Key points are not available for this paper at this time.
When translating natural language questions into SQL queries to answer from a database, contemporary semantic parsing models struggle to to unseen database schemas. The generalization challenge lies in (a) the database relations in an accessible way for the semantic parser, (b) modeling alignment between database columns and their mentions in a query. We present a unified framework, based on the relation-aware-attention mechanism, to address schema encoding, schema linking, and representation within a text-to-SQL encoder. On the challenging Spider this framework boosts the exact match accuracy to 57. 2%, surpassing its counterparts by 8. 7% absolute improvement. Further augmented with BERT, it the new state-of-the-art performance of 65. 6% on the Spider. In addition, we observe qualitative improvements in the model's of schema linking and alignment. Our implementation will be-sourced at https: //github. com/Microsoft/rat-sql.
Wang et al. (Sun,) studied this question.