A model for Free Recall
MetadataShow full item record
Abstract A model for free recall of words is presented and applied to representative free-recall experiments. According to the model, a subject's lexicon contains a representation for each word and for the association between each pair of words. Studying a word in a list strengthens that word's representation in the lexicon, but reduces the strength of previously studied words. In addition, while studying the list, the subject associates pairs of items and, thereby, strengthens the corresponding associations in the lexicon. A subject's efficiency at forming associations drops off as the number of pairs increases. Retrieval is initiated with a report cue. Selection of an item for report is based on its strength in the lexicon plus the strength of its association with the retrieval cue. Selecting an item for report changes the strength of its association with the cue in the lexicon. To test the model, parameters were obtained by fitting it to serial-position curves taken from the archival literature. The model predicted three additional dependent measures: order of report, items correct per trial, and the number of intrusions per trial. In addition, I applied the model to phenomena associated with the free-recall task and showed that it captures the list-strength effect, interference in part-list cuing, clustering with categorized lists, and distraction effects. The model itself does not change to capture these experimental data. Word lists are presented to the model according to the experimental protocol used in the original experiments, and the model captures output measures derived from the experimental data. The model demonstrates that simple mechanisms can capture a wide range of apparently complex behaviour if we allow for a large enough knowledge base. In effect, the complexity of behaviour in free-recall paradigms lies in the interactions in the lexicon, not in the complexity of the recall mechanisms.