Computers in the study of learning: namer—A pattern-recognition system for generating sentences about relations between line drawings

THIS PAPER reports on a series of experimental programs which are used to recognize line drawings and the spatial relationships between them. At the pattern-recognition level the programs learn to associate a name with a generalized bit pattern representing a line drawing. At the relation-learning level, the programs abstract characteristics which relate to such spatial concepts as “above,” “left,” “thicker than,” etc. The names that have been learned in association with a drawing, and relation names which the program selects as true for the spatial relations between two drawings, are substituted into a simple generation grammar; variations of sentences that are true for the picture are then generated.