Much of our research has focused on the formalization of inductive inference processes and on the mathematical and philosophical foundations of inductive learning. We began rsuch work some years ago in connection with a specific problem of (formal language) learning and learnability. There we sought to develop a learning technique applicable to any member of a particular (linguistic) knowledge class. The class included both finite and infinite elements: successful learning in either case required characterizing the knowledge by finite means. The solution to that specific learning problem employed techniques of inductive inference to discover learnable models of (possibly infinite) knowledge from suitable, finite knowledge samples. The key to establishing leamability was establishing the existence of a finite information sample from which a model for an entire body of knowledge could be found. When we determined a suitable sample that would lead a learning system to discover a correct result, we dealt with the issue of relevance.