Apparatus and method for processing natural language and apparatus and method for speech recognition

• US 5,991,721 A
• Filed: 05/29/1996
• Issued: 11/23/1999
• Est. Priority Date: 05/31/1995
• Status: Expired due to Term

• Current Assignee: Sony Corporation (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.)
• Inventors: Asano, Yasuharu, Ogawa, Hiroaki, Kagami, Tetsuya, Akabane, Makoto, Ishii, Kazuo, Kakuda, Hiroshi, Tanaka, Miyuki, Kato, Yasuhiko, Watari, Masao
• Primary Examiner(s): Isen, Forester W.
• Assistant Examiner(s): EDOUARD, PATRICK NESTOR
• Application Number:
  US08/654,893
• Time in Patent Office: 1,273 Days
• Field of Search: 704/1-2, 704/9, 704/239, 704/240, 704/241, 704/251, 704/255, 704/257
• US Class Current: 704/257
• CPC Class Codes: G06F 40/45 Example-based machine translation; Alignment
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models

Abstract

An apparatus and a method for processing a natural language arranged so as to improve the speech recognition rate. In an example search section, the degree of similarity between each of a plurality of examples of the actual use of the language stored in an example data base and each of a plurality of probable recognition results output from a recognition section, and one of the examples corresponding to the highest degree of similarity is selected. A final speech recognition result is obtained by using the selected example. The example search section calculates the degree of similarity by weighting the degree of similarity on the basis of a context according to at least one of the examples previously selected.

Claims

1. An apparatus for processing a natural language in which an input sentence is processed by using an example of an actual use of a language most similar to the input sentence, said apparatus comprising:

input means for inputting an input sentence;

conversion means for converting said input sentence into input sentence data;

example storage means for storing a plurality of examples of actual uses of a language; and

selection means for calculating a degree of similarity between the input sentence data and each of the examples stored in said example storage means and for selecting an example corresponding to a highest degree of similarity,

wherein said selection means is further configured to calculate the degree of similarity by weighting some of the examples, said weighting being performed based on a context according to at least one of the examples previously selected, and

wherein at least one of said plurality of examples comprises more than one word.

2. An apparatus according to claim 1, wherein said example storage means is further configured to store prediction information for predicting some of the examples likely to be used for a next processing, and

wherein said selection means is further configured to perform said weighting based on the prediction information.

3. An apparatus according to claim 2, wherein said example storage means is further configured to store as the prediction information a group of information in which the example presently used for processing is related to some of the examples likely to be used for the next processing, and

wherein said selection means is further configured to perform weighting so that the degree of similarity of some of the examples related the example used in a preceding processing increases or the degree of similarity of some of the examples not related the example used in a preceding processing decreases.

4. An apparatus according to claim 2, wherein said example storage means is further configured to store a group of information in which some of the examples likely to be used for the next processing are related to one of the examples presently used, the group of information being obtained by learning.

5. An apparatus according to claim 2, wherein said example storage means is further configured to store the prediction information by sorting the examples in the prediction information into groups of examples used in correspondence with predetermined places, and

wherein said selection means is further configured to use the prediction information according to each of the predetermined places.

6. A method of processing a natural language using a natural language processing apparatus which has example storage means having a plurality of examples of an actual use of a language stored therein, and which processes an input sentence by using one of the examples most similar to the input sentence, said method comprising the steps of:

inputting an input sentence;

converting said input sentence into input sentence data;

calculating a degree of similarity between the input sentence data and each of the examples stored in said example storage means; and

selecting an example corresponding to the highest degree of similarity,

wherein in said calculation step the degree of similarity is calculated by performing weighting based on a context according to at least one of the examples previously selected, and

wherein at least one of said plurality of examples comprises more than one word.

