Deep model statistics method for machine translation

• US 8,805,676 B2
• Filed: 09/25/2012
• Issued: 08/12/2014
• Est. Priority Date: 10/10/2006
• Status: Active Grant

• Current Assignee: ABBYY Development LLC
• Original Assignee: ABBYY InfoPoisk LLC
• Inventors: Anisimovich, Konstantin, Tuganbaev, Diar, Selegey, Vladimir, Zuev, Konstantin
• Primary Examiner(s): Vo, Huyen X.
• Application Number:
  US13/626,480
• Publication Number:
  US 20130024186A1
• Time in Patent Office: 686 Days
• Field of Search: 704 1- 10, 704/251, 704/255, 704/257, 704/277, 704/270
• US Class Current: 704/9
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/55 Rule-based translation
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation

Abstract

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.

Claims

1. A computer method of representing in a computer memory a natural language sentence in a language-independent format, the method comprising:

receiving one or more ratings generated based on analysis of sentences from a corpus of texts comprising parallel texts, wherein the parallel texts comprising texts in two or more languages in which each fragment in one language having a corresponding fragment in another language;

performing by a computer processor a syntactic analysis of a natural language sentence including selecting a syntactic structure for the natural language sentence based on the one or more ratings;

performing a semantic analysis on the natural language sentence;

storing in the computer memory the selected syntactic structure in a language-independent semantic structure; and

selecting a semantic structure based on the one or more ratings.

2. The method of claim 1 wherein the ratings for the selecting of the syntactic structure or semantic structure includes an a priori assessment.

3. The method of claim 1 wherein the ratings for the selecting of the syntactic structure or semantic structure are based on statistics reflecting the frequency of constructions and combinability of the syntactic structure elements or semantic structure elements.

4. The method of claim 1 wherein the ratings for the selecting of the syntactic structure or semantic structure are based on statistics reflecting the frequency of constructions and combinability of the syntactic structure elements and semantic structure elements.

5. The method of claim 1 wherein the ratings for the selecting of the syntactic structure or semantic structure are based on statistics concerning one or more of the following:

semantic and lexical classes, lexemes, deep and surface slots, syntforms, communicative forms, diathesis relations, grammatical and semantic values.

6. The method of claim 1, wherein the ratings for the selecting of the syntactic structure or semantic structure comprise combinatorial ratings for the surface elements and deep elements.

7. The method of claim 6, wherein the combinatorial ratings comprise a probability or combinability of objects of certain semantic classes being combined with objects of the same or another semantic class.

8. The method of claim 6, wherein the combinatorial ratings comprise a probability of objects of certain semantic classes being combined with certain deep slots.

9. The method of claim 1, further comprising generating statistics.

10. The method of claim 9, wherein generating statistics comprises analyzing the corpus of texts to calculate probability and combinability of occurrences of syntactic structure elements or semantic structure elements.

11. The method of claim 10, wherein the corpus of texts includes tagged texts and domain-specific texts.

12. The method of claim 9, wherein generating statistics comprises analyzing the corpus of texts to calculate probability and combinability of occurrences of syntactic structure elements and semantic structure elements.

13. The method of claim 9, wherein generating statistics includes analyzing natural-language sentences of parallel texts from the corpus of texts and comparing syntactic and semantic structures.

14. The method of claim 9, wherein generating statistics based on the corpus of texts comprises matching a deep structure of a sentence from the corpus of text in a source language to a deep structure of the sentence in a target language and establishing a correspondence between elements of the deep structures.

15. The method of claim 14 further comprising matching a surface structure of the sentence in the source language to a surface structure of the sentence in the target language derived from previously established correspondence between the deep structure elements.

16. The method of claim 14, wherein the statistics are generated from structure elements for which a correspondence has been found in the deep structures of the source and target sentences.

17. The method of claim 14, wherein the statistics are generated on possible structures corresponding to the sentence in the source language and possible structures corresponding to the sentence in the target language.

18. The method of claim 1 further comprising using database of terms and corresponding rules for detection of systematic errors.

19. A computer system for representing a natural language sentence in a language-independent format, the computer system comprising:

a memory;

a processor; and

a storage medium electrically coupled to the processor and the memory, the storage medium storing instructions which when executed by the processor cause the computer system to;

receiving one or more ratings generated based on analysis of sentences from a corpus of texts comprising parallel texts, wherein the parallel texts comprising texts in two or more languages in which each fragment in one language having a corresponding fragment in another language;

perform a syntactic analysis of a natural language sentence;

select a syntactic structure based on the one or more ratings; and

perform a semantic analysis so as to transition the selected syntactic structure into a language-independent semantic structure, including selecting a semantic structure based on the one or more ratings.

20. The computer system of claim 19 wherein the ratings for selecting a syntactic structure or semantic structure includes an a priori assessment.

21. The computer system of claim 19 wherein the ratings for selecting a syntactic structure or semantic structure are based on statistics concerning one or more of the following:

semantic and lexical classes, lexemes, deep and surface slots, syntforms, communicative forms, diathesis relations, grammatical and semantic values.

22. The computer system of claim 19 wherein the ratings for selecting a syntactic structure or semantic structure include combinatorial ratings for the surface elements and deep elements.

23. The computer system of claim 22, wherein the combinatorial ratings comprise a probability or combinability of objects of certain semantic classes being combined with objects of the same or another semantic class.

24. The computer system of claim 19, wherein the instructions further cause the computer system to generate statistics.

25. The computer system of claim 24, wherein generating statistics includes analyzing natural-language sentences of the parallel texts and comparing syntactic and semantic structures.

26. The computer system of claim 24, wherein generating statistics includes matching a deep structure of a sentence in a source language to a deep structure of the sentence in a target language and establishing a correspondence between elements of the deep structures.

27. A 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 in a computer memory a natural language sentence in a language-independent format, the instructions comprising:

receiving one or more ratings generated based on analysis of sentences from a corpus of texts comprising parallel texts, wherein the parallel texts comprising texts in two or more languages in which each fragment in one language having a corresponding fragment in another language;

performing by the computer system a syntactic analysis of a natural language sentence including selecting a syntactic structure for the natural language sentence based on the one or more ratings;

performing a semantic analysis on the natural language sentence;

storing in a computer memory the selected syntactic structure in a language-independent semantic structure; and

selecting a semantic structure based on the one or more ratings.

28. The non-transitory computer storage medium of claim 27 wherein the ratings for selecting the syntactic structure or semantic structure are based on statistics concerning one or more of the following:

semantic and lexical classes, lexemes, deep and surface slots, syntforms, communicative forms, diathesis relations, grammatical and semantic values.

29. The non-transitory computer storage medium of claim 28, wherein the combinatorial ratings comprise a probability or combinability of objects of certain semantic classes being combined with objects of the same or another semantic class.

30. The non-transitory computer storage medium of claim 27, wherein the ratings for selecting the syntactic structure or semantic structure include combinatorial ratings for the surface elements and deep elements.

31. The non-transitory computer storage medium of claim 27, further comprising generating statistics.

32. The non-transitory computer storage medium of claim 31, wherein generating statistics includes analyzing natural-language sentences of the parallel texts and comparing syntactic and semantic structures.

33. The non-transitory computer storage medium of claim 31, wherein generating statistics includes matching a deep structure of a sentence in a source language to a deep structure of the sentence in a target language and establishing a correspondence between elements of the deep structures.