Chaining models of serial recall can produce positional errors

Abstract

A major argument for positional-coding over associative chaining models of immediate serial recall has been the high probability that an error from a prior list will appear in its correct serial-position, so-called “protrusions.” Here we show that a chaining model can produce protrusions if it includes three characteristics that have been incorporated into published chaining models: (a) a “start-signal” item is associated with all first list-items, (b) memory is not cleared following each list, and (c) the retrieval cue for each item is always the full non-redintegrated retrieved information, regardless of the response. The model covertly recalls all studied lists in parallel (weighted by recency), such that when prior-list items intrude, they predominantly occur at the correct output position. In addition to fitting prior protrusion data, we report two new data sets that question the ubiquity of the simple protrusion-dominance characteristic. These findings show that protrusions cannot falsify an associative basis for serial-order memory and speak to the plausibility of mixture models.