@inproceedings{zeng-etal-2026-mechanistic,
title = "Mechanistic Insights into Deferred Semantic Drift in {LLM}s",
author = "Zeng, Jingjie and
Li, Huayang and
Yang, Liang and
Zhang, Shaowu and
Sun, Yuanyuan and
Lin, Hongfei",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.57/",
pages = "1116--1137",
ISBN = "979-8-89176-395-1",
abstract = "Large Language Models (LLMs) face a fundamental challenge with delayed disambiguation: **How is the meaning of an ambiguous word updated when clarifying context arrives only after it has been processed?** While LLMs possess the latent capacity to resolve such ambiguities{---}as revealed when a full, non-causal context is provided{---}their unidirectional architecture prevents immediate updates. We investigate the underlying computational mechanism and show this semantic re-evaluation is deferred to subsequent tokens in a process we term ``Deferred Semantic Drift (DSD)''. Through targeted analysis of attentional pathways, we find that later tokens actively retrieve context-dependent ``informational packets'' from the ambiguous word{'}s value vector to steer the final interpretation. We demonstrate this mechanism in metaphor comprehension and provide causal validation by steering model outputs towards literal or metaphorical meanings via targeted activation interventions. This research uncovers a key computational strategy for meaning construction, offering crucial insights for understanding and guiding the behavior of LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zeng-etal-2026-mechanistic">
<titleInfo>
<title>Mechanistic Insights into Deferred Semantic Drift in LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jingjie</namePart>
<namePart type="family">Zeng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huayang</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liang</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaowu</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuanyuan</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongfei</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Large Language Models (LLMs) face a fundamental challenge with delayed disambiguation: **How is the meaning of an ambiguous word updated when clarifying context arrives only after it has been processed?** While LLMs possess the latent capacity to resolve such ambiguities—as revealed when a full, non-causal context is provided—their unidirectional architecture prevents immediate updates. We investigate the underlying computational mechanism and show this semantic re-evaluation is deferred to subsequent tokens in a process we term “Deferred Semantic Drift (DSD)”. Through targeted analysis of attentional pathways, we find that later tokens actively retrieve context-dependent “informational packets” from the ambiguous word’s value vector to steer the final interpretation. We demonstrate this mechanism in metaphor comprehension and provide causal validation by steering model outputs towards literal or metaphorical meanings via targeted activation interventions. This research uncovers a key computational strategy for meaning construction, offering crucial insights for understanding and guiding the behavior of LLMs.</abstract>
<identifier type="citekey">zeng-etal-2026-mechanistic</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.57/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>1116</start>
<end>1137</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mechanistic Insights into Deferred Semantic Drift in LLMs
%A Zeng, Jingjie
%A Li, Huayang
%A Yang, Liang
%A Zhang, Shaowu
%A Sun, Yuanyuan
%A Lin, Hongfei
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F zeng-etal-2026-mechanistic
%X Large Language Models (LLMs) face a fundamental challenge with delayed disambiguation: **How is the meaning of an ambiguous word updated when clarifying context arrives only after it has been processed?** While LLMs possess the latent capacity to resolve such ambiguities—as revealed when a full, non-causal context is provided—their unidirectional architecture prevents immediate updates. We investigate the underlying computational mechanism and show this semantic re-evaluation is deferred to subsequent tokens in a process we term “Deferred Semantic Drift (DSD)”. Through targeted analysis of attentional pathways, we find that later tokens actively retrieve context-dependent “informational packets” from the ambiguous word’s value vector to steer the final interpretation. We demonstrate this mechanism in metaphor comprehension and provide causal validation by steering model outputs towards literal or metaphorical meanings via targeted activation interventions. This research uncovers a key computational strategy for meaning construction, offering crucial insights for understanding and guiding the behavior of LLMs.
%U https://aclanthology.org/2026.findings-acl.57/
%P 1116-1137
Markdown (Informal)
[Mechanistic Insights into Deferred Semantic Drift in LLMs](https://aclanthology.org/2026.findings-acl.57/) (Zeng et al., Findings 2026)
ACL
- Jingjie Zeng, Huayang Li, Liang Yang, Shaowu Zhang, Yuanyuan Sun, and Hongfei Lin. 2026. Mechanistic Insights into Deferred Semantic Drift in LLMs. In Findings of the Association for Computational Linguistics: ACL 2026, pages 1116–1137, San Diego, California, United States. Association for Computational Linguistics.