We report an application for extending a fuzzy set with new information to improve recognition rates in a sign language recognition system. The fuzzy sets in the rule base are provided by experts, based on linguistic models of sign languages, which are then extended by fuzzy sets estimated from actual data. The extension algorithm unites an initial knowledge entity and a piece of new information, which is iteratively incorporated until convergence is reached. Experiments show that combining prior information and new evidence improves recognition rates beyond what can be achieved using either body of knowledge by itself.