In the early 19th century, Goethe placed a prism against his eye and saw coloured fringes along object boundaries (see Bouma, 1971), where the fringes seen on one side of objects (e.g., left) were complementary to those seen on the other side (i.e., right). The same observations were made by an ophthalmologist later in the century (Macfarland, 1883), who also reported that the illusory colours faded and disappeared within a few weeks. The adaptation to boundary colours was left unexplained until Kohler (1951/1970) undertook the first systematic study of the phenomenon. Based on his findings, he argued that contingent colour adaptation arose from the formation of associations between edges and colour. The empirical studies of Kohler (1951/1970) are long forgotten by all but a few specialists, however his views on contingent colour adaptation remain current in the literature. The tradition he began is continued in associative learning accounts of a more common example of contingent colour adaptation, the orientation-contingent colour aftereffect (CAE) demonstrated by McCollough (1965a).
McCollough's studies were undertaken as a response to her dissatisfaction with Kohler's account of contingent colour adaptation. Her report to Science presented evidence to support her view that CAEs arise from the 'colour adaptation' of edge detectors in the visual cortex. McCollough's account became pre-eminent throughout the 1960's and 1970's, however more recent articles report evidence that seems to vindicate associative learning accounts of contingent colour adaptation. Much of this evidence is based on comparative studies designed to emulate the protocols of experiments that established the basic properties of associative learning. These studies claim that CAEs and conditioned responses (CRs) exhibit similar properties under variations in the temporal parameters of the training session (Murch, 1976; but see McCarter & Silver, 1977), variations in the correlation(s) between edges and colours (see Siegel et al., 1992; but see Skowbo & Forster, 1983; Siegel & Allan, 1987) and exposure to unreinforced CSs before (i.e. latent inhibition) or after (i.e. extinction) training (see Skowbo, 1988).
None of the studies discussed above address the essential difference between associative learning accounts of CAEs and the proposals forwarded after McCollough (1965a), which is concerned with the a priori relationship between edges and colour. McCollough's original proposal suggested that edges and colour are already associated in the neural mechanisms (i.e., double-duty cells, see Micheals, 1978). In contrast, associative learning accounts of presume that the mechanisms coding edges and colour are dissociated, and only become associated after the stimuli have been paired repeatedly. Issues concerning the coding of stimulus events are difficult to address in psychophysical studies, however the reports of blocking by Westbrook and Harrison (1984) seem consistent with the position on stimulus coding assumed by associative learning accounts of CAEs. This study found that pairing, say, horizontals with red pre-emptively 'blocks' the formation of associations between left-obliques and red in a following training session. That is, CAEs elicited by the left-obliques, if observed at all, are weaker than otherwise expected. Not only is this phenomenon found in other forms of learning, but it also suggests that the association between an edge CS and a colour US maintains the dissociation between other edges and the same colour.
The current thesis develops an alternative interpretation for reports of 'blocking' in relation to CAEs based on the observation that they share an important characteristic with a phenomenon that has recently gained prominence; namely, the 'indirect' orientation-contingent CAE. The indirect CAE is similar to the direct form of the illusion reported by McCollough (1965) except it is seen on test edges that are orthogonal to those viewed during training (see Dodwell & Humphrey, 1990). The main argument of the present thesis is simple: If viewing a coloured grating gives rise to CAEs on differently-oriented test edges, then the procedure would also affect the outcome of pairing the edges with colour in a subsequent training session. An advantage of this proposal is that it provides a unified account for reports of 'blocking' and 'indirect' CAEs, findings that have been treated as separate phenomena in the past (e.g., Siegel & Allan, 1992). However, the proposal rests on some controversial assumptions about the nature of indirect CAEs (see Eissenberg, Siegel & Allan, 1995). Accordingly, the experiments reported in the present thesis investigate the specific nature of indirect CAEs, along with their role in reports of blocking.
