Example-based translation method and system which calculates word similarity degrees, a priori probability, and transformation probability to determine the best example for translation

• US 6,161,083 A
• Filed: 04/24/1997
• Issued: 12/12/2000
• Est. Priority Date: 05/02/1996
• Status: Expired due to Fees

• Current Assignee: Sony Corporation (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.)
• Inventors: Franz, Alexander, Horiguchi, Keiko
• Primary Examiner(s): Thomas, Joseph
• Application Number:
  US08/839,989
• Time in Patent Office: 1,328 Days
• Field of Search: 704/2, 704/3, 704/4, 704/5, 704/6, 704/7, 704/8, 704/10, 707/535, 707/536, 434/156, 434/157, 434/167, 434/169
• US Class Current: 704/4
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/45 Example-based machine translation; Alignment

Abstract

A translating apparatus and a translating method wherein a first language sentence is divided into syntax units consisting of predetermined units of sentence structure such as clauses and phrases in stages from large syntax units into small syntax units and at each stage stored examples most similar to these syntax units are detected using probability models taking into account grammatical attributes of the syntax units and of the examples and using generalized linguistic knowledge and with reference to a thesaurus and the syntax units are translated on the basis of these detected examples and the results of translation of the syntax units are compounded to generate a second language sentence. The invention makes it possible to carry out high-quality translating efficiently and therefore at a high processing speed while eliminating the need for many superficially different examples having the same meaning content to be prepared and the need for the thesaurus to be regularly structured or have uniform distances between concepts.

Claims

1. A translating method for translating a first language sentence expressed in a first language into a second language sentence expressed in a second language using example memory means storing examples expressed in said first language and corresponding translations of said examples in said second language, comprising the steps of:

detecting from among said examples an example similar to said first language sentence on the basis of a priori probability of said example being used for translating said first language sentence and a transformation probability of said example being transformed into said first language sentence and used; and

translating said first language sentence on the basis of a translation among said translations corresponding to said example detected.

2. A translating apparatus for translating a first language sentence expressed in a first language into a second language sentence expressed in a second language, comprising:

example memory means storing examples expressed in said first language and corresponding translations of said examples in said second language;

detecting means for detecting from among said examples an example similar to said first language sentence on the basis of a priori probability of said example being used for translating said first language sentence and a transformation probability of said example being transformed into said first language sentence and used; and

translating means for translating said first language sentence on the basis of a translation among said translations corresponding to said example detected by said detecting means.

3. A translating apparatus according to claim 2, wherein said transformation probability is calculated from a grammatical attribute transformation probability of grammatical attributes of said example being transformed and a word transformation probability of words of said example being transformed.

4. A translating apparatus according to claim 3, wherein said grammatical attribute transformation probability is calculated from a grammatical attribute copy probability of a grammatical attribute of said example being used unchanged, a grammatical attribute deletion probability of a grammatical attribute of said example being deleted, a grammatical attribute addition probability of a grammatical attribute of said first language sentence being added, and a grammatical attribute replacement probability of a grammatical attribute of said example being replaced with a grammatical attribute of said first language sentence.

5. A translating apparatus according to claim 3, wherein said word transformation probability is calculated from a word copy probability of a word of said example being used unchanged, a word deletion probability of a word of said example being deleted, a word addition probability of a word of said first language sentence being added, and a word replacement probability of a word of said example being replaced with a word of said first language sentence.

6. A translating apparatus according to claim 5, wherein said word transformation probability is calculated further using at least one among a word scramble probability of a word of said example unchanged or replaced with a word of said first language sentence being used with its position changed, a word copy and compound probability of a word of said example having a modifier added thereto and being used as a compound word, and a word replace and compound probability of a word of said example replaced with a word of said first language sentence having a modifier added thereto and being used as a compound word.

7. A translating apparatus according to claim 5, wherein said word transformation probability is calculated according to an optimum transformation path for transforming said example into said first language sentence.

8. A translating apparatus according to claim 7, wherein in finding said optimum transformation path instead of an actual deletion path corresponding to actually deleting a word of said example a provisional deletion path provisionally deleting a word of said example is used and instead of an actual addition path corresponding to actually adding a word of said first language sentence a provisional addition path provisionally adding a word of said first language sentence is used,

further comprising word memory means for storing a provisionally deleted word and a provisionally added word provisionally deleted or added according to said provisional deletion path or said provisional addition path.

9. A translating apparatus according to claim 8, wherein a scramble path corresponding to using a word of said example unchanged or replaced with a word of said first language sentence with its position changed is selected only when either a word of said example has been stored in said word memory means as said provisionally deleted word or a word of said first language sentence has been stored in said word memory means as said provisionally added word, and thereafter said word of said example stored as said provisionally deleted word or said word of said syntax unit stored as said provisionally added word is erased from said word memory means, and

a copy and compound path corresponding to a word of said example having a modifier added thereto and being used as a compound word and a replace and compound path corresponding to a word of said example replaced with a word of said first language sentence having a modifier added to it and being used as a compound word are selected only when a word to become said modifier has been stored in said word memory means as said provisionally added word, and thereafter said word to become a modifier stored as said provisionally added word is erased from said word memory means.

