Using semantic information measures to evaluate learning strategies

Carnap (1) proposed measures for assigning probabilities as degrees of conformation to sentences in a formal language previously. This development in inductive logic permitted Carnap and Bar-Hillel (2) to outline semantic information measures (SIM) utilizing logical probabilities. Stapleton, Siegmann, and Coulter (3) demonstrated that certain learning experiments in both artificial intelligence and cognitive psychology can be characterized within the framework of languages specified by Carnap and Bar-Hillel. These experiments are called "dialogue experiments". The SIM can be employed to monitor information transfer during the experiment. This formalization also allows a dynamic representation of the problem space within the language.This research is extended here by a proof that specifies the relative efficiences of members of a class of learning strategies in dialogue experiments. These experiments concern the rote learning of associations among randomly matched predicates and individuals. The proof reveals a hierarchy of strategies that are ranked by rates of expected information gain relative to the SIM.