attention. what is attention? "everyone knows what attention is. it is the taking possession...
TRANSCRIPT
Attention
What is attention? "Everyone knows what attention is. It is the
taking possession by the mind in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought...It implies withdrawal from some things in order to deal effectively with others."
William James
What is attention? Though we may talk about it in the following
ways… “I can only do so much at once!” “I missed the play on the field, I wasn’t paying
attention.” “Pay attention to me!” “I need to focus.” “She has attention deficit disorder.”
Are they all referring to the same thing?
What is attention? What was first viewed as simple bottlenecks is really
much more complex Attention determines how information is processed
by the cognitive system… …and vice versa
Cognitive Control of Attention Limits to attention may be processing limits In some ways however we really don’t know much
more about it than folk theories. So what do we know?
Attention What is Attention?
Attention is the control of sensory input and cognitive resources Input = how much/ what gets in Control = what guides attention
Internal External
Other distinctions Task-defined and Maintenance of activity
Descriptive notion of attention defined by the task used in the study (e.g. if they answer a certain way they are attending to that stimulus), or continued action
Doesn’t speak to underlying mechanisms Process-oriented
Attention as a psychological process Involves selection among alternatives and improving the
effectiveness of mental processes
Other distinctions Perceptual Attention vs. Attention in
Complex Tasks Perceptual
Attention involved in the enhancement and selection of stimulus input from the environment
Complex Tasks Attention used for non-automatic tasks and task
selection
Five functions of attention (from Medin) Perceptual
1. Focusing
2. Perceptual enhancement
3. Binding
Complex Tasks4. Automaticity (sustaining behavior)
5. Task Selection
Paradigms of Attention Research Cuing
Analyze the orienting process and comparison of processing attended vs unattended stimuli
Search Analyze how attention eliminates interference from
irrelevant stimuli Filtering
Analyze how attention eliminates interference from irrelevant stimuli and the stages at which such stimuli are suppressed
Dual-Task Analyze how attention is involved in the coordination of
multiple tasks
Cuing Participants are led to expect a specific stimulus to
be presented in a particular manner Example: Spatial cuing Types of cues
Valid vs. Invalid Does the target appear at the cues location or not
Peripheral (exogenous) vs. Central (endogenous or symbolic)
Predictive (target is more often than not consistent with cue) vs. Non-Predictive (target is as likely to appear anywhere else same as the cued location) Ex. 80% predictive or 50-50% chance at being valid or not
Inhibition of return: bias against previously attended to areas
Cuing Predictive/Symbolic
Elicits long-lasting orienting of attention but takes a bit (100-200ms) for attention to shift due to being a symbolic cue
Predictive/Peripheral Rapid (50-100ms after cue onset) and long-lasting
Non-predictive/Symbolic No real orienting effect as there is no real motivation to shift attention
Non-predictive/Peripheral Rapid orientation but doesn’t last long If cue does appear at location (valid) this will be detected much more
quickly with shorter cue-target onset With more delay, attention shifts elsewhere and actually RT is slowed
for valid trials Inhibition of return
Search Look for stimuli embedded
among non-target stimuli Slopes and set size
Flat slopes indicate stimuli are processed independently (parallel search or automatic), no interference from non-targets
Steep slopes Serial search: attention
shifted from one item to the next until the target is found. Or…
Limited-capacity parallel search (parallel but slower due to set size increase)
Filtering Dichotic Listening Task
See how well we attend to message or how much of ignored gets through
Stroop task Blue Red Word processing ok in general, but if reporting color,
word can interfere
Global/Local Navon letters: slower to report local features if global
does not match However size can affect
Filtering Flankers
H T H T T T Respond what middle is (one response for T, a different
one for H) RT is slowed when flankers have a different response associated
with them (HTH), unless far enough apart Can show the spread of attention to nearby areas
Negative Priming B Particular letter requires a response RT slowed when unattended letter later is the to-be-
attended letter Previous inhibition affects later processing
A
Dual-Task How much interference? If involve the same cognitive processes, paying
attention to one will lead to a decrease in performance on the other
If independent, no interference However, in some cases increased difficulty in one task
will result in the other beginning to interfere So not necessarily same resources involved, but a reflection of
cognitive ‘load’
Focusing Perceptual Enhancement
Lu & Dosher (1998) Attention acts as sensory amplifier in general, not just signal
amplifier Performance should improve in low noise situations If lots of noise, attention will amplify that as well and so
performance will not improve with focus of attention
What gets in? Early Selection
Attention operates to help prevent sensory/perceptual overload Late Selection
Attention serves to protect higher level cognitive processes (e.g. working memory)
Sensory Memory (Sperling 1960) Participants view a briefly presented array of
letters. Change the duration between presentation of
array and the recall tone. Report as many characters as possible.
Sperling Array of 12 letters 50 msec. presentation
7 I V F
X L 5 3
B 4 W 7
Sperling Full-report Report as many items as possible
Recall (no delay) = ~4 items Recall decreased dramatically with tone delay
Suggests a limit (‘span of apprehension’) to what can be perceived.
Sperling
Partial-report paradigm Tone cued which line of the array to recall
High = top line Medium = middle line Low = bottom line
Compare recall across rows
7 I V F
X L 5 3
B 4 W 7
Sperling Recall with no delay
Regardless of row asked to recall, about 3/4 of the items would be, or 9 on average for 12 item presentation
Conclusions: Lots of information gets in and
receives some initial processing Lasts a short time
Same pattern of results as full report with tone delay
Sensory memory is rather large but has a short duration.
