@inproceedings{wu-etal-2026-pragmatic,
title = "Pragmatic Semantic Communication of Scene Description for Blind and Low-Vision Users",
author = "Wu, Peng and
Chen, Ruxiao and
Guo, Shuaishuai",
editor = "T.Y.S.S., Santosh and
Ghosh, Sreyan and
Duraiswami, Ramani and
Watanabe, Shinji and
Arora, Siddhant and
Kumar, Sonal and
Adi, Yossi and
Manocha, Dinesh and
Garcia Perera, Leibny Paola",
booktitle = "The Proceedings of Speech and Audio Language Models Workshop (2nd Edition)",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.salma-1.5/",
pages = "45--52",
ISBN = "979-8-89176-496-5",
abstract = "Artificial intelligence (AI) assistants for blind and low-vision (BLV) users deliver scene descriptions through a serial auditory channel, often followed by spoken user queries. A central challenge is that scene-description pipelines typically prioritize visually or corpus-salient content, whereas users' follow-up questions often depend on low-salience attributes that are omitted during generation or compression. To diagnose where this mismatch arises, we introduce an encoder-agnostic framework that decomposes scene-description systems into four stages: visual encoding, length control, content selection, and audio scheduling. Applying this framework to 200 blind-user-captured VizWiz images and 500 questions reveals a \textit{salience{--}relevance mismatch}: salience-based compression preserves identity information but discards low-salience attributes needed for attribute-bound questions. Improving the visual-encoding stage by replacing BLIP-base with stronger open-weight MLLMs raises semantic retention but does not eliminate this mismatch. As a lightweight repair, we propose CA-Keyword, which explicitly preserves attribute terms and recovers color-question answerability by 0.23 while producing descriptions 37{\%} shorter than the sentence/"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wu-etal-2026-pragmatic">
<titleInfo>
<title>Pragmatic Semantic Communication of Scene Description for Blind and Low-Vision Users</title>
</titleInfo>
<name type="personal">
<namePart type="given">Peng</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruxiao</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuaishuai</namePart>
<namePart type="family">Guo</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>The Proceedings of Speech and Audio Language Models Workshop (2nd Edition)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Santosh</namePart>
<namePart type="family">T.Y.S.S.</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sreyan</namePart>
<namePart type="family">Ghosh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ramani</namePart>
<namePart type="family">Duraiswami</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shinji</namePart>
<namePart type="family">Watanabe</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siddhant</namePart>
<namePart type="family">Arora</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sonal</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yossi</namePart>
<namePart type="family">Adi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dinesh</namePart>
<namePart type="family">Manocha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leibny</namePart>
<namePart type="given">Paola</namePart>
<namePart type="family">Garcia Perera</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">979-8-89176-496-5</identifier>
</relatedItem>
<abstract>Artificial intelligence (AI) assistants for blind and low-vision (BLV) users deliver scene descriptions through a serial auditory channel, often followed by spoken user queries. A central challenge is that scene-description pipelines typically prioritize visually or corpus-salient content, whereas users’ follow-up questions often depend on low-salience attributes that are omitted during generation or compression. To diagnose where this mismatch arises, we introduce an encoder-agnostic framework that decomposes scene-description systems into four stages: visual encoding, length control, content selection, and audio scheduling. Applying this framework to 200 blind-user-captured VizWiz images and 500 questions reveals a salience–relevance mismatch: salience-based compression preserves identity information but discards low-salience attributes needed for attribute-bound questions. Improving the visual-encoding stage by replacing BLIP-base with stronger open-weight MLLMs raises semantic retention but does not eliminate this mismatch. As a lightweight repair, we propose CA-Keyword, which explicitly preserves attribute terms and recovers color-question answerability by 0.23 while producing descriptions 37% shorter than the sentence/</abstract>
<identifier type="citekey">wu-etal-2026-pragmatic</identifier>
<location>
<url>https://aclanthology.org/2026.salma-1.5/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>45</start>
<end>52</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Pragmatic Semantic Communication of Scene Description for Blind and Low-Vision Users
%A Wu, Peng
%A Chen, Ruxiao
%A Guo, Shuaishuai
%Y T.Y.S.S., Santosh
%Y Ghosh, Sreyan
%Y Duraiswami, Ramani
%Y Watanabe, Shinji
%Y Arora, Siddhant
%Y Kumar, Sonal
%Y Adi, Yossi
%Y Manocha, Dinesh
%Y Garcia Perera, Leibny Paola
%S The Proceedings of Speech and Audio Language Models Workshop (2nd Edition)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-496-5
%F wu-etal-2026-pragmatic
%X Artificial intelligence (AI) assistants for blind and low-vision (BLV) users deliver scene descriptions through a serial auditory channel, often followed by spoken user queries. A central challenge is that scene-description pipelines typically prioritize visually or corpus-salient content, whereas users’ follow-up questions often depend on low-salience attributes that are omitted during generation or compression. To diagnose where this mismatch arises, we introduce an encoder-agnostic framework that decomposes scene-description systems into four stages: visual encoding, length control, content selection, and audio scheduling. Applying this framework to 200 blind-user-captured VizWiz images and 500 questions reveals a salience–relevance mismatch: salience-based compression preserves identity information but discards low-salience attributes needed for attribute-bound questions. Improving the visual-encoding stage by replacing BLIP-base with stronger open-weight MLLMs raises semantic retention but does not eliminate this mismatch. As a lightweight repair, we propose CA-Keyword, which explicitly preserves attribute terms and recovers color-question answerability by 0.23 while producing descriptions 37% shorter than the sentence/
%U https://aclanthology.org/2026.salma-1.5/
%P 45-52
Markdown (Informal)
[Pragmatic Semantic Communication of Scene Description for Blind and Low-Vision Users](https://aclanthology.org/2026.salma-1.5/) (Wu et al., SALMA 2026)
ACL