Table-Text Alignment: Explaining Claim Verification Against Tables in Scientific Papers

2025-06-12

AI Summary

Scientific paper table-based claim verification suffers from an explainability bottleneck: existing methods yield only binary labels without localizing critical evidence cells or providing reasoning justifications. Method: We reformulate the task as a cell-level table-text alignment explanation problem and propose the first explainable verification paradigm for scientific tables. We introduce SciTab-X—the first dataset with human-annotated minimal rationale cells—and establish a taxonomy for ambiguous cases. Our approach integrates data augmentation, cell-level supervised training, and LLM-aligned evaluation. Results: Explicit cell-level alignment significantly improves verification accuracy. Moreover, experiments reveal that while mainstream LLMs often produce correct binary labels, their identified rationale cells diverge substantially from human annotations—exposing severe non-faithfulness in their reasoning processes. This highlights a critical gap between label accuracy and reasoning fidelity in current LLM-based table verification systems.

Technology Category

• Knowledge Representation and Reasoning: Argumentation
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Scientific claim verification against tables typically requires predicting whether a claim is supported or refuted given a table. However, we argue that predicting the final label alone is insufficient: it reveals little about the model's reasoning and offers limited interpretability. To address this, we reframe table-text alignment as an explanation task, requiring models to identify the table cells essential for claim verification. We build a new dataset by extending the SciTab benchmark with human-annotated cell-level rationales. Annotators verify the claim label and highlight the minimal set of cells needed to support their decision. After the annotation process, we utilize the collected information and propose a taxonomy for handling ambiguous cases. Our experiments show that (i) incorporating table alignment information improves claim verification performance, and (ii) most LLMs, while often predicting correct labels, fail to recover human-aligned rationales, suggesting that their predictions do not stem from faithful reasoning.

Problem

• Predicting claim verification labels lacks interpretability and reasoning insight
• Identifying essential table cells for claim verification improves explanation
• Most LLMs fail to align rationales with human reasoning despite correct labels

Innovation

• Reframes table-text alignment as explanation task
• Extends SciTab with human-annotated cell rationales
• Proposes taxonomy for handling ambiguous table cases