Exploring Generative Process Reward Modeling for Semi-Structured Data: A Case Study of Table Question Answering

2025-10-23

AI Summary

This work systematically investigates, for the first time, the applicability of Process Reward Models (PRMs) to Table-based Question Answering (TQA)—a semi-structured reasoning task characterized by abundant distractors, weak inter-step dependency, and strong domain-knowledge coupling. Method: To jointly assess answer correctness and stepwise reasoning validity, we propose a generative PRM that integrates textual understanding with executable code execution. Contribution/Results: Experiments reveal only weak correlation between step-level rewards and final answer accuracy, exposing PRMs’ insufficient modeling of causal reasoning chains; moreover, cross-domain generalization remains limited, and solution selection gains plateau. Beyond establishing the first PRM evaluation benchmark for TQA, this work identifies the critical need for enhanced causal reasoning mechanisms in PRMs, thereby proposing a novel paradigm for trustworthy evaluation of semi-structured reasoning.

Technology Category

• Natural Language Processing: Question Answering
• Knowledge Representation and Reasoning: Qualitative Reasoning
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Process reward models (PRMs) improve complex reasoning in large language models (LLMs) by grading candidate solutions step-by-step and selecting answers via aggregated step scores. While effective in domains such as mathematics, their applicability to tasks involving semi-structured data, like table question answering (TQA) remains unexplored. TQA poses unique challenges for PRMs, including abundant irrelevant information, loosely connected reasoning steps, and domain-specific reasoning. This work presents the first systematic study of PRMs for TQA. We evaluate state-of-the-art generative PRMs on TQA from both answer and step perspectives. Results show that PRMs that combine textual and code verification can aid solution selection but struggle to generalize to out-of-domain data. Analysis reveals a weak correlation between performance in step-level verification and answer accuracy, possibly stemming from weak step dependencies and loose causal links. Our findings highlight limitations of current PRMs on TQA and offer valuable insights for building more robust, process-aware verifiers.

Problem

• Evaluating process reward models for table question answering tasks
• Addressing challenges like irrelevant information and weak step dependencies
• Assessing generalization limitations of current verification methods

Innovation

• Process reward models grade step-by-step solutions
• Combines textual and code verification for tables
• Studies generalization challenges in semi-structured data