All approaches to automatic pattern recognition or machine perception involve the initial assignment of a description to the pattern by the machine system in accordance with some predetermined rules or measurement procedures. The justification for adopting any particular set of rules or measurement procedures inevitably depends upon the uses to which the description is to be put. The main purpose of the lectures was to review systems of pattern description and to trace the evolution of "syntactic" descriptions.
Early property-list descriptions of patterns were motivated primarily by a view of the problem as that of assigning the pattern in tow to one of a finite number of classes. For tasks like solving picture puzzles calculating the properties of electrical circuits from their diagrams, "recognizing" twodimensional algebraic expressions, and so on, descriptions which exhibit the articular nature of the pattern are more appropriate. The similarity of this problem to that of the linguist concerned with assig~dng structural descriptions to sentences led many workers to formulate generative grammars as rule schemata by which to specify the range of descriptions to be assigned to a pattern.
However, generative grammars for a natural language are motivated by considerations of meaning as well as form. The principal thesis of ~hese lectures was that a similar concern for the meaning of a picture must also enter into the motivation for any syntactic descriptions to be assigned to that picture. By way of illustration, a brief account of a scheme for interpreting line diagrams as pictures of three-dimensional scenes was presented. (A more Artificial Intelligence I (1970), 291--298