View article

Theories of cognition often assume that a single type of rep1esentation underlies knowledge.. Traditionally, most theories have assumed that amodal symbols provide unifo1m knowledge representation (e.. g.., Collins and Loftus 1975; Fodor 1975; Newell and Simon 1972; Pylyshyn 1984).. More recently, theories have adopted statistical representations (e.. g., McClelland et al.. 1986; O'Reilly and Munakata, 2000; Rumelhart et al.. 1986) Most recently, theories have proposed that knowledge is grounded in modal simulations, embodiments, and situations (e.. g.., Allport 1985; Barsalou 1999, 2008a; Damasio 1989; Glenberg 1997; Martin 2001, 2007; Thompson-Schill 2003), while other theories have proposed that knowledge is grounded in linguistic context-vectors (e.. g.., Burgess and Lund 1997; Landauer and Dumais 1997). Our theme in this chapter is that multiple systems-not just one-represent knowledge.. We focus on two sources of knowledge that we believe have strong empirical support: linguistic forms in the brain's language systems, and situated simulations in the brain's modal systems.. Although we focus on these two sources of knowledge, we do not exclude the possibility that other types are important as well In particula1, we believe that statistical representations play central roles throughout the brain, and that they underlie linguistic forms and situated simulations.. At this point, we are somewhat skeptical that completely amodal representations exist in the brain, for both theoretical and empirical reasons (Barsalou 1999, 2008a; Simmons and Barsalou, 2003), but we are open to compelling arguments othe1wise

We begin by …