Training 515Figure 18.4 Performance of the control and componential-training groups
on the Space Fortress game. Source:Frederiksen and White (1989).
developed and evaluated. If a task consists of components
with clear starting and stopping points, it can simply be seg-
mented into the different components. If the last step in a seg-
mented task is practiced first, with earlier components added
later, the procedure is called backward chaining. Whether
backward chaining is more effective than forward chaining,
in which segments are trained sequentially, starting with the
first one, will depend on the type of feedback needed for per-
formance. For complex tasks in which the initial steps are far
removed from the goal, there might be a benefit for backward
chaining because this begins by emphasizing the steps closest
to the goal. When feedback from one component influences
performance on the next, forward chaining might be more
effective (Wightman & Lintern, 1985).
Marmie and Healy (1995) showed that the benefits of part-
task training using a segmentation and backward-chaining
strategy can show long-lasting effects in a simulated tank-
gunnery task. In the relevant experiment, participants prac-
ticed either the whole task (searching for a target, sighting
it, and firing) or, for several sessions, only the sighting and
firing components. Performance in whole-task retention ses-
sions given immediately after training or one month later
showed no difference between the groups in overall perfor-
mance (proportion of kills) or in time to identify the target.
However, the part-task training group, which was able to de-
vote more resources to the sighting and firing components of
the task during training, showed a long-lasting benefit in time
to fire.
If different task components are performed in parallel, it
is not possible to segment them. In this case, we speak of
fractionationof the task. This involves practicing some com-
ponents, such as perceptual skills, in isolation and then
combining them with other aspects of the task, such as
making responses. It has been argued that fractionation can
only be effective if there is relatively little time sharing or
interdependence between the components (W. Schneider &
Detweiler, 1988). In some cases, such as when multiple-task
components must be carried out in parallel, the demands
imposed by the need to recombine the separate skills counter-
act any benefits of part-task training. However, the view that
part-task training is ineffective for tasks that must be time
shared may be overly pessimistic.
Fabiani et al. (1989) compared the hierarchical training
tasks developed by Frederiksen and White (1989) with
whole-task training and with a so-called integrative training,
in which the whole task was practiced, but performers were
instructed to emphasize certain of the skills identified by
Frederiksen and White. If time sharing must be practiced in
order to be learned, one might expect better performance in a
whole-task transfer condition for the integrative- than for the
hierarchical-training group. However, although the integra-
tive group showed more learning than a control group who
practiced the whole task under normal instructions for the
same amount of time, they did not do any better than the hier-
archical group. A possible benefit for the integrative group
was, however, found when a variety of secondary tasks were
added to the game. The integrative group proved to be better
in coping with these new task demands.
Another method of training is to simplify the task, teach
the simplified version, and then release the constraints placed
on the task until the task is restored to its original complexity.
This method has been used successfully in teaching the use
of software and has led to the concept ofminimal training.
Carroll (1997) argues that step-by-step manuals and com-
puter tutorials are often frustrating and ineffective because
they do not match the way people approach learning.
According to Carroll, learners want to get started fast,
which often leads them to omit critical steps, and neglect
to plan tasks or predict the outcomes of their explorations.
They also prefer not to follow procedures, often reason from
inference—even when the similarity to the current situation
is only superficial—and, finally, are often poor at recogniz-
ing, diagnosing, and recovering from errors. Recognition
of these characteristics of learners led Carroll (Carroll &
Carrithers, 1984) to develop a training wheels interface for a
word processor that restricted what learners could do and,
hence, the errors that they could make. They found a substan-
tial benefit for the use of the training wheels interface on
transfer to the full word processor. They attributed the benefit
to the fact that training wheels users spent less time on error
recovery and more time learning useful tasks. The lessons
to be learned from Carroll and his colleagues’ work on
minimalist training (summarized in Carroll, 1997) are that