@inproceedings{nakhle-etal-2026-vertical,
title = "Vertical: Quality Estimation of Machine Translation from Universit{\'e} Grenoble Alpes",
author = "Nakhl{\'e}, Mariam and
Qader, Raheel and
Dinarelli, Marco and
Blanchon, Herv{\'e}",
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.105/",
pages = "1716--1722",
ISBN = "979-8-89176-474-3",
abstract = "This paper describes the submission of Universit{\'e} Grenoble Alpes and OVHai LLM to the Eleventh Conference on Machine Translation (WMT26) Shared Task on Automated Translation Quality Evaluation Systems. We participate in the Task 2 - Segment-Level Quality Score Prediction. We present our system that is trained to predict a single quality score per input segment. Our contributions are the following: 1) we present a metric that uses a Large Language Model (LLM) as backbone, namely gemma-3-1b-it, thus leveraging its long context size and its large-scale training on 2 trillion tokens, 2) we propose a method to adapt a decoder-only model to a downstream task by training a special token and using it as the sequence representation and 3) we propose a training curriculum designed to boost metric performance on long inputs. Our metric is trained using publicly available data and it can predict a score with or without a reference. Preliminary results show that our method outperforms baseline models when provided with a reference on the system level. The reference-free variant ranks second on the segment level."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="nakhle-etal-2026-vertical">
<titleInfo>
<title>Vertical: Quality Estimation of Machine Translation from Université Grenoble Alpes</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mariam</namePart>
<namePart type="family">Nakhlé</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Raheel</namePart>
<namePart type="family">Qader</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Dinarelli</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hervé</namePart>
<namePart type="family">Blanchon</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>This paper describes the submission of Université Grenoble Alpes and OVHai LLM to the Eleventh Conference on Machine Translation (WMT26) Shared Task on Automated Translation Quality Evaluation Systems. We participate in the Task 2 - Segment-Level Quality Score Prediction. We present our system that is trained to predict a single quality score per input segment. Our contributions are the following: 1) we present a metric that uses a Large Language Model (LLM) as backbone, namely gemma-3-1b-it, thus leveraging its long context size and its large-scale training on 2 trillion tokens, 2) we propose a method to adapt a decoder-only model to a downstream task by training a special token and using it as the sequence representation and 3) we propose a training curriculum designed to boost metric performance on long inputs. Our metric is trained using publicly available data and it can predict a score with or without a reference. Preliminary results show that our method outperforms baseline models when provided with a reference on the system level. The reference-free variant ranks second on the segment level.</abstract>
<identifier type="citekey">nakhle-etal-2026-vertical</identifier>
<location>
<url>https://aclanthology.org/2026.wmt-main.105/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>1716</start>
<end>1722</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Vertical: Quality Estimation of Machine Translation from Université Grenoble Alpes
%A Nakhlé, Mariam
%A Qader, Raheel
%A Dinarelli, Marco
%A Blanchon, Hervé
%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 nakhle-etal-2026-vertical
%X This paper describes the submission of Université Grenoble Alpes and OVHai LLM to the Eleventh Conference on Machine Translation (WMT26) Shared Task on Automated Translation Quality Evaluation Systems. We participate in the Task 2 - Segment-Level Quality Score Prediction. We present our system that is trained to predict a single quality score per input segment. Our contributions are the following: 1) we present a metric that uses a Large Language Model (LLM) as backbone, namely gemma-3-1b-it, thus leveraging its long context size and its large-scale training on 2 trillion tokens, 2) we propose a method to adapt a decoder-only model to a downstream task by training a special token and using it as the sequence representation and 3) we propose a training curriculum designed to boost metric performance on long inputs. Our metric is trained using publicly available data and it can predict a score with or without a reference. Preliminary results show that our method outperforms baseline models when provided with a reference on the system level. The reference-free variant ranks second on the segment level.
%U https://aclanthology.org/2026.wmt-main.105/
%P 1716-1722
Markdown (Informal)
[Vertical: Quality Estimation of Machine Translation from Université Grenoble Alpes](https://aclanthology.org/2026.wmt-main.105/) (Nakhlé et al., WMT 2026)
ACL