Declarative Memory and Skill Acquisition 501stimulus-driven mode of performance that is not dependent
on conscious control. Posner and Snyder (1975) described
automatic processes as those that may occur “without
intention, without any conscious awareness and without in-
terference with other mental activity” (p. 81). A great deal of
research has been directed to exploring and confirming this
view of dichotomous processing modes. For example, W.
Schneider and Shiffrin (1977; Shiffrin & Schneider, 1977)
performed an extensive series of hybrid memory and visual
search experiments that seemed to support the idea that there
are two different modes of processing and that controlled
processing gives way to automatic processing if only enough
practice is given.
The view that controlled and automatic processing are qual-
itatively distinct has, to some extent, fallen out of favor. Within
the realm of visual search, where Shiffrin and Schneider
(1977) carried out their influential work supporting such a di-
chotomy, researchers now tend speak about the efficiency of
search, rather than pre-attentive and attentive search, and
the role of attention in processing remains present across
search types. Rather than considering it a form of processing,
Neumann (1987) describes automaticity as a phenomenon
arising from a conjunction of input stimuli, skill, and the
desired action. In his view, it is appropriate to speak of auto-
maticity when all the information for performing a task is
present in the input information (stimulus information avail-
able in the environment) or in long-term memory. This view is
not too different from Logan’s (1988, 1990), described above,
in which automatic processing is based on memory retrieval,
and attention forms the cues necessary for the retrieval pro-
cessing. Attention remains an important process even in highly
practiced tasks.
As will be discussed at more length in the section on train-
ing, automatic processing, as assessed by an apparent insensi-
tivity to attentional resources or demands, can develop with
learning when the right conditions are provided. The important
conditions seem to be the consistency of the discrimination
and interpretation of the stimuli, and the stimulus-to-response
mapping (W. Schneider & Fisk, 1982). The development of
automaticity can be shown for a range of tasks. The idea that it
depends more on consistency than on properties of the stimuli,
such as perceptual salience, is supported by the finding that au-
tomatic processing can also be produced by training with stim-
uli divided into arbitrary classes (Shiffrin & Schneider, 1977).
According to the instance theory, “attention drives both
the acquisition of automaticity and the expression of auto-
maticity in skilled performance” (Logan & Compton, 1998,
p. 114). Selected information enters into the instances that
come to drive performance, but ignored information does
not. Moreover, if attention is not paid to the right cues,
associations dependent on those cues will not be retrieved
(Logan & Etherton, 1994). Logan and Compton describe at-
tention as an interface between memory and events in the
world. The dependence of memory on attention means that
knowing (or learning) what to attend to is a critical compo-
nent in the development of skill. Other authors have empha-
sized that learning not to attend to irrelevant information is
also a component of skill acquisition.Learning to Ignore Irrelevant InformationOne hypothesis about how learning to ignore irrelevant infor-
mation contributes to performance changes with practice is
theinformation reduction hypothesis(Haider & Frensch,
1996). According to this hypothesis, performance improve-
ments can be attributed to learning to distinguish task-
relevant information from task-redundant (and, therefore,
task-irrelevant) information and then learning to ignore the
task-irrelevant information. Evidence for this hypothesis
comes largely from tasks in which participants verified al-
phabetic strings such as E [4] J K L. The task is to determine
whether the letters follow in alphabetic order, where the num-
ber in brackets corresponds to the number of letters left out of
the alphabetic sequence. In most conditions, the length of the
string was varied by changing the number of letters following
the digit, which always occupied the second position in the
string. If there was an error in the stimulus, the error was in
the number of letters that was skipped (e.g., E [4] K L M).
Early in practice, Haider and Frensch found an effect of
string length on performance, such that verification times
were slower when the number of letters after the number in
brackets was increased. With practice, however, the slope of
the function relating performance time to string length de-
creased. This finding suggests that participants in the study
learned that the extra letters were not important for the task
and should be ignored. Additional evidence for this hypothe-
sis was found in a transfer condition in which errors could
occur in the letters to the right of the gap (e.g., E [4] J K M).
Consistent with the supposition that participants learned to
ignore the extra letters during training, the error rate in de-
tecting these invalid sequences increased as a function of
practice. Haider and Frensch also showed that learning in this
task was not stimulus specific by demonstrating transfer from
one half of the alphabet to the other.
Haider and Frensch (1996) showed that learners were able
to distinguish relevant from redundant task information and
to limit their processing to the relevant information. They
also showed that learning to reduce the amount of informa-
tion that is processed takes time, developing over the course
of practice, and that this ability appears to be largely stimulus