Creating New Sandwiches From Old

Earl is a case-based reasoning system which works in the domain of sandwich recipes. It has a case base of 9,507 cases, each being a sandwich recipe which we scraped from the web. We show how Earl can compose new sandwiches from the ingredients of sandwiches that it retrieves from its case base. On each iteration, it inserts into the sandwich the candidate ingredient which results in greatest coherence. Earl computes coherence with reference to the cases in the case base. A preliminary experiment with human participants shows that Earl’s sandwiches are competitive with human-authored sandwiches.