System and method for generating morphemes

• US 7,139,698 B1
• Filed: 11/14/2003
• Issued: 11/21/2006
• Est. Priority Date: 11/05/1999
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Riccardi, Giuseppe, Wright, Jeremy Huntley, Gorin, Allen Louis, Petrovska-Delacretaz, Dijana
• Primary Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US10/713,840
• Time in Patent Office: 1,103 Days
• Field of Search: 704/10, 704/243, 704/245
• US Class Current: 704/10
• CPC Class Codes: G06F 40/268 Morphological analysis

Abstract

The invention concerns a method of generating morphemes for speech recognition and understanding. The method may include receiving training speech, selecting candidate sub-morphemes from the training speech, selecting salient sub-morphemes from the candidate sub-morphemes based on salience measurements, and clustering the salient sub-morphemes based on semantic and syntactic similarities into morphemes. The morphemes may be acoustic and/or non-acoustic. The sub-morphemes may represent any sub-unit of communication including phones, phone-phrases, grammars, diphones, words, gestures, tablet strokes, body movements, mouse clicks, etc. The training speech may be verbal, non-verbal, a combination of verbal and non-verbal, or multimodal.

Claims

1. A method of generating morphemes from received speech, the method comprising:

selecting candidate sub-morphemes from the received speech;

selecting salient sub-morphemes from the candidate sub-morphemes based on salience measurements; and

clustering the salient sub-morphemes into morphemes based on using a combination of distortion measures.

2. The method of claim 1, wherein the generated morphemes are one of acoustic or non-acoustic morphemes.

3. The method of claim 1, further comprising representing the clustered morphemes as finite state machines.

4. The method of claim 1, wherein the generated morphemes are used by a speech recognition and understanding system.

5. The method of claim 1, wherein the received speech is training speech.

6. The method of claim 5, wherein the step of selecting candidate sub-morphemes further comprises:

filtering the training speech;

selecting all observed phone sequences of a predetermined length; and

selecting as candidate sub-morphemes the phone sequences that are of at least the predetermined length.

7. The method of claim 5, wherein the training speech comprises at least one of verbal and non-verbal speech.

8. The method of claim 7, wherein the non-verbal speech comprises the use of at least one of gestures, body movements, head movements, non-responses, text, keyboard entries, keypad entries, mouse clicks, DTMP codes, pointers, stylus, cable set-top box entries, graphical user interface entries or touchscreen entries.

9. The method of claim 1, wherein the speech includes multimodal forms.

10. The method of claim 1, wherein the speech is one of transcribed or untranscribed.

11. The method of claim 1, wherein the salient sub-morphemes are selected using a test for significance.

12. The method of claim 1, wherein the salient sub-morphemes are clustered into morphemes using at least one of a string distortion measure or a semantic distortion measure between the salient sub-morphemes.

13. The method of claim 1, wherein selecting candidate sub-morphemes from received speech further comprises:

using an automatic speech recognizer having a phonotactic language model.

14. A computer-readable medium storing a database of morphemes generated from received speech, the database generated according to a method comprising:

selecting candidate sub-morphemes from the received speech;

selecting salient sub-morphemes from the candidate sub-morphemes based on salience measurements; and

clustering the salient sub-morphemes into morphemes based on using a combination of distortion measures.

15. The computer-readable medium of claim 14, wherein the generated morphemes are one of acoustic or non-acoustic morphemes.

16. The computer-readable medium of claim 14, wherein the method further comprises representing the clustered morphemes as finite state machines.

17. The computer-readable medium of claim 14, wherein the generated morphemes are used by a speech recognition and understanding system.

18. The computer-readable medium of claim 14, wherein the method further comprises:

using an automatic speech recognizer having a phonotactic language model.

19. A natural spoken language system having a speech recognition and speech understanding modules, the natural language system using morphemes generated by a method comprising:

selecting candidate sub-morphemes from received speech;

selecting salient sub-morphemes from the candidate sub-morphemes based on salience measurements; and

clustering the salient sub-morphemes into morphemes based on using a combination of distortion measures.

20. The natural language system of claim 19, wherein the morphemes are one of acoustic or non-acoustic morphemes.

21. The natural language system of claim 19, wherein the received speech is training speech that includes at least one multimodal component.

22. The natural language system of claim 21, wherein the at least one multimodel component comprises one of gestures, body movements, head movements, non-responses, text, keyboard entries, keypad entries, mouse clicks, DTMF codes, pointers, stylus, cable set-top box entries, graphical user interface entries and touchscreen entries.

23. The natural spoken language system of claim 19, wherein selecting candidate sub-morphemes from received speech further comprises:

using an automatic speech recognizer having a phonotactic language model.