Attention: Theories of

Early and Late Selection

Inspired by analogies between humans and computers, the first of the modern theories of attention tried to explain how humans selected some sensory inputs for further processing while ignoring others. These theories also emphasized the extent to which sensory inputs could be processed without attention. Early selection theories claimed that the attention system acted like a filter that could select some inputs for additional processing based on visual physical characteristics such as location, color, or texture or based on auditory characteristics such as location, pitch, or the speaker's unique voice. Processing for identification or memory was claimed to occur only for attended-to inputs—unselected inputs were filtered out completely. At the other end of the continuum, late selection theories argued that considerable processing could be carried out on many inputs without limitations, and that the bottleneck in processing occurred just before the choice and execution of a response.

Initial tests of these theories were similar in essence to the “cocktail party” phenomenon, where a listener must pick out a single speaker in a room full of competing conversations. Experiments present different messages, sometimes by different speakers, in each ear (“dichotic” listening) using headphones. Listeners might be asked to immediately repeat (“shadow”) a target speaker, as the message is heard in one ear, while ignoring messages in the other ear. However, attention filtering was ultimately shown to be by no means complete or absolute. Unattended messages, such as those presented in the other ear, were at least partially processed, especially salient or high-valence messages, as in hearing one's name in the unattended message. People notice if a word in an unattended message is the same or related to the attended message, so long as they occur close together in time. Consistent with this, physiological measures have shown that the effects of attention seem to begin at the first stages of sensory processing. Event-related potentials (ERPs), or electric activity of the brain in response to visual inputs, show stereotypical differences in the responses for attended compared with unattended stimuli. Attention can modify the brain's responses to stimuli, amplifying relevant inputs or attenuating irrelevant ones, even the first and fastest responses. For example, the response over the visual cortex to an attended visual stimulus (“pay attention to targets in the upper right quadrant”) causes a larger response within the first 100 milliseconds (ms) in the P100 wave of the ERP, and this extends through early visual responses in cases where attention focuses on one location over longer blocks of many trials. Ignoring inputs does not, however, eradicate the sensory responses. In short, a modified form of the early selection [Page 108]model provides a good account: selection affects the earliest forms of sensory processing, but ignored inputs are attenuated rather than filtered out completely. The magnitude of the difference in response between attended and unattended inputs depends on exactly how attention is controlled. Attention has smaller effects on the earliest brain responses in cases where the focus of attention is changed to a new location by a cue on each trial. So, the impact of attention on perceptual processing can be more profound with a sustained period of attending to one location or feature than if selective attention is shifted moment to moment by changing cues.

...