Adapter for allowing both online and offline training of a text to text system

• US 20070094169A1
• Filed: 09/09/2005
• Published: 04/26/2007
• Est. Priority Date: 09/09/2005
• Status: Active Grant

• Current Assignee: SDL PLC (RWS Holdings Plc)
• Original Assignee: Language Weaver, Inc. (RWS Holdings Plc)
• Inventors: Yamada, Kenji, Knight, Kevin, Langmead, Greg
• Granted Patent: US 7,624,020 B2
• Time in Patent Office: Days
• US Class Current: 706/15
• CPC Class Codes: G06F 40/42 Data-driven translation

Abstract

An adapter for a text to text training. A main corpus is used for training, and a domain specific corpus is used to adapt the main corpus according to the training information in the domain specific corpus. The adaptation is carried out using a technique that may be faster than the main training. The parameter set from the main training is adapted using the domain specific part.

Claims

1. A method, comprising:

First carrying out a first training using at least one corpus of language information, using a first training operation to obtain a first parameter set;

second carrying out a second training using a domain specific corpus, using a second training operation which operates faster than said first training operation, and which is less accurate than said first training operation, to obtain a second parameter set; and

using said second parameter set to adapt said first parameter set to carry out a text-to-text operation.

2. A method as in claim 1, further comprising merging said first and second parameter sets into a merged parameter set, and using said merged parameter set for said text to text operation.

3. A method as in claim 1, wherein said text to text operation is a translation between first and second languages.

4. A method as in claim 2, wherein said merging comprises a weighted merge between the first and second databases.

5. A method as in claim 2, wherein said merging comprises an adaptive training merge between the first and second databases.

6. A method as in claim 4, wherein said weighted merge is sensitive to frequency of specified terms in the corpus.

7. A method as in claim 1, wherein said second training operation uses parameters from said first training operation.

8. A method as in claim 7, wherein said second training operation uses a basic seed probability from the first training operation

9. A method as in claim 7, wherein said second training operation uses alignment counts from the first training operation.

10. A method as in claim 2, wherein said merging uses an adaptive merging.

11. A method as in claim 10, wherein said adaptive merging uses a merge which is proportional to a frequency of a specified term in a training database.

12. A method as in claim 2, wherein said merging comprises adding indications of counts.

13. A method as in claim 2, wherein said merging comprises adding information that represent counts related to alignment.

14. A method as in claim 1, wherein said first carrying out is carried out at a first location, and said second carrying out is carried out at a second location, different than said first location.

15. A method as in claim 1, wherein said using comprises using partial information from the first training and partial information from the second training.

16. A method as in claim 15, wherein said using comprises forming an original table and an override table, and using both said original table and said override table as part of the text to text application.

17. A method as in claim 16, wherein said override table represents information which is present in both the first and second training corpuses.

18. A method as in claim 17, wherein the override table includes precomputed versions of specified formulas.

19. A method as in claim 17, wherein the information includes probabilities.

20. A method as in claim 17, wherein the information includes counts.

21. An apparatus, comprising:

a first training computer at a first location, carrying out a first training using at least one corpus of information, using a first training operation, to obtain a first parameter set;

a second training computer, at a second location, different than the first location, carrying out a second training using a domain specific corpus that has different information than said at least one corpus, using a second training operation which operates faster than said first training operation, and which is less accurate than said first training operation, to obtain a second parameter set, and using said first and second parameter sets together for a text-to-text operation.

22. An apparatus as in claim 21, wherein said second training computer also operates to merge said first and second parameter sets into a merged parameter set, and to use said merged parameter set for said text to text operation.

23. An apparatus as in claim 21, wherein said text to text operation is a translation between first and second languages.

24. An Apparatus as in claim 22, wherein said second computer carries out a weighted merge between the first and second databases.

25. An Apparatus as in claim 22, wherein said second computer carries out an adaptive training merge between the first and second databases.

26. An Apparatus as in claim 24, wherein said weighted merge is sensitive to frequency of specified terms in the corpus.

27. An Apparatus as in claim 22, wherein said second computer forms an original table and an override table, and uses both said original table and said override table as part of the text to text application.

28. An Apparatus as in claim 27, wherein said override table represents information which is present in both the first and second training corpuses.

29. An Apparatus as in claim 27, wherein the override table includes precomputed versions of specified formulas.

30. An Apparatus as in claim 27, wherein the information includes probabilities.

31. An apparatus, comprising:

a training part, including at least one computer, which carries out a first training for a text-to-text application using at least one corpus of training information, to obtain a first parameter set, and at a different time than said first training, carrying out a second training using a domain specific corpus that has different information than said at least one corpus, using a second training operation which operates faster than said first training operation, and which is less accurate than said first training operation, to obtain a second parameter set, and using said second parameter set to adapt said first parameter set to create an adapted parameter set, and to use the adapted parameter set for a text-to-text operation.

32. An apparatus as in claim 31, wherein said at least one training computer merges said first and second parameter sets into a merged parameter set, and uses said merged parameter set for said text to text operation.

33. An apparatus as in claim 31, wherein said text to text operation is a translation between first and second languages.

34. An Apparatus as in claim 32, wherein said training computer carries out a weighted merge between the first and second databases.

35. An Apparatus as in claim 32, wherein said training computer carries out an adaptive training merge between the first and second databases.

36. An Apparatus as in claim 34, wherein said weighted merge is sensitive to frequency of specified terms in the corpus.