How well do LLMs reason over tabular data, really?

TL;DR

This paper investigates the true reasoning capabilities of large language models over tabular data, revealing significant performance deficits and robustness issues under realistic data variations, and proposes more reliable evaluation methods.

Contribution

It critically assesses existing evaluation strategies for LLMs on tabular reasoning, introduces improved assessment methods, and highlights the impact of real-world data variations on LLM performance.

Findings

LLMs show significant deficits in tabular reasoning performance.

Current evaluation metrics may overestimate LLM capabilities.

Robustness of LLMs decreases with realistic data variations.

Abstract

Large Language Models (LLMs) excel in natural language tasks, but less is known about their reasoning capabilities over tabular data. Prior analyses devise evaluation strategies that poorly reflect an LLM's realistic performance on tabular queries. Moreover, we have a limited understanding of the robustness of LLMs towards realistic variations in tabular inputs. Therefore, we ask: Can general-purpose LLMs reason over tabular data, really?, and focus on two questions 1) are tabular reasoning capabilities of general-purpose LLMs robust to real-world characteristics of tabular inputs, and 2) how can we realistically evaluate an LLM's performance on analytical tabular queries? Building on a recent tabular reasoning benchmark, we first surface shortcomings of its multiple-choice prompt evaluation strategy, as well as commonly used free-form text metrics such as SacreBleu and BERT-score. We show that an LLM-as-a-judge procedure yields more reliable performance insights and unveil a significant deficit in tabular reasoning performance of LLMs. We then extend the tabular inputs reflecting three common characteristics in practice: 1) missing values, 2) duplicate entities, and 3) structural variations. Experiments show that the tabular reasoning capabilities of general-purpose LLMs suffer from these variations, stressing the importance of improving their robustness for realistic tabular inputs.