Deep model statistics method for machine translation
View Patent ↗In one embodiment, the invention provides a method for machine translation of a source document in an input language to a target document in an output language, comprising generating translation options corresponding to at least portions of each sentence in the input language; and selecting a translation option for the sentence based on statistics associated with the translation options.
1. A computer-implemented method of representing combinability of elements of linguistic descriptions, at least some of the linguistic descriptions are organized as hierarchies, wherein two of the elements are members of respective hierarchies, comprising:
performing by a processor a semantic analysis of electronically-stored texts to determine combinability statistics between two or more elements of linguistic descriptions present in the texts; and
storing one or more combinability statistics between the two elements of linguistic descriptions of the hierarchies as a sum of logarithms of combinabilities of one or more elements from the first hierarchy with one or more elements from the second hierarchy.
2. The computer-implemented method of claim 1 wherein the combinability statistics determined for higher levels of the hierarchies are used as combinability statistics for lower levels.
3. The computer-implemented method of claim 1 , wherein said performing semantic analysis comprises:
performing semantic analysis on a text in a source language; and
performing semantic analysis on a text in a target language, which is a translation of the text in the source language.
4. The computer-implemented method of claim 3 , wherein the method further comprises:
matching one or more deep structures of sentences in the text in the source language to one or more deep structures of sentences in the text in the target language;
determining a correspondence between two or more deep structure elements for sentences with similar deep structures; and
identifying word combinations in the source language and in the target language that correspond to each other through corresponding deep structure elements as likely translations.
5. The method of claim 4 further comprising using the likely translations to generate one or more translation rules.
6. The method of claim 4 further comprising:
identifying instances where translations of texts deviate from translations performed by a machine translation system so as to detect errors in language models and translation rules of the machine translation system.
7. The computer-implemented method of claim 1 , wherein the elements from the second hierarchy have a same ancestor element.
8. A computer system comprising:
a processor; and
a storage medium coupled to the processor, the storage medium storing instructions which when executed by the processor cause the computer system to perform a method for representing combinability of elements of linguistic descriptions, where at least some of the linguistic descriptions are organized as hierarchies, wherein two of the elements are members of respective hierarchies, the instructions comprising:
performing semantic analysis of electronically-stored texts to determine combinability statistics between two or more elements of linguistic descriptions present in the texts; and
storing one or more combinability statistics between the two elements of linguistic descriptions of the hierarchies as a sum of logarithms of combinabilities of one or more elements from the first hierarchy with one or more elements from the second hierarchy.
9. The computer system of claim 8 wherein the combinability statistics determined for higher levels of the hierarchies are used as combinability statistics for lower levels.
10. The computer system of claim 8 , wherein said performing semantic analysis comprises further comprises:
performing semantic analysis on a text in a source language; and
performing semantic analysis on a text in a target language, which is a translation of the text in the source language.
11. The computer system of claim 10 , wherein the instructions further comprise:
matching one or more deep structures of sentences in the text in the source language to one or more deep structures of sentences in the text in the target language;
determining a correspondence between two or more deep structure elements for sentences with similar deep structures; and
identifying word combinations in the source and in the target languages that correspond to each other through corresponding deep structure elements as likely translations.
12. The computer system of claim 11 , wherein the instructions further comprise:
using the likely translations to generate one or more translation rules.
13. The computer system of claim 11 , wherein the instructions further comprise:
identifying instances where translations of texts deviate from translations performed by a machine translation system so as to detect errors in language models and translation rules of the machine translation system.
14. The computer system of claim 8 , wherein the elements from the second hierarchy have a same ancestor element.
15. A physical, non-transitory computer storage medium having stored thereon a sequence of instructions which when executed by a computer system cause said computer system to perform a method for representing combinability of elements of linguistic descriptions, at least some of the linguistic descriptions are organized as hierarchies, wherein two of the elements are members of respective hierarchies, the instructions comprising:
performing semantic analysis of electronically-stored texts to determine combinability statistics between two or more elements of linguistic descriptions present in the texts; and
storing one or more combinability statistics between the two elements of linguistic descriptions of the hierarchies as a sum of logarithms of combinabilities of one or more elements from the first hierarchy with one or more elements from the second hierarchy, wherein the elements from the second hierarchy have the same ancestor element.
16. The physical, non-transitory computer storage medium of claim 15 , wherein the combinability statistics determined for higher levels of the hierarchies are used as combinability statistics for lower levels.
17. The physical, non-transitory computer storage medium of claim 15 , wherein performing semantic analysis comprises:
performing semantic analysis on a text in a source language; and
performing semantic analysis on a text in a target language, which is a translation of the text in the source language.
18. The physical, non-transitory computer storage medium of claim 17 , wherein the instructions further comprise:
matching one or more deep structures of sentences in the text in the source language to one or more deep structures of sentences in the text in the target language;
determining a correspondence between two or more deep structure elements for sentences with similar deep structures; and
identifying word combinations in the source and in the target languages that correspond to each other through corresponding deep structure elements as likely translations.
19. The physical, non-transitory computer storage medium of claim 18 , wherein the instructions further comprise:
using the likely translations to generate one or more additional translation rules.
20. The physical, non-transitory computer storage medium of claim 18 , wherein the instructions further comprise:
identifying instances where translations of texts deviate from translations performed by a machine translation system so as to detect errors in language models and translation rules of the machine translation system.