MRT at SemEval-2025 Task 8: Maximizing Recovery from Tables with Multiple Steps

2025-05-28

AI Summary

This work addresses SemEval-2025 Task 8, a table-based question answering challenge requiring precise semantic understanding and reasoning over structured data. Method: We propose a multi-step Python code generation framework comprising three tightly coupled stages: (1) semantic parsing of the input table, (2) generation of executable natural-language reasoning steps, and (3) translation into robust, executable Python code with dynamic execution and adaptive error recovery. The framework establishes a closed-loop pipeline—“instruction generation → code translation → dynamic execution”—leveraging fine-grained prompt engineering and collaborative inference among open-source large language models. Contribution/Results: Our approach requires no model fine-tuning, relying solely on program synthesis, prompt engineering, and built-in exception handling to achieve end-to-end table understanding and self-correcting execution. On Task 8 Subtask 1, it achieves 70.50% accuracy—substantially outperforming baseline methods—and demonstrates the effectiveness and robustness of stepwise, controllable code generation for table QA.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

In this paper we expose our approach to solve the extit{SemEval 2025 Task 8: Question-Answering over Tabular Data} challenge. Our strategy leverages Python code generation with LLMs to interact with the table and get the answer to the questions. The process is composed of multiple steps: understanding the content of the table, generating natural language instructions in the form of steps to follow in order to get the answer, translating these instructions to code, running it and handling potential errors or exceptions. These steps use open source LLMs and fine grained optimized prompts for each task (step). With this approach, we achieved a score of $70.50%$ for subtask 1.

Problem

• Solving question-answering over tabular data challenge
• Generating Python code with LLMs for table interaction
• Optimizing multi-step process for accurate answer retrieval

Innovation

• Python code generation with LLMs
• Multi-step natural language instructions
• Fine-grained optimized prompts