SemEval 2025

Task 8: Question Answering on Tabular Data

• Task Overview: SemEval 2025 focuses on Question Answering over Tabular Data using the DataBench benchmark.
• DataBench Description:
  • Composed of real-world table datasets.
  • Features large sizes with numerous rows and columns.
  • Contains a variety of data types to assess different question types.
• Participant Objectives:
  • Develop systems to answer questions based on day-to-day datasets.
  • Answers can be numerical, categorical, boolean, or lists.
• Data Usage:
  • DataBench can be used for training and validation.
  • A separate test set will be provided for competition.
• Task Format:
  • Participants will receive (dataset, question) pairs.
  • Answers will be compared against a gold standard.
• Methods:
  • Two example approaches: In-Context Learning and Code Generation.
  • Participants can use these or develop their own methods.

Dataset Link

• https://huggingface.co/datasets/cardiffnlp/databench