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 method of identifying word combinations in a source language and corresponding likely translations in a target language, performed by one or more processors, the method comprising:
acquiring, by one or more processors, a text in the source language and a text in the target language, wherein the text in the source language is a translation of the text in the target language;
performing, by one or more processors, semantic analysis on the acquired text in the source language to build deep semantic structures of one or more sentences of the acquired text in the source language, where the deep semantic structures comprise language-independent semantic classes and deep slots;
performing, by one or more processors, semantic analysis on the acquired text in the target language to build deep semantic structures of one or more sentences of the acquired text in the target language, where the deep semantic structures comprise language-independent semantic classes and deep slots;
matching, by one or more processors, the deep semantic structures of the sentences in the text in the source language to the deep semantic structures of the sentences in the text in the target language;
determining, by one or more processors, a correspondence between deep structure elements for sentences with essentially matching deep structures; and
identifying, by one or more processors, word combinations in the source and the target languages that substantially often match into each other through matching the deep structure of the source sentence to the deep structure of the target sentence as likely translations.
2. The method of claim 1 further comprising using the likely translations to generate additional translation rules.
3. The method of claim 1 further comprising identifying instances where the likely translations based on analysis of translated 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.
4. The method of claim 1 , further comprising
generating a machine translation of the source sentence into the target language based on the deep semantic structure of the source sentence;
comparing the identified word combinations in the machine translation with the identified word combinations in the target sentence to identify the differing word combinations in the machine translation and the target sentence;
generating additional translation rules for the word combinations in the source language corresponding to the differing word combination in the machine translation.
5. 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 steps for identifying word combinations in a source language and corresponding likely translations in a target language, the instructions comprising:
acquiring, by one or more processors, a text in the source language and a text in the target language, wherein the text in the source language is a translation of the text in the target language;
performing, by one or more processors, semantic analysis on the acquired text in the source language to build deep semantic structures of one or more sentences of the acquired text in the source language, where the deep semantic structures comprise language-independent semantic classes and deep slots;
performing, by one or more processors, semantic analysis on the acquired text in the target language to build deep semantic structures of one or more sentences of the acquired text in the target language, where the deep semantic structures comprise language-independent semantic classes and deep slots;
matching, by one or more processors, the deep semantic structures of the sentences in the text in the source language to the deep semantic structures of the sentences in the text in the target language;
determining, by one or more processors, a correspondence between deep structure elements for sentences with essentially matching deep structures; and
identifying, by one or more processors, word combinations in the source and the target languages that substantially often match into each other through corresponding deep structure elements as likely translations.
6. The computer system of claim 5 further comprising:
using the likely translations to generate additional translation rules.
7. The computer system of claim 5 further comprising:
identifying instances where the likely translations based on analysis of translated 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.
8. The system of claim 5 , further comprising
generating a machine translation of the source sentence into the target language based on the deep semantic structure of the source sentence;
comparing the identified word combinations in the machine translation with the identified word combinations in the target sentence to identify the differing word combinations in the machine translation and the target sentence;
generating additional translation rules for the word combinations in the source language corresponding to the differing word combination in the machine translation.
9. 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 identifying word combinations in a source language and corresponding likely translations in a target language, the instructions comprising:
acquiring, by one or more processors, a text in the source language and a text in the target language, wherein the text in the source language is a translation of the text in the target language;
performing, by one or more processors, semantic analysis on the acquired text in the source language to build deep semantic structures of one or more sentences of the acquired text in the source language, where the deep semantic structures comprise language-independent semantic classes and deep slots;
performing, by one or more processors, semantic analysis on the acquired text in the target language to build deep semantic structures of one or more sentences of the acquired text in the target language, where the deep semantic structures comprise language-independent semantic classes and deep slots;
matching, by one or more processors, the deep semantic structures of the sentences in the text in the source language to the deep semantic structures of the sentences in the text in the target language;
determining, by one or more processors, a correspondence between deep structure elements for sentences with essentially matching deep structures; and
identifying, by one or more processors, word combinations in the source and the target languages that substantially often match into each other through corresponding deep structure elements as likely translations.
10. The physical, non-transitory computer storage medium of claim 9 further comprising using the likely translations to generate additional translation rules.
11. The physical, non-transitory computer storage medium of claim 9 further comprising identifying instances where the likely translations based on analysis of translated 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.
12. The storage medium of claim 9 , further comprising
generating a machine translation of the source sentence into the target language based on the deep semantic structure of the source sentence;
comparing the identified word combinations in the machine translation with the identified word combinations in the target sentence to identify the differing word combinations in the machine translation and the target sentence;
generating additional translation rules for the word combinations in the source language corresponding to the differing word combination in the machine translation.