We describe an ongoing work in informationextraction which is seen as a text normalizationtask. The normalized representationcan be used to detect paraphrasesin texts. Normalization and paraphrasedetection tasks are built on top of a robustanalyzer for English and are exclusivelyachieved using symbolic methods.Both grammar development rules and informationextraction rules are expressedwithin the same formalism and are developedin an integrated way. The experimentwe describe in the paper is evaluated andpresents encouraging results.
展开▼
