Releasing dev set of AiTQA dataset

This is version 0.1 of the dev split of the dataset that contains the raw tables extracted from documents (unprocessed into any transformations mentioned in the paper) and associated questions. We hope the dev set gives a good indication of how the dataset is and the challenges of tableQA in the real world beyond Wikipedia tables and text.

We will soon release a full dataset containing a test split as well. Several variations of the dataset with transformations simplifying assumptions and strategies that we used to get better numbers on TaPas, TaBERT, and RCI models. See the paper for more details. If you have questions, comments, or if you are in a hurry to get the full dataset, please reach out to the authors.