Using a first natural language parser to train a second parser

• US 7,970,600 B2
• Filed: 11/03/2004
• Issued: 06/28/2011
• Est. Priority Date: 11/03/2004
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Wu, Tianlei, Chen, John T., Zhou, Ming
• Primary Examiner(s): YEN, ERIC L
• Application Number:
  US10/980,646
• Publication Number:
  US 20060095250A1
• Time in Patent Office: 2,428 Days
• Field of Search: 704/9, 704/1, 704/10
• US Class Current: 704/9
• CPC Class Codes: G06F 40/205 Parsing

Abstract

A computer-implemented method for developing a parser is provided. The method includes accessing a corpus of sentences and parsing the sentences to generate a structural description of each sentence. The parser is trained based on the structural description of each sentence.

Claims

1. A computer-implemented method for developing a parser, comprising:

accessing a corpus of sentences;

parsing the sentences with a computer to generate a structural description of each sentence using an existing parser, each structural description identifying noun phrases and a verb phrase from its corresponding sentence;

transforming each structural description to a skeleton relation that only identifies one relationship between one subject and one verb in the skeleton relation'"'"'s corresponding structural description'"'"'s corresponding sentence, and one other relationship between one subject and the one verb in the skeleton relation'"'"'s corresponding structural description'"'"'s corresponding sentence;

training another parser based on only the skeleton relations transformed from each structural description for use in identifying skeleton relations from sentences independent from the corpus of sentences;

utilizing said another parser to parse another corpus of sentences.

2. The method of claim 1 and further comprising using part-of-speech tags to generate the structural descriptions.

3. The method of claim 1 wherein the existing parser is a hand-crafted parser.

4. The method of claim 1 and further comprising generating a maximum entropy model to identify skeleton relations in a sentence outside of the corpus of sentences.

5. The method of claim 1 wherein the noun phrases comprises a first noun phrase including a subject and a second noun phrase comprising an object wherein transforming includes replacing a link in the structural description between the first noun phrase including the subject and the second noun phrase including the object with two links, one link between the subject and a verb and the other link between the object and the verb.

6. A non-transitory computer-readable storage medium having instructions that develop a parser for use in natural language processing, the instructions comprising:

an existing parser comprising a statistical parser that parses a corpus of sentences and generates a structural description of each sentence, each structural description identifying noun phrases and a verb phrase from its corresponding sentence

a training module transforming each structural description to a skeleton relation that only identifies one relationship between one subject and one verb in the skeleton relation'"'"'s corresponding structural description'"'"'s corresponding sentence, and one other relationship between one subject and the one verb in the skeleton relation'"'"'s corresponding structural description'"'"'s corresponding sentence; and

training another parser based only on the skeleton relations transformed from each structural description for use in identifying skeleton relations from sentences independent from the corpus of sentences.

7. The non-transitory computer readable storage medium of claim 6 wherein the training module is adapted to use part-of-speech tags to generate the structural descriptions.

8. The non-transitory computer readable storage medium of claim 6 wherein the existing parser is a hand-crafted parser.

9. The non-transitory computer readable storage medium of claim 6 wherein the training module utilizes a maximum entropy model to identify skeleton relations in a sentence.

10. The non-transitory computer readable storage medium of claim 6 wherein the training module is adapted to replace a link in the structural description between a subject and an object with two links, one link between the subject and a verb and the other link between the object and the verb.

11. A computer-implemented method for developing a parser, comprising:

accessing a corpus of sentences;

parsing the sentences with a computer to generate a structural description of each sentence using an existing parser, each structural description identifying noun phrases and a verb phrase from its corresponding sentence;

transforming each structural description to a corresponding skeleton relation that only identifies one relationship between one subject and one verb in the skeleton relation'"'"'s corresponding structural description'"'"'s corresponding sentence, and one other relationship between one subject and the one verb in the skeleton relation'"'"'s corresponding structural description'"'"'s corresponding sentence by replacing a link in the skeleton relation'"'"'s corresponding structural description between a subject and an object with two links, one link between the subject and the verb and the other link between the object and the verb;

training another parser based on only the skeleton relations transformed from each structural description for use in identifying skeleton relations from sentences independent from the corpus of sentences;

utilizing said another parser to parse another corpus of sentences to identify skeleton relations therein.