@inproceedings{ahmed-2026-closed,
title = "Closed-Book Knowledge Base Construction with Calibrated Abstention and Numeric Self-Consistency",
author = "Ahmed, Tanvir",
editor = "Kalo, Jan-Christoph and
Nakashole, Ndapa and
Razniewski, Simon and
Suchanek, Fabian M. and
Vlachos, Andreas and
McCallum, Andrew",
booktitle = "Proceedings of the 11th Workshop on Automated Knowledge Base Construction ({AKBC} 2026)",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.akbc-1.29/",
pages = "417--426",
ISBN = "979-8-89176-464-4",
abstract = "Large language models (LMs) store vast factual knowledge, yet turning it into a reliable, set-valued knowledge base, where a subject may have zero, one, or many objects, remains hard. The AKBC 2026 shared task (5th LM-KBC edition) poses this in a strict closed-book setting: an open-weight model of at most 32B parameters, with no retrieval, no external knowledge, and no fine-tuning. We start from the metric and observe that it rewards abstention: since an empty prediction scores a perfect F1 against an empty gold set, predicting nothing everywhere already reaches 0.169 relation-averaged macro-F1, and on the null-heavy relations even beats the official baseline. We therefore treat when to answer as a first-class decision and build a per-relation pipeline that grounds each prompt in the relation{'}s exact scope, decides abstention from the sampled set rather than by instruction, and takes a median over samples for numeric relations. Our Gemma-4-31B system (30.7B, the strongest in-budget model we could access) scores 0.461/0.488 (relation-averaged/micro) on validation from a single sample, beating the official baseline on five of the six relations, while a smaller 4-bit Qwen2.5-14B, even with self-consistency voting, reaches only 0.415: in this comparison a larger single-sample model outscores a smaller voted one. Generation is cached, so decoding reproduces without a GPU. We release all code, prompts, and cached generations: https://github.com/minhaz-42/lm-kbc-2026"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ahmed-2026-closed">
<titleInfo>
<title>Closed-Book Knowledge Base Construction with Calibrated Abstention and Numeric Self-Consistency</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tanvir</namePart>
<namePart type="family">Ahmed</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 11th Workshop on Automated Knowledge Base Construction (AKBC 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jan-Christoph</namePart>
<namePart type="family">Kalo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ndapa</namePart>
<namePart type="family">Nakashole</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Simon</namePart>
<namePart type="family">Razniewski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fabian</namePart>
<namePart type="given">M</namePart>
<namePart type="family">Suchanek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andreas</namePart>
<namePart type="family">Vlachos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andrew</namePart>
<namePart type="family">McCallum</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">979-8-89176-464-4</identifier>
</relatedItem>
<abstract>Large language models (LMs) store vast factual knowledge, yet turning it into a reliable, set-valued knowledge base, where a subject may have zero, one, or many objects, remains hard. The AKBC 2026 shared task (5th LM-KBC edition) poses this in a strict closed-book setting: an open-weight model of at most 32B parameters, with no retrieval, no external knowledge, and no fine-tuning. We start from the metric and observe that it rewards abstention: since an empty prediction scores a perfect F1 against an empty gold set, predicting nothing everywhere already reaches 0.169 relation-averaged macro-F1, and on the null-heavy relations even beats the official baseline. We therefore treat when to answer as a first-class decision and build a per-relation pipeline that grounds each prompt in the relation’s exact scope, decides abstention from the sampled set rather than by instruction, and takes a median over samples for numeric relations. Our Gemma-4-31B system (30.7B, the strongest in-budget model we could access) scores 0.461/0.488 (relation-averaged/micro) on validation from a single sample, beating the official baseline on five of the six relations, while a smaller 4-bit Qwen2.5-14B, even with self-consistency voting, reaches only 0.415: in this comparison a larger single-sample model outscores a smaller voted one. Generation is cached, so decoding reproduces without a GPU. We release all code, prompts, and cached generations: https://github.com/minhaz-42/lm-kbc-2026</abstract>
<identifier type="citekey">ahmed-2026-closed</identifier>
<location>
<url>https://aclanthology.org/2026.akbc-1.29/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>417</start>
<end>426</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Closed-Book Knowledge Base Construction with Calibrated Abstention and Numeric Self-Consistency
%A Ahmed, Tanvir
%Y Kalo, Jan-Christoph
%Y Nakashole, Ndapa
%Y Razniewski, Simon
%Y Suchanek, Fabian M.
%Y Vlachos, Andreas
%Y McCallum, Andrew
%S Proceedings of the 11th Workshop on Automated Knowledge Base Construction (AKBC 2026)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-464-4
%F ahmed-2026-closed
%X Large language models (LMs) store vast factual knowledge, yet turning it into a reliable, set-valued knowledge base, where a subject may have zero, one, or many objects, remains hard. The AKBC 2026 shared task (5th LM-KBC edition) poses this in a strict closed-book setting: an open-weight model of at most 32B parameters, with no retrieval, no external knowledge, and no fine-tuning. We start from the metric and observe that it rewards abstention: since an empty prediction scores a perfect F1 against an empty gold set, predicting nothing everywhere already reaches 0.169 relation-averaged macro-F1, and on the null-heavy relations even beats the official baseline. We therefore treat when to answer as a first-class decision and build a per-relation pipeline that grounds each prompt in the relation’s exact scope, decides abstention from the sampled set rather than by instruction, and takes a median over samples for numeric relations. Our Gemma-4-31B system (30.7B, the strongest in-budget model we could access) scores 0.461/0.488 (relation-averaged/micro) on validation from a single sample, beating the official baseline on five of the six relations, while a smaller 4-bit Qwen2.5-14B, even with self-consistency voting, reaches only 0.415: in this comparison a larger single-sample model outscores a smaller voted one. Generation is cached, so decoding reproduces without a GPU. We release all code, prompts, and cached generations: https://github.com/minhaz-42/lm-kbc-2026
%U https://aclanthology.org/2026.akbc-1.29/
%P 417-426
Markdown (Informal)
[Closed-Book Knowledge Base Construction with Calibrated Abstention and Numeric Self-Consistency](https://aclanthology.org/2026.akbc-1.29/) (Ahmed, AKBC 2026)
ACL