Pattern recognition programs with level adaptation

This paper describes different types of patterning which exist in such diverse types of data as speech, text, and pictorial scenes. Several programs which process such data are discussed in terms of contextual, syntactic, semantic, and problem-domain dependent factors. Each of these factors constitutes a potential level of computer processing. By introducing different processing levels each having certain special characteristics (success on some data, inability to resolve other data) we isolate a general technique for producing a level adaptive program system.