FeTaQA: Free-form Table Question Answering

Abstract Existing table question answering datasets contain abundant factual questions that primarily evaluate a QA system’s comprehension of query and tabular data. However, restricted by their short-form answers, these fail to include question–answer interactions represent more advanced naturally occurring information needs: ask for reasoning integration pieces retrieved from structured knowledge source. To complement the existing reveal challenging nature table-based task, we introduce FeTaQA, new dataset with 10K Wikipedia-based {table, question, free-form answer, supporting cells} pairs. FeTaQA is collected noteworthy descriptions Wikipedia tables people tend seek; generation requires processing humans perform on daily basis: Understand table, retrieve, integrate, infer, conduct text planning surface realization generate an answer. We provide two benchmark methods proposed task: pipeline method based semantic parsing-based systems end-to-end large pretrained models, show poses challenge both methods.