Granted Patent B1

US 7,472,060 · App. 11/219,629 · Granted Dec 30, 2008

Abstract

Claims (34)

1. A method for use in an automated dialog system in which a dialog comprising dialog exchanges is conducted between the system and a user, at least ones of the dialog exchanges of the dialog including an input communication from the user, the input communication including one or more communication symbols, the method comprising

recognizing at least one of the communication symbols of the input communication of a present dialog exchange to determine the input communication, and

determining a probability of conducting a successful dialog with the user, a successful dialog being a dialog exchange between the automated dialog system and the user that results in processing of input communications of the user to a stage consistent with the user's intent, said probability being determined based, at least in part, on understanding data from at least one prior dialog exchange of the dialog, said understanding data being data indicative of a probability that the input communication of said at least one prior dialog exchange was correctly understood,

wherein said one or more communication symbols includes at least one of words, phrases, morphemes, and gestures,

wherein said understanding data for a particular dialog exchange of the dialog includes at least one confidence score, said confidence score being a measure of a confidence that a particular task associated with said confidence score was a task that the user was requesting the system to perform in said particular dialog exchange,

wherein said at least one confidence score is a top-confidence score, said top-confidence score being the highest confidence score among a plurality of confidence scores, each of the plurality of confidence scores being a measure of a confidence that a particular respective task associated with said each confidence score was a task that the user was requesting the system to perform in said particular dialog exchange,

and wherein said understanding data for a particular dialog exchange of the dialog includes a confpertime feature, said confpertime feature being said top-confidence score normalized by the duration of the input communication.

2. A method for use in an automated dialog system in which a dialog comprising dialog exchanges is conducted between the system and a user, at least ones of the dialog exchanges of the dialog including an input communication from the user, the input communication including one or more communication symbols, the method comprising

recognizing at least one of the communication symbols of the input communication of a present dialog exchange to determine the input communication, and

determining a probability of conducting a successful dialog with the user, a successful dialog being a dialog exchange between the automated dialog system and the user that results in processing of input communications of the user to a stage consistent with the user's intent, said probability being determined based, at least in part, on understanding data from at least one prior dialog exchange of the dialog, said understanding data being data indicative of a probability that the input communication of said at least one prior dialog exchange was correctly understood,

wherein said one or more communication symbols includes at least one of words, phrases, morphemes, and gestures,

wherein the input communication is an utterance and wherein said understanding data for a particular dialog exchange of the dialog includes a salience-coverage feature, said salience-coverage feature being a measure of the proportion of the utterance that is covered by salient grammar fragments in said utterance,

and wherein said understanding data for a particular dialog exchange of the dialog includes a salpertime feature, said salpertime feature being said salience-coverage feature normalized by the duration of the input communication.

3. A method for use by an automated dialog system, the method comprising

determining a probability that the system will correctly understand a user's intent based on a dialog between the user and the system, said probability being determined based, at least in part, on understanding data of a) an input communication of a present dialog exchange between said system and said user and b) an input communication of one or more dialog exchanges between said system and said user that were prior to said present dialog exchange,

each said understanding data being data associated with a respective input communication from the user and being indicative of the probability that the system has correctly understood the associated input communication wherein said user's intent is to ask the system to perform a particular task,

wherein said understanding data for a particular input communication includes at least one confidence score, said confidence score being a measure of a confidence that a particular task associated with said confidence score was a task that the user was requesting the system to perform in said particular input communication,

wherein said at least one confidence score is a top-confidence score, said top-confidence score being the highest confidence score among a plurality of confidence scores, each of the plurality of confidence scores being a measure of a confidence that a particular respective task associated with said confidence score was a task that the user was requesting the system to perform in said particular input communication,

and wherein said understanding data for a particular input communication includes a confpertime feature, said confpertime feature being said top-confidence score normalized by the duration of said particular input communication.

4. A method for use by an automated dialog system, the method comprising

determining a probability that the system will correctly understand a user's intent based on a dialog between the user and the system, said probability being determined based, at least in part, on understanding data of a) an input communication of a present dialog exchange between said system and said user and b) an input communication of one or more dialog exchanges between said system and said user that were prior to said present dialog exchange,

each said understanding data being data associated with a respective input communication from the user and being indicative of the probability that the system has correctly understood the associated input communication,

wherein said user's intent is to ask the system to perform a particular task,

wherein at least one of said input communications is an utterance and wherein said understanding data for that input communication includes a salience-coverage feature, said salience-coverage feature being a measure of the proportion of the utterance that is covered by salient grammar fragments in said utterance,

and wherein said understanding data for said at least one of said input communications includes a salpertime feature, said salpertime feature being said salience-coverage feature normalized by the duration of said at least one of the input communications.

5. A method for use in an automated dialog system that conducts a dialog including input communications from a user, the automated dialog system being adapted to recognize the input communications, the method comprising

determining a probability of the dialog being successful, the dialog being successful when it results in the automated dialog system correctly understanding the intended natural language meaning of the input communications from the user, said probability being determined based, at least in part, on natural language understanding data indicative of whether each of a present one of the input communications and at least one prior one of the input communication have been correctly understood,

wherein said natural language understanding data for said each input communication includes at least one confidence score, said confidence score being a measure of a confidence that a particular task associated with said confidence score was a task that the user was requesting the system to perform in said each input communication,

wherein said confidence score is a top-confidence score, said top-confidence score being the highest confidence score among a plurality of confidence scores associated with said each input communication, each of the plurality of confidence scores being a measure of a confidence that a particular respective task associated with said confidence score was a task that the user was requesting the system to perform in said each input communication,

and wherein said natural language understanding data for said each input communication includes a confpertime feature, said confpertime feature being said top-confidence score normalized by the duration of said each input communication.

6. A method for use in an automated dialog system that conducts a dialog including input communications from a user, the automated dialog system being adapted to recognize the input communications, the method comprising

determining a probability of the dialog being successful, the dialog being successful when it results in the automated dialog system correctly understanding the intended natural language meaning of the input communications from the user, said probability being determined based, at least in part, on natural language understanding data indicative of whether each of a present one of the input communications and at least one prior one of the input communication have been correctly understood,

wherein at least one of said each input communication is an utterance and wherein said natural language understanding data for that input communication includes a salience-coverage feature, said salience-coverage feature being a measure of the proportion of the utterance that is covered by salient grammar fragments in said utterance,

and wherein said natural language understanding data for said each input communication includes a salpertime feature, said salpertime feature being said salience-coverage feature normalized by the duration of said each input communication.

Assignments (7)