A Computational Model of Conceptual Combination

We describe the Interactional-Constraint (ICON) model of conceptual combination. This model is based on the idea that combinations are interpreted by incrementally constraining the range of interpretation according to the interacting influence of both constituent nouns. ICON consists of a series of discrete stages, combining data from the British National Corpus, the WordNet lexicon and the Web to predict the dominant interpretation of a combination and a range of factors relating to ease of interpretation. One of the major advantages of the model is that it does not require a tailored knowledge base, thus broadening its scope and utility. We evaluate ICON’s reliability and find that it is accurate in predicting word senses and relations for a wide variety of combinations. However, its ability to predict ease of interpretation is poor. The implications for models of conceptual combination are discussed