MACHINE TRANSLATION DEVICE AND MACHINE TRANSLATION METHOD IN WHICH A SYNTAX CONVERSION MODEL AND A WORD TRANSLATION MODEL ARE COMBINED

• US 20130226556A1
• Filed: 07/20/2011
• Published: 08/29/2013
• Est. Priority Date: 11/05/2010
• Status: Active Grant

• Current Assignee: Eleven Street Co., Ltd. (SK Telecom Co., Ltd.)
• Original Assignee: SK Planet Co., Ltd. (SK Telecom Co., Ltd.)
• Inventors: Hwang, Young Sook, Kim, Sang-Bum, Yin, Chang Hao, Lee, Seong Wook, Lee, Yeon Su, Rim, Hae Chang
• Granted Patent: US 10,198,437 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• 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

The present invention relates to statistical machine translation, and provides a machine translation device and a machine translation method that acquire a creation probability for a target language from a single corpus while extracting respective conversion probabilities by extracting syntax conversion knowledge and word translation knowledge from a parallel corpus, model a weighted translation model by allowing each of the conversion knowledge and each of the probabilities to learn using a translation model learning device, and generate a target sentence through decoding processes of a syntax converter and a word translator by applying the translation model to a source sentence input in real time, thereby resolving disadvantages of the existing phrase-based SMT and syntax-based SMT and combining advantages thereof.

Claims

1. A statistical machine translation device, comprising:

a translation model constructor extracting syntax conversion knowledge and word translation knowledge of a target sentence by using word reordering information and syntax analysis information between a source sentence and the target sentence in a plurality of parallel corpora, and calculating conversion probabilities with respect to the respective extracted knowledges;

a translation model learning device generating a syntax conversion model and a word translation model by learning the respective translation knowledges and conversion probabilities extracted through the translation model constructor; and

a translated sentence generator decoding the source sentence into the target sentence by applying the syntax conversion model and the word translation model learned through the translation model learning device with respect to a source sentence input in real time.

2. The statistical machine translation device of claim 1, wherein:

the translation model constructor further includes

a language model generator generating a language model by extracting a creation probability of a target vocabulary string from a single corpus of the target sentence.

3. The statistical machine translation device of claim 1, wherein:

the translated sentence generator generates a target vocabulary string having a high probability into a final translation sentence by combining a plurality of syntax conversion probabilities and a plurality of word translation probabilities output by the syntax conversion model and the word translation model.

4. A translation model constructing apparatus, comprising:

a syntax conversion knowledge extractor extracting the syntax conversion knowledge for a target sentence by using word reordering information between a source sentence and a target sentence, and syntax analysis information of the source sentence in a plurality of parallel corpora and calculating a conversion probability with respect to the extracted knowledge; and

a word translation knowledge extractor extracting the word translation knowledge by using the word reordering information between the source sentence and the target sentence, and the syntax analysis information of the source sentence in the plurality of parallel corpora and calculating the conversion probability with respect to the extracted knowledge.

5. The translation model constructing apparatus of claim 4, wherein:

the syntax conversion extractor includes

a tree generator generating a syntax tree (target tree) of the target sentence by using the word reordering information of the source sentence and the target sentence in the plurality of parallel corpora;

a tree node reorderer reordering nodes based on the target tree generated in the tree generator and a source tree depending on the syntax analysis information of the source sentence;

a tree conversion knowledge extractor extracting conversion knowledge of a sub-tree in each node; and

a probability calculator statistically calculating the conversion probability with respect to the knowledge extracted by the tree conversion knowledge extractor.

6. The translation model constructing apparatus of claim 4, wherein:

the word translation extractor extracts the word translation probability by applying a feature function in which a predetermined constraint condition is defined in the word reordering information between the source sentence and the target sentence and the syntax analysis information of the source sentence.

7. The translation model constructing apparatus of claim 6, wherein:

the feature function is a function constraining a part of speech string of the target sentence and a translation order of words included in the source sentence from a syntax of the target sentence and a syntax of the source sentence, and intersyntax arrangement information and outputting the constrained part of speech string and translation order as a feature.

8. The translation model constructing apparatus of claim 6, wherein:

the feature function is a function to output only the translation order of the words included in the source sentence from the syntax of the target sentence and the syntax of the source sentence, and the intersyntax arrangement information as the feature.

