The report considers the effect that specific language descriptions have on the efficiency of pattern recognition and problem solving methods. The efficiency of a language for the description of a given set is viewed in terms of the size, in some sense, of the shortest expression that denotes the set. Central to the discussion are questions of how the description of a concept should be stored to use the smallest amount of memory, and how the description should be stored and processed so that, given an object and a concept, an efficient determination can be made on containment of the object in the concept. The languages are essentially non-numeric, enabling pre-processing to be described in the same format as that used for pattern description. Algorithms for learning and generalization are presented that use two of the languages. The property of succinctness is considered for the algorithms, and the effect of lack of succinctness on the statistical degree of confidence in the learned description is indicated. Analogies are made to descriptions in terms of discriminant functions and maximum likelihood ratios.