There are several immediate objections
to chaining models.For example, how do
they cope with repeated items, where two
different responses will share the same
cue? Chaining models also have difficulty
handling erroneous responses, after which
the cue for subsequent items will be
incorrect, leading to a cascade of further
errors – after all, a ‘chain is only as strong
as its weakest link’. More sophisticated
models can overcome these problems by
including remote associations between
items, an approach exemplified by recent
neural network models (Elman, 1990).