Method and apparatus for training a translation disambiguation classifier
First Claim
Patent Images
1. A method of training a classifier, the method comprising:
- applying a first classifier to a first set of unlabeled data to form a first set of labeled data, the first classifier capable of assigning data to classes in a first set of classes;
applying a second classifier to a second set of unlabeled data to form a second set of labeled data, the second classifier capable of assigning data to classes in a second set of classes that is different from the first set of classes; and
using the first set of labeled data and the second set of labeled data to retrain the first classifier to form a retrained classifier that can be used to assign data to classes, retraining comprising;
determining a first probability component from the first set of labeled data;
determining a second probability component from the second set of labeled data; and
combining the first probability component and the second probability component to form a probability term used to define the first classifier.
2 Assignments
0 Petitions
Accused Products
Abstract
A method of training a classifier includes applying a first classifier to a first set of unlabeled data to form a first set of labeled data. The first classifier is able to assign data to classes in a first set of classes. A second classifier is applied to a second set of unlabeled data to from a second set of labeled data. The second classifier is able to assign data to classes in a second set of classes that is different from the first set of classes. The first and second sets of labeled data are used to retrain the first classifier.
-
Citations
29 Claims
-
1. A method of training a classifier, the method comprising:
-
applying a first classifier to a first set of unlabeled data to form a first set of labeled data, the first classifier capable of assigning data to classes in a first set of classes; applying a second classifier to a second set of unlabeled data to form a second set of labeled data, the second classifier capable of assigning data to classes in a second set of classes that is different from the first set of classes; and using the first set of labeled data and the second set of labeled data to retrain the first classifier to form a retrained classifier that can be used to assign data to classes, retraining comprising; determining a first probability component from the first set of labeled data; determining a second probability component from the second set of labeled data; and combining the first probability component and the second probability component to form a probability term used to define the first classifier. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13)
-
-
14. A computer-readable storage medium storing computer-executable instructions for performing steps comprising:
-
generating first language labeled data that indicates a sense of at least one word in a first language; generating second language labeled data that indicates a sense of at least one word in a second language; and using the first language labeled data and the second language labeled data to train a classifier for the first language, where the classifier can be used to identify a sense of a word in the first language, wherein using the first language labeled data and the second language labeled data to train a classifier comprises; determining a first probability component comprising a probability using the first language labeled data; determining a second probability component comprising a probability using the second language labeled data; and combining the first probability component and the second probability component to form a probability term for the classifier. - View Dependent Claims (15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29)
-
Specification