@inproceedings{saad-falcon-etal-2025-lmunit,
title = "{LMUNIT}: Fine-grained Evaluation with Natural Language Unit Tests",
author = "Saad-Falcon, Jon and
Vivek, Rajan Pathe and
Berrios, William and
Naik, Nandita Shankar and
Franklin, Matija and
Vidgen, Bertie and
Singh, Amanpreet and
Kiela, Douwe and
Mehri, Shikib",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-emnlp.176/",
pages = "3303--3324",
ISBN = "979-8-89176-335-7",
abstract = "As language models become integral to critical workflows, assessing their behavior remains a fundamental challenge {--} human evaluation is costly and noisy, while automated metrics provide only coarse, difficult-to-interpret signals. We introduce \textbf{natural language unit tests}, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring model, LMUnit, which combines multi-objective training across preferences, direct ratings, and natural language rationales. Through controlled human studies, we show this paradigm significantly improves inter-annotator agreement and enables more effective LLM development workflows. LMUnit achieves state-of-the-art performance on evaluation benchmarks including FLASK, BigGenBench, and RewardBench 2, while maintaining competitive results on the original RewardBench. These results validate both our proposed paradigm and scoring model, suggesting a promising path forward for language model evaluation and development. Our code has been released at github.com/ContextualAI/LMUnit with an MIT license."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="saad-falcon-etal-2025-lmunit">
<titleInfo>
<title>LMUNIT: Fine-grained Evaluation with Natural Language Unit Tests</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jon</namePart>
<namePart type="family">Saad-Falcon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rajan</namePart>
<namePart type="given">Pathe</namePart>
<namePart type="family">Vivek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">William</namePart>
<namePart type="family">Berrios</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nandita</namePart>
<namePart type="given">Shankar</namePart>
<namePart type="family">Naik</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Matija</namePart>
<namePart type="family">Franklin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bertie</namePart>
<namePart type="family">Vidgen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amanpreet</namePart>
<namePart type="family">Singh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Douwe</namePart>
<namePart type="family">Kiela</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shikib</namePart>
<namePart type="family">Mehri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-335-7</identifier>
</relatedItem>
<abstract>As language models become integral to critical workflows, assessing their behavior remains a fundamental challenge – human evaluation is costly and noisy, while automated metrics provide only coarse, difficult-to-interpret signals. We introduce natural language unit tests, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring model, LMUnit, which combines multi-objective training across preferences, direct ratings, and natural language rationales. Through controlled human studies, we show this paradigm significantly improves inter-annotator agreement and enables more effective LLM development workflows. LMUnit achieves state-of-the-art performance on evaluation benchmarks including FLASK, BigGenBench, and RewardBench 2, while maintaining competitive results on the original RewardBench. These results validate both our proposed paradigm and scoring model, suggesting a promising path forward for language model evaluation and development. Our code has been released at github.com/ContextualAI/LMUnit with an MIT license.</abstract>
<identifier type="citekey">saad-falcon-etal-2025-lmunit</identifier>
<location>
<url>https://aclanthology.org/2025.findings-emnlp.176/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>3303</start>
<end>3324</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T LMUNIT: Fine-grained Evaluation with Natural Language Unit Tests
%A Saad-Falcon, Jon
%A Vivek, Rajan Pathe
%A Berrios, William
%A Naik, Nandita Shankar
%A Franklin, Matija
%A Vidgen, Bertie
%A Singh, Amanpreet
%A Kiela, Douwe
%A Mehri, Shikib
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Findings of the Association for Computational Linguistics: EMNLP 2025
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-335-7
%F saad-falcon-etal-2025-lmunit
%X As language models become integral to critical workflows, assessing their behavior remains a fundamental challenge – human evaluation is costly and noisy, while automated metrics provide only coarse, difficult-to-interpret signals. We introduce natural language unit tests, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring model, LMUnit, which combines multi-objective training across preferences, direct ratings, and natural language rationales. Through controlled human studies, we show this paradigm significantly improves inter-annotator agreement and enables more effective LLM development workflows. LMUnit achieves state-of-the-art performance on evaluation benchmarks including FLASK, BigGenBench, and RewardBench 2, while maintaining competitive results on the original RewardBench. These results validate both our proposed paradigm and scoring model, suggesting a promising path forward for language model evaluation and development. Our code has been released at github.com/ContextualAI/LMUnit with an MIT license.
%U https://aclanthology.org/2025.findings-emnlp.176/
%P 3303-3324
Markdown (Informal)
[LMUNIT: Fine-grained Evaluation with Natural Language Unit Tests](https://aclanthology.org/2025.findings-emnlp.176/) (Saad-Falcon et al., Findings 2025)
ACL
- Jon Saad-Falcon, Rajan Pathe Vivek, William Berrios, Nandita Shankar Naik, Matija Franklin, Bertie Vidgen, Amanpreet Singh, Douwe Kiela, and Shikib Mehri. 2025. LMUNIT: Fine-grained Evaluation with Natural Language Unit Tests. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 3303–3324, Suzhou, China. Association for Computational Linguistics.