Method and apparatus for training a recognizer

• US 5,680,480 A
• Filed: 07/29/1994
• Issued: 10/21/1997
• Est. Priority Date: 07/29/1994
• Status: Expired due to Term

• Current Assignee: Apple Inc.
• Original Assignee: Apple Computer Incorporated (Apple Inc.)
• Inventors: Temkin, David T., Auguste, Donna M., Beernink, Ernest H.
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): KELLEY, CHRISTOPHER S
• Application Number:
  US08/282,199
• Time in Patent Office: 1,180 Days
• Field of Search: 382/13, 382/57, 382/59, 382/187, 382/188, 382/189, 382/229, 382/155, 382/156, 382/157, 382/158, 382/159, 382/160, 382/161
• US Class Current: 382/187
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 10/987 with the intervention of an operator

Abstract

A method and apparatus for training a recognizer includes displaying a practice word on a display screen of a pen-based computer system and entering a handwritten word that is similar to the practice word on the display screen. A recognized candidate list that includes candidate words and an associated confidence level for each candidate word is derived from the handwritten word by a recognizer. The confidence level indicates how confidently the handwritten word has been matched to the associated candidate word. The candidate list is displayed and a training database is updated with training data derived from the handwritten word. The training data is used by the recognizer to recognize other handwritten words with greater accuracy. A practice word that is either easily recognized by the recognizer or recognized with difficulty can be displayed, and the easily recognized practice words are randomly retrieved from a dictionary tree of characters to provide words of reasonable length and a variety of types and combinations of characters.

Claims

1. A method for training a recognizer of a pen-based computer system, the method comprising the steps of:

having a pen-based computer system display a practice word on a display screen of said pen-based computer, said practice word being easily recognized accurately by said recognizer when entered as a handwritten word by a user, said easily recognized practice word being displayed when an indication from said user is received to display a practice word that is easily recognized;

having a user of said pen-based computer system enter a first handwritten word that said user desires to be recognized as said practice word into said pen-based computer system using a stylus;

obtaining a recognized candidate list from a recognizer, said candidate list being derived from said first handwritten word and including a plurality of candidate words and an associated confidence level for each of said plurality of candidate words, said confidence level indicating how confidently said first handwritten word has been matched to said associated candidate word, wherein each of said candidate words includes a sequence of a plurality of characters, wherein each of said candidate words has a different recognized meaning from all other candidate words in said candidate list, and wherein each of said candidate words includes a different number, sequence, or identity of characters from said other candidate words;

displaying said candidate list including said plurality of candidate words and said associated confidence levels on said display screen of said pen-based computer system, said candidate words being displayed in an order from the most confidently recognized candidate word to the least confidently recognized candidate word; and

updating a training database with training data derived from said first handwritten word such that when a second handwritten word is entered, said training data may be utilized to recognize said second handwritten word with greater accuracy.

3. A method as recited in claim 1 wherein only the candidate word, if present in said candidate list, which corresponds to said practice word is visually distinguished from other candidate words in said candidate list to indicate the position of said corresponding candidate word in said order of candidate words.

4. A method as recited in claim 1 wherein said step of displaying a practice word that is easily recognized includes retrieving a random word from a dictionary tree.

5. A method as recited in claim 4 wherein said step of displaying a practice word that is easily recognized includes retrieving a plurality of random words from said dictionary tree and selecting one of said random words to be said practice word.

6. A method as recited in claim 1 wherein said step of displaying a practice word includes displaying a practice word that is recognized accurately with difficulty by said recognizer when said practice word is entered as a handwritten word by said user and when said practice word is selected by said user to be a difficult practice word.

7. A method as recited in claim 6 wherein said step of displaying a word recognized with difficulty includes selecting a random word from a predetermined list of difficult words.

8. A method as recited in claim 1 wherein said step of entering a first handwritten word includes entering said first handwritten word on a display screen of said pen-based computer system with a stylus.

9. A method as recited in claim 8 wherein said step of obtaining a recognized candidate list from a recognizer includes obtaining training data associated with each candidate word.

10. A method as recited in claim 8 wherein said step of displaying said candidate list includes displaying only a predetermined number of candidate words having the highest confidence levels generated by said recognizer.

