Rapid acquisition of long spatial sequences in long-term memory

Learning complex movement sequences requires an active, attentional
selection of the content that is learned. The selection mechanism
can not be investigated in classical stimulus-guided sequence
learning paradigms because it requires a movement sequence
production that is not triggered by external stimuli. In deferred
imitation learning the whole stimulus sequence is presented and
reproduction is started only after the presentation has ended. In
order to investigate how the selective control of the learning
process proceeds in natural learning situations and to investigate
all influencing parameters we developed a new paradigm in which
long sequences were learned by deferred imitation learning. In this
task a long sequence of stimuli was presented on a graphic tablet
and reproduced by manual pointing after the stimulus presentation
was finished. Since the sequence exceeded the capacity of working
memory because of its length it had to be reproduced and learned in
several trials. Therefore, an attentional selection was required
during learning. In our first study a method for evaluating
reproduction performance in the new learning paradigm was
developed. The assignment of reproductions to target positions
posed a major methodological difficulty. This problem was solved by
introducing an assignment algorithm that takes the order of
reproduction into account. The algorithm was explained, it was
further compared to an algorithm that performs a nearest neighbor
assignment and finally validated by a comparison to a human
operator assignment. The results showed that the assignment
algorithm is an appropriate method for analyzing long sequences of
pointing movements and is suitable for evaluating reproduction
performance and learning progress in deferred imitation learning of
long sequences. In the second study we investigated further how
long sequences of pointing movements are acquired. Long-term
retention tests showed that the sequences were retained for at
least two weeks in long-term memory. A transfer test showed that
the sequences were represented in an effector independent
representation. The distributions of pointing positions were
analyzed in detail in order to characterize the control signal of
the pointing movements. The analysis showed that position errors to
successive target positions were not dependent on the movement
direction and further, that directional error did not propagate to
reproductions of successive target positions. These results suggest
that end points rather than movement trajectories are memorized in
this learning task. Our third study evaluated the organization and
reorganization of the sequence representation in memory. The change
in sequence reproduction without intermediate presentations showed
that the remembered target positions drifted away from the initial
representation, where the target drift saturated after about 5
trials. The analysis of the drift direction of representations of
single target positions showed that there was no systematic drift
direction for single subjects. Further it indicated that the
representation did not drift to similar, but to different patterns
across subjects. In order to investigate whether sequences are
encoded in chunks or as single target positions we performed an
experiment in which two target positions in a well learned sequence
were exchanged. We analyzed the effect of the target exchange on
target positions neighboring the exchanged target position. The
target exchange effected neither the position nor the variance of
neighboring memorized target positions. These results support the
view that single target positions rather than chunks of target
positions are memorized. Thus our study suggests that the sequence
acquisition is guided by an active selection process which is able
to quickly acquire abstract movement plans. Our findings further
support the view that these movement plans are represented as
strings of independent, absolute target positions.