Is shallow parsing useful for unsupervised learning of semantic clusters?

The context of this paper is the application of unsupervised Machine Learning techniques to building ontology extraction tools for Natural Language Processing. Our method relies on exploiting large amounts of linguistically annotated text, and on linguistic concepts such as selectional restrictions and co-composition. We work with a corpus of medical texts in English. First we apply a shallow parser to the corpus to get subject-verb-object structures. We then extract verb-noun relations, and apply a clustering algorithm to them to build semantic classes of nouns. We have evaluated the adequacy of the clustering method when applied to a syntactically tagged corpus, and the relevance of the semantic content of the resulting clusters.