@inproceedings{liu-etal-2026-escaping,
title = "Escaping the Ambiguity Trap: Content-Level Evidence Tracing for Open-Domain Table {QA}",
author = "Liu, Yibin and
Chen, Yewang and
Li, Zhe",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.109/",
pages = "2568--2583",
ISBN = "979-8-89176-459-0",
abstract = "Open-domain table QA requires reliable evidence localization beyond candidate retrieval. Compact table fingerprints reduce irrelevant signals during representation learning, but retrieval based on learned representations may still have limited ability to distinguish evidence segments among retrieved candidates. This issue is especially pronounced when tables are partitioned into segments that share similar metadata and schema patterns. We propose Probe-Grounded Reasoning (PGR), a coarse-to-fine framework for content-level evidence localization. PGR generates query-specific probes with metadata cues and cell-level evidence, then reranks candidate segments to improve fine-grained evidence localization. It then derives answers via a decoupled Semantic-Plan-Execute workflow with GRPO-optimized planning and execution. On Open-WikiTable, under a controlled evaluation setting with fixed BM25-retrieved candidates, PGR achieves 85.96{\%} segment-level evidence accuracy at k=5 and reaches 69.0{\%} end-to-end QA accuracy, suggesting that content-level verification supports more reliable evidence grounding for downstream reasoning. Code is available at https://github.com/XFastDataLab/PGR/."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-etal-2026-escaping">
<titleInfo>
<title>Escaping the Ambiguity Trap: Content-Level Evidence Tracing for Open-Domain Table QA</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yibin</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yewang</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhe</namePart>
<namePart type="family">Li</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Open-domain table QA requires reliable evidence localization beyond candidate retrieval. Compact table fingerprints reduce irrelevant signals during representation learning, but retrieval based on learned representations may still have limited ability to distinguish evidence segments among retrieved candidates. This issue is especially pronounced when tables are partitioned into segments that share similar metadata and schema patterns. We propose Probe-Grounded Reasoning (PGR), a coarse-to-fine framework for content-level evidence localization. PGR generates query-specific probes with metadata cues and cell-level evidence, then reranks candidate segments to improve fine-grained evidence localization. It then derives answers via a decoupled Semantic-Plan-Execute workflow with GRPO-optimized planning and execution. On Open-WikiTable, under a controlled evaluation setting with fixed BM25-retrieved candidates, PGR achieves 85.96% segment-level evidence accuracy at k=5 and reaches 69.0% end-to-end QA accuracy, suggesting that content-level verification supports more reliable evidence grounding for downstream reasoning. Code is available at https://github.com/XFastDataLab/PGR/.</abstract>
<identifier type="citekey">liu-etal-2026-escaping</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.109/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2568</start>
<end>2583</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Escaping the Ambiguity Trap: Content-Level Evidence Tracing for Open-Domain Table QA
%A Liu, Yibin
%A Chen, Yewang
%A Li, Zhe
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F liu-etal-2026-escaping
%X Open-domain table QA requires reliable evidence localization beyond candidate retrieval. Compact table fingerprints reduce irrelevant signals during representation learning, but retrieval based on learned representations may still have limited ability to distinguish evidence segments among retrieved candidates. This issue is especially pronounced when tables are partitioned into segments that share similar metadata and schema patterns. We propose Probe-Grounded Reasoning (PGR), a coarse-to-fine framework for content-level evidence localization. PGR generates query-specific probes with metadata cues and cell-level evidence, then reranks candidate segments to improve fine-grained evidence localization. It then derives answers via a decoupled Semantic-Plan-Execute workflow with GRPO-optimized planning and execution. On Open-WikiTable, under a controlled evaluation setting with fixed BM25-retrieved candidates, PGR achieves 85.96% segment-level evidence accuracy at k=5 and reaches 69.0% end-to-end QA accuracy, suggesting that content-level verification supports more reliable evidence grounding for downstream reasoning. Code is available at https://github.com/XFastDataLab/PGR/.
%U https://aclanthology.org/2026.emnlp-main.109/
%P 2568-2583
Markdown (Informal)
[Escaping the Ambiguity Trap: Content-Level Evidence Tracing for Open-Domain Table QA](https://aclanthology.org/2026.emnlp-main.109/) (Liu et al., EMNLP 2026)
ACL