Enriching WordNet with Folk Knowledge and Stereotypes

The knowledge that is needed to understand everyday language is not necessarily the knowledge one finds in an encyclopedia or dictionary. Much of this is “folk” knowledge, based on stereotypes and culturally-inherited associations that do not hold in all situations, or which may, strictly speaking, be false. We can open a linguistic window onto this knowledge through simile, since explicit similes make use of highly evocative and inference-rich concepts to ground comparisons and make the unfamiliar seem familiar. This paper describes a means of enriching WordNet with commonly ascribed cultural properties by mining explicit similes of the form "as ADJ as a NOUN" from the internet. We also show how these properties can be leveraged, through further web search, into rich frame structures for the most evocative WordNet concepts.