Feedback-to-Rubrics: Can We Learn Expert Criteria from Inline Comments?

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yoshida, Kotaro, Kuroki, So, Imajuku, Yuki, Nakamura, Taishi, Iwai, Ryunosuke, Goda, Haruki, Akiba, Takuya
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916061048209408
author Yoshida, Kotaro
Kuroki, So
Imajuku, Yuki
Nakamura, Taishi
Iwai, Ryunosuke
Goda, Haruki
Akiba, Takuya
author_facet Yoshida, Kotaro
Kuroki, So
Imajuku, Yuki
Nakamura, Taishi
Iwai, Ryunosuke
Goda, Haruki
Akiba, Takuya
contents Large language models (LLMs) are increasingly used for writing and review support, but their usefulness depends on context-dependent criteria, such as expert preferences or organization-specific conventions, that are often tacit, undocumented, and difficult to elicit directly. We propose a problem setting for learning reusable natural-language rubrics from accumulated inline comments on artifacts such as human-written or LLM-generated drafts. Our method infers rubrics from these comments and iteratively refines them by observing comment-wise mismatches between rubric-conditioned predictions and reference comments. We evaluate the proposed method in real-world review settings and in controlled settings with reference rubrics. These results show that inline comments can be distilled into reusable rubrics that support comment prediction, rubric understanding, and automatic artifact revision.
format Preprint
id arxiv_https___arxiv_org_abs_2605_29857
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Feedback-to-Rubrics: Can We Learn Expert Criteria from Inline Comments?
Yoshida, Kotaro
Kuroki, So
Imajuku, Yuki
Nakamura, Taishi
Iwai, Ryunosuke
Goda, Haruki
Akiba, Takuya
Machine Learning
Large language models (LLMs) are increasingly used for writing and review support, but their usefulness depends on context-dependent criteria, such as expert preferences or organization-specific conventions, that are often tacit, undocumented, and difficult to elicit directly. We propose a problem setting for learning reusable natural-language rubrics from accumulated inline comments on artifacts such as human-written or LLM-generated drafts. Our method infers rubrics from these comments and iteratively refines them by observing comment-wise mismatches between rubric-conditioned predictions and reference comments. We evaluate the proposed method in real-world review settings and in controlled settings with reference rubrics. These results show that inline comments can be distilled into reusable rubrics that support comment prediction, rubric understanding, and automatic artifact revision.
title Feedback-to-Rubrics: Can We Learn Expert Criteria from Inline Comments?
topic Machine Learning
url https://arxiv.org/abs/2605.29857