When LLMs Read Tables Carelessly: Measuring and Reducing Data Referencing Errors

Hugging Face Daily Papers Papers

Summary

This paper systematically evaluates data referencing errors (DREs) in LLMs processing tables and proposes critic-based filtering and rejection sampling to improve answer accuracy. A lightweight 4B-parameter critic model achieves high detection accuracy for DREs.

While large language models (LLMs) perform well on table tasks, they still make data referencing errors (DREs), i.e., incorrectly citing or omitting table values, despite understanding the table structure. Beyond final-answer accuracy, DREs directly compromise the correctness and reliability of intermediate reasoning steps. Yet prior studies have only offered limited, small-scale analyses. In this work, we present the first systematic evaluation of tabular data referencing errors across different models and tasks. Our results show that DREs occur across all tested models (1.7B to 20B parameters). Furthermore, we demonstrate that incorporating data referencing as a critic significantly improves answer accuracy up to 12.0%, through critic-based filtering and rejection sampling. Finally, we trained a lightweight 4B-parameter critic model that achieves an average F1 score of 78.2% in detecting both in-distribution and out-of-distribution DREs, and effectively assists inference for larger models.
Original Article
View Cached Full Text

Cached at: 07/02/26, 07:47 AM

Paper page - When LLMs Read Tables Carelessly: Measuring and Reducing Data Referencing Errors

Source: https://huggingface.co/papers/2606.32029

Abstract

Large language models exhibit data referencing errors when processing tables, which can be mitigated through critic-based filtering and rejection sampling, with a lightweight 4B-parameter model achieving high detection accuracy.

Whilelarge language models(LLMs) perform well ontable tasks, they still makedata referencing errors(DREs), i.e., incorrectly citing or omitting table values, despite understanding the table structure. Beyond final-answer accuracy, DREs directly compromise the correctness and reliability of intermediate reasoning steps. Yet prior studies have only offered limited, small-scale analyses. In this work, we present the first systematic evaluation of tabulardata referencing errorsacross different models and tasks. Our results show that DREs occur across all tested models (1.7B to 20B parameters). Furthermore, we demonstrate that incorporating data referencing as a critic significantly improvesanswer accuracyup to 12.0%, throughcritic-based filteringandrejection sampling. Finally, we trained a lightweight 4B-parameter critic model that achieves an averageF1 scoreof 78.2% in detecting bothin-distributionandout-of-distributionDREs, and effectively assists inference for larger models.

View arXiv pageView PDFGitHub2Add to collection

Get this paper in your agent:

hf papers read 2606\.32029

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.32029 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.32029 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.32029 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

LLM-as-a-Discriminator: When Synthetic Tables Still Look Real

arXiv cs.LG

This paper proposes an LLM-as-Discriminator method to audit privacy of synthetic tabular data by asking an LLM to classify samples as real or synthetic, showing that LLM discrimination can serve as a practical privacy audit signal.

LLM Judges Can Be Too Generous When There Is No Reference Answer

arXiv cs.CL

This paper shows that LLM judges tend to over-credit incorrect answers when no reference answer is provided, and adding a reference can flip verdicts by up to 85%, aligning more with human judgments. The authors propose calibration steps for using LLM judges in reference-free settings.

Auditing LLM Benchmarks with Item Response Theory

arXiv cs.CL

This paper introduces an Item Response Theory-based method to detect mislabeled examples in LLM benchmarks at 95% precision, tracing errors to labeling heuristics and annotation issues.

Can LLMs Take Retrieved Information with a Grain of Salt?

arXiv cs.CL

This paper investigates how large language models adapt to the certainty of retrieved information, identifying systematic limitations in handling uncertainty. It proposes an interaction strategy that reduces obedience errors by 25% without modifying model weights.