Training-Free Table Question-Answering via TableGrid Navigation and Progressive Inference Prompting

A new training-free approach for Table Question-Answering (TQA) has been proposed, leveraging two structured prompting frameworks: TableGrid Navigation (TGN) and Progressive Inference Prompting (PIP). TGN iteratively navigates rows and columns through a three-module loop to locate evidence and refine answers, while PIP enforces column identification for explicit progressive row selection based on the query. The method was evaluated on 17 LLMs against 6 baselines on the TableBench and FeTaQa datasets. On TableBench, TGN improves over the strongest baseline, demonstrating effective table navigation without fine-tuning or task-specific training. The work addresses the need for verifiable control in LLM-based TQA, offering a transparent reasoning process.