Key points are not available for this paper at this time.
Recent advances in natural language processing and deep learning have opened thousands of new dimensions for transforming dreams into reality. One such dimension is Table Question-Answering (Q/A). It's an old fantasy of computer science to query tables in natural human language and extract the desired information. Currently, the subject is trending because a huge chunk of modern business data still exists in structured and semi-structured tables. The computational and hardware requirements of the problem linearly increase with the size and number of the provided tables, which makes it complex for real-time production environments. This challenging nature has raised the interest of new aspiring researchers in efficiently applying modern deep-learning tactics and developing systematically optimized solutions. In this paper, we aim to explore a sub-field of Table Question-Answering (Q/A) called Non-Semantic Closed-Domain Free-Form Table Question-Answering (Q/A) and develop a novel solution by decoupling the tabular data from linguistic knowledge. The goal of such decoupling is to enable the computational and resource optimizations that will benefit the scalability of the proposed system and allow the solution to function independently of the table size.
Dhanani et al. (Tue,) studied this question.