Extracting Tabular data for Question-Answering from Documents

In this paper, we present the designs of a system that is targeted at automated data curation from tables appearing in large documents like reports, technical articles etc.. There is an increasing demand for applications that can support question answering over tables. It can also be used to automatically fill-up data-sheets in a regular fashion. The task is complex since there are no fixed formats for these tables. Inferring table structure and extracting tabular data are known to be difficult problems. In this paper we present methods to retrieve tabular data from approximately inferred structures. We present QuATab, an end-to-end system that can do Q&A from tables contained in documents.