How do we remember the order of a novel sequence of items, such as the digits in a telephone number? This problem is addressed by eight experiments on serial recall of temporal sequences. These experiments are used to develop a new model of short-term memory for serial order, the Start-End Model (SEM).
Existing approaches to the problem of serial order include chaining, positional and ordinal theories. Chaining theories, which store order in a chain of interitem associations, face problems explaining the pattern of errors in serial recall (Experiment 1). Positional theories, which store order by associations between items and their positions, are consistent with these errors, particularly those between sequences that maintain their position within a sequence (Experiments 2 and 3). Such positional errors cannot be explained by ordinal theories.
Previous positional models cannot explain the detailed pattern of errors however, as revealed by a meta-analysis of serial recall experiments. These errors are summarised as a set of empirical constraints, which are used to develop SEM, a computational model that simulates serial recall. SEM assumes (a) position is coded relative to the start and end of a sequence, (b) these codes are stored together with items as position-sensitive tokens, and (c) items are retrieved in order by cuing with codes for each position. SEM produces excellent quantitative fits to data from Experiments 1, 2 and 3.
SEM predicts that errors between sequences of different length will maintain position relative to the end of those sequences, in contrast to other positional models, which predict that such errors will maintain absolute position. SEM's predictions are confirmed in Experiments 4 and 5. SEM also predicts effects of repeated items in serial recall, which are examined in Experiments 6, 7 and 8. These complex effects of repetition pose important challenges for models of serial recall.
SEM is a model of short-term memory. However, the associated theory of relative position extends beyond short-term memory, and is a promising approach to the problem of serial order in general, particularly with respect to temporal order in episodic memory.