@inproceedings{de-vergnette-amblard-2026-sema,
title = "Sema System for the {DMR} 2026 Shared Task: Multistage {UMR} Parsing with Qwen3-4{B}",
author = "de Vergnette, R{\'e}mi and
Amblard, Maxime",
editor = "Zhao, Jin and
Post, Claire Benet and
Hoefer, Elizabeth",
booktitle = "Proceedings of The Seventh International Workshop on Designing Meaning Representations ({DMR} 2026) @ {LREC} 2026",
month = may,
year = "2026",
address = "Palma, Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.dmr-1.14/",
doi = "10.63317/24cughzsw4ae",
pages = "155--159",
abstract = "We present the Sema system for the DMR 2026 shared task on parsing from natural language to . Our approach relies on parameter-efficient fine-tuning of Qwen3-4B with a multistage training procedure. We first train on a capped subset of the noisy training data, then continue training on the clean split, and finally fine-tune a dedicated stage for word-to-node alignment prediction. The system generates sentence-level graphs, selected document-level information, and alignments in separate steps, followed by rule-based post-processing to satisfy the official evaluation format. Results show that the approach is viable across several languages and exhibits promising transfer to Italian despite the absence of Italian data for fine-tuning , while very low-resource languages remain challenging."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="de-vergnette-amblard-2026-sema">
<titleInfo>
<title>Sema System for the DMR 2026 Shared Task: Multistage UMR Parsing with Qwen3-4B</title>
</titleInfo>
<name type="personal">
<namePart type="given">Rémi</namePart>
<namePart type="family">de Vergnette</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maxime</namePart>
<namePart type="family">Amblard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of The Seventh International Workshop on Designing Meaning Representations (DMR 2026) @ LREC 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jin</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Claire</namePart>
<namePart type="given">Benet</namePart>
<namePart type="family">Post</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Elizabeth</namePart>
<namePart type="family">Hoefer</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca (Spain)</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>We present the Sema system for the DMR 2026 shared task on parsing from natural language to . Our approach relies on parameter-efficient fine-tuning of Qwen3-4B with a multistage training procedure. We first train on a capped subset of the noisy training data, then continue training on the clean split, and finally fine-tune a dedicated stage for word-to-node alignment prediction. The system generates sentence-level graphs, selected document-level information, and alignments in separate steps, followed by rule-based post-processing to satisfy the official evaluation format. Results show that the approach is viable across several languages and exhibits promising transfer to Italian despite the absence of Italian data for fine-tuning , while very low-resource languages remain challenging.</abstract>
<identifier type="citekey">de-vergnette-amblard-2026-sema</identifier>
<identifier type="doi">10.63317/24cughzsw4ae</identifier>
<location>
<url>https://aclanthology.org/2026.dmr-1.14/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>155</start>
<end>159</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Sema System for the DMR 2026 Shared Task: Multistage UMR Parsing with Qwen3-4B
%A de Vergnette, Rémi
%A Amblard, Maxime
%Y Zhao, Jin
%Y Post, Claire Benet
%Y Hoefer, Elizabeth
%S Proceedings of The Seventh International Workshop on Designing Meaning Representations (DMR 2026) @ LREC 2026
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F de-vergnette-amblard-2026-sema
%X We present the Sema system for the DMR 2026 shared task on parsing from natural language to . Our approach relies on parameter-efficient fine-tuning of Qwen3-4B with a multistage training procedure. We first train on a capped subset of the noisy training data, then continue training on the clean split, and finally fine-tune a dedicated stage for word-to-node alignment prediction. The system generates sentence-level graphs, selected document-level information, and alignments in separate steps, followed by rule-based post-processing to satisfy the official evaluation format. Results show that the approach is viable across several languages and exhibits promising transfer to Italian despite the absence of Italian data for fine-tuning , while very low-resource languages remain challenging.
%R 10.63317/24cughzsw4ae
%U https://aclanthology.org/2026.dmr-1.14/
%U https://doi.org/10.63317/24cughzsw4ae
%P 155-159
Markdown (Informal)
[Sema System for the DMR 2026 Shared Task: Multistage UMR Parsing with Qwen3-4B](https://aclanthology.org/2026.dmr-1.14/) (de Vergnette & Amblard, DMR 2026)
ACL