When an object is more than a binding of its features: Evidence for two mechanisms of visual feature integration

Abstract

People spontaneously integrate the features of visual events into episodic structures that are reactivated if their ingredients match aspects of the current input. Feature integration has been attributed to either the detection of feature conjunctions or the ad hoc binding of feature codes (e.g., by neural synchronization). We report evidence suggesting that both kinds of integration mechanisms coexist. Replicating earlier findings, repeating one visual feature facilitated performance but only if other visual features were also repeated. However, this effect was more pronounced with real objects as compared to arbitrary combinations of shapes and colours. Moreover, the real-object effect was restricted to visual feature integration but did not extend to visuomotor integration, suggesting that the underlying mechanism subserves perception only. We suggest a dual-process feature-integration model that distinguishes between ad hoc binding, which operates on any possible combination of features alike, and conjunction detection, which selectively operates on familiar feature combinations.

Keywords:

• Object perception
• Event file
• Feature integration

Notes

¹More specifically, our idea was to get the hypothesized object representations in long-term memory involved—which required the use of stimuli that were likely to have memory representations—without letting them do the integration job on their own (i.e., without the need for ad hoc feature binding)—which required the use of feature combinations that were unlikely to be covered by these representations. Following Colzato et al. (2006), we thus used shapes of real objects (which should suffice to activate the memory representations) but presented them in colours that were unlikely to be part of the memory representation—using two colours that according to the findings of Colzato et al. are not associated with either of the two object shapes. Theoretically speaking, we expected that this manipulation would activate object representations and the corresponding conjunction detectors but still require ad hoc binding of the uncommon shape–colour conjunction. The former was considered to provide top-down priming of the latter with real objects but not with arbitrary feature conjunctions. As an example, facing a banana should activate a banana-related conjunction detector, which would lead to top-down priming of all identity-relevant features belonging to the present banana (i.e., to both the familiar shape and the in this case unfamiliar colour). This would prime the colour and increase the likelihood that it is being integrated. As the geometric shapes were not considered to have conjunction detectors linking them to particular colours, no top-down priming should occur for these shapes and present their colours.

²Effect sizes were computed by subtracting the mean RT for complete repetitions and alternations from the mean RT for partial repetitions (i.e., shape repetition and colour alternation or shape alternation and colour repetition). Note that this amounts to the interaction term corrected for possible main effects.