McClelland and Elman initially implemented TRACE as a C program that has been the basis of all TRACE research to date. There is ample evidence that information sources (acoustic cues, semantic context, etc.) are used to recognize words and the phonemes they contain. Indeed, as Cole and Rudnicky (1983) noted, these fundamental facts were described in the first experiments of Bagley (1900) more than 80 years ago. Cole and Rudnicky pointed out that the Trace model work has added clarity to these basic findings, but has not resulted in a theoretical synthesis providing a satisfactory account of these and many other fundamental aspects of perception of speech. The Trace model approach grew out of a number of initial ideas, some originating first from research into spoken language recognition (Marslen-Wilson) and Welsh, 1978; Morton, 1969; Reddy, 1976) and others arising from more general considerations of interactive parallel processing (Anderson, 1977; Grossberg, 1978; McClelland ,1979 )