Does task decomposition improve automatic NLG evaluation?

Sebastian Steindl; Nikos Voskarides; Alberto Gasparin; Diego Marcheggiani

Does task decomposition improve automatic NLG evaluation?

Sebastian Steindl, Nikos Voskarides, Alberto Gasparin, Diego Marcheggiani

Correct Metadata for

Use this form to create a GitHub issue with structured data describing the correction. You will need a GitHub account. Once you create that issue, the correction will be reviewed by a staff member.

⚠️ Mobile Users: Submitting this form to create a new issue will only work with github.com, not the GitHub Mobile app.

Important: The Anthology treat PDFs as authoritative. Please use this form only to correct data that is out of line with the PDF. See our corrections guidelines if you need to change the PDF.

Title Adjust the title. Retain tags such as <fixed-case>.

Authors Adjust author names and order to match the PDF.

Abstract Correct abstract if needed. Retain XML formatting tags such as <tex-math>. You may use ... for bold, ... for italic, ... for underline, <sc>...</sc> for small-caps, <tt>...<tt> for typewriter text, <url>...</url> for URLs, <a href=...> for hyperlinks, and <par/> for paragraph breaks.

Verification against PDF Ensure that the new title/authors match the snapshot below. (If there is no snapshot or it is too small, consult the PDF.)

Authors concatenated from the text boxes above:

ALL author names match the snapshot above—including middle initials, hyphens, and accents.

Abstract

The LLM-as-a-judge (LLMaJ) framework has emerged as a promising solution for cheap, reproducible, reference-free Natural Language Generation (NLG) evaluation. Prior work seeks to improve LLMaJ by decomposing evaluation tasks into simpler sub-tasks. In this work, we systematically compare LLMaJ methods with and without decomposition on multiple NLG datasets. We find no evidence that LLMaJ with task decomposition leads to performance gains over a fair baseline that does not use decomposition. Instead, we find that previously reported performance gains in decomposition-based LLMaJ stem from using human labels as training data, and not task decomposition itself. Also, we find that, when human labels are available, LLMaJ without using task decomposition can perform comparably to human annotators.

Anthology ID:: 2026.emnlp-main.447
Volume:: Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
Month:: October
Year:: 2026
Address:: Budapest, Hungary
Editors:: Hung-Yi Lee, Sunipa Dev, Jessy Li, Desmond Elliott
Venue:: EMNLP
SIG:: SIGDAT
Publisher:: Association for Computational Linguistics
Note:
Pages:: 10288–10301
Language:
URL:: https://aclanthology.org/2026.emnlp-main.447/
DOI:
Bibkey:
Cite (ACL):: Sebastian Steindl, Nikos Voskarides, Alberto Gasparin, and Diego Marcheggiani. 2026. Does task decomposition improve automatic NLG evaluation?. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 10288–10301, Budapest, Hungary. Association for Computational Linguistics.
Cite (Informal):: Does task decomposition improve automatic NLG evaluation? (Steindl et al., EMNLP 2026)
Copy Citation:
PDF:: https://aclanthology.org/2026.emnlp-main.447.pdf

PDF Cite Search Fix data

Export citation

BibTeX
MODS XML
Endnote
Preformatted

@inproceedings{steindl-etal-2026-task,
    title = "Does task decomposition improve automatic {NLG} evaluation?",
    author = "Steindl, Sebastian  and
      Voskarides, Nikos  and
      Gasparin, Alberto  and
      Marcheggiani, Diego",
    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.447/",
    pages = "10288--10301",
    ISBN = "979-8-89176-459-0",
    abstract = "The LLM-as-a-judge (LLMaJ) framework has emerged as a promising solution for cheap, reproducible, reference-free Natural Language Generation (NLG) evaluation. Prior work seeks to improve LLMaJ by decomposing evaluation tasks into simpler sub-tasks. In this work, we systematically compare LLMaJ methods with and without decomposition on multiple NLG datasets. We find no evidence that LLMaJ with task decomposition leads to performance gains over a fair baseline that does not use decomposition. Instead, we find that previously reported performance gains in decomposition-based LLMaJ stem from using human labels as training data, and not task decomposition itself. Also, we find that, when human labels are available, LLMaJ without using task decomposition can perform comparably to human annotators."
}

Download as File

<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="steindl-etal-2026-task">
    <titleInfo>
        <title>Does task decomposition improve automatic NLG evaluation?</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Sebastian</namePart>
        <namePart type="family">Steindl</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Nikos</namePart>
        <namePart type="family">Voskarides</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Alberto</namePart>
        <namePart type="family">Gasparin</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Diego</namePart>
        <namePart type="family">Marcheggiani</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>The LLM-as-a-judge (LLMaJ) framework has emerged as a promising solution for cheap, reproducible, reference-free Natural Language Generation (NLG) evaluation. Prior work seeks to improve LLMaJ by decomposing evaluation tasks into simpler sub-tasks. In this work, we systematically compare LLMaJ methods with and without decomposition on multiple NLG datasets. We find no evidence that LLMaJ with task decomposition leads to performance gains over a fair baseline that does not use decomposition. Instead, we find that previously reported performance gains in decomposition-based LLMaJ stem from using human labels as training data, and not task decomposition itself. Also, we find that, when human labels are available, LLMaJ without using task decomposition can perform comparably to human annotators.</abstract>
    <identifier type="citekey">steindl-etal-2026-task</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.447/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>10288</start>
            <end>10301</end>
        </extent>
    </part>
</mods>
</modsCollection>

Download as File

%0 Conference Proceedings
%T Does task decomposition improve automatic NLG evaluation?
%A Steindl, Sebastian
%A Voskarides, Nikos
%A Gasparin, Alberto
%A Marcheggiani, Diego
%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 steindl-etal-2026-task
%X The LLM-as-a-judge (LLMaJ) framework has emerged as a promising solution for cheap, reproducible, reference-free Natural Language Generation (NLG) evaluation. Prior work seeks to improve LLMaJ by decomposing evaluation tasks into simpler sub-tasks. In this work, we systematically compare LLMaJ methods with and without decomposition on multiple NLG datasets. We find no evidence that LLMaJ with task decomposition leads to performance gains over a fair baseline that does not use decomposition. Instead, we find that previously reported performance gains in decomposition-based LLMaJ stem from using human labels as training data, and not task decomposition itself. Also, we find that, when human labels are available, LLMaJ without using task decomposition can perform comparably to human annotators.
%U https://aclanthology.org/2026.emnlp-main.447/
%P 10288-10301

Download as File

Markdown (Informal)

[Does task decomposition improve automatic NLG evaluation?](https://aclanthology.org/2026.emnlp-main.447/) (Steindl et al., EMNLP 2026)

Does task decomposition improve automatic NLG evaluation? (Steindl et al., EMNLP 2026)

ACL

Sebastian Steindl, Nikos Voskarides, Alberto Gasparin, and Diego Marcheggiani. 2026. Does task decomposition improve automatic NLG evaluation?. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 10288–10301, Budapest, Hungary. Association for Computational Linguistics.