@inproceedings{siani-2026-reference,
title = "Reference-Free Consensus Evaluation of Machine Translation: Ensembling {LLM} Error-Span Judgments with Neutral Cross-Verification and {STAPLE} Reliability Fusion",
author = "Siani, Assaf",
editor = "Haddow, Barry and
Kocmi, Tom and
Koehn, Philipp and
Monz, Christof",
booktitle = "Proceedings of the Eleventh Conference on Machine Translation",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wmt-main.108/",
pages = "1739--1744",
ISBN = "979-8-89176-474-3",
abstract = "We present a reference-free system for the WMT 2026 automated translation-quality-evaluation shared task, covering all three subtasks: character-level error-span detection with Minor/Major severity, segment level quality scoring on a 0{--}100 scale, and binary error-free classification. Rather than treating a single large language model (LLM) as an oracle, the system treats LLMs as fallible annotators. Three heterogeneous judges independently propose target-side error spans; a neutral cross-verification round converts these open-ended proposals into a complete, anonymized candidate-by-judge vote matrix; a character-level adaptation of STAPLE estimates latent error truth and per-judge reliability without references; and a two-state HMM imposes minimal span structure. Omissions are handled by a dedicated source-coverage pass rather than by scanning the target. Calibrating the single free prior against the official metric with human gold yields several findings: plain majority voting matches or beats STAPLE; the error rate prior barely matters once outputs are schema-conformant; precision{---}judge over-flagging at roughly 3{\texttimes} the human rate{---}is the binding constraint; and LLM-derived silver labels cannot calibrate the prior without circularity. The system was deployed across six language pairs with partial-failure-tolerant, resumable execution."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="siani-2026-reference">
<titleInfo>
<title>Reference-Free Consensus Evaluation of Machine Translation: Ensembling LLM Error-Span Judgments with Neutral Cross-Verification and STAPLE Reliability Fusion</title>
</titleInfo>
<name type="personal">
<namePart type="given">Assaf</namePart>
<namePart type="family">Siani</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 Eleventh Conference on Machine Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Barry</namePart>
<namePart type="family">Haddow</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tom</namePart>
<namePart type="family">Kocmi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philipp</namePart>
<namePart type="family">Koehn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christof</namePart>
<namePart type="family">Monz</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">979-8-89176-474-3</identifier>
</relatedItem>
<abstract>We present a reference-free system for the WMT 2026 automated translation-quality-evaluation shared task, covering all three subtasks: character-level error-span detection with Minor/Major severity, segment level quality scoring on a 0–100 scale, and binary error-free classification. Rather than treating a single large language model (LLM) as an oracle, the system treats LLMs as fallible annotators. Three heterogeneous judges independently propose target-side error spans; a neutral cross-verification round converts these open-ended proposals into a complete, anonymized candidate-by-judge vote matrix; a character-level adaptation of STAPLE estimates latent error truth and per-judge reliability without references; and a two-state HMM imposes minimal span structure. Omissions are handled by a dedicated source-coverage pass rather than by scanning the target. Calibrating the single free prior against the official metric with human gold yields several findings: plain majority voting matches or beats STAPLE; the error rate prior barely matters once outputs are schema-conformant; precision—judge over-flagging at roughly 3× the human rate—is the binding constraint; and LLM-derived silver labels cannot calibrate the prior without circularity. The system was deployed across six language pairs with partial-failure-tolerant, resumable execution.</abstract>
<identifier type="citekey">siani-2026-reference</identifier>
<location>
<url>https://aclanthology.org/2026.wmt-main.108/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>1739</start>
<end>1744</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Reference-Free Consensus Evaluation of Machine Translation: Ensembling LLM Error-Span Judgments with Neutral Cross-Verification and STAPLE Reliability Fusion
%A Siani, Assaf
%Y Haddow, Barry
%Y Kocmi, Tom
%Y Koehn, Philipp
%Y Monz, Christof
%S Proceedings of the Eleventh Conference on Machine Translation
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-474-3
%F siani-2026-reference
%X We present a reference-free system for the WMT 2026 automated translation-quality-evaluation shared task, covering all three subtasks: character-level error-span detection with Minor/Major severity, segment level quality scoring on a 0–100 scale, and binary error-free classification. Rather than treating a single large language model (LLM) as an oracle, the system treats LLMs as fallible annotators. Three heterogeneous judges independently propose target-side error spans; a neutral cross-verification round converts these open-ended proposals into a complete, anonymized candidate-by-judge vote matrix; a character-level adaptation of STAPLE estimates latent error truth and per-judge reliability without references; and a two-state HMM imposes minimal span structure. Omissions are handled by a dedicated source-coverage pass rather than by scanning the target. Calibrating the single free prior against the official metric with human gold yields several findings: plain majority voting matches or beats STAPLE; the error rate prior barely matters once outputs are schema-conformant; precision—judge over-flagging at roughly 3× the human rate—is the binding constraint; and LLM-derived silver labels cannot calibrate the prior without circularity. The system was deployed across six language pairs with partial-failure-tolerant, resumable execution.
%U https://aclanthology.org/2026.wmt-main.108/
%P 1739-1744
Markdown (Informal)
[Reference-Free Consensus Evaluation of Machine Translation: Ensembling LLM Error-Span Judgments with Neutral Cross-Verification and STAPLE Reliability Fusion](https://aclanthology.org/2026.wmt-main.108/) (Siani, WMT 2026)
ACL