Positional coding models represent an alternative to chaining accounts of serial order memory. Rather than linking items to one another, these models associate each item with a distinct position marker or positional code. At recall, the positional codes are reinstated in sequence, and each code cues retrieval of its associated item. This class of models naturally explains why transposition errors tend to involve nearby positions and why order can be recalled even when specific items are forgotten.
The Start-End Model (SEM)
Henson's (1998) Start-End Model represents positions using two opposing gradients: a start marker that decreases in strength across positions and an end marker that increases. The positional code for each position is a weighted combination of these two markers:
where S is the start marker, E is the end marker, and n is the list length. This generates a set of positional codes in which neighboring positions have similar representations, naturally producing the locality constraint on transposition errors. The model also explains the primacy and recency advantages in serial recall through the distinctiveness of endpoint positions.
Oscillator-Based Positional Codes
Brown, Preece, and Hulme (2000) proposed that positional codes are generated by a bank of oscillators at different frequencies. Each position corresponds to a unique pattern of oscillator phases. The similarity between positional codes follows a damped cosine function of positional separation, producing approximately local transposition gradients:
This approach connects to the neural basis of serial order, as oscillatory timing signals are ubiquitous in the brain, particularly in the hippocampal theta rhythm.
Competitive Queuing
Many positional coding models use competitive queuing as the output mechanism (Grossberg, 1978; Houghton, 1990). All items are activated in parallel according to their positional strengths, the most active item is selected for output, and that item is then suppressed (inhibited) to allow the next strongest item to be selected. This winner-take-all plus suppression mechanism produces the primacy gradient observed in serial recall and accounts for repetition and omission errors.
Advantages Over Chaining
Positional coding models account for several findings that challenge chaining models: (1) transposition errors respect position, with items migrating to nearby serial positions; (2) the fill-in effect, where skipping an item causes subsequent items to shift forward; (3) protrusion errors, where items from previous lists intrude at the same serial position; and (4) the ability to recall sequences from arbitrary starting points or in reverse order. These findings collectively provide strong evidence that serial order memory relies heavily on positional information.
An ongoing debate concerns whether positional codes represent ordinal position (first, second, third) or temporal position (the time at which an item occurred). Temporal grouping experiments, where pauses are inserted within sequences, suggest that both types of information are used: items are coded relative to both their ordinal position within a group and the temporal structure of the sequence.