FREUD'S CONTINUITY THESIS

Abstract

The motivating force for the present paper comes from a recent paperby Dell, Schwartz, Martin, Saffran, and Gagnon (1997), whose model isconnectionist and incorporates interactive activation, cascade processing,and feedback among the levels. It is a two-step/three-level computationalsystem of lexical retrieval for simulating picture naming. The three hierarchi-cal levels are semantic, lexical, and phonological. The semantic level com-prises the units that represent the concept of a pictured object. Test wordsfor naming in the Dell study were drawn from The Philadelphia NamingTest (Roach, Schwartz, Martin, Greweal, & Brecher, 1996). Dell et al. (1997)arbitrarily assumed (i.e., set the values) that each object would correspondto 10 mathematical units. The 10 units for each picture to be named connectto that object’s word node by ‘‘excitatory bidirectional connections’’ (p.805). Each word node connects to its phoneme nodes in a similar fashion.[See Dell et al. (1997, pp. 805–809) for the mathematical details of the linearactivation function equation and how the numerical values for connectionweights and decay rates are set.]As an example, a picture of a cat is presented (see Fig. 1). The picture isperceived by the visual system (outside Dell’s model). The 10 units (‘‘seman-tic nodes’’) are each given a ‘‘jolt’’ of activation of 10, for a total jolt of100. The metaphor jolt is a numerical value—not an electrical shock! Theword node for ‘‘cat’’ will be activated and will subsequently activate itscomponent phonemes /k/, /ae/, and /t/. Figure 1 demonstrates the lexicalassociates such as ‘‘dog’’ and ‘‘rat.’’ The former bears only a semantic rela-tion with cat, while he latter bears both a semantic and a phonological rela-tion with cat. Furthermore, Fig. 1 demonstrates phonological associativestructure in that rat and mat share the coda (final vowel1 consonant, the