connectionism (n.)

An application in LINGUISTICS of a computational framework for modelling cognitive functions, based on numerical computation rather than symbol manipulation. A connectionist network (or neural network) is devised which models the kinds of structures and processes thought to operate in the brain: the processing units in the network are called ‘neurons’ (in an abstract sense) or ‘nodes’, each being excited or inhibited (according to certain numerical formulae) by information obtained from the other units to which it is connected. The pattern of neuronal activity represents the data being processed by the network. A particular interpretation (e.g. of speech input data) is likely to depend on the activity pattern of a large number of related units (‘distributed representation’), the properties of which can be demonstrated only through statistical analysis. Because all the processing units compute at the same time, the approach is also known as parallel distributed processing. This approach contrasts with the view that people process sentences by TRANSFORMING REPRESENTATIONS according to a set of RULES, and rejects the notion that speakers INTERNALIZE grammars, in the GENERATIVE sense. Areas of application include the modelling of the non-DISCRETE and statistical properties of language use, and the study of language processing within PSYCHOLINGUISTICS, NEUROLINGUISTICS, and COMPUTATIONAL LINGUISTICS (e.g. automatic SPEECH RECOGNITION).