@inproceedings{ran-etal-2026-ruler,
title = "{RULER}: Instance-aware Rubric Rewards for {SVG} Generation",
author = "Ran, Hangyu and
Zheng, Yuhao and
Zhang, Yingying and
Lin, Kevin Qinghong and
Peng, Han",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1850/",
pages = "39506--39524",
ISBN = "979-8-89176-462-0",
abstract = "Generating Scalable Vector Graphics (SVG) code from natural-language instructions is an open-ended task without absolute visual ground truth, leaving both evaluation and policy optimization without a faithful signal. Scalar metrics (CLIP, Aesthetic) calibrated on natural images transfer poorly to stylized vector content, and reusing them as RL rewards triggers reward hacking. We address both limitations with rubric-based scoring. We first establish empirically that prompting a vision{--}language judge with a multi-axis rubric correlates with human judgments far better than scalar metrics, both across samples and within instructions. Building on this finding, we introduce **RULER** (Instance-aware **Ru**bric Rewards for Reinforcement **Le**a**r**ning), which converts each instruction into an instance-aware rubric of six items spanning semantic, visual, and stylistic axes; a judge VLM scores rendered rollouts item-by-item, and the weighted satisfactions form a fine-grained reward optimized via Group Relative Policy Optimization. Because the rubric is derived from text alone, RULER requires neither paired SVG ground truth nor human preference labels. On MMSVG-Illustration and MMSVG-Icon, RULER lifts the rubric score from $0.432/0.395$ to $0.693/0.683$, surpassing dedicated SVG specialists and matching the substantially larger DeepSeek-V3, with ablations identifying rubric design as the active lever for RL on open-ended SVG generation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ran-etal-2026-ruler">
<titleInfo>
<title>RULER: Instance-aware Rubric Rewards for SVG Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hangyu</namePart>
<namePart type="family">Ran</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuhao</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yingying</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kevin</namePart>
<namePart type="given">Qinghong</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Han</namePart>
<namePart type="family">Peng</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Generating Scalable Vector Graphics (SVG) code from natural-language instructions is an open-ended task without absolute visual ground truth, leaving both evaluation and policy optimization without a faithful signal. Scalar metrics (CLIP, Aesthetic) calibrated on natural images transfer poorly to stylized vector content, and reusing them as RL rewards triggers reward hacking. We address both limitations with rubric-based scoring. We first establish empirically that prompting a vision–language judge with a multi-axis rubric correlates with human judgments far better than scalar metrics, both across samples and within instructions. Building on this finding, we introduce **RULER** (Instance-aware **Ru**bric Rewards for Reinforcement **Le**a**r**ning), which converts each instruction into an instance-aware rubric of six items spanning semantic, visual, and stylistic axes; a judge VLM scores rendered rollouts item-by-item, and the weighted satisfactions form a fine-grained reward optimized via Group Relative Policy Optimization. Because the rubric is derived from text alone, RULER requires neither paired SVG ground truth nor human preference labels. On MMSVG-Illustration and MMSVG-Icon, RULER lifts the rubric score from 0.432/0.395 to 0.693/0.683, surpassing dedicated SVG specialists and matching the substantially larger DeepSeek-V3, with ablations identifying rubric design as the active lever for RL on open-ended SVG generation.</abstract>
<identifier type="citekey">ran-etal-2026-ruler</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1850/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>39506</start>
<end>39524</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T RULER: Instance-aware Rubric Rewards for SVG Generation
%A Ran, Hangyu
%A Zheng, Yuhao
%A Zhang, Yingying
%A Lin, Kevin Qinghong
%A Peng, Han
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F ran-etal-2026-ruler
%X Generating Scalable Vector Graphics (SVG) code from natural-language instructions is an open-ended task without absolute visual ground truth, leaving both evaluation and policy optimization without a faithful signal. Scalar metrics (CLIP, Aesthetic) calibrated on natural images transfer poorly to stylized vector content, and reusing them as RL rewards triggers reward hacking. We address both limitations with rubric-based scoring. We first establish empirically that prompting a vision–language judge with a multi-axis rubric correlates with human judgments far better than scalar metrics, both across samples and within instructions. Building on this finding, we introduce **RULER** (Instance-aware **Ru**bric Rewards for Reinforcement **Le**a**r**ning), which converts each instruction into an instance-aware rubric of six items spanning semantic, visual, and stylistic axes; a judge VLM scores rendered rollouts item-by-item, and the weighted satisfactions form a fine-grained reward optimized via Group Relative Policy Optimization. Because the rubric is derived from text alone, RULER requires neither paired SVG ground truth nor human preference labels. On MMSVG-Illustration and MMSVG-Icon, RULER lifts the rubric score from 0.432/0.395 to 0.693/0.683, surpassing dedicated SVG specialists and matching the substantially larger DeepSeek-V3, with ablations identifying rubric design as the active lever for RL on open-ended SVG generation.
%U https://aclanthology.org/2026.findings-emnlp.1850/
%P 39506-39524
Markdown (Informal)
[RULER: Instance-aware Rubric Rewards for SVG Generation](https://aclanthology.org/2026.findings-emnlp.1850/) (Ran et al., Findings 2026)
ACL
- Hangyu Ran, Yuhao Zheng, Yingying Zhang, Kevin Qinghong Lin, and Han Peng. 2026. RULER: Instance-aware Rubric Rewards for SVG Generation. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 39506–39524, Budapest, Hungary. Association for Computational Linguistics.