Dr Oren Griffiths

Elemental models of associative learning typically employ a common prediction-error term. Following a conditioning trial, they predict that the change in the strength of an association between a cue and an outcome is dependent upon how well the outcome was predicted. When multiple cues are present, they each contribute to that prediction. The same rule applies both to increases in associative strength during excitatory conditioning and the loss of associative strength during extinction. In five experiments using an allergy prediction task, we tested the involvement of a common error term in the extinction of causal learning. Two target cues were each paired with an outcome prior to undergoing extinction in compound either with a second excitatory cue or with a cue that had previously undergone extinction in isolation. At test, there was no difference in the causal ratings of the two target cues. Manipulations designed to bias participants toward elemental processing of cue compounds, to promote the acquisition of inhibitory associations, or to reduce generalization decrement between training and test were each without effect. These results are not consistent with common error term models of associative learning.

The phenomenon of sensory self-suppression - also known as sensory attenuation - occurs when a person generates a perceptible stimulus (such as a sound) by performing an action (such as speaking). The sensorimotor control system is thought to actively predict and then suppress the vocal sound in the course of speaking, resulting in lowered cortical responsiveness when speaking than when passively listening to an identical sound. It has been hypothesized that auditory hallucinations in schizophrenia result from a reduction in self-suppression due to a disruption of predictive mechanisms required to anticipate and suppress a specific, self-generated sound. It has further been hypothesized that this suppression is evident primarily in theta band activity. Fifty-one people, half of whom had a diagnosis of schizophrenia, were asked to repeatedly utter a single syllable, which was played back to them concurrently over headphones while EEG was continuously recorded. In other conditions, recordings of the same spoken syllables were played back to participants while they passively listened, or were played back with their onsets preceded by a visual cue. All participants experienced these conditions with their voice artificially shifted in pitch and also with their unaltered voice. Suppression was measured using event-related potentials (N1 component), theta phase coherence and power. We found that suppression was generally reduced on all metrics in the patient sample, and when voice alteration was applied. We additionally observed reduced theta coherence and power in the patient sample across all conditions. Visual cueing affected theta coherence only. In aggregate, the results suggest that sensory self-suppression of theta power and coherence is disrupted in schizophrenia.

Previous studies of human associative learning have demonstrated that people's experience with a cueing stimulus will change how that cue is treated during subsequent learning. Typically, studies have shown that people pay more attention to cues that were informative in the past, and learn new information about these cues more rapidly (these cues are said to have a higher associability). It has recently been shown that to-be-predicted events (outcomes) can also differ in their associability as a consequence of prior experience. However, to date there is no direct evidence that this change in associability is accompanied by a change in attention, which would provide stronger evidence of a parallel with the effects observed previously with cueing stimuli. In two experiments, we examined this question by tracking eye-gaze to provide a measure of participants' overt attention, as they completed a cued visual search task in which outcome predictability was manipulated. The prior predictability of an outcome stimulus biased eye-gaze and learning rate, in a manner reminiscent of the gaze biases observed in tasks that manipulate cue associability. The present results support the view that outcomes, like cues, can vary in the degree to which they attract both attention and learning resources, as a function of their associative history.