalize order novel domain its represent multiple aspects. addition, new, innovative approach builds top state-of-the-art method Our utilizes trait-attention mechanism multi-task architecture that leverages relationships between each simultaneously individual trait. We conduct extensive experiments widely used ASAP ASAP++ datasets demonstrate our outperform leading methods.