@inproceedings{tian-etal-2026-rubric,
title = "Rubric-Aware Multi-Expert Online Policy Distillation for {C}hinese Elementary School Essay Writing",
author = "Tian, Wei and
Zhou, Yuhao and
Sun, Xiayu and
Zhang, Yadong and
Zhou, Hao and
Liu, Shu and
Lan, Man",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1115/",
pages = "25543--25565",
ISBN = "979-8-89176-459-0",
abstract = "Chinese elementary-school writing is rubric-driven. We study it as a teacher-facing instructional support task: given a classroom prompt and requirements, generate a teacher-reviewable example matching the target genre, rubric, length, and age-appropriate register. Directly prompted frontier LLMs often mis-route genres or adopt an adult register.\par We present **RA-MOPD**, a four-stage pipeline that post-trains a 4B model into a unified essay writer across four genres and eight rubric dimensions. Its core stage augments on-policy reverse-KL distillation with *Rubric-Weighted KL*, which up-weights rubric-critical tokens via reward-model attribution, and *Cross-Genre Mixing*, a parameter-free expert blend guided by an empirical transfer prior.\par On a new 6,000-essay benchmark with eight reward models (QWK = 0.864), RA-MOPD reaches 4.21/5.00 and 97.8{\%} routing{---}within 0.09 of GPT-5.3 3-shot (p {\ensuremath{<}} 0.01). No statistically significant difference is detected relative to a 4{\texttimes}-cost four-expert logit-averaging baseline ({\ensuremath{\Delta}} = +0.03, p = 0.18).\par A GPT-4.1 judge, BERTScore, a 1,200-essay teacher study (62.8{\%} preference), and register analysis provide converging evidence beyond RM metrics. GPT-5.5 holistic evaluation further supports the component-level gains, which remain stable on Qwen3-1.7B without model-specific retuning.\par Code and data: [https://github.com/TW-NLP/MOPD](https://github.com/TW-NLP/MOPD)"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tian-etal-2026-rubric">
<titleInfo>
<title>Rubric-Aware Multi-Expert Online Policy Distillation for Chinese Elementary School Essay Writing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">Tian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuhao</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiayu</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yadong</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hao</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shu</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Man</namePart>
<namePart type="family">Lan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Chinese elementary-school writing is rubric-driven. We study it as a teacher-facing instructional support task: given a classroom prompt and requirements, generate a teacher-reviewable example matching the target genre, rubric, length, and age-appropriate register. Directly prompted frontier LLMs often mis-route genres or adopt an adult register.\par We present **RA-MOPD**, a four-stage pipeline that post-trains a 4B model into a unified essay writer across four genres and eight rubric dimensions. Its core stage augments on-policy reverse-KL distillation with *Rubric-Weighted KL*, which up-weights rubric-critical tokens via reward-model attribution, and *Cross-Genre Mixing*, a parameter-free expert blend guided by an empirical transfer prior.\par On a new 6,000-essay benchmark with eight reward models (QWK = 0.864), RA-MOPD reaches 4.21/5.00 and 97.8% routing—within 0.09 of GPT-5.3 3-shot (p \ensuremath< 0.01). No statistically significant difference is detected relative to a 4×-cost four-expert logit-averaging baseline (\ensuremathΔ = +0.03, p = 0.18).\par A GPT-4.1 judge, BERTScore, a 1,200-essay teacher study (62.8% preference), and register analysis provide converging evidence beyond RM metrics. GPT-5.5 holistic evaluation further supports the component-level gains, which remain stable on Qwen3-1.7B without model-specific retuning.\par Code and data: [https://github.com/TW-NLP/MOPD](https://github.com/TW-NLP/MOPD)</abstract>
<identifier type="citekey">tian-etal-2026-rubric</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1115/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>25543</start>
<end>25565</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Rubric-Aware Multi-Expert Online Policy Distillation for Chinese Elementary School Essay Writing
%A Tian, Wei
%A Zhou, Yuhao
%A Sun, Xiayu
%A Zhang, Yadong
%A Zhou, Hao
%A Liu, Shu
%A Lan, Man
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F tian-etal-2026-rubric
%X Chinese elementary-school writing is rubric-driven. We study it as a teacher-facing instructional support task: given a classroom prompt and requirements, generate a teacher-reviewable example matching the target genre, rubric, length, and age-appropriate register. Directly prompted frontier LLMs often mis-route genres or adopt an adult register.\par We present **RA-MOPD**, a four-stage pipeline that post-trains a 4B model into a unified essay writer across four genres and eight rubric dimensions. Its core stage augments on-policy reverse-KL distillation with *Rubric-Weighted KL*, which up-weights rubric-critical tokens via reward-model attribution, and *Cross-Genre Mixing*, a parameter-free expert blend guided by an empirical transfer prior.\par On a new 6,000-essay benchmark with eight reward models (QWK = 0.864), RA-MOPD reaches 4.21/5.00 and 97.8% routing—within 0.09 of GPT-5.3 3-shot (p \ensuremath< 0.01). No statistically significant difference is detected relative to a 4×-cost four-expert logit-averaging baseline (\ensuremathΔ = +0.03, p = 0.18).\par A GPT-4.1 judge, BERTScore, a 1,200-essay teacher study (62.8% preference), and register analysis provide converging evidence beyond RM metrics. GPT-5.5 holistic evaluation further supports the component-level gains, which remain stable on Qwen3-1.7B without model-specific retuning.\par Code and data: [https://github.com/TW-NLP/MOPD](https://github.com/TW-NLP/MOPD)
%U https://aclanthology.org/2026.emnlp-main.1115/
%P 25543-25565
Markdown (Informal)
[Rubric-Aware Multi-Expert Online Policy Distillation for Chinese Elementary School Essay Writing](https://aclanthology.org/2026.emnlp-main.1115/) (Tian et al., EMNLP 2026)
ACL