@inproceedings{salazar-etal-2026-long,
title = "Long Story Short: Disentangling Compositionality and Long-Caption Understanding in Contrastive {VLM}s",
author = "Salazar, Israfel and
Elliott, Desmond and
Kementchedjhieva, Yova",
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.2131/",
pages = "42994--43008",
ISBN = "979-8-89176-395-1",
abstract = "Contrastive vision-language models (VLMs) have made significant progress in binding visual and textual information, yet understanding long, compositional captions remains an open challenge. While these capabilities are often assumed to be closely related, the conditions under which they reinforce each other remain unclear. In this paper, we empirically analyze when compositional reasoning and long-caption understanding transfer across tasks, and when this relationship fails. Through controlled experiments across diverse training objectives, datasets, and architectural designs, we find a bidirectional but sensitive relationship between the two capabilities. Models trained on poorly grounded captions or with limited parameter updates fail to generalize, while high-quality long-caption data with strong visual grounding promotes both capabilities simultaneously. We further show that architectural choices aimed at preserving general alignment, such as frozen positional embeddings, can inadvertently limit compositional learning. Our analysis provides actionable guidelines for data selection and model design to improve VLM generalization."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="salazar-etal-2026-long">
<titleInfo>
<title>Long Story Short: Disentangling Compositionality and Long-Caption Understanding in Contrastive VLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Israfel</namePart>
<namePart type="family">Salazar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yova</namePart>
<namePart type="family">Kementchedjhieva</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>Contrastive vision-language models (VLMs) have made significant progress in binding visual and textual information, yet understanding long, compositional captions remains an open challenge. While these capabilities are often assumed to be closely related, the conditions under which they reinforce each other remain unclear. In this paper, we empirically analyze when compositional reasoning and long-caption understanding transfer across tasks, and when this relationship fails. Through controlled experiments across diverse training objectives, datasets, and architectural designs, we find a bidirectional but sensitive relationship between the two capabilities. Models trained on poorly grounded captions or with limited parameter updates fail to generalize, while high-quality long-caption data with strong visual grounding promotes both capabilities simultaneously. We further show that architectural choices aimed at preserving general alignment, such as frozen positional embeddings, can inadvertently limit compositional learning. Our analysis provides actionable guidelines for data selection and model design to improve VLM generalization.</abstract>
<identifier type="citekey">salazar-etal-2026-long</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.2131/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>42994</start>
<end>43008</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Long Story Short: Disentangling Compositionality and Long-Caption Understanding in Contrastive VLMs
%A Salazar, Israfel
%A Elliott, Desmond
%A Kementchedjhieva, Yova
%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 salazar-etal-2026-long
%X Contrastive vision-language models (VLMs) have made significant progress in binding visual and textual information, yet understanding long, compositional captions remains an open challenge. While these capabilities are often assumed to be closely related, the conditions under which they reinforce each other remain unclear. In this paper, we empirically analyze when compositional reasoning and long-caption understanding transfer across tasks, and when this relationship fails. Through controlled experiments across diverse training objectives, datasets, and architectural designs, we find a bidirectional but sensitive relationship between the two capabilities. Models trained on poorly grounded captions or with limited parameter updates fail to generalize, while high-quality long-caption data with strong visual grounding promotes both capabilities simultaneously. We further show that architectural choices aimed at preserving general alignment, such as frozen positional embeddings, can inadvertently limit compositional learning. Our analysis provides actionable guidelines for data selection and model design to improve VLM generalization.
%U https://aclanthology.org/2026.findings-acl.2131/
%P 42994-43008
Markdown (Informal)
[Long Story Short: Disentangling Compositionality and Long-Caption Understanding in Contrastive VLMs](https://aclanthology.org/2026.findings-acl.2131/) (Salazar et al., Findings 2026)
ACL