@inproceedings{wang-etal-2026-tarq,
title = "{TARQ}: Tail-Aware Reconstruction Quantization for Rare-Word Robust Automatic Speech Recognition",
author = "Wang, Xinyu and
Zhao, Ziyu and
Bai, Ke and
Meng, Silin and
Shen, Dongming and
Chang, Xiao-Wen and
He, Yixuan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.664/",
pages = "13928--13952",
ISBN = "979-8-89176-462-0",
abstract = "Data-aware post-training quantization (PTQ) minimizes a per-token reconstruction loss on a small calibration corpus{---}an objective natural for average-token preservation, but one that implicitly weights positions by their empirical frequency. For **Automatic Speech Recognition (ASR)**, this weighting is misaligned with tail-sensitive recognition risk: names, numerals, technical terms, and domain-specific words are practically important but receive proportionally little calibration mass.\par We propose **Tail-Aware Reconstruction Quantization (TARQ)**, a label-free PTQ framework that shifts calibration toward the lexical tail. **TARQ** introduces **rareBAL**, a closed-form per-Linear-layer rule that equalizes the calibration mass contributed by common and tail positions, plus a metric-consistent residual correction that aligns the sequential PTQ sweep with the reweighted objective. **TARQ** requires no entity labels, no curated calibration set, no validation decoding, and no additional training.\par Across eight ASR backbones and eight datasets under 4-bit weight-only quantization, **TARQ** improves mean rare-WER, the **Word Error Rate (WER)** restricted to rare-word reference positions, while maintaining competitive aggregate WER. It achieves the lowest cross-corpus rare-WER swing, **0.63 percentage points (pp)**, among compared methods, and transfers without modification to entity-rich evaluation on ProfASR and ContextASR-Speech-En with no entity supervision."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-tarq">
<titleInfo>
<title>TARQ: Tail-Aware Reconstruction Quantization for Rare-Word Robust Automatic Speech Recognition</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xinyu</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ziyu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ke</namePart>
<namePart type="family">Bai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Silin</namePart>
<namePart type="family">Meng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dongming</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiao-Wen</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yixuan</namePart>
<namePart type="family">He</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Data-aware post-training quantization (PTQ) minimizes a per-token reconstruction loss on a small calibration corpus—an objective natural for average-token preservation, but one that implicitly weights positions by their empirical frequency. For **Automatic Speech Recognition (ASR)**, this weighting is misaligned with tail-sensitive recognition risk: names, numerals, technical terms, and domain-specific words are practically important but receive proportionally little calibration mass.\par We propose **Tail-Aware Reconstruction Quantization (TARQ)**, a label-free PTQ framework that shifts calibration toward the lexical tail. **TARQ** introduces **rareBAL**, a closed-form per-Linear-layer rule that equalizes the calibration mass contributed by common and tail positions, plus a metric-consistent residual correction that aligns the sequential PTQ sweep with the reweighted objective. **TARQ** requires no entity labels, no curated calibration set, no validation decoding, and no additional training.\par Across eight ASR backbones and eight datasets under 4-bit weight-only quantization, **TARQ** improves mean rare-WER, the **Word Error Rate (WER)** restricted to rare-word reference positions, while maintaining competitive aggregate WER. It achieves the lowest cross-corpus rare-WER swing, **0.63 percentage points (pp)**, among compared methods, and transfers without modification to entity-rich evaluation on ProfASR and ContextASR-Speech-En with no entity supervision.</abstract>
<identifier type="citekey">wang-etal-2026-tarq</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.664/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>13928</start>
<end>13952</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T TARQ: Tail-Aware Reconstruction Quantization for Rare-Word Robust Automatic Speech Recognition
%A Wang, Xinyu
%A Zhao, Ziyu
%A Bai, Ke
%A Meng, Silin
%A Shen, Dongming
%A Chang, Xiao-Wen
%A He, Yixuan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F wang-etal-2026-tarq
%X Data-aware post-training quantization (PTQ) minimizes a per-token reconstruction loss on a small calibration corpus—an objective natural for average-token preservation, but one that implicitly weights positions by their empirical frequency. For **Automatic Speech Recognition (ASR)**, this weighting is misaligned with tail-sensitive recognition risk: names, numerals, technical terms, and domain-specific words are practically important but receive proportionally little calibration mass.\par We propose **Tail-Aware Reconstruction Quantization (TARQ)**, a label-free PTQ framework that shifts calibration toward the lexical tail. **TARQ** introduces **rareBAL**, a closed-form per-Linear-layer rule that equalizes the calibration mass contributed by common and tail positions, plus a metric-consistent residual correction that aligns the sequential PTQ sweep with the reweighted objective. **TARQ** requires no entity labels, no curated calibration set, no validation decoding, and no additional training.\par Across eight ASR backbones and eight datasets under 4-bit weight-only quantization, **TARQ** improves mean rare-WER, the **Word Error Rate (WER)** restricted to rare-word reference positions, while maintaining competitive aggregate WER. It achieves the lowest cross-corpus rare-WER swing, **0.63 percentage points (pp)**, among compared methods, and transfers without modification to entity-rich evaluation on ProfASR and ContextASR-Speech-En with no entity supervision.
%U https://aclanthology.org/2026.findings-emnlp.664/
%P 13928-13952
Markdown (Informal)
[TARQ: Tail-Aware Reconstruction Quantization for Rare-Word Robust Automatic Speech Recognition](https://aclanthology.org/2026.findings-emnlp.664/) (Wang et al., Findings 2026)
ACL