@inproceedings{le-2026-kvochurhegel-stancenakba,
title = "{K}vochur{H}egel at {S}tance{N}akba: Robust Stance Detection with Regularized Natural Language Inference",
author = "Le, Minh-Hoang",
editor = "Jarrar, Mustafa and
El-Haj, Mo and
Haddad, Amal and
Atiani, Serin and
Abudalfa, Shadi and
Regier, Terry and
Rayson, Paul and
Sima{'}an, Khalil and
Mansour, Camille",
booktitle = "Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ {LREC} 2026",
month = may,
year = "2026",
address = "Palma, Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.nakbanlp-1.13/",
doi = "10.63317/3qqbomcjihz4",
pages = "113--117",
abstract = "Actor-level stance detection over noisy, politically sensitive data can present challenges that standard training procedures fail to handle reliably. This paper presents KvochurHegel, our submission to the StanceNakba 2026 Shared Task, which addresses these challenges by framing stance classification as Natural Language Inference (NLI) to capture actor-level granularity. The official StanceNakba dataset contains high label noise and topic-correlated spurious features, such as texts discussing unrelated global conflicts using in-domain political vocabulary. To handle these conditions within a three-class schema, we construct templates encoding stance hypotheses for specific actors (e.g., ``The author expresses support for Palestine'') and introduce a broadened neutral class designed to absorb spurious out-of-domain inputs. A DeBERTa-v3 Cross-Encoder independently evaluates the entailment between the input text and each class-specific hypothesis. Because standard cross-entropy training tends to memorize contradictory annotations under these conditions, we regularize the training procedure with R-Drop and label smoothing. This regularized setup likely contributed to robustness against distribution shifts between the competition{'}s evaluation phases (the public leaderboard and private test set), allowing our model to improve from a Macro-F1 of 0.9094 to 0.9384 without requiring large generative models, cross-validation, or inference-time ensembling."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="le-2026-kvochurhegel-stancenakba">
<titleInfo>
<title>KvochurHegel at StanceNakba: Robust Stance Detection with Regularized Natural Language Inference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Minh-Hoang</namePart>
<namePart type="family">Le</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 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mustafa</namePart>
<namePart type="family">Jarrar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mo</namePart>
<namePart type="family">El-Haj</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amal</namePart>
<namePart type="family">Haddad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Serin</namePart>
<namePart type="family">Atiani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shadi</namePart>
<namePart type="family">Abudalfa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Terry</namePart>
<namePart type="family">Regier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paul</namePart>
<namePart type="family">Rayson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Khalil</namePart>
<namePart type="family">Sima’an</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Camille</namePart>
<namePart type="family">Mansour</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>Actor-level stance detection over noisy, politically sensitive data can present challenges that standard training procedures fail to handle reliably. This paper presents KvochurHegel, our submission to the StanceNakba 2026 Shared Task, which addresses these challenges by framing stance classification as Natural Language Inference (NLI) to capture actor-level granularity. The official StanceNakba dataset contains high label noise and topic-correlated spurious features, such as texts discussing unrelated global conflicts using in-domain political vocabulary. To handle these conditions within a three-class schema, we construct templates encoding stance hypotheses for specific actors (e.g., “The author expresses support for Palestine”) and introduce a broadened neutral class designed to absorb spurious out-of-domain inputs. A DeBERTa-v3 Cross-Encoder independently evaluates the entailment between the input text and each class-specific hypothesis. Because standard cross-entropy training tends to memorize contradictory annotations under these conditions, we regularize the training procedure with R-Drop and label smoothing. This regularized setup likely contributed to robustness against distribution shifts between the competition’s evaluation phases (the public leaderboard and private test set), allowing our model to improve from a Macro-F1 of 0.9094 to 0.9384 without requiring large generative models, cross-validation, or inference-time ensembling.</abstract>
<identifier type="citekey">le-2026-kvochurhegel-stancenakba</identifier>
<identifier type="doi">10.63317/3qqbomcjihz4</identifier>
<location>
<url>https://aclanthology.org/2026.nakbanlp-1.13/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>113</start>
<end>117</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T KvochurHegel at StanceNakba: Robust Stance Detection with Regularized Natural Language Inference
%A Le, Minh-Hoang
%Y Jarrar, Mustafa
%Y El-Haj, Mo
%Y Haddad, Amal
%Y Atiani, Serin
%Y Abudalfa, Shadi
%Y Regier, Terry
%Y Rayson, Paul
%Y Sima’an, Khalil
%Y Mansour, Camille
%S Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F le-2026-kvochurhegel-stancenakba
%X Actor-level stance detection over noisy, politically sensitive data can present challenges that standard training procedures fail to handle reliably. This paper presents KvochurHegel, our submission to the StanceNakba 2026 Shared Task, which addresses these challenges by framing stance classification as Natural Language Inference (NLI) to capture actor-level granularity. The official StanceNakba dataset contains high label noise and topic-correlated spurious features, such as texts discussing unrelated global conflicts using in-domain political vocabulary. To handle these conditions within a three-class schema, we construct templates encoding stance hypotheses for specific actors (e.g., “The author expresses support for Palestine”) and introduce a broadened neutral class designed to absorb spurious out-of-domain inputs. A DeBERTa-v3 Cross-Encoder independently evaluates the entailment between the input text and each class-specific hypothesis. Because standard cross-entropy training tends to memorize contradictory annotations under these conditions, we regularize the training procedure with R-Drop and label smoothing. This regularized setup likely contributed to robustness against distribution shifts between the competition’s evaluation phases (the public leaderboard and private test set), allowing our model to improve from a Macro-F1 of 0.9094 to 0.9384 without requiring large generative models, cross-validation, or inference-time ensembling.
%R 10.63317/3qqbomcjihz4
%U https://aclanthology.org/2026.nakbanlp-1.13/
%U https://doi.org/10.63317/3qqbomcjihz4
%P 113-117
Markdown (Informal)
[KvochurHegel at StanceNakba: Robust Stance Detection with Regularized Natural Language Inference](https://aclanthology.org/2026.nakbanlp-1.13/) (Le, NakbaNLP 2026)
ACL