Method and apparatus for performing plan-based dialog
AssigneeMicrosoft Corporation
InventorKuansan Wang
Life of this patent
Estimating the term…
Abstract
The present invention provides a dialog system in which the subsystems are integrated under a single technology model. In particular, each of the sub-systems uses stochastic modeling to identify a probability for its respective output. The combined probabilities identify a most probable action to be taken by the dialog system given the latest input from the user and the past dialog states. An additional aspect of the present invention is an embodiment in which the sub-systems communicate with one another through XML pages, thus allowing the sub-systems to be distributed across a distributed network.
Claims32
A method of providing a dialog interface in a computer system, the method comprising:
- receiving user input;
- applying the user input to a language model to determine a score for at least one surface semantic that describes the content of the user input;
- applying a surface semantic to a discourse model to determine a score for at least one discourse semantic that describes the current state of a dialog with the user;
- applying a discourse semantic to a behavior model to determine a score for at least one action to be taken by the computer system;
- and
- taking an action based in part on the score provided by the discourse model and based in part on the score provided by the behavior model.
A system for providing a computer-based dialog interface to a user, the system comprising:
- a dialog state engine that receives input from the user and that uses at least one model to generate at least one score for a current dialog state that is represented by a discourse semantic structure comprising semantic tokens that provide a general representation of specific entities, the score being based on the user's input and a previous dialog state, the dialog state engine comprising:
- a recognition engine that uses a model to determine a score for at least one surface semantic based on the user input;
- a discourse engine that uses a model to determine a score for at least one current dialog state based on the surface semantic from the recognition engine and a previous dialog state, the discourse engine expanding the discourse semantic structure based on the surface semantic and attempting to collapse a portion of the discourse semantic structure before using the model to determine a score for the discourse semantic structure;
- and
- a database containing specific entities that are represented by the semantic tokens;
- and
- a rendering engine that uses a model to identify a score for at least one action based on a current dialog state.
A system for providing a computer-based dialog interface to a user, the system comprising:
- a dialog state engine that receives input from the user and that uses at least one model to generate at least one score for a current dialog state based on the user's input and a previous dialog state, wherein the dialog state engine describes the current dialog state using a markup language;
- and
- a rendering engine that uses a model to identify a score for at least one action based on a current dialog state and that converts the markup language describing the current dialog state into a second markup language that describes an action.
A method of providing a dialog interface, the method comprising:
- receiving input generated by a user;
- determining a current dialog state based on the received input and a past dialog state;
- formatting the current dialog state into an extensible markup language page;
- converting the markup language page into an output markup language page based on the current dialog state and an available output user interface by applying the extensible markup language page to an extensible stylesheet language transformation;
- and
- passing the output markup language page to the available output user interface.
A system for providing a computer-based dialog interface to a user, the system comprising:
- a dialog state engine that receives input from the user and that generates a markup language page representing a current dialog state;
- and
- a rendering engine that converts the markup language page representing the current dialog state into a markup language page representing an action, wherein the rendering engine comprises extensible stylesheet language transformations that convert the markup language page representing the current dialog state into the markup language page representing an action.
A system for providing a computer-based dialog interface to a user, the system comprising:
- a dialog state engine that receives input from the user and that generates a markup language page representing a current dialog state;
- and
- a rendering engine that converts the markup language page representing the current dialog state into a markup language page representing an action, wherein the rendering engine utilizes a document matching portion of extensible stylesheet language transformations to convert the markup language page representing the current dialog state into machine instructions representing actions that are not described in a markup languages.
A system for providing a computer-based dialog interface to a user, the system comprising:
- a dialog state engine that receives input from the user and that generates a markup language page representing a current dialog state, wherein the dialog state engine comprises:
- a recognition engine that receives user input and generates a markup language page representing the surface semantics of the user input, wherein the recognition engine applies a document generating portion of extensible stylesheet language transformations to produce the markup language page representing the surface semantics;
- and
- a discourse engine that converts the markup language page representing the surface semantics into the markup language page representing the dialog state;
- and
- a rendering engine that converts the markup language page representing the current dialog state into a markup language page representing an action.
Dependencies are inferred from the claim text.