@inproceedings{varriale-etal-2026-latent,
title = "Latent Bridges for Multi-Table Question Answering",
author = "Varriale, Simone and
Cucumides, Tamara and
Geerts, Floris and
Papotti, Paolo",
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.2073/",
pages = "47201--47228",
ISBN = "979-8-89176-459-0",
abstract = "We introduce GRAB, a constructor{--}encoder{--}bridge pipeline for table question answering. Our method lifts relational data into a heterogeneous graph, encodes it via message passing, and transfers the signals to an LLM through a small set of query-conditioned latent tokens. The LLM receives these tokens alongside the serialized table, and the two play different roles: the graph shows where the relevant evidence is, while the text keeps the exact cell values that the graph does not encode. The graph supplements the text rather than replacing it, and our ablations confirm that both are required. Crucially, the LLM remains strictly frozen to preserve its general reasoning capabilities; we train only the lightweight graph encoder and latent bridge (91M parameters), allowing the entire pipeline to be trained efficiently. Our pipeline significantly improves performance on relational Question Answering, with the largest gains in demanding multi-table settings, offering an efficient, principled way to connect relational deep learning with LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="varriale-etal-2026-latent">
<titleInfo>
<title>Latent Bridges for Multi-Table Question Answering</title>
</titleInfo>
<name type="personal">
<namePart type="given">Simone</namePart>
<namePart type="family">Varriale</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tamara</namePart>
<namePart type="family">Cucumides</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Floris</namePart>
<namePart type="family">Geerts</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paolo</namePart>
<namePart type="family">Papotti</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>We introduce GRAB, a constructor–encoder–bridge pipeline for table question answering. Our method lifts relational data into a heterogeneous graph, encodes it via message passing, and transfers the signals to an LLM through a small set of query-conditioned latent tokens. The LLM receives these tokens alongside the serialized table, and the two play different roles: the graph shows where the relevant evidence is, while the text keeps the exact cell values that the graph does not encode. The graph supplements the text rather than replacing it, and our ablations confirm that both are required. Crucially, the LLM remains strictly frozen to preserve its general reasoning capabilities; we train only the lightweight graph encoder and latent bridge (91M parameters), allowing the entire pipeline to be trained efficiently. Our pipeline significantly improves performance on relational Question Answering, with the largest gains in demanding multi-table settings, offering an efficient, principled way to connect relational deep learning with LLMs.</abstract>
<identifier type="citekey">varriale-etal-2026-latent</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2073/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>47201</start>
<end>47228</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Latent Bridges for Multi-Table Question Answering
%A Varriale, Simone
%A Cucumides, Tamara
%A Geerts, Floris
%A Papotti, Paolo
%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 varriale-etal-2026-latent
%X We introduce GRAB, a constructor–encoder–bridge pipeline for table question answering. Our method lifts relational data into a heterogeneous graph, encodes it via message passing, and transfers the signals to an LLM through a small set of query-conditioned latent tokens. The LLM receives these tokens alongside the serialized table, and the two play different roles: the graph shows where the relevant evidence is, while the text keeps the exact cell values that the graph does not encode. The graph supplements the text rather than replacing it, and our ablations confirm that both are required. Crucially, the LLM remains strictly frozen to preserve its general reasoning capabilities; we train only the lightweight graph encoder and latent bridge (91M parameters), allowing the entire pipeline to be trained efficiently. Our pipeline significantly improves performance on relational Question Answering, with the largest gains in demanding multi-table settings, offering an efficient, principled way to connect relational deep learning with LLMs.
%U https://aclanthology.org/2026.emnlp-main.2073/
%P 47201-47228
Markdown (Informal)
[Latent Bridges for Multi-Table Question Answering](https://aclanthology.org/2026.emnlp-main.2073/) (Varriale et al., EMNLP 2026)
ACL
- Simone Varriale, Tamara Cucumides, Floris Geerts, and Paolo Papotti. 2026. Latent Bridges for Multi-Table Question Answering. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 47201–47228, Budapest, Hungary. Association for Computational Linguistics.