Abstract

Semantic parsing processes natural language queries to convert them into a structured parse. This disclosure describes a two-stage scheme for semantic parsing, comprising a plan stage and a fill stage. In the plan stage, the intent or plan behind an input query is identified. In the fill stage, a parse is generated by filling the plan with the relevant span from the query. The separation of parsing into plan and fill enables decoupling losses corresponding to basic intent generation (plan) and span identification (fill) stages. The described techniques provide the flexibility to decouple model parameters that correspond to the two stages. The described techniques provide an efficient alternative to sequence-to-sequence models that use both an encoder and a decoder for parsing.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS