Trained data input system
First Claim
Patent Images
1. A data input system comprising:
- a processor;
a language model implemented using the processor, which computes candidate next items in an input sequence of one or more items;
a training engine implemented using the processor which performs training of the language model using training data comprising a plurality of true words and at least one alternative candidate word for each of the plurality of true words,wherein the plurality of true words comprises respective words intended to be input with a virtual keyboard,wherein the at least one alternative candidate word for each true word comprises at least one word received from imperfect entry during attempted input of the true word with the virtual keyboard, andwherein, as a result of the training, the language model is trained to reward discriminating between the plurality of true words and the at least one alternative candidate word for each of the plurality of true words.
2 Assignments
0 Petitions
Accused Products
Abstract
A data input system has a processor which receives user input comprising a sequence of one or more items and a language model which computes candidate next items in the sequence using the user input. A training engine trains the language model using data about a plurality of true words which a user intended to input using the data input system, and for each true word, at least one alternative candidate, being a word computed assuming imperfect entry of the true word to the data input system.
18 Citations
20 Claims
-
1. A data input system comprising:
-
a processor; a language model implemented using the processor, which computes candidate next items in an input sequence of one or more items; a training engine implemented using the processor which performs training of the language model using training data comprising a plurality of true words and at least one alternative candidate word for each of the plurality of true words, wherein the plurality of true words comprises respective words intended to be input with a virtual keyboard, wherein the at least one alternative candidate word for each true word comprises at least one word received from imperfect entry during attempted input of the true word with the virtual keyboard, and wherein, as a result of the training, the language model is trained to reward discriminating between the plurality of true words and the at least one alternative candidate word for each of the plurality of true words. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
16. A computer-implemented method performed by an electronic device, the method comprising:
-
receiving, at a processor, user input comprising a sequence of one or more items; computing, using a language model, candidate next items in the sequence of one or more items; training the language model using training data comprising a plurality of true words and at least one alternative candidate word for each of the plurality of true words, wherein the plurality of true words comprises respective words intended to be input with a virtual keyboard, wherein the at least one alternative candidate word for each true word comprises at least one word received from imperfect entry during attempted input of the true word with the virtual keyboard, and wherein, as a result of the training, the language model is trained to reward discriminating between the plurality of true words and the at least one alternative candidate word for each of the plurality of true words. - View Dependent Claims (17, 18, 19)
-
-
20. An electronic device comprising:
-
a processor which receives user input from a virtual keyboard, the user input comprising a sequence of one or more items; a language model, implemented using the processor, which computes candidate next items in the sequence using the user input; the language model having been trained using training data comprising a plurality of true words and at least one alternative candidate word for each of the plurality of true words, wherein the plurality of true words comprises respective words intended to be input with the virtual keyboard, wherein the at least one alternative candidate word for each true word comprises at least one word received from imperfect entry during attempted input of the true word with the virtual keyboard, and wherein the language model is trained to reward discriminating between the plurality of true words and the at least one alternative candidate word for each of the plurality of true words.
-
Specification