Robust information extraction from utterances
The performance of traditional speech recognition systems (as applied to information extraction or translation) decreases significantly with, larger domain size, scarce training data as well as under noisy environmental conditions. This invention mitigates these problems through the introduction of a novel predictive feature extraction method which combines linguistic and statistical information for representation of information embedded in a noisy source language. The predictive features are combined with text classifiers to map the noisy text to one of the semantically or functionally similar groups. The features used by the classifier can be syntactic, semantic, and statistical.
1. A classification system comprising:
a non-transitory storage medium;
a classification engine comprising a set of executable instructions stored on the non-transitory storage medium and configured to:
receive an input associated with semantic information;
generate an action-based concept tagged document of the input;
classify the input into one or more semantic classes based at least in part on the action-based concept tagged document;
predict paraphrased representations of the input with respect to a source language based on the one or more semantic classes; and
generate a system response based on the paraphrased representations.
2. The classification system of claim 1 , wherein the input comprises at least one of a speech utterance, and a text string.
3. The classification system of claim 1 , wherein the classification engine is further configured to classify the input into one or more semantic classes based at least in part on at least one of feature sets and classification models.
4. The classification system of claim 3 , wherein the classification engine is further configured to extract and combine at least one of statistical features, syntactic features, and semantic features from a training corpus to generate the feature sets.
5. The classification system of claim 3 , wherein the classification engine is further configured to generate the classification models as a function of extracted features of a training corpus.
6. The classification system of claim 5 , wherein the classification engine is further configured to balance the training corpus before generating the feature sets.
7. The classification system of claim 1 , wherein the classification engine is further configured to extract feature sets from the input.
8. The classification system of claim 7 , wherein the classification engine is further configured to classify the input into one or more semantic classes based at least in part on the feature sets.
9. The classification system of claim 7 , wherein the classification engine is further configured to weigh features in the feature sets according to an information metric model from a training corpus.
10. The classification system of claim 1 , wherein the classification engine is further configured to classify the input into one or more semantic classes according to a hierarchy.
11. The classification system of claim 10 , wherein the hierarchy is based on clustering of the semantic classes.
12. The classification system of claim 10 , wherein the classification engine is further configured to classify the input into one or more semantic classes by performing a top-down classification of the one or more semantic classes according to the hierarchy.
13. The classification system of claim 1 , wherein the classification engine is further configured to classify the input into one or more semantic classes by refining classification of the one or more semantic classes via rule-base template matching.
14. The classification system of claim 1 , wherein the classification engine is further configured to classify the input into one or more semantic classes by rejecting out-of-domain inputs based on a rejection model.
15. The classification system of claim 1 , wherein the classification engine is further configured to translate the classified input based on the one or more semantic classes.
16. The classification system of claim 1 , further comprising a speech recognition engine configured to generate a word recognition lattice from a speech utterance.
17. The classification system of claim 16 , wherein the input received by the classification engine comprises the word recognition lattice.
18. The classification system of claim 1 , wherein the system response comprises a command with respect to a task.
19. The classification system of claim 12 , wherein the task relates to control behavior of an IVR system.
20. The classification system of claim 1 , wherein the system response comprises a system action.
21. The classification system of claim 1 , wherein the input comprises a noisy language.
22. The classification system of claim 1 , wherein the input comprises an input query.
23. The classification system of claim 22 , wherein the system response comprises a rejection of the input query.
24. The classification system of claim 22 , wherein the system response comprises a prompt to a user.
25. The classification system of claim 24 , wherein the prompt to the user comprises a request for the user to rephrase the input query.
26. The classification system of claim 1 , wherein the system response relates to travel.
27. The classification system of claim 1 , wherein the system response relates to a purchase.
28. The classification system of claim 1 , wherein the system response relates to security.
29. The classification system of claim 1 , wherein the system response relates to billing.