7. A speech recognition apparatus for performing sentence speech recognition, said apparatus comprising:

extraction means for extracting at least one feature parameter of a speech in a language by acoustically analyzing the speech;

probable recognition result calculation means for obtaining a plurality of probable recognition results and a likelihood of each of the probable recognition results by recognizing the speech based on the feature parameter output from said extraction means;

example storage means for storing a plurality of examples of an actual use of the language;

linguistic similarity calculation means for calculating a degree of linguistic similarity between each of the plurality of probable recognition results and each of the examples stored in the example storage means;

integrated similarity calculation means for calculating a degree of integrated similarity between each of the plurality of probable recognition results and each of the examples stored in the example storage means based on the degree of linguistic similarity and the likelihood;

selection means for selecting one of the examples corresponding to a highest degree of integrated similarity; and

recognition result calculation means for obtaining a result of recognition of the speech based on the example selected by said selection means.

8. An apparatus according to claim 7, wherein said integrated similarity calculation means calculates the degree of integrated similarity by adding the degree of linguistic similarity and the likelihood in a predetermined ratio.

9. An apparatus according to claim 8, wherein said integrated similarity calculation means dynamically changes the ratio in which the degree of linguistic similarity and the likelihood are added.

10. A speech recognition method using a speech recognition apparatus which has example storage means for storing a plurality of examples of an actual use of a language and which performs sentence speech recognition, said method comprising the steps of:

extracting at least one feature parameter of a speech by acoustically analyzing the speech;

obtaining a plurality of probable recognition results and a likelihood of each of the probable recognition results by recognizing the speech based on the feature parameter;

calculating a degree of linguistic similarity between each of the plurality of probable recognition results and each of the examples stored in the example storage means;

calculating a degree of integrated similarity between each of the plurality of probable recognition results and each of the examples stored in the example storage means based on the degree of linguistic similarity and the likelihood; and

selecting one of the examples corresponding to a highest degree of integrated similarity and obtaining a result of recognition of the speech based on the example selected.

11. An apparatus for processing a natural language in which an input sentence is processed by using an example of an actual use of a language most similar to the input sentence, said apparatus comprising:

example storage means for storing a plurality of examples of actual uses of a language;

selection means for calculating a degree of similarity between the input sentence and each of the examples stored in said example storage means and for selecting an example corresponding to a highest degree of similarity; and

output means for generating an output sound corresponding to the selected example,

wherein said selection means is further configured to calculate the degree of similarity by weighting some of the examples, said weighting being performed based on a context according to at least one of the examples previously selected, and

wherein at least one of said plurality of examples comprises more than one word.

12. An apparatus according to claim 11, wherein said example storage means is further configured to store prediction information for predicting some of the examples likely to be used for a next processing, and

wherein said selection means is further configured to perform said weighting based on the prediction information.

13. An apparatus according to claim 12, wherein said example storage means is further configured to store as the prediction information a group of information in which the example presently used for processing is related to some of the examples likely to be used for the next processing, and

wherein said selection means is further configured to perform weighting so that the degree of similarity of some of the examples related the example used in a preceding processing increases or the degree of similarity of some of the examples not related the example used in a preceding processing decreases.

14. An apparatus according to claim 12, wherein said example storage means is further configured to store a group of information in which some of the examples likely to be used for the next processing are related to one of the examples presently used, the group of information being obtained by learning.

15. An apparatus according to claim 12, wherein said example storage means is further configured to store the prediction information by sorting the examples in the prediction information into groups of examples used in correspondence with predetermined places, and

wherein said selection means is further configured to use the prediction information according to each of the predetermined places.

16. A method of processing a natural language using a natural language processing apparatus which has example storage means having a plurality of examples of an actual use of a language stored therein, and which processes an input sentence by using one of the examples most similar to the input sentence, said method comprising the steps of:

calculating a degree of similarity between the input sentence and each of the examples stored in said example storage means; and

selecting an example corresponding to the highest degree of similarity; and

generating an output sound corresponding to the selected example,

wherein in said calculation step the degree of similarity is calculated by performing weighting based on a context according to at least one of the examples previously selected, and

wherein at least one of said plurality of examples comprises more than one word.

17. An apparatus for processing a natural language in which an input sentence is processed by using an example of an actual use of a language most similar to the input sentence, said apparatus comprising:

a microphone configured to input an input sentence;

an A/D converter section configured to convert said input sentence into input sentence data;

an example data base configured to store a plurality of examples of actual uses of a language; and

an example search section configured to calculate a degree of similarity between the input sentence data and each of the examples stored in said example data base and to select an example corresponding to a highest degree of similarity,

wherein said example search section is further configured to calculate the degree of similarity by weighting some of the examples, said weighting being performed based on a context according to at least one of the examples previously selected, and

wherein at least one of said plurality of examples comprises more than one word.

