Predicting Cognitively Salient Modifiers of the Constitutive Parts of Concepts

When subjects describe concepts in terms of their characteristic properties, they often produce composite properties, e. g., rabbits are said to have long ears, not just ears. We present a set of simple methods to extract the modifiers of composite properties (in particular: parts) from corpora. We achieve our best performance by combining evidence about the association between the modifier and the part both within the context of the target concept and independently of it. We show that this performance is relatively stable across languages (Italian and German) and for production vs. perception