Systems and methods for multi-path language translation

Patent metadata

Field	Value
Publication number	US-9928236-B2
Application number	US-201514858470-A
Country	US
Kind code	B2
Filing date	Sep 18, 2015
Priority date	Sep 18, 2015
Publication date	Mar 27, 2018
Grant date	Mar 27, 2018

Abstract

Systems, apparatus, computer-readable media, and methods to provide translation of words or phrases from an initial language to a target language using multiple pathways are disclosed. The multiple pathways may have independent or near independent errors and the use of multiple pathways may reduce the errors that may be encountered in semantic vector based language translation. Cost values may be determined for translation to various potential words in the target language based at least in part on the multiple translation pathways between the initial language and the final language. The cost values may be used to select from among the various potential words in the target language.

First claim

The claimed invention is:

1. One or more non-transitory computer-readable media comprising computer-executable instructions that, when executed by one or more processors associated with a server, cause the one or more processors to:
identify, a document received from a device over a network;
identify a word from the document in an initial language to be translated to a target language;
determine a first target semantic vector of the word in a target language semantic vector space;
determine a second target semantic vector of the word in the target language semantic vector space based at least in part on a first translation matrix;
identify, based on the first target semantic vector, a first potential word in the target language and a second potential word in the target language;
identify a first potential semantic vector in the target language semantic vector space corresponding to the first potential word in the target language based on the second target semantic vector;
identify a second potential semantic vector in the target language semantic vector space corresponding to the second potential word in the target language based on the second target semantic vector;
determine a first cost value corresponding to the first potential semantic vector based at least in part on a cost function, the first target semantic vector, and the second target semantic vector;
determine a second cost value corresponding to the second potential semantic vector based at least in part on the cost function, the first target semantic vector, and the second target semantic vector;
determine that the first potential word is to be mapped to the word as the translation of the word in the target language based on the first cost value and the second cost value;
generate a second document including the translation of the word in the target language; and
cause the second document to be transmitted over the network to the device.

2. The one or more non-transitory computer-readable media of claim 1 , wherein the instructions further cause the one or more processors to:
determine an initial semantic vector of the word in an initial language semantic vector space; and
identify an initial language to target language translation matrix,
the instructions cause the one or more processors to determine the first target semantic vector of the word by multiplying the initial semantic vector of the word by the initial language to target language translation matrix.

3. The one or more non-transitory computer-readable media of claim 1 , wherein the first translation matrix is a translation matrix between an initial language semantic vector space and an intermediate language semantic vector space, and wherein the instructions cause the one or more processors to determine the second target semantic vector of the word by:
determining an initial semantic vector of the word in the initial language semantic vector space; and
multiplying the initial semantic vector by (1) the first translation matrix and (2) a second translation matrix, wherein the second translation matrix is between the intermediate language semantic vector space and the target language semantic vector space.

4. The one or more non-transitory computer-readable media of claim 1 , wherein the instructions cause the one or more processors to identify the first potential semantic vector by:
determining a distance between the first target semantic vector and the first potential semantic vector; and
determining that the distance is less than a threshold distance.

5. The one or more non-transitory computer-readable media of claim 4 , wherein the instructions cause the one or more processors to determine the distance between the first target semantic vector and the first potential semantic vector by determining at least one of: (i) a cosine distance between the first target semantic vector and the first potential semantic vector, or (ii) a Euclidean distance between the first target semantic vector and the first potential semantic vector.

6. The one or more non-transitory computer-readable media of claim 4 , wherein the distance is a first distance and the instructions cause the one or more processors to determine the first cost value corresponding to the first potential semantic vector by:
determining a second distance between the second target semantic vector and the first potential semantic vector; and
combining the first distance with the second distance according to the cost function.

7. The one or more non-transitory computer-readable media of claim 1 , wherein the instructions cause the one or more processors to:
identify a third potential semantic vector in the target language semantic vector space corresponding to a third potential word in the target language;
determine a third cost value corresponding to the third potential semantic vector based at least in part on the cost function, the first target semantic vector, and the second target semantic vector; and
determine that the third cost value is greater than the first cost value.

8. The one or more non-transitory computer-readable media of claim 1 , wherein the instructions cause the one or more processors to:
determine a third target semantic vector of the word in the target language semantic vector space based at least in part on an third translation matrix and a fourth translation matrix; and
determine the first cost value corresponding to the first potential semantic vector based at least in part on the third target semantic vector.

9. A server
to receive a document from a device over a network, the server comprising:
at least one memory that stores computer-executable instructions; and
at least one processor to access the at least one memory and execute the computer-executable instructions to:
determine a first target semantic vector of a word from the document in a target language semantic vector space via a direct path, the word to be translated from an initial language to a target language, the target language corresponding to the target language semantic vector space;
determine a second target semantic vector of the word in the target language semantic vector space via an indirect path;
identify, based on the first target semantic vector, a first potential word in the target language and a second potential word in the target language;
identify a first potential semantic vector in the target language semantic vector space corresponding to the first potential word in the target language;
identify a second potential semantic vector in the target language semantic vector space corresponding to the second potential word in the target language;
determine a first distance between the first target semantic vector and the first potential semantic vector;
determine a second distance between the second target semantic vector and the first potential semantic vector;
determine a third distance between the first target semantic vector and the second potential semantic vector;
determine a fourth distance between the second target semantic vector and the second potential semantic vector;
determine that the first potential word is to be mapped to the word as a translation of the word in the target language based at least in part on the first distance, the second distance, the third distance, and the fourth distance;
generate a second document including the translation of the word in the target language; and
cause the second document to be transmitted over the network to the device.

10. The server of claim 9 , wherein the at least one processor is further to execute the computer-executable instructions to:
determine a first cost value based at least in part on the first distance, the second dista

Assignees

• Mcafee Inc
• Mcafee Llc

Inventors

• Dixon Edward
• Dziduch Marcin
• Olinsky Craig

Classifications

• G06F40/47 Primary

  Machine-assisted translation, e.g. using translation memory · CPC title

• G06F40/44 Primary

  Statistical methods, e.g. probability models · CPC title

• G06F17/2836 Primary

  Physics · mapped topic

• G06F17/2818

  Physics · mapped topic