Recursive and Clause-Wise Decoding for Complex and Cross-Domain Text-to-SQL Generation

Most deep learning approaches for text-to-SQL generation are limited to the WikiSQL dataset, which only supports very simple queries over a single table. We focus on the Spider dataset, a complex and cross-domain text-to-SQL task, which includes complex queries over multiple tables. In this paper, we propose a SQL clause-wise decoding neural architecture with a self-attention based database schema encoder to address Spider task. Each of the clause-specific decoders consists of a set of sub-modules, which is defined by the syntax of each clause. Additionally, our model works recursively to support nested queries. The experimental result shows that our model outperforms the previous state-of-the-art model by 9.8% in the exact matching accuracy on the Spider dev dataset. In addition, we show that our model is significantly more effective to predict complex and nested queries than previous works.