@inproceedings{romanski-picek-2026-beyond,
title = "Beyond Fooling: Model Manipulation Under Explanation-aware Training",
author = "Romanski, Tabio and
Picek, Stjepan",
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.2089/",
doi = "10.18653/2026.findings-emnlp.2089",
pages = "44490--44509",
ISBN = "979-8-89176-462-0",
abstract = "Feature-level explanations are commonly used to interpret transformer-based NLP models. Still, little is known about how explanation-aware objectives influence model behaviour during training. While prior work has demonstrated training-time manipulation of explanations in vision models, its implications for transformers and token-level explainability remain unexplored. We study training-time manipulation of token-level explanations in transformer-based NLP classifiers and introduce sequence-aware objectives suited to text input. We show that explanation-aware training systematically alters token relevance patterns while largely preserving task accuracy. Importantly, masking and cross-method evaluations reveal that these attribution changes can coincide with shifts in functional model reliance rather than isolated failures of specific explanation methods. Our results suggest that apparent vulnerabilities of feature-level explanations can reflect deeper model adaptations, underscoring the need to consider learning dynamics when interpreting explanation robustness."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="romanski-picek-2026-beyond">
<titleInfo>
<title>Beyond Fooling: Model Manipulation Under Explanation-aware Training</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tabio</namePart>
<namePart type="family">Romanski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stjepan</namePart>
<namePart type="family">Picek</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>Feature-level explanations are commonly used to interpret transformer-based NLP models. Still, little is known about how explanation-aware objectives influence model behaviour during training. While prior work has demonstrated training-time manipulation of explanations in vision models, its implications for transformers and token-level explainability remain unexplored. We study training-time manipulation of token-level explanations in transformer-based NLP classifiers and introduce sequence-aware objectives suited to text input. We show that explanation-aware training systematically alters token relevance patterns while largely preserving task accuracy. Importantly, masking and cross-method evaluations reveal that these attribution changes can coincide with shifts in functional model reliance rather than isolated failures of specific explanation methods. Our results suggest that apparent vulnerabilities of feature-level explanations can reflect deeper model adaptations, underscoring the need to consider learning dynamics when interpreting explanation robustness.</abstract>
<identifier type="citekey">romanski-picek-2026-beyond</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.2089</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2089/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>44490</start>
<end>44509</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Beyond Fooling: Model Manipulation Under Explanation-aware Training
%A Romanski, Tabio
%A Picek, Stjepan
%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 romanski-picek-2026-beyond
%X Feature-level explanations are commonly used to interpret transformer-based NLP models. Still, little is known about how explanation-aware objectives influence model behaviour during training. While prior work has demonstrated training-time manipulation of explanations in vision models, its implications for transformers and token-level explainability remain unexplored. We study training-time manipulation of token-level explanations in transformer-based NLP classifiers and introduce sequence-aware objectives suited to text input. We show that explanation-aware training systematically alters token relevance patterns while largely preserving task accuracy. Importantly, masking and cross-method evaluations reveal that these attribution changes can coincide with shifts in functional model reliance rather than isolated failures of specific explanation methods. Our results suggest that apparent vulnerabilities of feature-level explanations can reflect deeper model adaptations, underscoring the need to consider learning dynamics when interpreting explanation robustness.
%R 10.18653/2026.findings-emnlp.2089
%U https://aclanthology.org/2026.findings-emnlp.2089/
%U https://doi.org/10.18653/2026.findings-emnlp.2089
%P 44490-44509
Markdown (Informal)
[Beyond Fooling: Model Manipulation Under Explanation-aware Training](https://aclanthology.org/2026.findings-emnlp.2089/) (Romanski & Picek, Findings 2026)
ACL