Zones of proximal development for models of spoken word recognition

Computational accounts of spoken word recognition have largely relied on the TRACE model (McClelland & Elman, 1986) and simple recurrent networks (SRNs) (Elman, 1990). It is a testament to their explanatory power that, 25 years later, these models remain the dominant computational frameworks for spoken word recognition. These models have been so successful that they have largely defi ned the agenda for models of spoken word recognition, but there are now welldocumented, important phenomena that are relevant to spoken word recognition and yet are (largely) not addressed by models of spoken word recognition. The goal of this chapter is to look forward from these models. Detailed reviews of spoken word recognition (e.g., Magnuson, Mirman, & Myers, 2013) and comprehensive historical reviews of models of spoken word recognition (e.g., Magnuson, Mirman, & Harris, 2012; Weber & Scharenborg, 2012) are available elsewhere; I will provide only a basic introduction to the computational problem of spoken word recognition and the architectures of the TRACE model and SRNs, then focus on a few important behavioral phenomena that these models have addressed and a few challenges for these models. The challenges take the form of neighboring domains that are important for understanding spoken word recognition and where a substantial behavioral literature has accrued but where computational accounts are still lacking. Other models offer more focused accounts of phenomena either within spoken word recognition or in related domains, but none offer a substantive account of spoken word recognition that also addresses the challenges discussed in this chapter. I will conclude with some suggestions for expanding the scope of models of spoken word recognition through the integration of computational accounts of spoken word recognition and other domains.