Supervised automatic text generation based on word classes for language modeling

• US 20030046078A1
• Filed: 09/04/2001
• Published: 03/06/2003
• Est. Priority Date: 09/04/2001
• Status: Active Grant

• Current Assignee: Sony Corporation (Sony Group Corp.), Sony Electronics Inc. (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.), Sony Electronics Inc. (Sony Group Corp.)
• Inventors: Abrego, Gustavo Hernandez, Menendez-Pidal, Xavier
• Granted Patent: US 7,035,789 B2
• Time in Patent Office: Days
• US Class Current: 704/260
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

A system and method is provided that randomly generates text with a given structure. The structure is taken from a number of learning examples. The structure of training examples is captured by word classification and the definition of the relationships between word classes in a given language. The text generated with this procedure is intended to replicate the information given by the original learning examples. The resulting text may be used to better model the structure of a language in a stochastic language model.

Claims

1. A method for use in supervised automatic text generation comprising:

inputting an original text of a language;

analyzing the original text to provide word classes; and

generating an N-gram of valid word classes using the word classes.

2. The method as claimed in claim 1 including:

parsing the original text to eliminate word class ambiguities in the original text.

3. The method as claimed in claim 1 wherein:

analyzing the original text includes;

decomposing the original text into words,

determining the roles of the words in the original text, and

placing the words with the same roles into the word classes; and

including;

providing relationships between the word classes.

4. The method as claimed in claim 1 including:

generating a string of random word classes; and

validating a group of word classes in the string of random word classes using the N-gram of the valid word classes; and

including;

providing a word for words in classes;

substituting the word in the group of word classes; and

outputting new text in the language.

5. The method as claimed in claim 4 wherein:

generating the string of random word classes, generates a random length string.

6. The method as claimed in claim 4 including:

substituting a plurality of words in a class in the group of word classes includes weighting the frequency of substitution.

7. The method as claimed in claim 4 including:

adding a new class to the string of random word classes to define a new partial string.

8. A method for use in supervised automatic text generation comprising:

providing an N-gram of valid word classes;

generating a string of random word classes;

validating a group of word classes in the string of random word classes using the N-gram of valid word classes;

providing a word in a language for words in classes;

substituting the word in the class in the group of word classes; and

outputting new text in the language.

9. The method as claimed in claim 8 wherein:

generating the string of random word classes, generates a random length string.

10. The method as claimed in claim 8 including:

substituting a plurality of words in a class in the group of word classes includes

weighting the frequency of substitution.

11. The method as claimed in claim 8 including:

adding a new class to the string of random word classes to define a new partial string.

12. A method for use in supervised automatic text generation comprising:

inputting an original text of a language;

analyzing the original text to provide word classes;

generating an N-gram of valid word classes using the word classes;

generating a string of random word classes;

validating a group of word classes in the string of random word classes using the N-gram of valid word classes;

providing a word for words in classes;

substituting the word in a word class in the group of word classes; and

outputting new text in the language.

13. In a supervised automatic text generation system, a system comprising:

an input for inputting an original text of a language;

a text analyzer for analyzing original text to provide word classes; and

an N-gram generator for generating an N-gram of valid word classes using the word classes.

14. The system as claimed in claim 13 including:

a parsing mechanism for parsing the original text to eliminate word class ambiguities in the original text.

15. The system as claimed in claim 13 wherein:

the text analyzer includes;

a first mechanism for decomposing the original text into words,

a second mechanism determining the roles of the words in the original text, and

a third mechanism placing the words with the same roles into the word classes; and

including;

an expert system for providing relationships between the word classes.

16. The system as claimed in claim 13 including:

a random class generator for generating a string of random word classes;

a validating mechanism for validating a group of word classes in the string of random word classes using the N-gram of the valid word classes; and

including;

a word in classes mechanism for providing a word for words in classes;

a substitution mechanism for substituting the word in the group of word classes; and

an output for outputting new text in the language.

17. The system as claimed in claim 16 wherein:

the random class generator generates a random length string.

18. The system as claimed in claim 16 wherein:

the word in classes mechanism provides a plurality of words; and

the substitution mechanism includes a mechanism for weighting the frequency of substitution.

19. The system as claimed in claim 16 including:

a mechanism for adding a new class to the string of random word classes to define a new partial string.

20. In a supervised automatic text generation system, a system comprising:

a random class generator for generating a string of random word classes;

a validating mechanism for validating a group of word classes in the string of random word classes using an N-gram of valid word classes;

a word in classes mechanism for providing a word in a language for words in classes;

a substitution mechanism substituting the word in the group of word classes; and

an output for outputting new text in the language.

21. The system as claimed in claim 20 wherein:

the random class generator generates a random length string.

22. The system as claimed in claim 20 wherein:

the word in classes mechanism provides a plurality of words; and

the substitution mechanism includes a mechanism for weighting the frequency of substitution.

23. The system as claimed in claim 20 including:

a mechanism for adding a new class to the string of random word classes to define a new partial string.

24. In a supervised automatic text generation system, a system comprising:

an input for inputting original text of a language;

a text analyzer for analyzing the original text to provide word classes;

an N-gram generator for generating an N-gram of valid word classes using the word classes;

a random class generator for generating a string of random word classes;

a validating mechanism for validating a group of word classes in the string of random word classes using the N-gram of valid word classes;

a word in classes mechanism for providing a word for words in classes;

a substitution mechanism for substituting the word in a word class in the group of word classes; and

an output outputting new text in the language.