ABSTRACT

In this chapter, I outline the historical overview of the development of connectionism–how it came about, how it waned temporarily, and how it came back in 1980s. The attraction of connectionism is its ability to deal with various cognitive functions which the traditional symbolic approach has difficulty handling. In particular, connectionist networks have flexibility and respond to partial system failure with some degraded performance, which is human-like (graceful degradation), while symbolic models, unlike humans, completely break down if part of the system fails (brittleness). Connectionism involves neuron-like processing units that go through training (i.e. learning) and in the case of distributed models, concepts are represented over many units and are acquired through iterated exposure that results in increase/decrease in connection strengths among units. Patterns of association/connection constitute knowledge representation and their change constitutes learning.

The connectionist emphasis on representational change has brought back focus on ‘learning’ in cognitive science; while traditional symbolic models tended to focus on understanding the steady state of knowing something, connectionists are more interested in developmental change. After 25 years since the PDP volumes were published, connectionism has dramatically changed how cognitive scientists view our cognition as well as our methodology in approaching cognitive science, including integration of neuroscience as a major area of focus, and the connectionist network has formed an integral part of our new understanding of brain function (‘computational neuroscience’). With respect to generative vs. functional approaches to language acquisition, connectionism is more compatible with a functional, cognitive, usage-based linguistic approach to developmental psycholinguistics, with its emphasis on learning via input processing, in which input frequency is very important, without necessarily presupposing innate, language-specific constraints such as universal grammar.