In this article, I explore how connectionism might expand its role in second language acquisition (SLA) theory by showing how some symbolic models of bilingual and second language lexical memory can be reduced to a biologically realistic (i.e., neurally plausible) connectionist model. This integration or hybridization of the two models follows the principles of what philosophers of science call intertheoretic reduction. Such a reduction serves two important purposes: It expands the explanatory scope of the symbolic models and it explains how some features of these models can actually emerge through learning in neural systems. To this end, I present a connectionist simulation of experimental data and show both the general feasibility of such a reduction and the specific manner in which the salient phenomenological distinction between form and meaning may be an emergent product of cortical memory processes. I argue this intertheoretic reduction of the symbolic to the neural serves an important goal of SLA, as these neural models can provide the theory of learning lacking in symbolic models of SLA.