T-REX: Table – Refute or Entail eXplainer

This chapter explores the critical task of table fact-checking, which involves classifying textual claims as entailed or refuted based on information contained in tables. Unlike traditional fact-checking, this task requires a combination of linguistic understanding, symbolic reasoning, and numerical computation. The chapter introduces T-REX, a pioneering tool for real-time table fact-checking that leverages state-of-the-art instruction-tuned reasoning LLMs. T-REX is designed to be accessible, with a user-friendly interface that supports multilingual input and output, and provides transparent reasoning outputs. The system architecture and design are modular, integrating table processing, LLM inference, Optical Character Recognition (OCR), and a web-based frontend for interactive verification and visualization. The verification pipeline supports various table input methods and employs a robust process to ensure accuracy. The chapter also discusses the interface design and user features, highlighting the tool's usability and the ability to handle real-time claim verification. The conclusion emphasizes T-REX's potential to extend table fact-checking into real-world applications, making accurate and accessible claim verification available to a broader audience.