Beyond Agreement: Diagnosing the Rationale Alignment of Automated Essay Scoring Methods based on Linguistically-informed Counterfactuals

TL;DR

This paper introduces a counterfactual intervention method using LLMs to diagnose and compare the rationale alignment of different AES models, revealing their focus areas and improving transparency.

Contribution

It presents a novel approach employing linguistically-informed counterfactuals with LLMs to analyze and enhance understanding of AES decision mechanisms.

Findings

BERT-like models focus mainly on sentence-level features

LLMs like GPT-3.5, GPT-4, and Llama-3 consider conventions, accuracy, and organization

LLMs can identify counterfactual interventions in essay feedback

Abstract

While current Automated Essay Scoring (AES) methods demonstrate high scoring agreement with human raters, their decision-making mechanisms are not fully understood. Our proposed method, using counterfactual intervention assisted by Large Language Models (LLMs), reveals that BERT-like models primarily focus on sentence-level features, whereas LLMs such as GPT-3.5, GPT-4 and Llama-3 are sensitive to conventions & accuracy, language complexity, and organization, indicating a more comprehensive rationale alignment with scoring rubrics. Moreover, LLMs can discern counterfactual interventions when giving feedback on essays. Our approach improves understanding of neural AES methods and can also apply to other domains seeking transparency in model-driven decisions.