@inproceedings{qingyun-etal-2026-reading,
title = "Reading the Geometry: A Mechanism for When Frozen {LLM} Text Helps Tabular Foundation Models",
author = "Qingyun, Zou and
Li, Yingze and
Chen, Nuo",
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.2028/",
pages = "43179--43194",
ISBN = "979-8-89176-462-0",
abstract = "Tabular foundation models like TabPFN and TabICL are often augmented with frozen large language model (LLM) embeddings when tables contain free-form text. Whether this helps is highly case-dependent: across 732 dataset{--}backbone{--}encoder configurations, 45.9{\%} improve, 30.9{\%} stay neutral, and 23.2{\%} degrade. Using a fixed, outcome-free raw score, we define high- and low-readability strata before applying target-covariance Gaussian sampling, row{--}text shuffling, and pre-projection whitening. Each intervention reduces the mean gain in the high-readability stratum to near zero, whereas the paired intervention effects in the low-readability stratum remain within sampling noise. The first two interventions show that neither a Gaussian sample from the target covariance nor row-misaligned text reproduces the high-stratum gain; whitening further shows sensitivity to broad pre-projection geometry. Together, the results support a dependence on row-aligned embedding structure without isolating the covariance spectrum as its unique cause. The high-stratum Gaussian-replacement effect replicates on a second backbone. A within-dataset encoder ranking, Decision-Boundary Alignment (DBA), also shows stronger rank alignment with gain than the Massive Text Embedding Benchmark (MTEB) rank in our eight-dataset diagnostic."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="qingyun-etal-2026-reading">
<titleInfo>
<title>Reading the Geometry: A Mechanism for When Frozen LLM Text Helps Tabular Foundation Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zou</namePart>
<namePart type="family">Qingyun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yingze</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nuo</namePart>
<namePart type="family">Chen</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>Tabular foundation models like TabPFN and TabICL are often augmented with frozen large language model (LLM) embeddings when tables contain free-form text. Whether this helps is highly case-dependent: across 732 dataset–backbone–encoder configurations, 45.9% improve, 30.9% stay neutral, and 23.2% degrade. Using a fixed, outcome-free raw score, we define high- and low-readability strata before applying target-covariance Gaussian sampling, row–text shuffling, and pre-projection whitening. Each intervention reduces the mean gain in the high-readability stratum to near zero, whereas the paired intervention effects in the low-readability stratum remain within sampling noise. The first two interventions show that neither a Gaussian sample from the target covariance nor row-misaligned text reproduces the high-stratum gain; whitening further shows sensitivity to broad pre-projection geometry. Together, the results support a dependence on row-aligned embedding structure without isolating the covariance spectrum as its unique cause. The high-stratum Gaussian-replacement effect replicates on a second backbone. A within-dataset encoder ranking, Decision-Boundary Alignment (DBA), also shows stronger rank alignment with gain than the Massive Text Embedding Benchmark (MTEB) rank in our eight-dataset diagnostic.</abstract>
<identifier type="citekey">qingyun-etal-2026-reading</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2028/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>43179</start>
<end>43194</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Reading the Geometry: A Mechanism for When Frozen LLM Text Helps Tabular Foundation Models
%A Qingyun, Zou
%A Li, Yingze
%A Chen, Nuo
%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 qingyun-etal-2026-reading
%X Tabular foundation models like TabPFN and TabICL are often augmented with frozen large language model (LLM) embeddings when tables contain free-form text. Whether this helps is highly case-dependent: across 732 dataset–backbone–encoder configurations, 45.9% improve, 30.9% stay neutral, and 23.2% degrade. Using a fixed, outcome-free raw score, we define high- and low-readability strata before applying target-covariance Gaussian sampling, row–text shuffling, and pre-projection whitening. Each intervention reduces the mean gain in the high-readability stratum to near zero, whereas the paired intervention effects in the low-readability stratum remain within sampling noise. The first two interventions show that neither a Gaussian sample from the target covariance nor row-misaligned text reproduces the high-stratum gain; whitening further shows sensitivity to broad pre-projection geometry. Together, the results support a dependence on row-aligned embedding structure without isolating the covariance spectrum as its unique cause. The high-stratum Gaussian-replacement effect replicates on a second backbone. A within-dataset encoder ranking, Decision-Boundary Alignment (DBA), also shows stronger rank alignment with gain than the Massive Text Embedding Benchmark (MTEB) rank in our eight-dataset diagnostic.
%U https://aclanthology.org/2026.findings-emnlp.2028/
%P 43179-43194
Markdown (Informal)
[Reading the Geometry: A Mechanism for When Frozen LLM Text Helps Tabular Foundation Models](https://aclanthology.org/2026.findings-emnlp.2028/) (Qingyun et al., Findings 2026)
ACL