System and method for automatic generation of a natural language understanding model
View Patent ↗A system and method is provided for rapidly generating a new spoken dialog application. In one embodiment, a user experience person labels the transcribed data (e.g., 3000 utterances) using a set of interactive tools. The labeled data is then stored in a processed data database. During the labeling process, the user experience person not only groups utterances in various call type categories, but also flags (e.g., 100-200) specific utterances as positive and negative examples for use in an annotation guide. The labeled data in the processed data database can also be used to generate an initial natural language understanding (NLU) model.
1. A method for generating a natural language understanding model, comprising:
a. collecting a plurality of unlabeled utterances;
b. generating via a processor a plurality of call types for a first natural language understanding model based on call types with high probabilities in an existing natural language understanding model for a different application, each generated call type being based on a first set of utterances selected from the plurality of unlabeled utterances;
c. generating the first natural language understanding model using call type information contained within the first set of utterances;
d. testing the first natural language understanding model;
e. modifying the plurality of call types based on the testing; and
f. generating a second natural language understanding model using the modified plurality of call types.
2. The method of claim 1 , further comprising generating an annotation guide using a second set of utterances which is a subset of the first set of utterances.
3. The method of claim 1 , further comprising generating call type data using at least one of data clustering, relevance feedback, string searching, data mining, and active learning tools.
4. The method of claim 3 , wherein the call type data is generated using a graphical user interface.
5. The method of claim 1 , wherein the first natural language understanding model is trained using a first text file containing utterances contained within the first set of utterances and a second text file containing call types assigned to the utterances in the first text file.
6. The method of claim 1 , wherein the natural language understanding model is tested using a subset of the first set of utterances.
7. The method of claim 1 , wherein the plurality of call types are modified using a graphical user interface.
8. The method of claim 1 , wherein the first natural language understanding model is created prior to an annotation guide.
9. A method for generating a natural language understanding model, comprising:
collecting a plurality of unlabeled utterances;
generating via a processor a plurality of call types for a natural language understanding model based on call types with high probabilities in an existing natural language understanding model for a different application, each of the plurality of call types having utterances selected from the plurality of unlabeled utterances, the utterances used to generate the plurality of call types representing a subset of the collection of utterances; and
generating the natural language understanding model using call type information contained within the subset of utterances, wherein the natural language understanding model is generated prior to receipt of manually labeled utterance data.
10. The method of claim 9 , wherein the manually labeled utterance data is generated using an annotation guide that is created using a portion of the subset of utterances.
11. The method of claim 9 , wherein the natural language understanding model is generated using a first text file containing utterances contained within the subset of utterances and a second text file containing call types assigned to the utterances in the first text file.
12. The method of claim 9 , wherein the natural language understanding model is tested using a second subset of the collection of utterances.