Context-based translation retrieval via multilingual space

1. A computing system comprising:

a storage configured to store a data file; and

a processor configured to

identify a source word from the data file,

determine a plurality of translation candidates for the source word which translate the source word from a source language into a different language,

generate a plurality of vectors for the plurality of translation candidates, respectively, each vector comprising a combination of values including a value for the source word, a value for context associated with the source word, a value for a different respective translation candidate from among the plurality of translation candidates, and a cosine similarity between the source word and the different respective translation candidate,

input the plurality of vectors into a neural network which determines for the plurality of translation candidates based on the plurality of input vectors a relevancy of each respective translation candidate to the source word,

remove less relevant translation candidates from the plurality of translation candidates based on the determined relevancy of each respective translation candidate, and

choose a target translation candidate that has a greatest cosine similarity from among remaining translation candidates which have not been removed,

wherein the processor is further configured to output the translated target translation candidate for display via a display device.

2. The computing system of claim 1 , wherein the storage comprises a multilingual text repository which stores prior translations of the source word in a plurality of different languages.

3. The computing system of claim 1 , wherein the processor is configured to predict the target translation candidate for the source word based on semantic clusters built from other entries in the data file.

4. The computing system of claim 3 , wherein a semantic cluster associated with a respective translation candidate comprises a cluster of words that are commonly used with the source word.

5. The computing system of claim 1 , wherein the plurality of translation candidates comprise a common spelling of the source word but with different contextual meaning.

6. The computing system of claim 1 , wherein each translation candidate corresponds to at least one of a different meaning of the source word in the source language and a different notation of the source word, and includes translations of the source word into a plurality of other languages.

7. The computing system of claim 6 , wherein the processor is further configured to map the plurality of translation candidates into a plurality of vectors in multilingual space.

8. The computing system of claim 7 , wherein the processor is configured to map the source word into a vector in the multilingual space.

9. The computing system of claim 1 , wherein the processor is configured to simultaneously output a plurality of translations of the target translation candidate in a plurality of different languages, respectively.

10. The computing system of claim 1 , wherein the value for the different respective translation candidate comprises a centroid value computed from an average of multiple vectors representing the translation candidate in multiple different languages.

11. A method comprising:

identifying a source word from a data file;

determining a plurality of translation candidates for the source word which translate the source word from a source language into a different language;

generating a plurality of vectors for the plurality of translation candidates, respectively, each vector comprising a combination of values including a value for the source word, a value for context associated with the source word, a value for a different respective translation candidate from among the plurality of translation candidates, and a cosine similarity between the source word and the different respective translation candidate;

inputting the plurality of vectors into a neural network which determines for the plurality of translation candidates based on the plurality of input vectors a relevancy of each respective translation candidate to the source word;

removing less relevant translation candidates from the plurality of translation candidates based on the determined relevancy of each respective translation candidate;

choosing a target translation candidate that has a greatest cosine similarity from among remaining translation candidates which have not been removed; and

outputting the target translation candidate for display via a display device.

12. The method of claim 11 , wherein the determining comprises predicting the target translation candidate for the source word based on based on semantic clusters built from other entries in the data file.

13. The method of claim 12 , wherein a semantic cluster associated with a respective target translation candidate comprises a cluster of words that are commonly used with the source word.

14. The method of claim 11 , wherein the plurality of translation candidates comprise a common spelling of the source word but with different contextual meaning.

15. The method of claim 11 , wherein each translation candidate corresponds to at least one of a different meaning of the source word in the source language and a different notation of the source word, and includes translations of the source word into a plurality of other languages.

16. The method of claim 15 , further comprising mapping the plurality of translation candidates into a plurality of vectors in multilingual space.

17. The method of claim 16 , wherein the identifying of the target translation candidate further comprises mapping the source word into a vector in the multilingual space.

18. The method of claim 11 , wherein the outputting comprises simultaneously outputting a plurality of translations of the target translation candidate in a plurality of different languages, respectively.

19. A non-transitory computer-readable medium storing instructions which when executed by a processor cause a computer to perform a method comprising:

identifying a source word from a data file;

determining a plurality of translation candidates for the source word which translate the source word from a source language into a different language;

generating a plurality of vectors for the plurality of translation candidates, respectively, each vector comprising a combination of values including a value for the source word, a value for context associated with the source word, a value for a different respective translation candidate from among the plurality of translation candidates, and a cosine similarity between the source word and the different respective translation candidate;

inputting the plurality of vectors into a neural network which determines for the plurality of translation candidates based on the input plurality of vectors a relevancy of each respective translation candidate to the source word;

removing less relevant translation candidates from the plurality of translation candidates based on the determined relevancy of each respective translation candidate;

choosing a target translation candidate that has a greatest cosine similarity from among remaining translation candidates which have not been removed; and

outputting the translated target translation candidate for display via a display device.