Using generic predictive models for slot values in language modeling

• US 8,032,375 B2
• Filed: 03/17/2006
• Issued: 10/04/2011
• Est. Priority Date: 03/17/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Paek, Timothy S., Chickering, David M.
• Primary Examiner(s): RIDER, JUSTIN W
• Application Number:
  US11/378,202
• Publication Number:
  US 20070219974A1
• Time in Patent Office: 2,027 Days
• Field of Search: 704/255, 704/270
• US Class Current: 704/255
• CPC Class Codes: G06Q 10/10 Office automation; Time management

Abstract

A generic predictive argument model that can be applied to a set of shot values to predict a target slot value is provided. The generic predictive argument model can predict whether or not a particular value or item is the intended target of the user command given various features. A prediction for each of the slot values can then be normalized to infer a distribution over all values or items. For any set of slot values (e.g., contacts), a number of binary variable s are created that indicate whether or not each specific slot value was the intended target. For each slot value, a set of input features can be employed to predict the corresponding binary variable. These input features are generic properties of the contact that are “instantiated” based o n properties of the contact (e.g., contact-specific features). These contact-specific features can be stored in a user data store.

Claims

1. A system for the prediction of a target slot value of a user goal, comprising:

a user data store that stores information regarding user behavior, the information comprising a plurality of features associated with a user'"'"'s behavior comprising at least one or more of a periodic feature, a contact-specific feature, a predicate-related feature, a time-related feature and a device-related feature, at least the periodic feature and the time-related feature being independent of content that is included in a user input; and

an argument model that employs statistical modeling and the information stored in the user data store, the argument model is applied to each of a set of slot values to predict the target slot value by;

for each of the set of slot values, determining a probability that the particular slot value is the target slot value,

normalizing the probabilities of all of the slot values to infer a probability distribution over all of the slot values, and

identifying the target slot value using the inferred probability distribution.

2. The system of claim 1, further comprising:

a predictive user model that employs statistical modeling, the statistical modeling of the predictive user model is performed without regard to the substantive content of the user input.

3. The system of claim 1, wherein the argument model for unseen values is uniform.

4. The system of claim 1, wherein the argument model comprises a conditional model.

5. The system of claim 4, wherein the conditional model is a decision tree.

6. The system of claim 1, wherein the argument model comprises a generative model.

7. The system of claim 6, wherein the generative model is a naï

ve-Bayes model.

8. The system of claim 1, wherein the argument model is learned for a general user population.

9. The system of claim 1, wherein at least one of a parameter or a structure of the argument model is updated in an online manner.

10. The system of claim 9, wherein the online update is performed according to a learning rate.

11. The system of claim 1 employed for command/control of a personal device, wherein the personal device comprises a desktop computer system, a notebook computer system, a personal digital assistant, a media player, a mobile phone, a smart phone or a remote control.

12. A method for prediction of a target slot value of a user goal for command/control of a personal device, comprising:

receiving a user input at the personal device, the user input comprises at least one or more of a periodic feature, a contact-specific feature, a predicate-related feature, a time-related feature and a device-related feature, at least the periodic feature and the time-related feature being independent of content that is included in a user input;

for each of a set of slot values, applying a generic argument model on the user input by the user device to obtain a probability that the slot value is the target slot value;

normalizing the probabilities of the slot values to infer a distribution over the set of slot values; and

employing the inferred probability distribution to identify the target slot value.

13. The method of claim 12, wherein the probability distribution determined by the argument model is independent of a speech input.

14. The method of claim 12, wherein the set of slot values comprises seen slot values.

15. A language model system for prediction of a user goal for command/control of a personal device, comprising:

a predictive user model that comprises an argument model that employs statistical modeling and information stored in a user data store, the information stored in the user data store comprises a plurality of features associated with the particular user'"'"'s behavior, the features comprising at least one or more of a periodic feature, a contact-specific feature, a predicate-related feature, a time-related feature and a device-related feature, at least the periodic feature and the time-related feature being independent of content that is included in a user input, the argument model is applied to each of a set of slot values to predict a target slot value by;

for each of the set of slot values utilized by a user, determining a probability that the particular slot value is the target slot value,

normalizing the probabilities of all of the slot values to infer a probability distribution over all of the slot values, and

identifying the target slot value using the inferred probability distribution;

a speech recognition component that provides a probability distribution associated with a speech input; and

a language model that determines the user goal based, at least in part, upon the probability distributions provided by the predictive user model and the speech recognition component.