System and method of semi-supervised learning for spoken language understanding using semantic role labeling
View Patent ↗A system and method are disclosed for providing semi-supervised learning for a spoken language understanding module using semantic role labeling. The method embodiment relates to a method of generating a spoken language understanding module. Steps in the method comprise selecting at least one predicate/argument pair as an intent from a set of the most frequent predicate/argument pairs for a domain, labeling training data using mapping rules associated with the selected at least one predicate/argument pair, training a call-type classification model using the labeled training data, re-labeling the training data using the call-type classification model and iteratively several of the above steps until training set labels converge.
1. A method comprising:
selecting, via a processor, an intent from a list of predicate/argument pairs associated with a spoken dialog system;
labeling training data using mapping rules associated with the intent, wherein the mapping rules specify rules for selecting a call-type label for an utterance; and
while the training data and a classification model associated with the call-type label have a divergence, iteratively:
training the classification model using the training data; and
re-labeling the training data using the classification model.
2. The method of claim 1 , further comprising assigning the verbs “be” and “have” as special predicates.
3. The method of claim 1 , further comprising distinguishing verbs from utterances which do not have a predicate by assigning the verbs to a special class.
4. The method of claim 1 , wherein the method is semi-supervised.
5. The method of claim 1 , further comprising capturing infrequent call types using an active-learning approach.
6. The method of claim 1 , wherein the selecting of the intent is performed independent of a domain.
7. The method of claim 1 , wherein the mapping rules specify that the call-type is represented by multiple predicate/argument pairs.
8. A system comprising:
a processor; and
a computer-readable storage medium having instructions stored which, when executed by the processor, perform operations comprising:
selecting an intent from a list of predicate/argument pairs associated with a spoken dialog system;
labeling training data using mapping rules associated with the intent, wherein the mapping rules specify rules for selecting a call-type label for an utterance; and
while the training data and a classification model associated with the call-type label have a divergence, iteratively:
training the classification model using the training data; and
re-labeling the training data using the classification model.
9. The system of claim 8 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising assigning the verbs “be” and “have” as special predicates.
10. The system of claim 8 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising distinguishing verbs from utterances which do not have a predicate by assigning the verbs to a special class.
11. The system of claim 8 , wherein the operations are semi-supervised.
12. The system of claim 8 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising capturing infrequent call types using an active-learning approach.
13. The system of claim 8 , wherein the selecting of the intent is performed independent of a domain.
14. The system of claim 8 , wherein the mapping rules specify that the call-type is represented by multiple predicate/argument pairs.
15. A computer-readable storage medium having instructions stored which, when executed by a computing device, cause the computing device to perform operations comprising:
selecting an intent from a list of predicate/argument pairs associated with a spoken dialog system;
labeling training data using mapping rules associated with the intent, wherein the mapping rules specify rules for selecting a call-type label for an utterance; and
while the training data and a classification model associated with the call-type label have a divergence, iteratively:
training the classification model using the training data; and
re-labeling the training data using the classification model.
16. The computer-readable storage medium of claim 15 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising assigning the verbs “be” and “have” as special predicates.
17. The computer-readable storage medium of claim 15 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising distinguishing verbs from utterances which do not have a predicate by assigning the verbs to a special class.
18. The computer-readable storage medium of claim 15 , wherein the operations are semi-supervised.
19. The computer-readable storage medium of claim 15 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising capturing infrequent call types using an active-learning approach.
20. The computer-readable storage medium of claim 15 , wherein the selecting of the intent is performed independent of a domain.