Scaling statistical language understanding systems across domains and intents

• US 9,292,492 B2
• Filed: 02/04/2013
• Issued: 03/22/2016
• Est. Priority Date: 02/04/2013
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Sarikaya, Ruhi, Deoras, Anoop, Celikyilmaz, Fethiye Asli, Janardhana, Ravikiran, Boies, Daniel
• Primary Examiner(s): SINGH, SATWANT K
• Application Number:
  US13/758,683
• Publication Number:
  US 20140222422A1
• Time in Patent Office: 1,142 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3338 Query expansion
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation

Abstract

A scalable statistical language understanding (SLU) system uses a fixed number of understanding models that scale across domains and intents (i.e. single vs. multiple intents per utterance). For each domain added to the SLU system, the fixed number of existing models is updated to reflect the newly added domain. Information that is already included in the existing models and the corresponding training data may be re-used. The fixed models may include a domain detector model, an intent action detector model, an intent object detector model and a slot/entity tagging model. A domain detector identifies different domains identified within an utterance. All/portion of the detected domains are used to determine associated intent actions. For each determined intent action, one or more intent objects are identified. Slot/entity tagging is performed using the determined domains, intent actions, and intent object detector.

Claims

1. A method for scaling a language understanding system, comprising:

receiving, by a computing device, an utterance using the language understanding system;

detecting, by the computing device, a plurality of domains that are associated with the utterance;

detecting, by the computing device, a plurality of actions from the utterance for each detected domain of the plurality of domains;

associating, by the computing device, each action of the plurality of actions with one or more of the detected plurality of domains;

detecting, by the computing device, a plurality of objects from the utterance for each detected action of the plurality of actions;

training at least one model by applying a threshold hypothesis to at least one stage from a model manager, whereby training data is shared across different domains supported by the language understanding system; and

performing, by the computing device, slot filling, wherein at least one slot is filled, the at least one slot further defining at least one of the plurality of domains or the plurality of objects.

2. The method of claim 1, wherein the language understanding system comprises:

a domain model;

an intent action model;

an intent object model; and

a slot filling model.

3. The method of claim 1, wherein detecting the plurality of domains that are associated with the utterance comprises accessing, by the computing device, a single domain model that is used to create a list of possible domains detected from the utterance.

4. The method of claim 3, wherein intent objects are shared across different domains.

5. The method of claim 1, wherein detecting the plurality of actions from the utterance comprises accessing, by the computing device, a single intent action model that is used to create a list of possible actions using the detected domains.

6. The method of claim 1, wherein detecting the plurality of objects from the utterance comprises accessing, by the computing device, a single intent object model that is used to create a list of possible objects using the detected domains.

7. The method of claim 1, wherein intent actions are shared across different domains.

8. The method of claim 1, wherein training data is shared across different domains supported by the language understanding system.

9. The method of claim 1, wherein an additional model is not added when a new domain is added to the language understanding system.

10. A computer storage device having computer-executable instructions that, when executed by at least one processor, cause the at least one processor to perform a method for scaling statistical language understanding systems, the method comprising:

receiving, by a computing device, an utterance using the language understanding system, wherein the language understanding system includes at least a domain model;

an intent action model;

an intent object model; and

a slot filling model;

detecting, by the computing device, a plurality of domains that are associated with the utterance;

detecting, by the computing device, a plurality of actions from the utterance for each detected domain of the plurality of domains;

associating, by the computing device, each action of the plurality of actions with one or more of the detected plurality of domains;

detecting by the computing device, a plurality of objects from the utterance for each detected action of the plurality of actions;

training at least one model by applying a threshold hypothesis to at least one stage from a model manager, whereby training data is shared across different domains supported by the language understanding system; and

performing, by the computing device, slot filling, wherein at least one slot is filled, the at least one slot further defining at least one of the plurality of domains or the plurality of objects.

11. The computer storage device of claim 10, wherein the domain model is a single domain model that is used to create a list of possible domains detected from the utterance.

12. The computer storage device of claim 10, wherein the intent action model is a single intent action model that is used to create a list of possible actions using the detected domains.

13. The computer storage device of claim 10, wherein the intent object model is a single intent object model that is used to create a list of possible objects using the detected domains.

14. The computer storage device of claim 10, wherein intent actions included in the intent action model are shared across different domains and wherein intent objects in the intent action model are shared across different domains.

15. The computer storage device of claim 10, wherein training data is shared across different domains supported by the language understanding system.

16. A system for scaling statistical language understanding systems, comprising:

a processor and memory;

an operating environment executing using the processor;

a display; and

a model manager that is configured to perform actions comprising;

receiving by a computing device, an utterance using the language understanding system, wherein the language understanding system includes at least a domain model;

an intent action model;

an intent object model; and

a slot filling model;

detecting, by the computing device, a plurality of domains that are associated with the utterance;

detecting, by the computing device, a plurality of actions from the utterance for each detected domain of the plurality of domains;

associating, by the computing device, each action of the plurality of actions with one or more of the detected plurality of domains;

detecting, by the computing device, a plurality of objects from the utterance for each detected action of the plurality of actions;

training at least one model by applying a threshold hypothesis to at least one stage from a model manager, whereby training data is shared across different domains supported by the language understanding system; and

performing, by the computing device, slot filling, wherein at least one slot is filled, the at least one slot further defining at least one of the plurality of domains or the plurality of objects.

17. The system of claim 16, wherein the domain model is a single domain model that is used to create a list of possible domains detected from the utterance.

18. The system of claim 16, wherein the intent action model is a single intent action model that is used to create a list of possible actions using the detected domains.

19. The system of claim 16, wherein the intent object model is a single intent object model that is used to create a list of possible objects using the detected domains.

20. The system of claim 16, wherein intent actions included in the intent action model are shared across different domains and wherein intent objects in the intent action model are shared across different domains.