Abstract
Intrusions are intriguing errors in which subjects report that a feature of a distractor stimulus belonged to a target. These errors can occur in a specific type of rapid serial visual presentation (RSVP) experiment in which all stimuli carry two features. A stimulus-stream in such an experiment could consist of coloured letters and a typical task would be to report the identity of the stream’s red letter. In this context, the feature that distinguishes target- from distractor-stimuli is called the key-feature (e.g. colour) and the feature that must be reported after each trial is called the response-feature (e.g. identity). Intrusions happen when, instead of the response feature of the target, that of a temporally neighbouring distractor is reported. For example, if the stream consisted of a blue S, a red E and a black P, and the task was to report the red letter, subjects would regularly report seeing the P as being red. Explanations of intrusions in previous research have typically only been informal in nature and the models available do not offer complete explanations. We present a computational model, which replicates response distributions of several experiments, for example after different empirical manipulations of key- & response-feature processing speeds. Our model additionally replicates counterintuitive reaction-time patterns and very recent Event Related Potential findings associated with the N2pc and P3. We further demonstrate, empirically, that the N2pc and P3 are temporally correlated components, which supports a hypothesis implied by our model-architecture. We propose that illusory intrusions are the result of mis-bindings of stimulus features to working-memory encoding episodes. The likelihood of intrusions is dependent on the relative timing between key- and response-feature processing, which opposes models that proposed correct percepts and intrusions are processed in a qualitatively different manner. Thus, in our model, correct bindings are just “fortunate” conjunctions.