@inproceedings{kulumba-etal-2026-language,
title = "Language-Switching Triggers Take a Latent Detour Through Language Models",
author = "Kulumba, Francis and
Antoun, Wissam and
Lasnier, Th{\'e}o and
Sagot, Beno{\^i}t and
Seddah, Djam{\'e}",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2664/",
pages = "60647--60661",
ISBN = "979-8-89176-459-0",
abstract = "Backdoor attacks on language models are widely studied, yet the internal mechanisms by which a trigger sequence hijacks model computations remain poorly understood. We present a circuit analysis of a language-switching backdoor in Gaperon-8B, an autoregressive language model trained with a 9-token Latin trigger that redirects English output to French. We identify a three-phase circuit: (1) distributed attention heads at early layers compose the trigger tokens into the last sequence position; (2) the resulting signal propagates through mid-layers in a subspace orthogonal to the model{'}s natural language-identity direction; (3) the MLP at the final layer converts this latent signal into French logits. The entire circuit flows through a serial bottleneck at a single position: corrupting that position at any layer entirely mitigate the trigger but also hinder the model{'}s capabilities. The orthogonal latent encoding suggests that defenses that search for language-like signals in intermediate representations would miss this trigger entirely."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kulumba-etal-2026-language">
<titleInfo>
<title>Language-Switching Triggers Take a Latent Detour Through Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Francis</namePart>
<namePart type="family">Kulumba</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wissam</namePart>
<namePart type="family">Antoun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Théo</namePart>
<namePart type="family">Lasnier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Benoît</namePart>
<namePart type="family">Sagot</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Djamé</namePart>
<namePart type="family">Seddah</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 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Backdoor attacks on language models are widely studied, yet the internal mechanisms by which a trigger sequence hijacks model computations remain poorly understood. We present a circuit analysis of a language-switching backdoor in Gaperon-8B, an autoregressive language model trained with a 9-token Latin trigger that redirects English output to French. We identify a three-phase circuit: (1) distributed attention heads at early layers compose the trigger tokens into the last sequence position; (2) the resulting signal propagates through mid-layers in a subspace orthogonal to the model’s natural language-identity direction; (3) the MLP at the final layer converts this latent signal into French logits. The entire circuit flows through a serial bottleneck at a single position: corrupting that position at any layer entirely mitigate the trigger but also hinder the model’s capabilities. The orthogonal latent encoding suggests that defenses that search for language-like signals in intermediate representations would miss this trigger entirely.</abstract>
<identifier type="citekey">kulumba-etal-2026-language</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2664/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>60647</start>
<end>60661</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Language-Switching Triggers Take a Latent Detour Through Language Models
%A Kulumba, Francis
%A Antoun, Wissam
%A Lasnier, Théo
%A Sagot, Benoît
%A Seddah, Djamé
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F kulumba-etal-2026-language
%X Backdoor attacks on language models are widely studied, yet the internal mechanisms by which a trigger sequence hijacks model computations remain poorly understood. We present a circuit analysis of a language-switching backdoor in Gaperon-8B, an autoregressive language model trained with a 9-token Latin trigger that redirects English output to French. We identify a three-phase circuit: (1) distributed attention heads at early layers compose the trigger tokens into the last sequence position; (2) the resulting signal propagates through mid-layers in a subspace orthogonal to the model’s natural language-identity direction; (3) the MLP at the final layer converts this latent signal into French logits. The entire circuit flows through a serial bottleneck at a single position: corrupting that position at any layer entirely mitigate the trigger but also hinder the model’s capabilities. The orthogonal latent encoding suggests that defenses that search for language-like signals in intermediate representations would miss this trigger entirely.
%U https://aclanthology.org/2026.emnlp-main.2664/
%P 60647-60661
Markdown (Informal)
[Language-Switching Triggers Take a Latent Detour Through Language Models](https://aclanthology.org/2026.emnlp-main.2664/) (Kulumba et al., EMNLP 2026)
ACL