@inproceedings{liu-etal-2026-cognitive-chunking,
title = "Cognitive Chunking for Soft Prompts: Accelerating Compressor Learning via Block-wise Causal Masking",
author = "Liu, Guojie and
Wang, Yiqi and
Yang, Yanfeng and
Fan, Wenqi and
Jian, Songlei and
Zhang, Jianfeng and
Yu, Jie",
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.1159/",
pages = "26550--26565",
ISBN = "979-8-89176-459-0",
abstract = "Providing extensive context via prompting is vital for leveraging the capabilities of Large Language Models (LLMs). However, lengthy contexts significantly increase inference latency. To mitigate this issue, context compression{---}particularly soft prompt compression{---}has emerged as a widely studied solution, which converts long contexts into shorter memory embeddings via a trained compressor. Existing methods typically compress the entire context indiscriminately into a set of memory tokens, requiring the compressor to capture global dependencies and necessitating extensive pre-training data to learn effective patterns. Inspired by the chunking mechanism in human working memory and our empirical observation of spatial specialization within memory embeddings, we propose Parallelized Iterative Compression (PIC). By simply modifying the Transformer{'}s attention mask, PIC explicitly restricts the receptive field of memory tokens to sequential local chunks, thereby significantly reducing the difficulty of compressor training. Experiments across multiple downstream tasks demonstrate that PIC outperforms competitive baselines while dramatically expediting the training process. Finally, we provide an investigation into the emergence of spatial specialization, revealing how tuning strategies and explicit structural priors shape the latent structure of the memory embeddings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-etal-2026-cognitive-chunking">
<titleInfo>
<title>Cognitive Chunking for Soft Prompts: Accelerating Compressor Learning via Block-wise Causal Masking</title>
</titleInfo>
<name type="personal">
<namePart type="given">Guojie</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiqi</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanfeng</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenqi</namePart>
<namePart type="family">Fan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Songlei</namePart>
<namePart type="family">Jian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianfeng</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jie</namePart>
<namePart type="family">Yu</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>Providing extensive context via prompting is vital for leveraging the capabilities of Large Language Models (LLMs). However, lengthy contexts significantly increase inference latency. To mitigate this issue, context compression—particularly soft prompt compression—has emerged as a widely studied solution, which converts long contexts into shorter memory embeddings via a trained compressor. Existing methods typically compress the entire context indiscriminately into a set of memory tokens, requiring the compressor to capture global dependencies and necessitating extensive pre-training data to learn effective patterns. Inspired by the chunking mechanism in human working memory and our empirical observation of spatial specialization within memory embeddings, we propose Parallelized Iterative Compression (PIC). By simply modifying the Transformer’s attention mask, PIC explicitly restricts the receptive field of memory tokens to sequential local chunks, thereby significantly reducing the difficulty of compressor training. Experiments across multiple downstream tasks demonstrate that PIC outperforms competitive baselines while dramatically expediting the training process. Finally, we provide an investigation into the emergence of spatial specialization, revealing how tuning strategies and explicit structural priors shape the latent structure of the memory embeddings.</abstract>
<identifier type="citekey">liu-etal-2026-cognitive-chunking</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1159/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>26550</start>
<end>26565</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Cognitive Chunking for Soft Prompts: Accelerating Compressor Learning via Block-wise Causal Masking
%A Liu, Guojie
%A Wang, Yiqi
%A Yang, Yanfeng
%A Fan, Wenqi
%A Jian, Songlei
%A Zhang, Jianfeng
%A Yu, Jie
%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 liu-etal-2026-cognitive-chunking
%X Providing extensive context via prompting is vital for leveraging the capabilities of Large Language Models (LLMs). However, lengthy contexts significantly increase inference latency. To mitigate this issue, context compression—particularly soft prompt compression—has emerged as a widely studied solution, which converts long contexts into shorter memory embeddings via a trained compressor. Existing methods typically compress the entire context indiscriminately into a set of memory tokens, requiring the compressor to capture global dependencies and necessitating extensive pre-training data to learn effective patterns. Inspired by the chunking mechanism in human working memory and our empirical observation of spatial specialization within memory embeddings, we propose Parallelized Iterative Compression (PIC). By simply modifying the Transformer’s attention mask, PIC explicitly restricts the receptive field of memory tokens to sequential local chunks, thereby significantly reducing the difficulty of compressor training. Experiments across multiple downstream tasks demonstrate that PIC outperforms competitive baselines while dramatically expediting the training process. Finally, we provide an investigation into the emergence of spatial specialization, revealing how tuning strategies and explicit structural priors shape the latent structure of the memory embeddings.
%U https://aclanthology.org/2026.emnlp-main.1159/
%P 26550-26565
Markdown (Informal)
[Cognitive Chunking for Soft Prompts: Accelerating Compressor Learning via Block-wise Causal Masking](https://aclanthology.org/2026.emnlp-main.1159/) (Liu et al., EMNLP 2026)
ACL