In order to achieve the long-range goalof semantic interpretation of noun compounds,it is often necessary to first determinetheir syntactic structure. This paperdescribes an unsupervised method fornoun compound bracketing which extractsstatistics fromWeb search engines using aχ2 measure, a new set of surface features,and paraphrases. On a gold standard, thesystem achieves results of 89.34% (baseline66.80%), which is a sizable improvementover the state of the art (80.70%).
展开▼