@inproceedings{rezkellah-etal-2026-encoders,
title = "Encoders Strike Back: Why and How a Strong Discriminative Baseline Outperforms {LLM}s in Argument Mining Across Three Heterogeneous Datasets",
author = "Rezkellah, Fatma-Zohra and
Karray, Ghassen and
Gubelmann, Reto",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2358/",
doi = "10.18653/2026.findings-emnlp.2358",
pages = "50696--50713",
ISBN = "979-8-89176-462-0",
abstract = "We present a comprehensive study of LLMs for argument mining, comparing four learning paradigms against a strong encoder-only baseline across three heterogeneous corpora to assess cross-domain generalisation. We evaluate in-context learning, supervised fine-tuning (SFT), and cold-start reinforcement fine-tuning (RFT) via GRPO. Our results demonstrate that BERT-style encoder-only transformers consistently outperform much larger decoder-only LLMs across all settings. Further mechanistic analysis reveals that the central bottleneck is not knowledge but expression: linear probing of frozen hidden states recovers argument relations at 69{--}80{\%} accuracy from base models, yet edge generation remains poor across all paradigms. Attention analysis confirms a universal structural blindness across 27 tested conditions."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rezkellah-etal-2026-encoders">
<titleInfo>
<title>Encoders Strike Back: Why and How a Strong Discriminative Baseline Outperforms LLMs in Argument Mining Across Three Heterogeneous Datasets</title>
</titleInfo>
<name type="personal">
<namePart type="given">Fatma-Zohra</namePart>
<namePart type="family">Rezkellah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ghassen</namePart>
<namePart type="family">Karray</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Reto</namePart>
<namePart type="family">Gubelmann</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>We present a comprehensive study of LLMs for argument mining, comparing four learning paradigms against a strong encoder-only baseline across three heterogeneous corpora to assess cross-domain generalisation. We evaluate in-context learning, supervised fine-tuning (SFT), and cold-start reinforcement fine-tuning (RFT) via GRPO. Our results demonstrate that BERT-style encoder-only transformers consistently outperform much larger decoder-only LLMs across all settings. Further mechanistic analysis reveals that the central bottleneck is not knowledge but expression: linear probing of frozen hidden states recovers argument relations at 69–80% accuracy from base models, yet edge generation remains poor across all paradigms. Attention analysis confirms a universal structural blindness across 27 tested conditions.</abstract>
<identifier type="citekey">rezkellah-etal-2026-encoders</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.2358</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2358/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>50696</start>
<end>50713</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Encoders Strike Back: Why and How a Strong Discriminative Baseline Outperforms LLMs in Argument Mining Across Three Heterogeneous Datasets
%A Rezkellah, Fatma-Zohra
%A Karray, Ghassen
%A Gubelmann, Reto
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F rezkellah-etal-2026-encoders
%X We present a comprehensive study of LLMs for argument mining, comparing four learning paradigms against a strong encoder-only baseline across three heterogeneous corpora to assess cross-domain generalisation. We evaluate in-context learning, supervised fine-tuning (SFT), and cold-start reinforcement fine-tuning (RFT) via GRPO. Our results demonstrate that BERT-style encoder-only transformers consistently outperform much larger decoder-only LLMs across all settings. Further mechanistic analysis reveals that the central bottleneck is not knowledge but expression: linear probing of frozen hidden states recovers argument relations at 69–80% accuracy from base models, yet edge generation remains poor across all paradigms. Attention analysis confirms a universal structural blindness across 27 tested conditions.
%R 10.18653/2026.findings-emnlp.2358
%U https://aclanthology.org/2026.findings-emnlp.2358/
%U https://doi.org/10.18653/2026.findings-emnlp.2358
%P 50696-50713
Markdown (Informal)
[Encoders Strike Back: Why and How a Strong Discriminative Baseline Outperforms LLMs in Argument Mining Across Three Heterogeneous Datasets](https://aclanthology.org/2026.findings-emnlp.2358/) (Rezkellah et al., Findings 2026)
ACL