@inproceedings{shin-lee-2026-mass,
title = "Mass-Tail Attribution Entropy: A Regime-Conditional Decoder for Vision-Language Model Hallucination",
author = "Shin, Jaegwang and
Lee, Suan",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.174/",
pages = "3466--3490",
ISBN = "979-8-89176-444-6",
abstract = "Decoding-time mitigation of vision-language model (VLM) hallucination applies one fixed intervention and typically trades recall for lower hallucination. We trace this trade-off to treating hallucination as a single failure mode. Measuring the normalised entropy H of leave-one-out attributions over panoptic image segments, we find that a VLM{'}s inference mass concentrates in one task-conditional tail of H: captioning in the over-concentration tail (low H, 53{--}66{\%}), binary question-answering in the over-spread tail (high H, 74{--}95{\%}). The pattern holds in 6/6 (backbone {\texttimes} benchmark) cells on LLaVA-1.5-7B and Qwen2.5-VL-7B, survives exact Shapley recomputation, reproduces within AMBER, and is shared by three further backbones. Because errors and correct outputs share the tail{---}H is a regime signal, not a hallucination detector{---}an intervention matched to the wrong tail actively hurts: blank-image contrastive decoding raises captioning CHAIR{\_}S from 49.3 to 60.3. The regime can be read without attribution at inference, from blank-image agreement and on-image token margin; SBC (Segment-Based Conditioning) uses them to route between segment boosting and probabilistic marginalisation. With one shared configuration, SBC lowers CHAIR{\_}I while preserving recall and raises POPE F1 on both backbones (+1.95 / +3.31), at 1.7{--}1.9{\texttimes} greedy cost on captioning and 4.9{\texttimes} on single-token QA."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shin-lee-2026-mass">
<titleInfo>
<title>Mass-Tail Attribution Entropy: A Regime-Conditional Decoder for Vision-Language Model Hallucination</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jaegwang</namePart>
<namePart type="family">Shin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Suan</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Decoding-time mitigation of vision-language model (VLM) hallucination applies one fixed intervention and typically trades recall for lower hallucination. We trace this trade-off to treating hallucination as a single failure mode. Measuring the normalised entropy H of leave-one-out attributions over panoptic image segments, we find that a VLM’s inference mass concentrates in one task-conditional tail of H: captioning in the over-concentration tail (low H, 53–66%), binary question-answering in the over-spread tail (high H, 74–95%). The pattern holds in 6/6 (backbone × benchmark) cells on LLaVA-1.5-7B and Qwen2.5-VL-7B, survives exact Shapley recomputation, reproduces within AMBER, and is shared by three further backbones. Because errors and correct outputs share the tail—H is a regime signal, not a hallucination detector—an intervention matched to the wrong tail actively hurts: blank-image contrastive decoding raises captioning CHAIR_S from 49.3 to 60.3. The regime can be read without attribution at inference, from blank-image agreement and on-image token margin; SBC (Segment-Based Conditioning) uses them to route between segment boosting and probabilistic marginalisation. With one shared configuration, SBC lowers CHAIR_I while preserving recall and raises POPE F1 on both backbones (+1.95 / +3.31), at 1.7–1.9× greedy cost on captioning and 4.9× on single-token QA.</abstract>
<identifier type="citekey">shin-lee-2026-mass</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.174/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3466</start>
<end>3490</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mass-Tail Attribution Entropy: A Regime-Conditional Decoder for Vision-Language Model Hallucination
%A Shin, Jaegwang
%A Lee, Suan
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F shin-lee-2026-mass
%X Decoding-time mitigation of vision-language model (VLM) hallucination applies one fixed intervention and typically trades recall for lower hallucination. We trace this trade-off to treating hallucination as a single failure mode. Measuring the normalised entropy H of leave-one-out attributions over panoptic image segments, we find that a VLM’s inference mass concentrates in one task-conditional tail of H: captioning in the over-concentration tail (low H, 53–66%), binary question-answering in the over-spread tail (high H, 74–95%). The pattern holds in 6/6 (backbone × benchmark) cells on LLaVA-1.5-7B and Qwen2.5-VL-7B, survives exact Shapley recomputation, reproduces within AMBER, and is shared by three further backbones. Because errors and correct outputs share the tail—H is a regime signal, not a hallucination detector—an intervention matched to the wrong tail actively hurts: blank-image contrastive decoding raises captioning CHAIR_S from 49.3 to 60.3. The regime can be read without attribution at inference, from blank-image agreement and on-image token margin; SBC (Segment-Based Conditioning) uses them to route between segment boosting and probabilistic marginalisation. With one shared configuration, SBC lowers CHAIR_I while preserving recall and raises POPE F1 on both backbones (+1.95 / +3.31), at 1.7–1.9× greedy cost on captioning and 4.9× on single-token QA.
%U https://aclanthology.org/2026.findings-aacl.174/
%P 3466-3490
Markdown (Informal)
[Mass-Tail Attribution Entropy: A Regime-Conditional Decoder for Vision-Language Model Hallucination](https://aclanthology.org/2026.findings-aacl.174/) (Shin & Lee, Findings 2026)
ACL