Flexible and expandable dialogue system
A system that allows non-engineers administrators, without programming, machine language, or artificial intelligence system knowledge, to expand the capabilities of a dialogue system. The dialogue system may have a knowledge system, user interface, and learning model. A user interface allows non-engineers to utilize the knowledge system, defined by a small set of primitives and a simple language, to annotate a user utterance. The annotation may include selecting actions to take based on the utterance and subsequent actions and configuring associations. A dialogue state is continuously updated and provided to the user as the actions and associations take place. Rules are generated based on the actions, associations and dialogue state that allows for computing a wide range of results.
1. A method for expanding a dialogue system, comprising:
displaying, to a human annotator via an annotation interface, a dialogue including a user utterance and a candidate action defining one or more computer-implemented steps to be performed responsive to the user utterance, the dialogue accessed from an application executing on a server computer, the candidate action provided by a learning model, and the candidate action including an error in a particular computer-implemented step of the one or more computer-implemented steps of the candidate action to be performed responsive to the user utterance;
receiving, from the human annotator via the annotation interface, an annotated action that corrects the error in the particular computer-implemented step of the candidate action by specifying a fix to the particular computer-implemented step;
maintaining a representation of a dialogue state on the server computer, the dialogue state including the utterance and the annotated action; and
training the learning model based on the utterance, the annotated action, and the representation of the dialogue state.
2. The method of claim 1 , wherein the training includes defining one or more features on input based on detected relevant state portions.
3. The method of claim 1 , wherein the error is a constraint violation.
4. The method of claim 3 , further comprising generating a question to submit to the human annotator based on the constraint violation.
5. The method of claim 1 , wherein the training includes inducing business logic rules from data.
6. The method of claim 1 , wherein the training includes inserting rules based on grammar induction.
7. The method of claim 1 , wherein the training includes inserting tags into the dialogue.
8. The method of claim 1 , wherein the training includes performing a dialogue check on the dialogue.
9. The method of claim 1 , wherein the training includes marginalizing actions over an equivalent action.
10. The method of claim 1 , further comprising displaying the representation of the dialogue state within the annotation interface, the representation of the dialogue state updated based on the annotated action received from the human annotator.
11. A computer readable storage device having embodied thereon a program, the program being executable by a processor to perform a method for expanding a dialogue system, the method comprising:
displaying, to a human annotator via an annotation interface, a dialogue including a user utterance and a candidate action defining one or more computer-implemented steps to be performed responsive to the user utterance, the dialogue accessed from an application executing on a server computer, the candidate action provided by a learning model, and the candidate action including an error in a particular computer-implemented step of the one or more computer-implemented steps of the candidate action to be performed responsive to the user utterance;
receiving, from the human annotator via the annotation interface, an annotated action that corrects the error in the particular computer-implemented step of the candidate action by specifying a fix to the particular computer-implemented step;
maintaining a representation of a dialogue state on the server computer, the dialogue state including the utterance and the annotated action; and
training the learning model based on the utterance, the annotated action, and the representation of the dialogue state.
12. The computer readable storage device of claim 11 , wherein the training includes defining one or more features on input based on detected relevant state portions.
13. The computer readable storage device of claim 11 , wherein the error is a constraint violation.
14. The computer readable storage device of claim 13 , further comprising generating a question to submit to the human annotator based on the constraint violation.
15. The computer readable storage device of claim 11 , wherein the training includes inducing business logic rules from data.
16. The computer readable storage device of claim 11 , wherein the training includes inserting rules based on grammar induction.
17. The computer readable storage device of claim 11 , wherein the training includes inserting tags into the dialogue.
18. The computer readable storage device of claim 11 , wherein the training includes performing a dialogue check on the dialogue.
19. The computer readable storage device of claim 11 , wherein the training includes marginalizing actions over an equivalent action.
20. A system for expanding a dialogue system, comprising:
a processor;
memory; and
one or more modules stored in memory and executable by the processor to:
display, to a human annotator via an annotation interface, a dialogue including a user utterance and a candidate action defining one or more computer-implemented steps to be performed responsive to the user utterance, the dialogue accessed from an application executing on a server computer, the candidate action provided by a learning model, and the candidate action including an error in a particular computer-implemented step of the one or more computer-implemented steps of the candidate action to be performed responsive to the user utterance;
receive, from the human annotator via the annotation interface, an annotated action that corrects the error in the particular computer-implemented step of the candidate action by specifying a fix to the particular computer-implemented step;
maintain a representation of a dialogue state on the server computer, the dialogue state including the utterance and the annotated action; and
train the learning model based on the utterance, the annotated action, and the representation of the dialogue state.