12. A method as recited in claim 9 wherein said step of updating a training database includes saving said training data associated with said candidate word which matches said practice word such that said practice word is associated with said training data.

13. A method as recited in claim 1 wherein said step of updating a training database includes saving training data which describes portions of said first handwritten word which match said practice word.

14. A method as recited in claim 10 wherein said practice word and said first handwritten word are comprised of characters, and wherein said step of updating a training database includes associating characters included in said first handwritten word with corresponding characters of said practice word.

15. A method as recited in claim 1 wherein said step of updating a training database includes updating a training database associated with a user of said pen-based computer system, wherein said training database is one of a plurality of training databases.

21. A pen-based computer system as recited in claim 1 wherein said means for displaying a practice word recognized with difficulty includes means for selecting a random word from a predetermined list of difficult words.

2. A method for training a recognizer of a pen-based computer system, the method comprising the steps of:

receiving a selection from a user indicating that a practice word be a word that is easily recognized by said pen based computer when entered as a handwritten word by said user;

displaying said practice word on a display screen of said pen-based computer;

receiving a first handwritten word entered by said user into said pen-based computer system using a stylus, said user desiring said handwritten word to be recognized as said practice word;

obtaining a recognized candidate list from a recognizer, said candidate list being derived from said first handwritten word and including a plurality of candidate words and an associated confidence level for each of said plurality of candidate words, said confidence level indicating how confidently said first handwritten word has been matched to said associated candidate word;

displaying said candidate list including said plurality of candidate words and said associated confidence levels on said display screen of said pen-based computer system; and

updating a training database with training data associated with a candidate word that corresponds to said practice word such that when a second handwritten word is entered, said training data may be utilized to recognize said second handwritten word with greater accuracy.

11. A method as recited in claim 2 wherein said step of displaying a practice word that is easily recognized includes retrieving a random word from a dictionary tree.

16. A pen-based computer system comprising:

a digital processor;

read/write memory coupled to said digital processor;

a display screen coupled to said digital processor;

a pointing device coupled to said digital processor, said pointing device being operative to enter a first ink object into said computer system;

means for displaying a practice word on said display screen, said practice word being accurately recognized with difficulty by said pen-based computer system when entered by a user as said first ink object, said difficult practice word being displayed when an indication from said user is received to display a practice word that is recognized with difficulty;

a recognizer implemented by said digital processor for analyzing said first ink object and providing a recognized candidate list, said candidate list including a plurality of candidate words and an associated confidence level for each of said plurality of candidate words, said confidence level indicating how closely said first ink object matches said associated candidate word, wherein said candidate list is displayed on said display screen such that the candidate word corresponding to said practice word, if present in said candidate list, is visually distinguished from other candidate words not corresponding to said practice word; and

means for updating a training database with training data derived from said first ink object such that when a second ink object is entered, said training data is utilized to recognize said second ink object more accurately, wherein said means for updating is implemented by said digital processor.

17. A pen-based computer system as recited in claim 16 wherein said pointing device is a stylus operative to engage said display screen to enter said first and second ink objects into said computer system.

18. A pen-based computer system as recited in claim 16 further comprising means for displaying a practice word that is easily recognized accurately by said recognizer when handwritten by a user, said easily recognized practice word being displayed when said user designates said practice word to be an easily recognized practice word.

19. A pen-based computer system as recited in claim 18 wherein said means for displaying a practice word that is easily recognized includes means for retrieving a plurality of random words from said dictionary tree and means for selecting one of said random words to be said practice word.

20. A pen-based computer system as recited in claim 19 wherein said means for displaying a practice word includes means for displaying a practice word that is accurately recognized with difficulty by said recognizer when handwritten by a user, said difficult practice word being displayed when said user designates said practice word to be an easily recognized practice word.

22. A pen-based computer system as recited in claim 16 further comprising means for displaying a predetermined number of candidate words from said candidate list provided by said recognizer.

23. A pen-based computer system as recited in claim 16 wherein said means for updating a training database includes means for saving training data which describes said first handwritten word such that said practice word is associated with said training data.

