Amazon Technologies, Inc.

Charles Elkan

Rashmi Gangadharaiah

Balakrishnan Narayanaswamy

In response to determining that a particular sequence of natural language input has been generated by a first entity participating in a multi-interaction dialog, a first representation of accumulated dialog state associated with the sequence is obtained from a machine learning model at an artificial intelligence service. Based on the first representation, a state response entry is selected from a collection of state response entries. The state response entry indicates a mapping between a second representation of accumulated dialog state, and a response recorded in a training example of the model. The recorded response is implemented.

Artificial intelligence system using hybrid technique for task-oriented dialog management

A goal-oriented dialog system interacts with a user over one or more turns of dialog to determine a goal expressed by the user; the dialog system may then act to fulfill the goal by, for example, calling an application-programming interface. The user may supply dialog via text, speech, or other communication. The dialog system includes a first trained model, such as a translation model, to encode the dialog from the user into a context vector; a second trained model, such as another translation model, determines a plurality of candidate probabilities of items in a vocabulary. A language model determines responses to the user based on the input from the user, the context vector, and the plurality of candidate probabilities.

Goal-oriented dialog systems and methods

Techniques for intelligent task-oriented multi-turn dialog system automation are described. A seq2seq ML model can be trained using a corpus of training data and a loss function that is based at least in part on a distance to a goal. The seq2seq ML model can be provided a user utterance as an input, and a vector of a plurality of values output by a plurality of hidden units of a decoder of the seq2seq ML model can be used to select one or more candidate responses to the user utterance via a nearest neighbor algorithm. In some embodiments, the specially adapted seq2seq ML model can be trained using unsupervised learning, and can be adapted to select intelligent, coherent agent responses that move a task-oriented dialog toward its completion.

Growing community of inventors

Seattle, WA, United States of America

Charles Elkan