Multiple device intelligent language model synchronization

• US 8,983,849 B2
• Filed: 03/15/2013
• Issued: 03/17/2015
• Est. Priority Date: 10/17/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Phillips, Andrew, Kay, David, Unruh, Erland, Fu, Eric Jun
• Primary Examiner(s): Opsasnick, Michael N
• Application Number:
  US13/834,575
• Publication Number:
  US 20140108003A1
• Time in Patent Office: 732 Days
• Field of Search: 704/270, 704/270.1, 704/275, 704/276
• US Class Current: 704/276
• CPC Class Codes: G06F 16/273 Asynchronous replication or reconciliation
• CPC Class Codes: G06F 3/0237 using prediction or retrieval techniques
• CPC Class Codes: G06F 40/263 Language identification
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

Systems and methods for intelligent language models that can be used across multiple devices are provided. Some embodiments provide for a client-server system for integrating change events from each device running a local language processing system into a master language model. The change events can be integrated, not only into the master model, but also into each of the other local language models. As a result, some embodiments enable restoration to new devices as well as synchronization of usage across multiple devices. In addition, real-time messaging can be used on selected messages to ensure that high priority change events are updated quickly across all active devices. Using a subscription model driven by a server infrastructure, utilization logic on the client side can also drive selective language model updates.

Claims

1. A method for operating a master language processing system, the method comprising:

receiving, at the master language processing system, update messages from multiple local language processing systems,

wherein the update messages include change events to local language models detected by the multiple local language processing systems;

updating a master language model in the master language processing system based on the change events included in the update messages received from the multiple local language processing systems;

generating synchronization messages for each of the multiple local language processing systems to integrate change events from other local language processing systems; and

transmitting the synchronization messages to each of the multiple local language processing systems,

such that updates to the master language model are propagated to the local language models.

2. The method of claim 1, wherein each update message further includes a priority level reflecting the priority of change events in the update message.

3. The method of claim 2, wherein the update messages received at the master language processing system include groups of batched update messages having a low priority level from the multiple local language processing systems.

4. The method of claim 2, wherein the master language model is updated with any update messages having a high priority level before the master language model is updated with any update messages having a low priority level.

5. The method of claim 4, wherein each synchronization message further includes a priority level reflecting the priority of change events in the synchronization message.

6. The method of claim 5, wherein synchronization messages having a high priority level are generated immediately after the master language model is updated with each update message having the high priority level.

7. The method of claim 2, wherein the update messages are transmitted through different network connections depending on priority level.

8. The method of claim 1, wherein the master language model system is a server-based system and update messages are received via a network connection.

9. The method of claim 1, wherein the one or more change events include new words, removed words, n-grams, or word weights.

10. A computer-readable medium, excluding a transitory propagating signal, containing instructions that, when executed cause a computing system to:

receive, at a master language processing system, update messages from multiple local language processing systems associated with devices of a user,

wherein each update message includes one or more change events to a local language model on a device, each of the change events having an associated priority;

update a master language model in the master language processing system based on the change events included in the update messages received from the multiple local language processing systems,

wherein the master language model is updated in an order based on the priority associated with the change event;

generate synchronization messages for each of the multiple local language processing systems to integrate change events from other local language processing systems; and

transmit the synchronization messages to each of the multiple local language processing systems such that updates to the master language model are propagated to the local language models on the devices of the user.

11. The computer-readable medium of claim 10, wherein each synchronization message further includes a priority level reflecting the priority of change events in the synchronization message.

12. The computer-readable medium of claim 11, wherein synchronization messages having a high priority level are generated immediately after the master language model is updated with each change message having a high priority level.

13. The computer-readable medium of claim 10, wherein the received one or more change events are categorized into groups of change events to allow the master language processor to process like events.

14. The computer-readable medium of claim 10, wherein the master language model includes a static language model and a dynamic language model.

15. A computer-implemented method for operating a local language processing system, the method comprising:

detecting, at a computing device running the local language processing system, one or more change events to a local language model,

wherein the one or more change events reflect an update to the local language model such that the local language model is different from a master language model;

modifying the local language model based on the one or more change events that have been detected;

communicating the one or more change events to a master language processing system, the master language processing system using the one or more change events to update a master language model,

wherein the master language processing system operates on a server-based computing system remote from the computing device.

16. The computer-implemented method of claim 15, further comprising receiving a synchronization message from the master language processing system, wherein the synchronization message includes changes to be incorporated in the local language model at the computing device based on event changes reported to the master language processing system by other local language processing systems on other computing devices.

17. The computer-implemented method of claim 15, wherein the local language processing system on the computing device and the other local language processing systems on other computing devices are associated with a common user.

18. The computer-implemented method of claim 15, wherein the one or more change events include new words, removed words, n-grams, or word weights.

19. The computer-implemented method of claim 15, further comprising:

determining whether the computing device is a new device associated with a user; and

if the computing device is a new device, receiving change events to the local language model so that the local language model is synchronized with the master language model.

20. The computer-implemented method of claim 15, further comprising:

determining a priority level associated with each of the one or more change events;

initially sending to the master language processing system any of the one or more change events with a high priority level; and

grouping any of the one or more change events with a low priority level into batches to be sent at a subsequent time to the master language processing system.