Translating texts between languages

• US 9,053,090 B2
• Filed: 06/20/2012
• Issued: 06/09/2015
• Est. Priority Date: 10/10/2006
• Status: Active Grant

• Current Assignee: ABBYY Development LLC
• Original Assignee: ABBYY InfoPoisk LLC
• Inventors: Anisimovich, Konstantin, Selegey, Vladimir, Zuev, Konstantin
• Primary Examiner(s): PULLIAS, JESSE SCOTT
• Application Number:
  US13/528,716
• Publication Number:
  US 20120259621A1
• Time in Patent Office: 1,084 Days
• Field of Search: 704 2- 8
• US Class Current: 1/1
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• 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 40/55 Rule-based translation

Abstract

Methods and computer systems for translating sentences between languages from an intermediate language-independent semantic representation are provided. Based on a comprehensive understanding about languages and semantics, exhaustive linguistic descriptions are used to analyze sentences, build syntactic structures and language independent semantic structures and representations, and synthesize one or more sentences in a natural or artificial language. A computer system is also provided to analyze and synthesize various linguistic structures and perform translation of a wide spectrum of various sentence types. As result, a generalized data structure, such as a semantic structure, is generated from a sentence of an input language and can be transformed into a natural sentence expressing its meaning correctly in an output language. The methods and systems can be applied to automated abstracting, machine translation, natural language processing, control systems, Internet information retrieval, etc.

Claims

1. A computer-implemented method of translating a source text in a source language into an output language, comprising:

obtaining, using a processor, the source text in the source language from an input detected in a device;

dividing, using a processor, the source text into sentences;

translating, using a processor, each sentence of the source text, said translating comprising;

executing a first syntactic analysis, said first syntactic analysis including

generating a set of possible relationships between words of the respective sentence and building a graph of generalized constituents;

executing a second syntactic analysis of each sentence, said second syntactic analysis comprising;

putting forward at least one hypothesis related to a syntactic structure of the sentence, wherein the putting forward the at least one hypothesis comprises

performing a lexical selection to select a first lexical meaning from a first lexical class for a node of a syntactic tree of the sentence from a set of possible lexical classes; and

verifying the hypothesis about the syntactic structure of the sentence;

generating a language-independent semantic structure of the sentence based at least in part on the verified hypothesis;

generating a syntactic structure of an output sentence in the output language based at least in part on the generated language-independent semantic structure; and

performing a morphological synthesis of the output sentence in the output language based at least in part on the generated syntactic structure of the output sentence;

generating, using a processor, a text in the output language based on said translating of each of the sentences of the source text;

detecting, using a processor, a translation error associated with incorrect lexical selection comprising

comparing the first lexical class of the selected first lexical meaning to a topic of the source text; and

repeating, using a processor, translation of the identified sentences with a second lexical meaning from a second lexical class, wherein the second lexical class is different from the first lexical class.

2. The method of claim 1, wherein verifying the hypothesis about the syntactic structure of the sentence further comprises performing a lexical disambiguation.

3. The method of claim 1, wherein performing a lexical selection comprises selecting a lexical model corresponding to the selected first lexical meaning.

4. The method of claim 3, wherein the lexical selection is performed on the basis of information about at least one of a style, genre, topic or field of the translated text in the output language.

5. The method of claim 1, wherein information about the lexical selection for each node of the syntactic tree is saved for possible use during other steps of translating the text from the source language into another language.

6. The method of claim 1, wherein performing the lexical selection is performed on a basis of rating for the first lexical meaning.

7. The method of claim 6, wherein the ratings are related to “

a priori”

ratings, statistical ratings, or “

a priori” and

statistical ratings.

8. The method of claim 6, wherein said rating for the first lexical meaning comprises ratings for the first lexical meaning according to a pragmatic class.

9. The method of claim 6, wherein an overall rating of the first lexical meaning is calculated based on a weighed sum of rating scores of the first lexical meaning regarding pragmatic classes, wherein said rating scores of the first lexical meaning are multiplied by current weights of respective pragmatic classes calculated for a corresponding portion of the translated text.

10. The method of claim 9, wherein the overall rating of the first lexical meaning is stored in a memory.

11. The method of claim 6, wherein information about the lexical selection and pragmatic characteristics of the first lexical meaning selected for a node of the syntactic tree of the sentence and its textual environment is processed and stored, wherein said information defines a pragmatic context.

12. The method of claim 1, wherein verifying the hypothesis about the syntactic structure of the sentence further comprises establishing possible non-tree links.

13. The method of claim 12, wherein the verifying the hypothesis about the syntactic structure of the sentence further comprises making another lexical selection when the establishing non-tree links is unsuccessful.

