Rumelhart, a psychologist at Stanford University, contributed to cognitive models of learning in artificial intelligence (with cognitive scientist Donald Norman, 1978) and parallel distributed processing (with fellow psychologist Jay McClelland, 1986).
In their landmark 1978 book, Rumelhart and Norman identify three separate phases of learning: accretion, restructuring, and tuning. Restructuring is of particular interest to second language researchers.
Founders of connectionist theory, which models the working of the brain as connections between neurons in neural networks, Rumelhart and McClelland (1986) contend that learning occurs not serially - one connection after another - but in parallel - many connections at the same time. Their key 1986 experiment involved training a computer to produce past tense forms for English verbs.