Abstract
We present a sequence-to-action parsing approach for the natural language toSQL task that incrementally fills the slots of a SQL query with feasibleactions from a pre-defined inventory. To account for the fact that typicallythere are multiple correct SQL queries with the same or very similar semantics,we draw inspiration from syntactic parsing techniques and propose to train oursequence-to-action models with non-deterministic oracles. We evaluate ourmodels on the WikiSQL dataset and achieve an execution accuracy of 83.7% on thetest set, a 2.1% absolute improvement over the model trained with traditionalstatic oracles assuming a single correct target SQL query. When furthercombined with the execution-guided decoding strategy, our model sets a newstate-of-the-art performance at an execution accuracy of 87.1%. This is a work-in-progress technical report.