| Summary: | Some researchers have argued that the category learning literature is conclusive: people learn to generalise from past experiences to novel ones using multiple learning systems (Ashby & Maddox, 2011). To the extent that this claim is true, it is due in no small part to work investigating the predictions of the dual-system model COVIS (COmpetition between Verbal and Implicit Systems; Ashby, Alfonso-Reese, Turken, & Waldron, 1998). The work presented here investigates the evidence for this model. In Chapter 1, I describe the main features of the COVIS model and briefly review some of the evidence that supports this model. This section highlights two main problems with this literature. First, that many of the studies that have purported to support a dualsystems description of category learning have been found to be flawed when re-examined by independent researchers. This observation is explored more deeply in the next two chapters, where I attempted to reproduce the work of two studies argued to support the COVIS literature. In Chapter 2, I re-examined the work reported by Ashby, Maddox, and Bohil (2002) that looked at the effect of training type on category learning. In Chapter 3, I re-examined work reported by Spiering and Ashby (2008) that looked at the effect of training order on category learning. Both these chapters failed to find evidence for two systems of category learning. The second issue raised in Chapter 1 is that none of the studies cited in support of COVIS critically examined the fundamental assumptions of the model. More specifically, Chapters 4 and 5 looked at how participants complete the categorisation tasks used in this literature. In Chapter 4, I conducted experimental work to determine whether some category structures are learned implicitly, as argued throughout the COVIS literature. Then, in Chapter 5 I conducted several simulations to investigate an analysis used ubiquitously in the COVIS literature to determine the strategies participants use to complete categorisation tasks. This analysis is a critical manipulation check for all these experiments. However, I found evidence that the analysis systematically over-estimates the evidence for dual-systems. Furthermore, both these chapters found evidence to suggest that the evidence for the COVIS model can be explained without needing to assume that participants can learn implicitly. Finally, in Chapter 6 I bring these threads together to discuss the implications for the COVIS model.
|
|---|