24. A method for providing an easily recognized practice word to a user of a pen-based computer system who desires to handwrite an ink word that can be easily recognized accurately as said practice word, the method comprising:

receiving input from said user indicating that said user has selected a practice word that is easily recognized when input as a handwritten word to said pen computer system;

retrieving a plurality of random words from a dictionary tree of characters, wherein each of said random words includes a plurality of characters;

calculating a word score for each of said random words based on the number of pairs of characters included in each of said random words; and

selecting a random word having the highest word score for said plurality of random words as said practice word for a user of a pen-based computer system, said user intending to handwrite an ink word that will be accurately and easily recognized as said selected practice word.

25. A method as recited in claim 24 wherein said step of retrieving a plurality of random words includes randomly selecting one of a plurality of possible next characters from said tree.

26. A method as recited in claim 25 wherein a retrieved random word is not used as said practice word if said retrieved word includes a number of characters outside a predetermined range of characters.

27. A method as recited in claim 26 wherein each character in said tree includes an indicator indicating if said character can be used to terminate said random word, and wherein said random word is terminated at a character when said character includes said terminator and a randomly generated number falls within a predetermined threshold.

28. A method as recited in claim 25 wherein said step of calculating a word score includes incrementing said word score of a random word when a pair of characters in said random word has not already been used in said random word.

29. A method as recited in claim 28 further comprising a step of incrementing a pair number for each pair of characters used in said random word, said pair number being stored in an array of character pairs, wherein said word score is incremented for each character pair in said random word having a pair number that has not been previously incremented.

30. A method as recited in claim 24 further comprising a step of displaying said practice word on a display screen of said pen-based computer system.

31. A method as recited in claim 30 further comprising a step of displaying an entry field in which said user writes a handwritten ink word with a stylus, said handwritten ink word being intended to represent said practice word.

32. A method as recited in claim 31 further comprising a step of recognizing said handwritten ink word and displaying an indication of how closely said handwritten ink word corresponds to said practice word.

33. A computer readable medium containing program instructions for training a recognizer, the program instructions performing the steps comprising:

receiving an indication from a user to display either a practice word that is easily recognized accurately when entered as a handwritten ink word, or a practice word that is accurately recognized with difficulty when entered as a handwritten ink word;

displaying a practice word on a display screen of said pen-based computer system corresponding to said indication;

receiving a first handwritten ink word from said user of said pen-based computer system, wherein said user desires said first handwritten ink word to be recognized as said practice word;

providing said first handwritten ink work to a recognizer and obtaining a recognized candidate list from said recognizer, said candidate list including a plurality of candidate words that are possible matches to said handwritten ink word, wherein each of said candidate words includes a plurality of characters, said candidate list also including an associated confidence level for each of said candidate words that indicates how confidently said first hand written ink word has been matched to said associated candidate word;

displaying said candidate list including said plurality of candidate words and said associated confidence levels on said display screen of said pen-based computer system; and

updating a training database with training data associated with a candidate word that corresponds to said practice word, such that when a second handwritten word is entered, said training data may be utilized to recognize said second handwritten word with greater accuracy.

34. A computer readable medium as recited in claim 33 wherein said candidate list has an order of said candidate words from the most confidently recognized candidate word to the least confidently recognized candidate word and wherein said program instructions further perform a step of highlighting only the candidate word in said candidate list that corresponds to said practice word to indicate the position of said corresponding candidate word in said order of candidate words.

35. A computer readable medium as recited in claim 33 wherein said step of displaying a practice word that is easily recognized includes retrieving a random word from a dictionary tree.

36. A computer readable medium as recited in claim 33 wherein said step of displaying a word recognized with difficulty includes selecting a random word from a predetermined list of difficult words.

37. A computer readable medium as recited in claim 33 wherein said first handwritten ink word is received on a display screen of said pen-based computer system by a user using a stylus.

38. A computer readable medium as recited in claim 33 wherein said step of displaying said candidate list includes displaying only a predetermined number of candidate words having the highest confidence levels generated by said recognizer.

39. A computer readable medium as recited in claim 37 wherein said step of updating a training database includes saving said training data associated with said candidate word which matches said practice word such that said practice word is associated with said training data.