CRAFT: A Unified Counterfactual Reasoning Framework for Tabular Question Answering and Fact Verification

2026-06-04

AI Summary

This work addresses the limitations of large language models in table reasoning, particularly their constrained ability to perform structured, multi-step, and bidirectional exploration of alternative hypotheses. To overcome this, the study introduces counterfactual reasoning into table understanding for the first time, proposing a unified bidirectional verification framework. This framework constructs both the original statement and its counterfactual variant, then extracts and adaptively fuses evidence along dual reasoning paths to generate answers. The approach significantly enhances performance on complex table-based question answering and fact verification tasks, consistently outperforming existing methods on the WikiTQ and TabFact benchmarks—especially in challenging scenarios. Moreover, it effectively narrows the performance gap across different large language models and improves reasoning consistency.

Abstract

Table reasoning remains challenging for large language models (LLMs), particularly in tasks that require multi-step inference over long and structured tables. Existing approaches predominantly rely on single-direction reasoning, which limits their ability to explore alternative hypotheses across tasks. In this work, we propose CRAFT, a unified Counterfactual Reasoning Framework that reformulates Tabular question answering and fact verification into a general bidirectional verification process. Our method explicitly constructs both declarative statements and their counterfactual variants. Evidence is then extracted from reasoning along both the original and counterfactual paths, and integrated via a weighted mechanism to arrive at the final answer. Experimental results show that our approach consistently surpasses representative baselines on table reasoning datasets such as WikiTQ and TabFact, achieving especially large improvements on complex question answering. Our framework also significantly mitigates performance gaps between different backbone LLMs. This indicates that counterfactual reasoning effectively overcomes the limitations of single-direction inference, guiding LLMs toward more discerning reasoning and establishing a more principled paradigm for structured reasoning tasks. Our code will be made publicly available upon acceptance.

Problem

• table reasoning
• counterfactual reasoning
• question answering
• fact verification
• large language models

Innovation

• Counterfactual Reasoning
• Tabular Question Answering
• Fact Verification
• Bidirectional Verification
• Structured Reasoning