Active labeling for spoken language understanding
View Patent ↗An active labeling process is provided that aims to minimize the number of utterances to be checked again by automatically selecting the ones that are likely to be erroneous or inconsistent with the previously labeled examples. In one embodiment, the errors and inconsistencies are identified based on the confidences obtained from a previously trained classifier model. In a second embodiment, the errors and inconsistencies are identified based on an unsupervised learning process. In both embodiments, the active labeling process is not dependent upon the particular classifier model.
1. A classifier associated with automatic speech recognition (ASR), the classifier comprising:
a processor;
a module configured to control the processor to select a first set of reference utterances, each labeled with at least one classification type;
a module configured to control the processor to generate, based on a trained classifier, a confidence score for a plurality of classification types for each of the first set of reference utterances; and
a module configured to control the processor to identify a second set of candidate utterances from the first set of reference utterances as having a potential classification error, wherein the identifying is based on an analysis of previously assigned classification types and generated confidence scores.
2. The classifier of claim 1 , wherein the confidence score is generated by the trained classifier.
3. The classifier of claim 1 , wherein the confidence score is calculated based on an output of the trained classifier.
4. A system associated with speech classification, the system comprising:
a processor;
a module configured to control the processor to select a first set of candidate utterances, each of the first set of candidate utterances labeled with at least one previously assigned classification type;
a module configured to control the processor to train a classifier using the first set of candidate utterances to produce a trained classifier;
a module configured to control the processor to classify the first set of candidate utterances using the trained classifier; and
a module configured to control the processor to identify a second set of candidate utterances from the first set of candidate utterances as having a potential classification error, wherein the identifying is based on an analysis of the previously assigned classification types and the results of the classifying.
5. The system of claim 4 , wherein the first set of candidate utterances includes labeled and unchecked data.
6. The system of claim 4 , wherein the classifier is an iterative classifier.
7. The system of claim 4 , further comprising a module configured to control the processor to generate, based on a trained classifier, a confidence score for a plurality of classification types for each of the first set of candidate utterances.
8. The system of claim 4 , wherein the classification types include call classification types.
9. A system that classifies use input, the system comprising:
a processor;
a module configured to control the processor to classify a set of candidate utterances using a classifier, the set of candidate utterances including labeled and unchecked data; and
a module configured to control the processor to automatically select a subset of the set of candidate utterances as likely including erroneous or inconsistent classifications, the automatic selection being based on an analysis of an output of the classifier.
10. The system of claim 9 , wherein the classifier is an iterative classifier.
11. The system of claim 9 , wherein the output of the classifier includes confidence scores.
12. The system of claim 9 , wherein the output of the classifier includes information used to derive confidence scores.
13. The system of claim 9 , wherein the output of the classifier includes classification labels.
14. A spoken language understanding system having a processor and modules configured to control the processor, the system generated by a method comprising:
retrieving a labeled and unchecked set of candidate utterances;
training a classifier using the candidate utterances;
classifying the set of candidate utterances;
sorting the classified utterances by determining which of the classified utterances have classifications that are distinct from labels in the labeled and unchecked set of candidate utterances; and
rechecking utterances that have classifications that do not match.