The implementation of later stages of information processing before the completion of earlier stages. For example, in retrieving the meaning of a printed word, a person may have to identify all the letters and match the word against an internal lexicon (2) in order to extract its meaning. According to discrete processing models, information is not passed on to the next stage until processing at the current stage is complete, but in a cascade processing model a word may, for example, be matched against items in the lexicon, by way of hypotheses or guesses, before all the letters have been identified, and conjectures about meaning may be made before a complete lexical match has been established. The concept was introduced in 1979 by the US cognitive scientist James L(loyd) McClelland (born 1948) and has been influential in connectionist models of information processing. See also parallel processing, serial processing.