Method and apparatus for exploiting human feedback in an intelligent automated assistant
First Claim
Patent Images
1. A method for conducting an interaction between a human user and a device, the method comprising:
- with the device, receiving input indicative of a user request for information;
with a microphone coupled to the device, receiving sensed data;
with a feature extraction processor coupled to the microphone, extracting a plurality of speech features from the sensed data;
with a classifier processor coupled to the feature extraction processor, inferring an affective state of the human user based on the plurality of speech features extracted from the sensed data;
with an interaction management system coupled to the classifier processor, inferring an intent from the received input by performing one or more of automated speech recognition and natural language understanding using a learned model;
formulating a proposed response to the received input in accordance with the intent, the proposed response comprising system-generated output;
determining a measure of certainty associated with one or more of the intent and the proposed response;
presenting a final response to the received input by an output device of the device when the measure of certainty satisfies a minimum acceptable level of certainty;
with an interface coupled to the interaction management system between the inferring of the intent and the presenting of the final response, when the measure of certainty does not satisfy the minimum acceptable level of certainty and prior to presenting the final response, communicating the intent and the proposed response and the inferred affective state to a wizard, receiving feedback on the intent and the proposed response and the affective state from the wizard, incorporating the feedback into the final response, updating a model used to generate the proposed response based on the feedback, wherein the wizard is a human person who is not a source of the received input.
1 Assignment
0 Petitions
Accused Products
Abstract
The present invention relates to a method and apparatus for exploiting human feedback in an intelligent automated assistant. One embodiment of a method for conducting an interaction with a human user includes inferring an intent from data entered by the human user, formulating a response in accordance with the intent, receiving feedback from a human advisor in response to at least one of the inferring and the formulating, wherein the human advisor is a person other than the human user, and adapting at least one model used in at least one of the inferring and the formulating, wherein the adapting is based on the feedback.
10 Citations
24 Claims
-
1. A method for conducting an interaction between a human user and a device, the method comprising:
-
with the device, receiving input indicative of a user request for information; with a microphone coupled to the device, receiving sensed data; with a feature extraction processor coupled to the microphone, extracting a plurality of speech features from the sensed data; with a classifier processor coupled to the feature extraction processor, inferring an affective state of the human user based on the plurality of speech features extracted from the sensed data; with an interaction management system coupled to the classifier processor, inferring an intent from the received input by performing one or more of automated speech recognition and natural language understanding using a learned model; formulating a proposed response to the received input in accordance with the intent, the proposed response comprising system-generated output; determining a measure of certainty associated with one or more of the intent and the proposed response; presenting a final response to the received input by an output device of the device when the measure of certainty satisfies a minimum acceptable level of certainty; with an interface coupled to the interaction management system between the inferring of the intent and the presenting of the final response, when the measure of certainty does not satisfy the minimum acceptable level of certainty and prior to presenting the final response, communicating the intent and the proposed response and the inferred affective state to a wizard, receiving feedback on the intent and the proposed response and the affective state from the wizard, incorporating the feedback into the final response, updating a model used to generate the proposed response based on the feedback, wherein the wizard is a human person who is not a source of the received input. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22)
-
-
23. At least one non-transitory computer readable medium containing an executable program for conducting an interaction between a human user and a computing device, where the executable program performs steps comprising:
-
with a microphone coupled to the computing device, receiving sensed data; with a feature extraction processor coupled to the microphone, extracting a plurality of features from the sensed data; with a classifier processor coupled to the feature extraction processor, inferring an affective state of the human user based on the plurality of features extracted from the sensed data; with an interaction management system coupled to the classifier processor, inferring an intent from a received input by performing one or more of automated speech recognition and natural language understanding using a learned model; formulating a proposed response in accordance with the intent and the affective state, the proposed response comprising system-generated output; determining a measure of certainty associated with one or more of the intent and the proposed response; presenting a final response by an output device of the computing device when the measure of certainty satisfies a minimum acceptable level of certainty; interposing an interface between the inferring of the intent and the presenting of the final response to, when the measure of certainty does not satisfy the minimum acceptable level of certainty and prior to presenting the final response, communicate the intent and the proposed response to a wizard, receive feedback on the intent and the proposed response from the wizard, incorporate the feedback into the final response, update a model used to generate the proposed response based on the feedback, wherein the wizard is not a source of the received input.
-
-
24. A system for conducting an interaction between a human user and a consumer computing device, the system comprising a plurality of processor-executable modules embodied in one or more non-transitory machine readable storage media, the system configured to:
-
with a microphone coupled to the consumer computing device, receiving sensed data; with a feature extraction processor coupled to the microphone, extracting a plurality of features from the sensed data; with a classifier processor coupled to the feature extraction processor, inferring an affective state of the human user based on the plurality of features extracted from the sensed data; with an interaction management system coupled to the classifier processor, infer an intent from a received input by performing one or more of automated speech recognition and natural language understanding using a learned model; formulate a proposed response in accordance with the intent and the affective state, the proposed response comprising system-generated output; determine a measure of certainty associated with one or more of the intent and the proposed response; present a final response by an output device of the consumer computing device when the measure of certainty satisfies a minimum acceptable level of certainty; with an interface coupled between the inferring of the intent and the presenting of the final response, when the measure of certainty does not satisfy the minimum acceptable level of certainty and prior to presenting the final response, communicate the intent and the proposed response to a wizard, receive feedback on the intent and the proposed response from the wizard, incorporate the feedback into the final response, update a model used to generate the proposed response based on the feedback, wherein the wizard is not a source of the received input.
-
Specification