Hierarchical Preferences in a Broad-Coverage Lexical Taxonomy

We investigate the problem of finding informative superordinates in a broad-coverage taxonomy of nominal concepts. We present results from a study which shows that speakers often exhibit strong preferences on what superordinate is more informative, together with a solid bias for specific classes. We then define the task of identifying the properties that characterize such concepts in the taxonomy as a ranking problem. We identify several such properties which are related to properties of basic concepts. While these properties provide accurate sources of information for identifying the most useful superordinate, their interaction remains obscure.