| 摘要 |
The present invention discloses a method and apparatus for improving the performance of a natural language parser. During training, natural language inputs are processed such that statistical information is extracted, concerning the frequency of attachment of certain phrase types of interest to certain other host phrase types. In one embodiment, only unambiguous attachments are examined during training. During runtime this statistical information is used in cases where the attachment of some phrase type of interest is ambiguous. In such cases, the statistical model is used to evaluate the relative likelihood of the alternative attachments resulting in a good interpretation. A syntactic representation can be produced for each alternative attachment that the statistical model deems likely to result in a good interpretation.
|
