@inproceedings{duan-etal-2026-mechanistic,
title = "Mechanistic Interpretability for Understanding Language Abilities in Large Language Models: A Survey",
author = "Duan, Xufeng and
Yao, Zhaoqian and
Zhou, Xinyu and
Fu, Zihao and
Kan, Beijia and
Wang, Yibo and
Xiao, Bei and
Zhang, Hengyuan and
Tang, Xuemei and
Nie, Ercong and
Cai, Zhenguang",
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.464/",
pages = "9735--9759",
ISBN = "979-8-89176-462-0",
abstract = "Large Language Models (LLMs) exhibit strong performance across language tasks, raising the question of whether this success reflects linguistic abstraction or reliance on surface-level statistical patterns. This survey reviews recent advances in mechanistic interpretability (MI) for analyzing candidate internal mechanisms underlying linguistic competence in LLMs. We trace the shift from model-agnostic analyses to transformer-specific methods and summarize core approaches including probing, neuron- and circuit-level analysis, and causal interventions. Synthesizing empirical findings, we argue that current evidence supports an intermediate view: LLMs appear to use reusable mechanisms for some grammatical behaviors, while these mechanisms remain frequency-conditioned, distributed, and only partially understood. We further review interpretability studies of multilingual models, highlighting evidence for both language-specific and shared representations. Finally, we outline open challenges in scalability, benchmarking, and ethics, and discuss directions toward a more cautious mechanistic account of language in LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="duan-etal-2026-mechanistic">
<titleInfo>
<title>Mechanistic Interpretability for Understanding Language Abilities in Large Language Models: A Survey</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xufeng</namePart>
<namePart type="family">Duan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhaoqian</namePart>
<namePart type="family">Yao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xinyu</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zihao</namePart>
<namePart type="family">Fu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Beijia</namePart>
<namePart type="family">Kan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yibo</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bei</namePart>
<namePart type="family">Xiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hengyuan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuemei</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ercong</namePart>
<namePart type="family">Nie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhenguang</namePart>
<namePart type="family">Cai</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>Large Language Models (LLMs) exhibit strong performance across language tasks, raising the question of whether this success reflects linguistic abstraction or reliance on surface-level statistical patterns. This survey reviews recent advances in mechanistic interpretability (MI) for analyzing candidate internal mechanisms underlying linguistic competence in LLMs. We trace the shift from model-agnostic analyses to transformer-specific methods and summarize core approaches including probing, neuron- and circuit-level analysis, and causal interventions. Synthesizing empirical findings, we argue that current evidence supports an intermediate view: LLMs appear to use reusable mechanisms for some grammatical behaviors, while these mechanisms remain frequency-conditioned, distributed, and only partially understood. We further review interpretability studies of multilingual models, highlighting evidence for both language-specific and shared representations. Finally, we outline open challenges in scalability, benchmarking, and ethics, and discuss directions toward a more cautious mechanistic account of language in LLMs.</abstract>
<identifier type="citekey">duan-etal-2026-mechanistic</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.464/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>9735</start>
<end>9759</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mechanistic Interpretability for Understanding Language Abilities in Large Language Models: A Survey
%A Duan, Xufeng
%A Yao, Zhaoqian
%A Zhou, Xinyu
%A Fu, Zihao
%A Kan, Beijia
%A Wang, Yibo
%A Xiao, Bei
%A Zhang, Hengyuan
%A Tang, Xuemei
%A Nie, Ercong
%A Cai, Zhenguang
%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 duan-etal-2026-mechanistic
%X Large Language Models (LLMs) exhibit strong performance across language tasks, raising the question of whether this success reflects linguistic abstraction or reliance on surface-level statistical patterns. This survey reviews recent advances in mechanistic interpretability (MI) for analyzing candidate internal mechanisms underlying linguistic competence in LLMs. We trace the shift from model-agnostic analyses to transformer-specific methods and summarize core approaches including probing, neuron- and circuit-level analysis, and causal interventions. Synthesizing empirical findings, we argue that current evidence supports an intermediate view: LLMs appear to use reusable mechanisms for some grammatical behaviors, while these mechanisms remain frequency-conditioned, distributed, and only partially understood. We further review interpretability studies of multilingual models, highlighting evidence for both language-specific and shared representations. Finally, we outline open challenges in scalability, benchmarking, and ethics, and discuss directions toward a more cautious mechanistic account of language in LLMs.
%U https://aclanthology.org/2026.findings-emnlp.464/
%P 9735-9759
Markdown (Informal)
[Mechanistic Interpretability for Understanding Language Abilities in Large Language Models: A Survey](https://aclanthology.org/2026.findings-emnlp.464/) (Duan et al., Findings 2026)
ACL
- Xufeng Duan, Zhaoqian Yao, Xinyu Zhou, Zihao Fu, Beijia Kan, Yibo Wang, Bei Xiao, Hengyuan Zhang, Xuemei Tang, Ercong Nie, and Zhenguang Cai. 2026. Mechanistic Interpretability for Understanding Language Abilities in Large Language Models: A Survey. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 9735–9759, Budapest, Hungary. Association for Computational Linguistics.