@inproceedings{prabhune-etal-2026-fast,
title = "Fast Listwise Passage Reranking via Semantic Tokens",
author = "Prabhune, Omkar and
Wang, Xijun and
Bi, Nan and
Wang, Haiyan and
Alizadeh, Shima and
Jain, Anoop and
Lou, Yu and
Liang, Junbang",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.217/",
pages = "4897--4910",
ISBN = "979-8-89176-439-2",
abstract = "Listwise passage reranking often achieves strong ranking performance, but it incurs substantial computational cost. We propose FLASH, an efficient listwise reranking framework based on semantic token compression and single-step listwise scoring. FLASH represents each query{--}passage pair with a small set of query-aware semantic tokens and jointly scores all candidates in a single decoder step, enabling query-conditioned joint scoring without expensive token-level listwise processing. We further introduce multi-cardinality training, which allows a single model to support different semantic-token budgets at inference time. We also propose FLASH-LITE, a lightweight variant with offline passage encoding for low-cost online inference. Across TREC and BEIR benchmarks, FLASH improves average ranking effectiveness at comparable online latency and slightly lower FLOPs, while FLASH-Lite substantially reduces query-time cost with a modest effectiveness trade-off. FLASH-LITE enables offline passage encoding, substantially reducing online query-time cost while maintaining competitive reranking performance. Code is provided in the supplementary material."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="prabhune-etal-2026-fast">
<titleInfo>
<title>Fast Listwise Passage Reranking via Semantic Tokens</title>
</titleInfo>
<name type="personal">
<namePart type="given">Omkar</namePart>
<namePart type="family">Prabhune</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xijun</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nan</namePart>
<namePart type="family">Bi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haiyan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shima</namePart>
<namePart type="family">Alizadeh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anoop</namePart>
<namePart type="family">Jain</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Lou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junbang</namePart>
<namePart type="family">Liang</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>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</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">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>Listwise passage reranking often achieves strong ranking performance, but it incurs substantial computational cost. We propose FLASH, an efficient listwise reranking framework based on semantic token compression and single-step listwise scoring. FLASH represents each query–passage pair with a small set of query-aware semantic tokens and jointly scores all candidates in a single decoder step, enabling query-conditioned joint scoring without expensive token-level listwise processing. We further introduce multi-cardinality training, which allows a single model to support different semantic-token budgets at inference time. We also propose FLASH-LITE, a lightweight variant with offline passage encoding for low-cost online inference. Across TREC and BEIR benchmarks, FLASH improves average ranking effectiveness at comparable online latency and slightly lower FLOPs, while FLASH-Lite substantially reduces query-time cost with a modest effectiveness trade-off. FLASH-LITE enables offline passage encoding, substantially reducing online query-time cost while maintaining competitive reranking performance. Code is provided in the supplementary material.</abstract>
<identifier type="citekey">prabhune-etal-2026-fast</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.217/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>4897</start>
<end>4910</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Fast Listwise Passage Reranking via Semantic Tokens
%A Prabhune, Omkar
%A Wang, Xijun
%A Bi, Nan
%A Wang, Haiyan
%A Alizadeh, Shima
%A Jain, Anoop
%A Lou, Yu
%A Liang, Junbang
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F prabhune-etal-2026-fast
%X Listwise passage reranking often achieves strong ranking performance, but it incurs substantial computational cost. We propose FLASH, an efficient listwise reranking framework based on semantic token compression and single-step listwise scoring. FLASH represents each query–passage pair with a small set of query-aware semantic tokens and jointly scores all candidates in a single decoder step, enabling query-conditioned joint scoring without expensive token-level listwise processing. We further introduce multi-cardinality training, which allows a single model to support different semantic-token budgets at inference time. We also propose FLASH-LITE, a lightweight variant with offline passage encoding for low-cost online inference. Across TREC and BEIR benchmarks, FLASH improves average ranking effectiveness at comparable online latency and slightly lower FLOPs, while FLASH-Lite substantially reduces query-time cost with a modest effectiveness trade-off. FLASH-LITE enables offline passage encoding, substantially reducing online query-time cost while maintaining competitive reranking performance. Code is provided in the supplementary material.
%U https://aclanthology.org/2026.aacl-long.217/
%P 4897-4910
Markdown (Informal)
[Fast Listwise Passage Reranking via Semantic Tokens](https://aclanthology.org/2026.aacl-long.217/) (Prabhune et al., AACL-IJCNLP 2026)
ACL
- Omkar Prabhune, Xijun Wang, Nan Bi, Haiyan Wang, Shima Alizadeh, Anoop Jain, Yu Lou, and Junbang Liang. 2026. Fast Listwise Passage Reranking via Semantic Tokens. In Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4897–4910, Hengqin, China. Association for Computational Linguistics.