Topological stochastic grammars

Abstract

King Sun Fu (1982) has discussed the need for recognising that human pattern recognition is syntactic. When the human eye reads a character, there is evidence that the character is decomposed into 'strokes' and that the relationship between the components is coded in some way. Similar observations have been made by Pavlidis (1977) and others. The program inaugurated by Fu demanded the use of string grammar inference and an arguably premature discretisation process, and has foundered on both technical theoretical and practical grounds. We argue that we can pursue Fu's quest by formulating a theory of topological stochastic grammars, in which the discretisation arises from the data, and where the model is a continuous one.<>