Understanding Exploitations of Familiar Conceptual Metaphors: An Approach and Artificial Intelligence System

We present an approach to the reasoning needed to handle a class of metaphorical utterances, and a computer program (ATT-Meta) implementing that approach. The approach emanates from artificial intelligence research but is offered also for consideration by cognitive scientists generally. The class includes utterances that exploit familiar conceptual metaphors by going beyond the between-domain mappings they involve. The crux in dealing with such utterances is possibly-extensive inferencing in the terms of the source (vehicle) domain of the conceptual metaphor, and in, generally, avoiding attempts to find or create a targetdomain correspondent for a source-domain concept used in the utterance but not yet mapped to the target by the metaphor. The approach is similar in flavor to those of a small number of other metaphor researchers, but we present new claims and a more extensive implementation. Significant claims arise from our close attention to reasoning uncertainty; notably, we take a distinctive stance on conflicts between metaphorical inferences and target-domain information.