A Multinomial Model-Based Analysis of Bindings in Working Memory

Abstract

We examined how multi-featured objects are represented in working memory (WM). Four contrasting conceptual models exist with regard to this question: (1) a Pure Object-Based model, consisting of a single integrated object representation which is retrieved in an all-or- none fashion; (2) an Object-With-Feature-Failure model, consisting of a single integrated object representation in which after retrieving an object, access to its features can still fail; (3) a Pairwise-Feature-Binding model, in which features of an object are represented separately and are bound together in pairs—the assumption that location mediates feature binding is a special case of this model, in which object features are bound to each other only indirectly via their shared location; (4) a Hybrid model that includes integrated object as well as individual feature representations. We developed four multinomial process tree (MPT) models to test these theories. In three experiments, participants memorized multiple objects which were random combinations of three features. They were given one feature as a cue and prompted to report the other two features. The model-comparison analysis revealed that both Hybrid and Object- With-Feature-Failure models provide reasonable explanations for the data, whereas a Pure Object-Based or Pairwise-Feature-Binding model did not. We conclude that object features are not retrieved in an all-or-none manner, and that locations are not essential for maintaining the binding of non-location features.Keywords: working memory, binding, multinomial modeling