Activations as Features: Probing LLMs for Generalizable Essay Scoring Representations
This work addresses the problem of diverse scoring criteria in automated essay scoring for educational applications, representing an incremental improvement by leveraging existing LLM activations in a novel way.
The paper tackled the challenge of automated essay scoring in cross-prompt settings by exploring the use of LLM activations as features, finding that these activations have strong discriminative power and allow LLMs to adapt evaluation perspectives to different traits and essay types.
Automated essay scoring (AES) is a challenging task in cross-prompt settings due to the diversity of scoring criteria. While previous studies have focused on the output of large language models (LLMs) to improve scoring accuracy, we believe activations from intermediate layers may also provide valuable information. To explore this possibility, we evaluated the discriminative power of LLMs' activations in cross-prompt essay scoring task. Specifically, we used activations to fit probes and further analyzed the effects of different models and input content of LLMs on this discriminative power. By computing the directions of essays across various trait dimensions under different prompts, we analyzed the variation in evaluation perspectives of large language models concerning essay types and traits. Results show that the activations possess strong discriminative power in evaluating essay quality and that LLMs can adapt their evaluation perspectives to different traits and essay types, effectively handling the diversity of scoring criteria in cross-prompt settings.