In asking what kind of model best fits the findings on reading by touch, three main factors have to be considered: perceptual processes, knowledge of the language, and knowledge of the orthographic rules and conventions which govern the translation between the sounds of language and the perception of the tactual patterns which symbolize them. The model must also be able to account for maturation and development, and specify how learning takes place.
The working model of reading by touch that I am proposing uses the overall metaphor of 'converging active processing in interrelated networks' (CAPIN for short) for human information processing (Millar, 1994). It stresses the inherent activity of the system which incoming patterns of impulses modify, and which also influences incoming patterns of activation. The notion of converging processes is needed to describe the intersensory nature of perception by touch. I used the description originally to account for the spatial organization of intersensory cues from touch, movement and posture in the absence of current external frame cues (Millar, 1981 a, 1994). The description applies to the initial perception of braille patterns, and the progressive spatial organization of scanning movements found here (Chapters 2 and 3). The metaphor of networks of interconnections also accounts for the variables involved in language acquisition. The connections between heard sounds, speech output, semantic and linguistic inputs become progressively more organized in converging networks. Incoming information that activates part of the system will activate other connections to some extent also.
I am also making assumptions about developmental processes. We know that the central nervous system is innately biased towards accepting and organizing some inputs more than others. The patterns of connections that serve hearing, speech and language must be assumed to have stronger connections with each other originally than with the patterns that serve the connections between touch and movement. At the same time, even 'dedicated' parts of the network have connections or potential connections with other parts of the network. There is increasing evidence that even inherently dedicated connections in the actual neural network can be