Attention in Discrimination Learning: a Stimulus Control Approach to the Intradimensional-Extradimensional Shift Paradigm
In the present thesis, the intradimensional-extradimensional shift effect was treated as a problem of two-dimensional stimulus control. Factors determining stimulus control in the ID-ED shift were explored over six experiments. In Experiment 1 adult students were trained to discriminate between successively presented stimuli that differed in both line length and orientation. For half the subjects the length dimension was relevant (ie: different stimuli on that dimension were correlated with different outcomes) and for half the subjects orientation was relevant (phase 1). All subjects were then shifted to a second discrimination between new line lengths and orientations (Phase 2). For half, this constituted an intradimensional (ID) shift in that the previously relevant dimension remained relevant; for the remaining subjects the previously irrelevant dimension was made relevant in an extradimensional (ED) shift. The ID shift required significantly fewer trials to establish strong stimulus control by the relevant dimension in Phase 2 than did the ED shift. Experiments 1 and 2 further established that such differences were not attributable to a dominance relationship between dimensions or to specific cue values. Experiment 3 examined the development of stimulus control by the two dimensions over trials in Phase 2. In the ED shift, two-dimensional generalisation gradients showed a systematic weakening and strengthening of control by the Phase 1 relevant and Phase 2 relevant dimensions respectively. In the ID shift there was no change in stimulus control by either dimension. Experiment 4 established that transfer to the orientation dimension following differential training on length (ED shift) was superior to orientation following non-differential training on length (PD shift). Learning that differences on an extradimensional dimension were relevant in Phase I therefore had a facilitatory effect on control by orientation. Experiments 5 and 6 investigated the effects of manipulating the number of cues on the irrelevant Phase 1 dimension (orientation) and/or the irrelevant phase 2 dimension (length), in an ED shift where orientation was relevant in Phase 2. Both orientation and length (Experiment 5) or orientation alone (Experiment 6) were varied in the generalisation test. The ED shift in Phase 2 was retarded by the irrelevant dimension in Phase 1. It was concluded that in general the phase I relevant dimension must lose control in Phase 2, and the phase 1 irrelevant dimension must gain control in Phase 2 (Experiment 3). However, the inverse relation between loss of control by one dimension and gaining of control by the other does not occur in a way consistent with the Inverse Hypothesis of some selective attention theories. In addition, the previously relevant dimension in an ED shift facilitates control by the new relevant dimension in phase 2 re1ative to non-differential training, consistent with attentional enhancement. The major factor found to be slowing down the development of control by the new relevant dimension in an ED shift is the presence of the irrelevant dimension in Phase 1, (Experiment 5). This is probably a 'learned irrelevance' effect.