Two-phase cross-prompt pre-finetuning approach for automated short answer scoring

Research Method · All research methods
Evidence · 3 claims (3 for) · 1 study (1 causal), q2 · 1 of 1 report an effect size · 3 claims rest on one study

Description

A training framework in which an SAS model is first pre-finetuned on already annotated prompts using rubric key phrases concatenated with each answer, so it learns "the task's general property (i.e., a principle of scoring): an answer receives a high score if it contains the information specified by the rubric." The pre-finetuned model is then finetuned on a new target prompt without requiring access to the cross-prompt data, resolving data accessibility because the second phase "needs access to only the parameters of the pre-finetuned model." The approach was instantiated with both a BERT regression model and open-source LLMs.

Accounts

• Two-phase cross-prompt pre-finetuning approach for automated short answer scoring: A training framework in which an SAS model is first pre-finetuned on already annotated prompts using rubric key phrases concatenated with each answer, so it learns "the task's general property (i.e., a principle of scoring): an answer receives a high score if it contains the information specified by the rubric." The pre-finetuned model is then finetuned on a new target prompt without requiring access to the cross-prompt data, resolving data accessibility because the second phase "needs access to only the parameters of the pre-finetuned model." The approach was instantiated with both a BERT regression model and open-source LLMs. (Funayama et al. (2025))

Claims

• Two-phase cross-prompt pre-finetuning with key phrases significantly improves short answer scoring accuracy when in-prompt training data is limited [+M]
• Pre-finetuning without key phrases does not improve scoring performance and slightly lowers it relative to the BERT baseline [+M]
• The pre-finetuned model learns a scoring principle: predicted scores correlate strongly and negatively with edit distance between answers and key phrases [+M]

Related Research Methods

-

Key Sources

• Funayama, H., Matsubayashi, Y., Asazuma, Y., Mizumoto, T., & Inui, K. (2025). Cross-prompt Pre-finetuning of Language Models for Short Answer Scoring. Int J Artif Intell Educ. https://doi.org/10.1007/s40593-025-00474-w

Page metadata

Field	Value
type	research-method
id	two-phase-cross-prompt-pre-finetuning-approach-for-automated-short-answer-scoring
title	Two-phase cross-prompt pre-finetuning approach for automated short answer scoring
description	A training framework in which an SAS model is first pre-finetuned on already annotated prompts using rubric key phrases concatenated with each answer, so it learns "the task's general property (i.e., a principle of scoring): an answer receives a high score if it contains the information specified by
status	review
generated	by process:settle-candidates on 2026-10-28
trust	unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.