@inproceedings{kowalyshyn-etal-2026-belief,
title = "Belief or Circuitry? Causal Evidence for In-Context Graph Learning",
author = "Kowalyshyn, Katharine and
Duggan, Timothy and
Little, Daniel and
Hughes, Michael C",
editor = "Arad, Dana and
Belinkov, Yonatan and
Chen, Hanjie and
Feldhus, Nils and
Mueller, Aaron and
Mohebbi, Hosein and
Sarti, Gabriele and
Tutek, Martin",
booktitle = "Proceedings of the 9th {B}lackbox{NLP} Workshop: Analyzing and Interpreting Neural Networks for {NLP}",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.blackboxnlp-1.29/",
pages = "477--486",
ISBN = "979-8-89176-476-7",
abstract = "How do LLMs learn in-context? Is it by pattern-matching recent tokens, or by inferring latent structure? We probe this question using a toy graph random-walk across two competing graph structures. We present behavioral, representational, and causal evidence that neither account alone is sufficient. First, a complexity-weighted sigmoid model captures topology-dependent evidence accumulation, and PCA shows that both graph topologies are visually recoverable at intermediate mixture ratios. Second, residual-stream activation patching and graph-difference steering causally intervene on this graph-family signal: late-layer patching almost fully transfers the clean graph preference, while linear steering moves predictions in the intended direction and fails under norm-matched and label-shuffled controls. An explicit transition-cache baseline reproduces the fitted asymmetry and held-out-edge transfer, however, so these results do not by themselves distinguish global inference from local copying. Taken together, our findings are consistent with a dual-mechanism account in which structure inference and induction circuits operate in parallel, but do not uniquely identify it."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kowalyshyn-etal-2026-belief">
<titleInfo>
<title>Belief or Circuitry? Causal Evidence for In-Context Graph Learning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Katharine</namePart>
<namePart type="family">Kowalyshyn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Timothy</namePart>
<namePart type="family">Duggan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daniel</namePart>
<namePart type="family">Little</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Michael</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Hughes</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 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dana</namePart>
<namePart type="family">Arad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yonatan</namePart>
<namePart type="family">Belinkov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanjie</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nils</namePart>
<namePart type="family">Feldhus</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Mueller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hosein</namePart>
<namePart type="family">Mohebbi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gabriele</namePart>
<namePart type="family">Sarti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Martin</namePart>
<namePart type="family">Tutek</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">979-8-89176-476-7</identifier>
</relatedItem>
<abstract>How do LLMs learn in-context? Is it by pattern-matching recent tokens, or by inferring latent structure? We probe this question using a toy graph random-walk across two competing graph structures. We present behavioral, representational, and causal evidence that neither account alone is sufficient. First, a complexity-weighted sigmoid model captures topology-dependent evidence accumulation, and PCA shows that both graph topologies are visually recoverable at intermediate mixture ratios. Second, residual-stream activation patching and graph-difference steering causally intervene on this graph-family signal: late-layer patching almost fully transfers the clean graph preference, while linear steering moves predictions in the intended direction and fails under norm-matched and label-shuffled controls. An explicit transition-cache baseline reproduces the fitted asymmetry and held-out-edge transfer, however, so these results do not by themselves distinguish global inference from local copying. Taken together, our findings are consistent with a dual-mechanism account in which structure inference and induction circuits operate in parallel, but do not uniquely identify it.</abstract>
<identifier type="citekey">kowalyshyn-etal-2026-belief</identifier>
<location>
<url>https://aclanthology.org/2026.blackboxnlp-1.29/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>477</start>
<end>486</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Belief or Circuitry? Causal Evidence for In-Context Graph Learning
%A Kowalyshyn, Katharine
%A Duggan, Timothy
%A Little, Daniel
%A Hughes, Michael C.
%Y Arad, Dana
%Y Belinkov, Yonatan
%Y Chen, Hanjie
%Y Feldhus, Nils
%Y Mueller, Aaron
%Y Mohebbi, Hosein
%Y Sarti, Gabriele
%Y Tutek, Martin
%S Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-476-7
%F kowalyshyn-etal-2026-belief
%X How do LLMs learn in-context? Is it by pattern-matching recent tokens, or by inferring latent structure? We probe this question using a toy graph random-walk across two competing graph structures. We present behavioral, representational, and causal evidence that neither account alone is sufficient. First, a complexity-weighted sigmoid model captures topology-dependent evidence accumulation, and PCA shows that both graph topologies are visually recoverable at intermediate mixture ratios. Second, residual-stream activation patching and graph-difference steering causally intervene on this graph-family signal: late-layer patching almost fully transfers the clean graph preference, while linear steering moves predictions in the intended direction and fails under norm-matched and label-shuffled controls. An explicit transition-cache baseline reproduces the fitted asymmetry and held-out-edge transfer, however, so these results do not by themselves distinguish global inference from local copying. Taken together, our findings are consistent with a dual-mechanism account in which structure inference and induction circuits operate in parallel, but do not uniquely identify it.
%U https://aclanthology.org/2026.blackboxnlp-1.29/
%P 477-486
Markdown (Informal)
[Belief or Circuitry? Causal Evidence for In-Context Graph Learning](https://aclanthology.org/2026.blackboxnlp-1.29/) (Kowalyshyn et al., BlackboxNLP 2026)
ACL