14. The method of claim 13, wherein the making another lexical selection further comprises:

storing a list of lexical variants for each word;

calculating new weights for pragmatic classes regarding respective translated text or text fragments;

detecting words for which lexical selection can be changed;

calculating bonuses to lexical classes for said words; and

repeating the second syntactic analysis for each sentence.

15. A computer-implemented method for disambiguation during translating a source text from a source language into an output language, comprising:

obtaining, using a processor, the source text in the source language from an input detected in a device;

dividing, using a processor, the source text into sentences;

translating, using a processor, each sentence of the source text, wherein translating comprises;

generating a set of possible relationships between words in respective sentences;

building a graph of generalized constituents for respective sentences;

putting forward at least one hypothesis related to a syntactic structure of each sentence; and

verifying the hypothesis about the syntactic structure of the sentence;

generating a language-independent semantic structure of the source text;

generating a syntactic structure of an output text in the output language, comprising;

making a lexical selection to select a first lexical meaning from a first lexical class for a node of the semantic structure from a set of possible lexical classes; and

performing a morphological synthesis of an output sentence in the output language;

obtaining, using a processor, a text in the output language based on the output sentences obtained as result of said translating;

identifying, using a processor, the output sentences that include one or more errors by comparing the first lexical class of the selected first lexical meaning to a topic of the source text;

repeating, using a processor, translation of the identified sentences with a second lexical meaning from a second lexical class, wherein the second lexical class is different from the first lexical class.

16. The computer-implemented method of claim 15, wherein the input text includes text detected in a device or on a Web site.

17. The method of claim 15, wherein the identifying of the output sentences that include one or more errors is at least in part based on analysis of information about the lexical selection wherein the information comprises:

information about words with a same lemma in a surrounding fragment or a lexical and semantic class selected for either the selected lexical class or the same lemma.

18. The method of claim 15, wherein repeating translation of identified sentences includes calculating new weights of pragmatic classes associated with the text in the output language or a fragment of the text in the output language.

19. One or more physical non-transitory computer accessible media encoded with instructions for performing a method, the method comprising:

obtaining, using a processor, a source text in a source language from an input detected in a device;

dividing, using a processor, the source text into sentences;

translating, using a processor, each sentence of the source text, said translating comprising;

executing a first syntactic analysis, said first syntactic analysis including generating a set of possible relationships between words of the respective sentence and building a graph of generalized constituents;

executing a second syntactic analysis of each sentence, said second syntactic analysis comprising;

putting forward at least one hypothesis related to a syntactic structure of the sentence, wherein the putting forward the at least one hypothesis comprises

performing a lexical selection to select a first lexical meaning from a first lexical class for a node of a syntactic tree of the sentence from a set of possible lexical classes; and

verifying the hypothesis about the syntactic structure of the sentence;

generating a language-independent semantic structure of the sentence based at least in part on the verified hypothesis;

generating a syntactic structure of an output sentence in an output language based at least in part on the generated language-independent semantic structure; and

performing a morphological synthesis of the output sentence in the output language based at least in part on the generated syntactic structure of the output sentence;

generating, using a processor, a text in the output language based on said translating of each of the sentences of the source text;

detecting, using a processor, a translation error associated with incorrect lexical selection comprising

comparing the first lexical class of the selected first lexical meaning to a topic of the source text; and

repeating, using a processor, translation of the identified sentences with a second lexical meaning from a second lexical class, wherein the second lexical class is different from the first lexical class.

20. A computer-implemented method of translating a source text in a source language into an output language, comprising:

obtaining, using a processor, the source text in the source language from an input detected in a device;

dividing, using a processor, the source text into sentences;

translating, using a processor, each sentence of the source text, said translating comprising;

executing a first syntactic analysis, said first syntactic analysis including generating a set of possible relationships between words of the respective sentence and building a graph of generalized constituents;

executing a second syntactic analysis of each sentence, said second syntactic analysis comprising;

putting forward at least one hypothesis related to a syntactic structure of the sentence, wherein the putting forward the at least one hypothesis comprises

performing a lexical selection to select a first lexical meaning from a first lexical class for a node of a syntactic tree of the sentence from a set of possible lexical classes; and

verifying the hypothesis about the syntactic structure of the sentence;

generating a language-independent semantic structure of the sentence based at least in part on the verified hypothesis;

generating a syntactic structure of an output sentence in the output language based at least in part on the generated language-independent semantic structure; and

performing a morphological synthesis of the output sentence in the output language based at least in part on the generated syntactic structure of the output sentence;

generating, using a processor, a text in the output language based on said translating of each of the sentences of the source text;

detecting, using a processor, a translation error associated with incorrect lexical selection comprising

comparing the first lexical class of the selected first lexical meaning to a lexical class of another word with the same lemma as a word corresponding to the node; and

repeating, using a processor, translation of the identified sentences with a second lexical meaning from a second lexical class, wherein the second lexical class is different from the first lexical class.