Cognition, development and language

Up  Home Page

The following papers on the above subject are available on this Web site:

This paper, given originally as early as 1976 as a talk at an Artificial Intelligence conference, suggests that the question 'how easy is it to make the step from one kind of skill to a more advanced one?' plays a crucial role in defining the developmental process.

This paper describes a new approach to understanding the functioning of the nervous system, unifying previous ideas of Josephson and Hauser (see above), Baas (the hyperstructure concept), and Brooks (layered design of complex systems). Its basis consists in analysing the total developmental process into basic components of development, whose corresponding mechanisms (skill constructors) are organised together into a coherent total system.

It is argued that the constructivist neural network approach to modelling the mind, which tends to reject the idea of innate neural mechanisms adapted to specific problem domains in favour of the idea that expertise in a given domain is acquired by exposure of a general purpose neural network to problems in that domain, is confused. This confusion comes from a failure to appreciate the fact that complex systems may have structure at a number of levels (in this case, both the neuronal level and a level connected with behaviour of an algorithmic character). The old-fashioned algorithmic approach to the mind, on the other hand, failed because it was based on an out of date serial approach to computation and lacked the insights concerning computation found in modern approaches such as that of object-oriented programming. The real nervous system will have equivalents to the 'classes' utilised in languages such as C++ and Java which will guide its constructive activities, and will get some of its power by having built in mechanisms that can provide the equivalent to devices such as inheritance and threads.

This report, written in 1982 but not made generally available until the mid 1990's when the IPPE archive made publication possible over the Internet, applies the multistage development concepts of Josephson and Hauser to language. It sketches out what mechanisms might be behind the developing child's ability to discover step by step what the phenomenon of language is all about. It transcends the misleading but still commonplace idea that the process of language acquisition occurs in a kind of a vacuum where the question of which sequences of words are grammatical is about all that really matters. Instead, in the approach to language proposed here, the use that can be made of language is what attention is primarily focussed upon.

(684K PDF file, 400 K gzipped PostScript file). This project investigates a concept of cognitive development due to N. Baas, according to which development consists primarily of joining simple structures into more complex ones, and these into still more complex ones and so on; specific observer mechanisms note the behaviour of various combinations and adjust the interactions that do the joining accordingly. The idea is applied to a problem involving the attempt to stabilise an unstable one-dimensional system. The simulation integrates a collection of elementary movements into more complex ones, and it is found that long-term stability results only when a correctly designed observer mechanism is used.

Up  Home Page