9. A translation sentence generating apparatus, comprising:

a syntax converter syntactically analyzing a source sentence input in real time, extracting syntax conversion knowledge of a target sentence from a syntax of the analyzed source sentence, and making the extracted knowledge to learn a conversion probability;

a word translator generating a target vocabulary string based on a word translation model in which a constraint condition is imposed to the syntax of the target sentence extracted through the syntax converter; and

a probability calculator combining a creation probability of the target vocabulary string generated through the word translator with the conversion probability learned through the syntax converter and thereafter, generating a target vocabulary string having the highest probability into a translation sentence.

10. The translation sentence generating apparatus of claim 9, wherein:

the word translator includes

a feature extractor extracting a feature based on word reordering information of the source sentence input from the syntax converter, syntax analysis information, and syntax analysis information of the target sentence;

a translation option constraining device constraining a translation option from a part of speech string and a translation order extracted by the feature extractor;

a translation distortion constraining device rearranging the translation order by constraining the word order extracted by the feature extractor;

a hypothesis searcher searching a hypothesis by reflecting constraint conditions determined through the translation option constraining device and the translation distortion constraining device to the feature extracted through the feature extractor; and

an M-Best tracker selecting the target vocabulary string having the high probability by calculating a probability for creation of the same target vocabulary string with respect to the hypothesis searched in the hypothesis searcher.

11. A machine translation method, comprising:

(a) syntactically analyzing a source sentence input in real time and extracting syntax conversion knowledge and a conversion probability of a target sentence from a syntax of the analyzed source sentence;

(b) generating a target vocabulary string based on a word translation model in which a constraint condition is imposed to the syntax of the target sentence extracted from the syntax conversion knowledge of the target sentence; and

(c) generating a target vocabulary string having a high probability into a translation sentence by combining the syntax conversion probability of the target sentence and a creation probability of the target vocabulary string.

12. The machine translation method of claim 11, wherein:

step (b) includes,

extracting a feature based on word reordering information of the source sentence input from the syntax converter, syntax analysis information, and syntax analysis information of the target sentence;

setting a translation option to constrain the feature;

searching a hypothesis by reflecting the translation option to the feature; and

selecting a target vocabulary string having a high probability by calculating a probability to generate the same target vocabulary string with respect to the searched hypothesis.

13. The machine translation method of claim 11, wherein:

the translation option selectively includes at least one of a part of speech string and a translation order.

19. A compute-readable recording medium in which a program to execute a process of claim 11 is recorded.

14. A translation model constructing method, comprising:

(a) extracting syntax conversion knowledge for a target sentence by using word reordering information between a source sentence and a target sentence and syntax analysis information of the source sentence in a plurality of parallel corpora;

(b) extracting word translation knowledge by using the word reordering information between the source sentence and the target sentence and the syntax analysis information of the source sentence in the plurality of parallel corpora; and

(c) calculating conversion probabilities for the syntax conversion knowledge and the word translation knowledge, respectively and making a weight to be learned with respect to each conversion probability.

15. The translation model constructing method of claim 14, wherein:

step (a) includes,

generating a syntax tree of the target sentence by using the word reordering information of the source sentence and word reordering information of the target sentence in the plurality of parallel corpora;

arranging nodes based on a syntax tree of the target sentence and a source tree according to syntax analysis information of the source sentence; and

extracting conversion knowledge of a sub-tree in each node.

16. The translation model constructing method of claim 14, wherein:

in the calculating of the word translation probability in step (c),

the information is extracted by applying a feature function in which a predetermined constraint condition is defined in the word reordering information of the source sentence and the word reordering information of the target sentence, and the syntax analysis information of the target sentence.

17. The translation model constructing method of claim 16, wherein:

the feature function uses a function constraining a part of speech string of the target sentence and a translation order of words included in the source sentence from a syntax of the target sentence and a syntax of the source sentence, and intersyntax arrangement information and outputting the constrained part of speech string and translation order as a feature.

18. The translation model constructing method of claim 16, wherein:

the feature function uses a function to output only the translation order of the words included in the source sentence from the syntax of the target sentence and the syntax of the source sentence, and the intersyntax arrangement information as the feature.