What is in a text, what isn't, and what this has to do with lexical semantics

This paper queries which aspects of lexical semantics can reasonably be expected to be modelled by corpus-based theories such as distributional semantics or techniques such as ontology extraction. We argue that a full lexical semantics theory must take into account the extensional potential of words. We investigate to which extent corpora provide the necessary data to model this information and suggest that it may be partly learnable from text-based distributions, partly inferred from annotated data, using the insight that a concept's features are extensionally interdependent.