@inproceedings{canby-etal-2026-analyzing,
title = "Analyzing Multilingual Parametric Knowledge Retrieval with Sparse Autoencoders",
author = "Canby, Marc E. and
Cho, Ikhyun and
Hockenmaier, Julia",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.163/",
pages = "3250--3264",
ISBN = "979-8-89176-444-6",
abstract = "Multilingual language models can answer factual questions across languages, but their answers are not always consistent across languages, and the internal pathways used to retrieve this knowledge remain unclear. A model may rely on language-specific shortcuts, shared multilingual representations, or another language as an internal pivot. We investigate these possibilities in Llama-3.1-8B and Gemma-2-9B with sparse autoencoders (SAEs), identifying, analyzing, and causally intervening on language-associated features during factual recall. On CUS-QA (CITATION), a parallel QA dataset spanning Czech, Slovak, Ukrainian, and English, we identify SAE features associated with particular subsets of languages. We then decompose each question{'}s activations into these subsets to measure reliance on input-language, multilingual, or pivot-like representations across layers. Across both models, factual recall is heterogeneous: different input languages and questions exhibit different language-associated pathways. Steering substantially changes cross-lingual answer alignment and sometimes increases correctness in Llama, while it produces only small changes in Gemma. These results show that multilingual factual recall is mediated by identifiable SAE features that can be used to diagnose and control how models access knowledge across languages."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="canby-etal-2026-analyzing">
<titleInfo>
<title>Analyzing Multilingual Parametric Knowledge Retrieval with Sparse Autoencoders</title>
</titleInfo>
<name type="personal">
<namePart type="given">Marc</namePart>
<namePart type="given">E</namePart>
<namePart type="family">Canby</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ikhyun</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Julia</namePart>
<namePart type="family">Hockenmaier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Multilingual language models can answer factual questions across languages, but their answers are not always consistent across languages, and the internal pathways used to retrieve this knowledge remain unclear. A model may rely on language-specific shortcuts, shared multilingual representations, or another language as an internal pivot. We investigate these possibilities in Llama-3.1-8B and Gemma-2-9B with sparse autoencoders (SAEs), identifying, analyzing, and causally intervening on language-associated features during factual recall. On CUS-QA (CITATION), a parallel QA dataset spanning Czech, Slovak, Ukrainian, and English, we identify SAE features associated with particular subsets of languages. We then decompose each question’s activations into these subsets to measure reliance on input-language, multilingual, or pivot-like representations across layers. Across both models, factual recall is heterogeneous: different input languages and questions exhibit different language-associated pathways. Steering substantially changes cross-lingual answer alignment and sometimes increases correctness in Llama, while it produces only small changes in Gemma. These results show that multilingual factual recall is mediated by identifiable SAE features that can be used to diagnose and control how models access knowledge across languages.</abstract>
<identifier type="citekey">canby-etal-2026-analyzing</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.163/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3250</start>
<end>3264</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Analyzing Multilingual Parametric Knowledge Retrieval with Sparse Autoencoders
%A Canby, Marc E.
%A Cho, Ikhyun
%A Hockenmaier, Julia
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F canby-etal-2026-analyzing
%X Multilingual language models can answer factual questions across languages, but their answers are not always consistent across languages, and the internal pathways used to retrieve this knowledge remain unclear. A model may rely on language-specific shortcuts, shared multilingual representations, or another language as an internal pivot. We investigate these possibilities in Llama-3.1-8B and Gemma-2-9B with sparse autoencoders (SAEs), identifying, analyzing, and causally intervening on language-associated features during factual recall. On CUS-QA (CITATION), a parallel QA dataset spanning Czech, Slovak, Ukrainian, and English, we identify SAE features associated with particular subsets of languages. We then decompose each question’s activations into these subsets to measure reliance on input-language, multilingual, or pivot-like representations across layers. Across both models, factual recall is heterogeneous: different input languages and questions exhibit different language-associated pathways. Steering substantially changes cross-lingual answer alignment and sometimes increases correctness in Llama, while it produces only small changes in Gemma. These results show that multilingual factual recall is mediated by identifiable SAE features that can be used to diagnose and control how models access knowledge across languages.
%U https://aclanthology.org/2026.findings-aacl.163/
%P 3250-3264
Markdown (Informal)
[Analyzing Multilingual Parametric Knowledge Retrieval with Sparse Autoencoders](https://aclanthology.org/2026.findings-aacl.163/) (Canby et al., Findings 2026)
ACL