Exploiting Connectivity for Case Construction in Learning by Reading

One challenge faced by cognitive systems is how to organize information that is learned by reading. Analogical reasoning provides a method for immediately using learned knowledge, and analogical generalization potentially provides a means to integrate knowledge across multiple sources. To use analogy on learned material requires organizing information, represented in predicate calculus, into effective cases. This paper argues that using connectivity in semantic interpretations to organize knowledge learned by reading into overlapping cases can support analogical reasoning with these structures. We describe two connectivity-based methods and compare their performance with two baselines for the task of comparing and contrasting topics included in material the system has read.