Grammar rule generation for virtual assistant applications is difficult to scale due to the need for manual labeling. This disclosure describes a scalable solution to automatically generate enhanced grammar rules in predefined verticals. The techniques enable incremental improvements to interpretations of voice commands by a virtual assistant application. Queries that have not already been processed for a target vertical are identified and extracted from a corpus of user queries, e.g., a time-limited corpus in a particular language. Queries are analyzed to discover arguments and patterns that are specific to the vertical. Grammar rules are generated based on the arguments and patterns.

Creative Commons License

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