Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Yupei Wang, Renfen Hu, Zhe Zhao

TL;DR

The proposed method, using counterfactual intervention assisted by Large Language Models, 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.

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.

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

TL;DR

The proposed method, using counterfactual intervention assisted by Large Language Models, 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.

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.
Paper Structure (45 sections, 5 equations, 4 figures, 14 tables)

This paper contains 45 sections, 5 equations, 4 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. AES based on Neural Language Models
  4. Interpretability and Robustness of AES Models
  5. Counterfactual Analysis
  6. Method
  7. Concepts for Intervention
  8. Measurement of Rationale Alignment
  9. Counterfactual Generation
  10. The Validity of LLM Generated Counterfactuals
  11. Experiments
  12. Settings
  13. Counterfactual Validation Results
  14. Scoring Results
  15. Feedback Analysis
  16. ...and 30 more sections

Figures (4)

  • Figure 1: The pipeline of our proposed method.
  • Figure 2: Cohen's $\mathcal{D}$ measured for seven linguistic metrics on three interventions.
  • Figure 3: Scoring performance of GPT-3.5 SFT models with varying size of training data. The models' performance improves as the number of training samples increases, reaching comparable or equivalent levels to BERT-like models.
  • Figure 7: The pipeline of our proposed method.