The results from the model show that it is possible to broadly
simulate participant performance across different span lists on
the basis of an associative learning mechanism combined with a
constraint on the extent to which new chunks can be formed and
accessed for a given sequential input. However, to make clear that
greater span size for digits over words is not an artefact of the particular
parameter values implemented in the model, we carried out
span tests at different chunk capacities. Table 7 shows that
regardless of chunk capacity, span size for digit lists is consistently
larger than span size for word lists. Performance in the model
is consistent with the hypothesis that a greater amount of
instance-based, associative learning occurs for random digit
sequences than for random word sequences, and this accounts
for why span size for random sequences of digits is greater than
span size for random sequences of words. We provide a final
empirical test of this proposal in Experiment 4.
5. Experiment