Dynamically updatable offline grammar model for resource-constrained offline device

• US 9,922,138 B2
• Filed: 05/27/2015
• Issued: 03/20/2018
• Est. Priority Date: 05/27/2015
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google LLC (Alphabet Inc.)
• Inventors: Sung, Sangsoo, Gao, Yuli, Murugesan, Prathab
• Primary Examiner(s): Abebe, Daniel
• Application Number:
  US14/723,305
• Publication Number:
  US 20160350320A1
• Time in Patent Office: 1,028 Days
• Field of Search: 704243
• CPC Class Codes: G06F 16/2228 Indexing structures
• CPC Class Codes: G06F 16/2425 Iterative querying; Query formulation based on the results of a preceding query
• CPC Class Codes: G06F 16/243 Natural language query formulation
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/90324 using system suggestions
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 17/22 Interactive procedures; Man-machine interfaces

Abstract

An offline semantic processor of a resource-constrained voice-enabled device such as a mobile device utilizes an offline grammar model with reduced resource requirements to parse voice-based queries received by the device. The offline grammar model may be generated from a larger and more comprehensive grammar model used by an online voice-based query processor, and the generation of the offline grammar model may be based upon query usage data collected from one or more users to enable a subset of more popular voice-based queries from the online grammar model to be incorporated into the offline grammar model. In addition, such a device may collect query usage data and upload such data to an online service to enable an updated offline grammar model to be generated and downloaded back to the device and thereby enable a dynamic update of the offline grammar model to be performed.

Claims

1. A method, comprising:

maintaining an online grammar model used by an online voice-based query processor to parse online voice-based queries, the online grammar model mapping a plurality of queries to actions,

wherein the actions include non-search actions performable by a computer system,

wherein each of the actions is mapped in the grammar model to one or more corresponding queries of the plurality of queries, and

wherein each of a plurality of the actions includes one or more corresponding parameters for constraining performance of the action;

analyzing query usage data for at least a subset of the plurality of queries to identify a subset of popular queries from among the plurality of queries mapped by the online grammar model, wherein the query usage data includes query usage data collected for queries issued by a plurality of users; and

building an offline grammar model that maps the subset of popular queries to actions among the actions for use by a resource-constrained offline device, wherein the offline grammar model has reduced resource requirements relative to the online grammar model and omits mappings for one or more queries among the plurality of queries.

2. The method of claim 1, further comprising communicating the offline grammar model to the resource-constrained offline device for storage by the resource-constrained offline device and for use by an offline semantic processor of the resource-constrained offline device, the offline semantic processor using the offline grammar model to map text, outputted by a voice to text module, to a corresponding action of the offline grammar model, and to identify one or more attributes, for the corresponding action, that constrain the corresponding action based on the corresponding parameters.

3. The method of claim 2, wherein communicating the offline grammar model to the resource-constrained offline device dynamically updates the offline grammar model stored in the resource-constrained offline device.

4. The method of claim 1, wherein the query usage data includes query usage data collected for voice-based queries issued by the plurality of users.

5. The method of claim 4, wherein the query usage data includes voice-based query usage data collected for queries processed by the online voice-based query processor.

6. The method of claim 4, wherein the query usage data includes query usage data collected for voice-based queries issued by a user of the resource-constrained offline device and processed by an offline semantic processor of the resource-constrained offline device.

7. The method of claim 4, wherein the query usage data includes query usage data collected for voice-based queries issued by a plurality of users of a plurality of resource-constrained offline devices and processed by offline semantic processors of the plurality of resource-constrained offline devices.

8. The method of claim 1, wherein the query usage data includes query usage data collected for text queries from the plurality of users.

9. The method of claim 1, wherein the query usage data includes one or more query logs.

10. The method of claim 1, wherein building the offline grammar model includes building a personalized offline grammar model for a user of the resource-constrained offline device based at least in part on query usage data collected from the resource-constrained offline device.

11. The method of claim 1, wherein analyzing the query usage data includes, for a first action among the actions:

determining a distribution of queries from among a plurality of queries mapped to the first action by the online grammar model using the collected query usage data; and

including a top N queries from among the plurality of queries mapped to the first action in the identified subset of popular queries.

12. The method of claim 1, wherein building the offline grammar model includes training the offline grammar model using the subset of popular queries.

13. The method of claim 1, wherein analyzing the query usage data includes, for each of multiple actions among the actions:

determining a corresponding distribution of queries from among a plurality of corresponding queries mapped to the action by the online grammar model using the collected query usage data; and

including, in the identified subset of popular queries, at least a top corresponding query from among the plurality of corresponding queries mapped to the action.

14. The method of claim 1, wherein the actions mapped by the offline grammar model include two or more of:

a setting an alarm action;

a setting a reminder action;

an initiating a communication action;

a playing a song action; and

a changing a device setting action.

15. The method of claim 1, further comprising:

using, by a resource-constrained offline device, the offline grammar model, wherein using the offline grammar model comprises;

mapping text, outputted by a voice to text module of the resource-constrained device, to a corresponding action of the offline grammar model, and

identifying attributes, for the corresponding action, that constrain the corresponding action based on the corresponding parameters.

16. The method of claim 15, further comprising:

performing, by the resource-constrained device, the action with the attributes.

17. The method of claim 15, wherein using the offline grammar model is in response to determining lack of connectivity to the online voice-based query processor.

18. An apparatus including memory and one or more processors operable to execute instructions stored in the memory, comprising instructions to:

maintain an online grammar model used by an online voice-based query processor to parse online voice-based queries, the online grammar model mapping a plurality of queries to actions,

wherein the actions include non-search actions performable by a computer system,

wherein each of the actions is mapped in the grammar model to one or more corresponding queries of the plurality of queries, and

wherein each of a plurality of the actions includes one or more corresponding parameters for constraining performance of the action;

analyze query usage data for at least a subset of the plurality of queries to identify a subset of popular queries from among the plurality of queries mapped by the online grammar model, wherein the query usage data includes query usage data collected for queries issued by a plurality of users; and

build an offline grammar model that maps the subset of popular queries to actions among the actions for use by a resource-constrained offline device, wherein the offline grammar model has reduced resource requirements relative to the online grammar model and omits mappings for one or more queries among the plurality of queries.

19. A non-transitory computer readable storage medium storing computer instructions executable by one or more processors to perform a method comprising:

maintaining an online grammar model used by an online voice-based query processor to parse online voice-based queries, the online grammar model mapping a plurality of queries to actions,

wherein the actions include non-search actions performable by a computer system,

wherein each of the actions is mapped in the grammar model to one or more corresponding queries of the plurality of queries, and

wherein each of a plurality of the actions includes one or more corresponding parameters for constraining performance of the action;

analyzing query usage data for at least a subset of the plurality of queries to identify a subset of popular queries from among the plurality of queries mapped by the online grammar model, wherein the query usage data includes query usage data collected for queries issued by a plurality of users; and

building an offline grammar model that maps the subset of popular queries to actions among the actions for use by a resource-constrained offline device, wherein the offline grammar model has reduced resource requirements relative to the online grammar model and omits mappings for one or more queries among the plurality of queries.