Microsoft Research Summit 2021
Return to Microsoft Research Summit 2021 Microsoft Research Summit 2021

Research talk: Closing the loop in natural language interfaces to relational databases

One of the most interesting tasks in semantic parsing is the translation of natural language sentences to database queries. As part of the Yale Spider project, researchers at Yale University have developed three new datasets and matching shared tasks, which will be covered in this session. Spider is a collection of 10,181 natural language questions and 5,693 matching database queries from 138 domains. SParC (Semantic Parsing in Context) consists of 4,298 coherent sequences of questions and the matching queries. Finally, CoSQL consists of a Wizard-of-Oz collection of 3,000 dialogues, a total of 30,000 turns, and their translations to SQL. This session will also introduce GraPPa, a pretraining approach for table semantic parsing that learns a compositional inductive bias in the joint representations of textual and tabular data. The researchers used GraPPa to obtain state-of-the-art performance on four popular fully supervised and weakly supervised table semantic parsing benchmarks. This is joint work with Tao Yu, Rui Zhang, Victoria Lin, Caiming Xiong, and many others.

Learn more about the 2021 Microsoft Research Summit:

Deep Learning & Large-Scale AI
Dragomir Radev
Yale University

Deep Learning & Large-Scale AI