A user’s interaction with a virtual assistant typically involves the issuance of commands to perform certain actions, e.g., “play Mozart,” “turn on the lights every night,” etc. To execute actions accurately, a virtual assistant is trained to understand or detect action intent. This disclosure describes techniques to obtain a wide variety of query examples, with user permission, to train a virtual assistant to understand action intents. Information-seeking queries submitted to a search engine are transformed into action intents, e.g., commands, based on the grammar of the language of the search-engine query. The techniques are especially useful when launching new features, surfaces, or locales for a virtual assistant.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Zomet, Asaf; Shmiel, Tomer; Keysar, Dvir; and Cohen, Amitai, "Obtaining Query Examples for Virtual Assistant Actions Using Search Logs", Technical Disclosure Commons, (March 11, 2020)