The experiments reported in Chapter 2 investigated whether the stimulus conditions that purportedly 'block' the acquisition of CAEs also yield reports of indirect CAEs. Experiment 1 demonstrated a form of 'reverse blocking' that is not consistent with accounts of 'blocking' in conventional associative learning theory. That is, inspecting a coloured grating in one session appeared to affect the apparent colour of edges that were paired with colour in a previous training session. Importantly, the findings of Experiment 1 also supported the main argument of the current thesis; the 'blocking' of CAEs may be explained as a summation of the direct and indirect CAEs generated in consecutive training sessions. The results of Experiment 2 replicated and extended those of Experiment 1. Taken together, the findings reported in Chapter 2 question the extent to which observations of 'blocking' can be considered prima facie evidence for associative learning accounts of CAEs.
The experiments reported in Chapter 3 tested the main argument of the thesis by attempting to find a blocking-related phenomenon under conditions that were uncontaminated by indirect CAEs. These conditions were implemented by designing a study in which observers viewed similar edge-colour contingencies in consecutive training sessions. Conventional associative learning theory suggests that robust edge-colour associations can be weakened if the edges were paired with a relatively weak colour US. While the procedure is novel, it is based on the same logic that led Rescoria (1970) to demonstrate overprediction. Experiment 3 tested this hypothesis but failed to demonstrate the expected effect. Experiment 4 investigated a potential artefact in the design of Experiment 3, however the results indicated that the design of Experiment 3 was sound. Experiment 5 investigated the logic of Experiment 3 by simulating the study with a software implementation of the Rescorla- Wagner (RW) model of associative learning. The results of this study verified the logic of the Experiment 3. The discussion of Chapter 3 discusses conceptual problems arising from operationalising the notion of 'US intensity' in relation to colour.
Based on the results reported in Chapter 3, one might conclude that the properties of CAEs are inconsistent with associative learning theory. However, the design of Experiment 3 has not been implemented in studies of human or animal learning, so it might not be a fair test for associative learning accounts of CAEs. Thus, the experiments in Chapter 4 returned to the main issue raised in the discussion of Chapter 2; namely, the nature of indirect CAEs. On the one hand, if the indirect CAE is a valid phenomenon, then one would conclude that at least some instances of 'blocking' involve the summation of direct and indirect CAEs generated during training. On the other hand, if the term 'indirect CAE' is a misnomer, one would conclude that the results reported in Chapter 2 were instances of blocking with an additional confound.
Experiment 6 tested associative accounts of indirect CAEs by manipulating the colour US presented during training. The results of Experiment 6 were not commensurate with an associative learning account of indirect CAEs; the illusion was reported when the colour US was absent during training. Experiment 7 investigated the spatial determinants of indirect CAEs by selectively adapting responses to a number of relevant stimuli before CAE training.
The results of Experiment 7 show that the indirect effect of viewing coloured edges are affected by pre-adapting with orthogonal edges. These results appear to confirm the account of CAEs proposed in the current thesis. According to this view, CAEs involve asymmetries in the strength of inhibition between units responding conjointly to colour and orientation. Direct CAEs involve increases in the inhibition between units that are stimulated during training, whilst indirect CAEs involve reductions in the inhibition between units that are suppressed by the training pattern. To verify this proposal, the procedures used in Experiments 6 and 7 were simulated with the aid of an artificial neural network (ANN). The outcomes of these simulations (Experiments 8 & 9) were consistent with the observations of human observers, and thus validated the model of CAEs proposed in the present thesis.
On one hand, the main findings reported in this thesis indicate that associative learning theory gives a misleading perspective on CAEs. On the other hand, the findings illusu-ate a role for associative principles in contingent colour adaptation; a role made explicit by the model proposed in Chapter 5. This model retains the principle of association in the form of modifiable (synaptic) links between neural mechanisms that code distinct stimulus events. However, the model makes assumptions about neural coding that differentiate it from conventional associative learning accounts of CAEs. As discussed earlier, conventional accounts assume that orientation and colour are dissociated in the neural code. The model outlined in Chapter 5 makes assumptions about stimulus coding more like those of McCollough, That is, neural mechanisms are said to encode combinations of stimulus properties (e.g., orientation, polarity and colour contrast), and CAEs involve associations in the activities of these mechanisms. These associations express the relative likelihood of specific edge events rather than the likelihood that a particular object (e.g., grating) is paired with a colour.