We present some novel machine learning techniques for the identification ofsubcategorization information for verbs in Czech. We compare three differentstatistical techniques applied to this problem. We show how the learningalgorithm can be used to discover previously unknown subcategorization framesfrom the Czech Prague Dependency Treebank. The algorithm can then be used tolabel dependents of a verb in the Czech treebank as either arguments oradjuncts. Using our techniques, we ar able to achieve 88% precision on unseenparsed text.
展开▼
