T5-SR: A Unified Seq-to-Seq Decoding Strategy for Semantic Parsing

However, Translating natural language queries into SQLs in a seq2seq to produce a correct SQL expression, a parser should not manner has attracted much attention recently. However, only understand the semantics of the input query but also produce compared with abstract-syntactic-tree-based SQL generation, predictions that satisfy the SQL grammar and database seq2seq semantic parsers face much more challenges, including schema restrictions. We experimentally find that with the help poor quality on schematical information prediction and of pre-trained language models, seq2seq models are capable poor semantic coherence between natural language queries of generating legal SQL skeletons, while detailed schematic and SQLs. This paper analyses the above difficulties and information prediction remains a big difficulty for seq2seq proposes a seq2seq-oriented decoding strategy called SR, parsers. To solve this problem, in this paper, we propose which includes a new intermediate representation SSQL and a new intermediate representation called SSQL (Semantic-a reranking method with score re-estimator to solve the above SQL) for seq2seq SQL generation based on standard SQL obstacles respectively.