10. A translating apparatus according to claim 9, wherein after the end point of a transformation path transforming said example into said first language sentence is reached a provisional deletion path or a provisional addition path provisionally deleting or adding said provisionally deleted word or said provisionally added word stored in said word memory means is converted into an actual deleting path or an actual addition path respectively and the path obtained is made the optimum transformation path.

11. A translating apparatus according to claim 2, wherein said word transformation probability is calculated according to an optimum transformation path for transforming said example into said first language sentence.

12. A translating apparatus according to claim 11, wherein said optimum transformation path is found by dynamic programming.

13. A translating apparatus for translating a first language sentence expressed in a first language into a second language sentence expressed in a second language, comprising:

example memory means storing examples expressed in said first language and corresponding translations of said examples in said second language;

dividing means for dividing said first language sentence into syntax units consisting of predetermined units of sentence structure;

detecting means for detecting for each of said syntax units of said first language sentence outputted by said dividing means an example similar to that syntax unit from among said examples;

translating means for translating said first language sentence in said syntax units on the basis of those of said translations corresponding to those of said examples detected by said detecting means; and

generating means for generating said second language sentence by compounding results of translation of said syntax units obtained by said translating means;

wherein said detecting means detects said example similar to said syntax unit on the basis of a priori probability of said example being used to translate said syntax unit and a transformation probability of said example being transformed into said syntax unit and used.

14. A translating apparatus according to claim 13, wherein said transformation probability is calculated from a grammatical attribute transformation probability of grammatical attributes of said example being transformed and a word transformation probability of words of said example being transformed.

15. A translating apparatus according to claim 14, wherein said grammatical attribute transformation probability is calculated from a grammatical attribute copy probability of a grammatical attribute of said example being used unchanged, a grammatical attribute deletion probability of a grammatical attribute of said example being deleted, a grammatical attribute addition probability of a grammatical attribute of said syntax unit being added, and a grammatical attribute replacement probability of a grammatical attribute of said example being replaced with a grammatical attribute of said syntax unit.

16. A translating apparatus according to claim 14, wherein said word transformation probability is calculated from a word copy probability of a word of said example being used unchanged, a word deletion probability of a word of said example being deleted, a word addition probability of a word of said syntax unit being added, and a word replacement probability of a word of said example being replaced with a word of said syntax unit.

17. A translating apparatus according to claim 16, further comprising:

thesaurus memory means storing a thesaurus wherein words are classified in a tree structure on the basis of their concepts; and

word similarity degree calculating means for calculating a word similarity degree expressing a similarity between a word of said syntax unit and a word of said example on the basis of said thesaurus and outputting said word similarity degree as said word replacement probability.

18. A translating apparatus according to claim 17, wherein said word similarity degree calculating means calculates said word similarity degree on the basis of the number of words included in a concept common to said word of said syntax unit and said word of said example in said thesaurus.

19. A translating apparatus according to claim 16, wherein said word transformation probability is calculated further using at least one among a word scramble probability of a word of said example unchanged or replaced with a word of said syntax unit being used with its position changed, a word copy and compound probability of a word of said example having a modifier added thereto and being used as a compound word, and a word replace and compound probability of a word of said example replaced with a word of said syntax unit having a modifier added thereto and being used as a compound word.

20. A translating apparatus according to claim 19, wherein said word transformation probability is calculated according to an optimum transformation path for transforming said example into said syntax unit.

21. A translating apparatus according to claim 20, wherein in finding said optimum transformation path instead of an actual deletion path corresponding to actually deleting a word of said example a provisional deletion path provisionally deleting a word of said example is used and instead of an actual addition path corresponding to actually adding a word of said syntax unit a provisional addition path provisionally adding a word of said syntax unit is used,

further comprising word memory means for storing a provisionally deleted word and a provisionally added word provisionally deleted or added according to said provisional deletion path or said provisional addition path.

22. A translating apparatus according to claim 21, wherein a scramble path corresponding to using a word of said example unchanged or replaced with a word of said syntax unit with its position changed is selected only when either a word of said example has been stored in said word memory means as said provisionally deleted word or a word of said syntax unit has been stored in said word memory means as said provisionally added word, and thereafter said word of said example stored as said provisionally deleted word or said word of said syntax unit stored as said provisionally added word is erased from said word memory means, and

a copy and compound path corresponding to a word of said example having a modifier added thereto and being used as a compound word and a replace and compound path corresponding to a word of said example replaced with a word of said syntax unit having a modifier added to it and being used as a compound word are selected only when a word to become said modifier has been stored in said word memory means as said provisionally added word, and thereafter said word to become a modifier stored as said provisionally added word is erased from said word memory means.

23. A translating apparatus according to claim 22, wherein after the end point of a transformation path transforming said example into said syntax unit is reached a provisional deletion path or a provisional adding path provisionally deleting or adding said provisionally deleted word or said provisionally added word stored in said word memory means is converted into an actual deleting path or an actual adding path respectively and the path obtained is made said optimum transformation path.

24. A translating apparatus according to claim 14, wherein said word transformation probability is calculated according to an optimum transformation path for transforming said example into said syntax unit.

25. A translating apparatus according to claim 24, wherein said optimum transformation path is found by dynamic programming.

26. A translating apparatus according to claim 13, wherein said priori probability or said transformation probability is calculated on the basis of corpus data.

27. A translating apparatus according to claim 13, wherein said priori probability or said transformation probability is calculated on the basis of said examples stored in said example memory means.