Including statistical NLU models within a statistical parser
View Patent ↗A method and system for statistical parsing. The method involves a series of steps. The system can apply a statistical natural language understanding (NLU) model to test input for identifying substrings within the text input. The statistical NLU model can be selected for identifying a particular class of substring. The system can examine each identified substring using an inventory of queries corresponding to the reusable statistical NLU model.
1. A statistical parsing method comprising:
applying at least one statistical natural language understanding (NLU) model to text input for identifying substrings within said text input, said statistical NLU model belonging to a group of one or more statistical NLU models included within a statistical parser and selected for identifying a particular class of substring; and,
examining said each identified substring using an inventory of queries corresponding to said selected statistical NLU model.
2. The method of claim 1 , further comprising iteratively applying different selected statistical NLU models to said text input.
3. The method of claim 1 , wherein said step of applying a statistical NLU model further includes comparing a probability value, which corresponds to said identified substrings, to a threshold probability value, which corresponds to said selected statistical NLU model.
4. The method of claim 1 , wherein said selected statistical NLU model comprises a maximum entropy direct channel model, a source-channel model trained by the expectation maximization algorithm, a n-gram model, a statistical parser, or a word spotter.
5. The method of claim 1 , wherein said step of applying a statistical NLU model further comprises identifying a parse tree which corresponds to each said identified substring.
6. The method of claim 1 , said inventory of queries having a hierarchy determined during training of an NLU system.
7. A maximum entropy parsing method comprising:
applying at least one statistical natural language understanding (NLU) model to text input for identifying substrings within said text input, said statistical NLU model belonging to a group of one or more statistical NLU models included within a maximum entropy parser and selected for identifying a particular class of substring; and,
examining each said possible substring using one or more features corresponding to said selected statistical NLU model.
8. The method of claim 7 , further comprising iteratively applying different selected statistical NLU models to said text input.
9. The method of claim 7 , wherein said step of applying a statistical NLU model further includes comparing a probability value, which corresponds to said identified substrings, to a threshold probability value, which corresponds to said selected statistical NLU model.
10. The method of claim 7 , wherein said selected statistical NLU model comprises a maximum entropy direct channel model, a source-channel model trained by the expectation maximization algorithm, an n-gram model, a statistical parser, or a word spotter.
11. The method of claim 7 , wherein said step of applying a statistical NLU model further comprises identifying a parse tree which corresponds to each said identified substring.
12. The method of claim 7 , said features having weights determined during training of an NLU system.
13. A statistical parsing system comprising:
a text buffer for storing text input;
at least one statistical NLU model included within a statistical parser for recognizing a substring within said text input; and,
an inventory of queries wherein each query within said inventory of queries corresponds to one of said at least one statistical NLU models.
14. The system of claim 13 , wherein said statistical NLU model includes a threshold probability value for comparing each said recognized substring to said threshold probability value.
15. The system of claim 13 , wherein said statistical NLU model comprises a maximum entropy direct channel model, a source-channel model trained by the expectation maximization algorithm, an n-gram model, a statistical parser, or a word spotter.
16. The system of claim 13 , wherein said statistical NLU model can identify a parse tree corresponding to each said recognized substring within said text input.
17. The system of claim 13 , said inventory of queries having a hierarchy determined during training of an NLU system.
18. A statistical parsing system comprising:
a text buffer for storing text input;
at least one statistical NLU model embedded within a statistical parser for recognizing a substring within said text input; and,
one or more features wherein each feature corresponds to one of said at least one statistical NLU models.
19. The system of claim 18 , wherein said statistical NLU model includes a threshold probability value for comparing each said recognized substring to said threshold probability value.
20. The system of claim 18 , wherein said statistical NLU model comprises a maximum entropy direct channel model, a source-channel model trained by the expectation maximization algorithm, an n-gram model, a statistical parser, or a word spotter.
21. The system of claim 18 , wherein said statistical NLU model can identify a parse tree corresponding to each said recognized substring within said text input.
22. The system of claim 18 , each said feature having a weight determined during training of an NLU system.
23. A machine readable storage, having stored thereon a computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:
applying at least one statistical natural language understanding (NLU) model to text input for identifying substrings within said text input, said statistical NLU model belonging to a group of one or more statistical NLU models embedded within a statistical parser and selected for identifying a particular class of substring; and,
examining said each identified substring using an inventory of queries corresponding to said selected statistical NLU model.
24. The machine readable storage of claim 23 , for causing the machine to perform the additional step of iteratively applying different selected statistical NLU models to said text input.
25. A machine readable storage, having stored thereon a computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:
applying at least one statistical natural language understanding (NLU) model to text input for identifying substrings within said text input, said statistical NLU model belonging to a group of one or more statistical NLU models included within a statistical parser and selected for identifying a particular class of substring; and,
examining each said possible substring using one or more features corresponding to said selected statistical NLU model.
26. The machine readable storage of claim 25 , for causing the machine to perform the additional step of iteratively applying different selected statistical NLU models to said text input.
27. In a natural language understanding (NLU) system, a direct channel method for determining a meaning for a text input comprising:
applying at least one statistical NLU model to a text input, said statistical NLU model belonging to a group of one or more statistical NLU models included within a direct channel model;
identifying one or more substrings within said text input, each said substring corresponding to said statistical NLU model; and,
determining a meaning for said text input based upon said identified substrings from possible meanings within said statistical NLU model.
28. The method of claim 27 , further comprising:
comparing a probability value corresponding to each said substring to one or more threshold probability values within said statistical NLU model.
29. The method of claim 27 , further comprising iteratively applying different selected statistical NLU models to said text input.
30. The method of claim 27 , wherein said NLU model comprises a maximum entropy direct channel model, a source-channel model trained by an expectation maximization algorithm, a n-gram model, a statistical parser, or a word spotter.
31. The method of claim 27 wherein said determining step comprises applying weighted features corresponding to said identified substrings to said text input.
32. A machine readable storage, having stored thereon a computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:
applying a statistical natural language understanding (NLU) model to a text input, said statistical NLU model belonging to a group of one or more statistical NLU models embedded within a statistical parser;
identifying one or more substrings within said text input, each said substring corresponding to said statistical NLU model; and,
determining a meaning for said text input based upon said identified substrings from possible meanings within said statistical NLU model.
33. The machine readable storage of claim 32 , for causing the machine to perform the additional step of:
comparing a probability value corresponding to each said substring to one or more threshold probability values within said statistical NLU model.
34. The machine readable storage of claim 32 , for causing the machine to perform the additional step of:
iteratively applying different selected statistical NLU models to said text input.
35. The machine readable storage of claim 32 , wherein said determining step comprises applying weighted features corresponding to said identified substrings to said text input.