Authoring speech grammars

• US 7,617,093 B2
• Filed: 06/02/2005
• Issued: 11/10/2009
• Est. Priority Date: 06/02/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Ramsey, William D.
• Primary Examiner(s): Hudspeth; David R
• Assistant Examiner(s): Rider; Justin W
• Application Number:
  US11/143,357
• Publication Number:
  US 20060277031A1
• Time in Patent Office: 1,622 Days
• Field of Search: 704 1- 10, 704/200, 704/231, 704/240, 704251-257, 704270-272, 704/277, 345/506, 345/619, 705/34, 706/46, 707/1, 707/100, 709/202, 709/219, 717/123, 717/124, 717/143
• US Class Current: 704/10
• CPC Class Codes: G06F 40/279 Recognition of textual entities
• CPC Class Codes: G06Q 30/04 Billing or invoicing
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules

Abstract

A method and apparatus are provided for automatically forming a grammar. Example text strings are received and N-grams are formed based on the text strings. A rule in the grammar is then generated automatically based in part on the n-grams.

Claims

1. A method of automatically constructing a grammar, the method comprising:

receiving sentences;

forming n-grams based on the sentences; and

a processor automatically generating a rule having a name attribute that identifies the rule in the grammar based in part on the n-grams, wherein generating a rule comprises generating a rule for a specific n-gram by including an entry in the rule to allow the specific n-gram to appear in a position where it did not appear in the sentences.

2. The method of claim 1 wherein forming n-grams comprises:

receiving at least one class definition comprising a semantic class and at least one text string associated with the semantic class;

replacing a text string in a sentence with a tag for a semantic class associated with the text string to form a modified sentence; and

forming n-grams from the modified sentence.

3. The method of claim 1 wherein forming n-grams comprises forming different orders of n-grams.

4. The method of claim 2 wherein generating a rule further comprises generating a separate rule for each specific n-gram found in the modified sentence.

5. The method of claim 1 wherein generating a rule for a specific n-gram further comprises placing an entry in the rule that includes a reference to a rule for the tag or word in the last position of the n-gram followed by a reference to a rule for at least one n+1-gram.

6. The method of claim 5 wherein the reference to a rule for the tag comprises a reference to a rule that lists all words associated with the semantic class of the tag.

7. The method of claim 2 wherein replacing a text string with a tag comprises replacing as many words as possible in the sentence.

8. The method of claim 2 wherein receiving a class definition comprises receiving a reference to a class defined in an existing grammar.

9. A computer-readable storage medium having computer-executable instructions encoded thereon for performing steps comprising:

receiving a list of text strings associated with a semantic class;

receiving at least one sentence in which a text string in the list of text strings appears with additional text; and

generating a grammar in a particular grammar format based on the at least one sentence and the semantic class by;

selecting the text string from the sentence;

comparing the text string to text strings that are listed for the semantic class;

determining that the text string exactly matches a text string listed for the semantic class;

replacing the text string in the at least one sentence with a tag for the semantic class associated with the text string to form a modified sentence comprising the tag and the additional text where the tag for the semantic class represents the multiple text strings associated with the semantic class;

forming n-grams using the modified sentence; and

generating rules for the grammar based on the n-grams.

10. The computer-readable storage medium of claim 9 wherein receiving a text string comprises:

receiving a text string parameter that identifies the text string;

receiving a semantic class parameter that identifies the semantic class; and

receiving a weight parameter that indicates the relative weight to be assigned to the text string in the semantic class.

11. The computer-readable storage medium of claim 9 wherein receiving at least one sentence comprises:

receiving a sentence parameter that identifies the sentence; and

receiving a weight parameter that indicates a weight for the sentence in the grammar.

12. The computer-readable storage medium of claim 9 wherein generating a grammar comprises:

receiving a grammar output file parameter that designates the name to be given to the file containing the grammar; and

receiving a grammar format parameter that designates the format for the grammar.

13. The computer-readable storage medium of claim 9 further comprising defining a semantic class.

14. The computer-readable storage medium of claim 13 wherein defining the semantic class comprises:

receiving a main semantic class parameter that provides a main semantic class for the semantic class;

receiving an indicator semantic class parameter that defines an indicator semantic class for the semantic class;

receiving an active node parameter that indicates whether the combination of the indicator semantic class and the main semantic class should be used as the default for the main semantic class; and

defining the semantic class based on the main semantic class parameter, the indicator semantic class parameter and the active node parameter.

15. The computer-readable medium of claim 14 wherein the main semantic class parameter comprises a reference to a semantic class in an external grammar.

16. A method of forming a grammar, the method comprising:

displaying a user interface that provides a text area for a user to enter sentences representing possible user responses and an area for associating text strings with a semantic class, at least one semantic class being associated with multiple text strings;

a processor forming modified sentences from the sentences entered through the user interface by performing steps for each sentence entered through the user interface, the steps comprising;

identifying a group of words in the sentence;

comparing the group of words to a list of the text strings associated with the semantic class;

determining that the group of words matches a text string in the list of text strings associated with the semantic class;

replacing the group of words in the sentences with a tag for a semantic class associated with the text string that matches the group of words while leaving other words in the sentence unchanged to form a modified sentence, the tag representing multiple text strings associated with the semantic class; and

using the modified sentences to form the grammar.

17. The method of claim 16 further comprising identifying n-grams of a plurality of orders in the modified sentences.

18. The method of claim 17 further comprising forming a separate rule in the grammar for each identified n-gram.

19. The method of claim 16 wherein using the modified sentences to form the grammar comprises providing rules in the grammar for word sequences that were not present in the sentences.