18. An apparatus according to claim 17, wherein said example data base is further configured to store prediction information for predicting some of the examples likely to be used for a next processing, and

wherein said example search section is further configured to perform said weighting based on the prediction information.

19. An apparatus according to claim 18, wherein said example data base is further configured to store as the prediction information a group of information in which the example presently used for processing is related to some of the examples likely to be used for the next processing, and

wherein said example search section is further configured to perform weighting so that the degree of similarity of some of the examples related the example used in a preceding processing increases or the degree of similarity of some of the examples not related the example used in a preceding processing decreases.

20. An apparatus according to claim 18, wherein said example data base is further configured to store a group of information in which some of the examples likely to be used for the next processing are related to one of the examples presently used, the group of information being obtained by learning.

21. An apparatus according to claim 18, wherein said example data base is further configured to store the prediction information by sorting the examples in the prediction information into groups of examples used in correspondence with predetermined places, and

wherein said example search section is further configured to use the prediction information according to each of the predetermined places.

22. A speech recognition apparatus for performing sentence speech recognition, said apparatus comprising:

an analysis section configured to extract at least one feature parameter of a speech in a language by acoustically analyzing the speech;

a recognition section configured to obtain a plurality of probable recognition results and a likelihood of each of the probable recognition results by recognizing the speech based on the feature parameter output from said analysis section;

an example data base configured to store a plurality of examples of an actual use of the language; and

an example search section configured to calculate a degree of linguistic similarity between each of the plurality of probable recognition results and each of the examples stored in the example data base,

said example search section being further configured to calculate a degree of integrated similarity between each of the plurality of probable recognition results and each of the examples stored in the example data base based on the degree of linguistic similarity and the likelihood,

said example search section being further configured to select one of the examples corresponding to a highest degree of integrated similarity, and

said example search section being further configured to obtain a result of recognition of the speech based on the example selected corresponding to the highest degree of integrated similarity.

23. An apparatus according to claim 22, wherein said example search section is further configured to calculate the degree of integrated similarity by adding the degree of linguistic similarity and the likelihood in a predetermined ratio.

24. An apparatus according to claim 23, wherein said example search section is further configured to dynamically change a ratio in which the degree of linguistic similarity and the likelihood are added.

25. An apparatus for processing a natural language in which an input sentence is processed by using an example of an actual use of a language most similar to the input sentence, said apparatus comprising:

an example data base configured to store a plurality of examples of actual uses of a language;

an example search section configured to calculate a degree of similarity between the input sentence and each of the examples stored in said example data base and to select an example corresponding to a highest degree of similarity; and

a speech synthesizer configured to generate an output sound corresponding to the selected example,

wherein said example search section is further configured to calculate the degree of similarity by weighting some of the examples, said weighting being performed based on a context according to at least one of the examples previously selected, and

wherein at least one of said plurality of examples comprises more than one word.

26. An apparatus according to claim 25, wherein said example data base is further configured to store prediction information for predicting some of the examples likely to be used for a next processing, and

wherein said example search section is further configured to perform said weighting based on the prediction information.

27. An apparatus according to claim 26, wherein said example data base is further configured to store as the prediction information a group of information in which the example presently used for processing is related to some of the examples likely to be used for the next processing, and

wherein said example search section is further configured to perform weighting so that the degree of similarity of some of the examples related the example used in a preceding processing increases or the degree of similarity of some of the examples not related the example used in a preceding processing decreases.

28. An apparatus according to claim 26, wherein said example data base is further configured to store a group of information in which some of the examples likely to be used for the next processing are related to one of the examples presently used, the group of information being obtained by learning.

29. An apparatus according to claim 26, wherein said example data base is further configured to store the prediction information by sorting the examples in the prediction information into groups of examples used in correspondence with predetermined places, and

wherein said example search section is further configured to use the prediction information according to each of the predetermined places.