Abstract: Systems and processes for optimizing dialogue policy decisions for digital assistants using implicit feedback are provided. In an example process, a user utterance is received. Based on a text representation of the user utterance, one or more user intents corresponding to the user utterance are determined. A policy action is selected from a plurality of candidate policy actions based on a belief state for the one or more user intents and a policy model. The policy action is performed, including outputting results of the policy action for presentation. A success score for the policy action is determined based on whether one or more predetermined types of implicit user feedback are detected after performing the policy action. A set of parameter values of the policy model is modified using the determined success score.

Abstract: Systems and processes for operating a digital assistant using a hierarchical belief state are disclosed. In an example process, a user utterance of a dialog is received. A belief state for the dialog is determined. The belief state comprises a plurality of dialog slots. Each dialog slot of the plurality of dialog slots includes a respective marginal certainty for a concept or property represented by the respective dialog slot. A first dialog slot of the plurality of dialog slots further includes one or more joint certainties for one or more interpretations arising from the first dialog slot. Based on the marginal certainty of each dialog slot of the plurality of dialog slots and the one or more joint certainties of the first dialog slot, a policy action is selected from a plurality of candidate policy actions that correspond to the belief state. The selected policy action is performed.

Abstract: Systems and processes for operating a digital assistant using a hierarchical belief state are disclosed. In an example process, a user utterance of a dialogue is received. A belief state for the dialogue is determined. The belief state comprises a plurality of dialogue slots. Each dialogue slot of the plurality of dialogue slots includes a respective marginal certainty for a concept or property represented by the respective dialogue slot. A first dialogue slot of the plurality of dialogue slots further includes one or more joint certainties for one or more interpretations arising from the first dialogue slot. Based on the marginal certainty of each dialogue slot of the plurality of dialogue slots and the one or more joint certainties of the first dialogue slot, a policy action is selected from a plurality of candidate policy actions that correspond to the belief state. The selected policy action is performed.

Abstract: Systems and processes for optimizing dialogue policy decisions for digital assistants using implicit feedback are provided. In an example process, a user utterance is received. Based on a text representation of the user utterance, one or more user intents corresponding to the user utterance are determined. A policy action is selected from a plurality of candidate policy actions based on a belief state for the one or more user intents and a policy model. The policy action is performed, including outputting results of the policy action for presentation. A success score for the policy action is determined based on whether one or more predetermined types of implicit user feedback are detected after performing the policy action. A set of parameter values of the policy model is modified using the determined success score.