Method for statistical language modeling in speech recognition

• US 7,197,457 B2
• Filed: 04/30/2003
• Issued: 03/27/2007
• Est. Priority Date: 04/30/2003
• Status: Active Grant

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Raghunathan, Badri, Schmidt, Hauke, Weng, Fuliang
• Primary Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US10/427,601
• Publication Number:
  US 20040220813A1
• Time in Patent Office: 1,427 Days
• Field of Search: 704/255, 704/257
• US Class Current: 704/255
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 2015/0633 using lexical or orthographic knowledge sources

Abstract

A system for generating language modeling data for a speech recognition system includes an expression extractor to extract expression from domain-specific data of an existing domain using a base of linguistic knowledge, a concept structure mapper to map extracted expression to expression in a new domain using vocabulary for the new domain, a concatenation module to concatenate extracted expression with domain-general data, and a filter arrangement to identify and filter out unrealistic expression in the mapped or concatenated expression.

Claims

1. A system for generating language modeling data for a speech recognition system, comprising:

an expression extractor to extract expression from domain-specific data of an existing domain using a base of linguistic knowledge;

a concept structure mapper to map extracted expression to expression in a new domain using vocabulary for the new domain;

a concatenation module to concatenate extracted expression with domain-general data; and

a filter arrangement to identify and filter out unrealistic expression in at least one of mapped and concatenated expression.

2. The system of claim 1, wherein the system is configured to generate the language modeling data without a pre-existing domain-specific grammar.

3. The system of claim 1, wherein the extracted expression includes a domain-invariant expression.

4. The system of claim 1, wherein the extracted expression includes at least one of an idiomatic, a temporal, a spatial, and a monetary expression.

5. The system of claim 1, wherein the extracted expression includes combined temporal and spatial expression.

6. The system of claim 1, wherein the base of linguistic knowledge includes a lexical reference source.

7. The system of claim 6, wherein the lexical reference source includes a dictionary.

8. The system of claim 7, wherein the dictionary includes an electronic dictionary.

9. The system of claim 8, wherein the electronic dictionary is available via a data network.

10. The system of claim 9, wherein the data network includes an Internet.

11. The system of claim 1, wherein the concept structure mapper includes a domain-specific word-pair mapping table to correlate at least one word in the existing domain with at least one word in the new domain.

12. A method for generating language modeling data for a speech recognition system, comprising:

extracting expression from domain-specific data for an existing domain using a base of linguistic knowledge;

mapping an extracted expression to an expression in a new domain using vocabulary for the new domain and a concept mapping table;

concatenating the extracted expression using domain-general data; and

filtering at least one of the mapped and concatenated expression.

13. The method of claim 12, wherein the step of mapping the extracted expression includes establishing a domain-specific word-pair mapping table to identify and correlate at least one word in the existing domain with at least one word in the new domain.

14. The method of claim 12, wherein the new domain includes at least one of a navigation system, a hotel information system, and a restaurant information system.

15. The method of claim 14, wherein the navigation system includes an automotive navigation system.

16. The method of claim 14, wherein the navigation system includes a hand-held device.

17. The method of claim 16, wherein the hand-held device includes a mobile telephone.

18. A method for generating language modeling data for a speech recognition system, comprising:

extracting expression from domain-specific data for an existing domain using a base of linguistic knowledge;

mapping an extracted expression to a expression in a new domain using vocabulary for the new domain;

concatenating the extracted expression using domain-general data; and

filtering at least one of the mapped and concatenated expression,

wherein the step of mapping the extracted expression includes performing a neighboring word collocation verification test on the mapped expression to verify a naturalness of the mapped expression.

19. A method for generating language modeling data for a speech recognition system, comprising:

extracting expression from domain-specific data for an existing domain using a base of linguistic knowledge;

mapping an extracted expression to a expression in a new domain using vocabulary for the new domain;

concatenating the extracted expression using domain-general data; and

filtering at least one of the mapped and concatenated expression,

wherein the step of concatenating includes performing a statistical collocation measurement of the concatenated expression to ensure a smoothness of at least one of neighboring words and neighboring phrases and chaining highly-collocated pairs to form candidate sentences for the new domain.

20. The method of claim 19, wherein the step of filtering includes performing an additional statistical collocation measurement of the candidate sentences and removing candidate sentences having a collocation value below a predefined value.

21. A storage medium having a set of instructions residing therein, the set of instructions being executable by a processor to implement a method for performing:

extracting expression from domain-specific data for an existing domain using a base of linguistic knowledge;

mapping an extracted expression to an expression in a new domain using vocabulary for the new domain and a concept mapping table;

concatenating the extracted expression using domain-general data; and

filtering at least one of the mapped and concatenated expression.

22. The storage medium of claim 21, wherein the step of mapping the extracted expression includes establishing a domain-specific word-pair mapping table to identify and correlate at least one word in the existing domain with at least one word in the new domain.

23. The storage medium of claim 21, wherein the new domain includes at least one of a navigation system, a hotel information system, and a restaurant information system.

24. The storage medium of claim 23, wherein the navigation system includes an automotive navigation system.

25. The storage medium of claim 23, wherein the navigation system includes a hand-held device.

26. The storage medium of claim 25, wherein the hand-held device includes a mobile telephone.

27. A storage medium having a set of instructions residing therein, the set of instructions being executable by a processor to implement a method for performing:

extracting expression from domain-specific data for an existing domain using a base of linguistic knowledge;

mapping an extracted expression to a expression in a new domain using vocabulary for the new domain;

concatenating the extracted expression using domain-general data; and

filtering at least one of the mapped and concatenated expression,

wherein the step of mapping the extracted expression includes performing a neighboring word collocation verification test on the mapped expression to verify a naturalness of the mapped expression.

28. A storage medium having a set of instructions residing therein, the set of instructions being executable by a processor to implement a method for performing:

extracting expression from domain-specific data for an existing domain using a base of linguistic knowledge;

mapping an extracted expression to a expression in a new domain using vocabulary for the new domain;

concatenating the extracted expression using domain-general data; and

filtering at least one of the mapped and concatenated expression,

wherein the step of concatenating includes performing a statistical collocation measurement of the concatenated expression to ensure a smoothness of at least one of neighboring words and neighboring phrases and chaining highly-collocated pairs to form candidate sentences for the new domain.

29. The storage medium of claim 28, wherein the step of filtering includes performing an additional statistical collocation measurement of the candidate sentences and removing candidate sentences having a collocation value below a predefined value.