Focusing: Selecting Channels Early Selection
Attention operates early on to protect low level processes from being overloaded
Late(r) Selection Operates after meaning has
been extracted from incoming stimuli
Working memory If we don’t use the
information…it is lost Cocktail Party Paradigm
Dichotic listening If early sensory systems do not
limit the information that is processed, when does selection take place?
What Gets In? Early idea: Only what is specifically attended
to gets in Broadbent’s bottleneck (1958)
One sensory input at a time processing Dichotic listening
Filter is flexible and can shift, but only what is focused on gets to later processing
Filter acts early after sensory stage Problem
Some ‘unattended’ info gets through Moray 1959 (can still hear our name in unattended channel)
INPUT
Sensoryregister
Selectivefilter
Detectiondevice
Short-termmemory
RESPONSE
Determine who is speaking
Danger signals, one’s name
Based on current goals
Revised conceptualization: some stable high priority info is checked without attention
Early/Late Selection Such a model comes from Triesman
(1960) All in but some in attenuated form
More relevant, less attenuated In the example here, Ss report hearing
the whole sentence Cocktail Party effect:
If meets certain criteria, will be attended (flexible bottleneck)
Capacity Limit for the amount of information
(and the amount of resources) available at any one time.
Although ‘early’ selection, key differences compared to Broadbent’s include: All info gets in initially for at least
some basic low level processing Possibility for flexible or multiple
filtering
Attended
InThe
PicnicBasket
SheHad
Peanut
ButterBookWoodLiveAtOn
Unattended
CatLargeDay
AppleFriendHouseSpoon
CapSandwiches
AndChocolate
CakeCrab
Treisman
Suggests that the filter/attenuator is occurring somewhat later but still before information reaches short-term/working memory
INPUT
Sensoryregister
Attenuationcontrol
Detectiondevice
Short-termmemory
RESPONSE
Late Selection Deutsch & Deutsch (1963), Norman (1968)
Proposed ideas for a late selection of attended information Essentially a different interpretation/version of Treisman
Both channels of information (in dichotic listening task) are recognized but are quickly forgotten unless they are relevant (or strong) Info makes it to short-term memory
Not really all that different from Treisman’s except the filter comes after meaning is fully processed for both channels
More on Late selection Mackay (1973).
Sophisticated meaning analysis of unattended channel
They threw stones towards the bank
… … … … … money
or
… … … … … river
Subject shadows this
Unattended ear
Late selectionPost-Shadowing Test Heard 26 ambiguous sentences. 26 “recognition trials” pick sentence that best
matches meaning of the sentences on the attended channel: “They threw stones toward the side of the river
yesterday.” vs. “They threw stones toward the savings and loan
association yesterday.”
Late Selection Result Choice of sentence influenced by word in unattended ear
Hear: money → More likely to pick “financial institution” Hear: river → More likely to pick “river bank”
When asked about the word in the unattended ear, participants entirely unaware of unattended word
Conclusion Unattended information was fully processed for meaning No attenuation early on, but rather is it relevant to the required
response?
Late selection
Information makes it to the detection/processing of meaning stage and passed on to STM for further processing and perhaps eventually to LTM
Both channels are processed fully for meaning, but only one of those reaches conscious awareness
Sensoryregister
Detectiondevice
Workingmemory
The end of early selection? Not so fast Evidence from
neurophysiological studies studies show the workings of attention very early on, before sensory/ perceptual processing is complete
Recap: Comparison of early/late selection So there is evidence for
both early and late May be that how attention
is utilized depends on the task and the current perceptual load, and instead be related to ‘attentional resources’ available rather than bottlenecks.
Other ideas Lavie (1995) suggested that it may have to do with
perceptual load If low load all information will be initially processed
and selection for further processing will take place after all relevant information has been analyzed E.g. flankers task
In high load conditions, attention acts as a perceptual filter Adding more stimuli to the flankers task suppresses the
effect of the flankers (i.e. they are not making it through the initial perceptual filter)
Capacity Model Kahneman (1973)
What gets in depends Attention is a resource
to be allocated across tasks
Practiced tasks require less resources Automaticity
Attentional Control Dual-Task Paradigm
Participant must perform more than one task at a time
In general, two tasks can be performed at once.. … with a detriment to one task … … depending on the type of tasks.
Driving and talking on the phone- which suffers?
Dual-Task Psychological Refractory Period and
Attentional Blink Refer to the same thing only usually in terms of
RT in the former and accuracy in the latter How long does it take a process to “prepare”
for additional work?
PRP Present two stimuli at about the same time. Each stimulus varies on some distinct
psychological dimension Example
Tone (high or Low) Letter (‘T’ or ‘Q’)
Make a forced-choice response to both stimuli Instructed to give one response first
PRP Measure the RT to the Second Response and compare it with
RT in a control situation (respond to second target alone) RT typically is longer in the dual task situation even for much
different stimuli Differences in RT patterns indicate the presence of central processes
that must be completed before response selection for the second stimulus can occur
RT varies as a function of a number of factors such as: Perceptual ambiguity of stimuli The nature of the response required Difficulty of tasks
Attentional blink Rapid serial visual
presentation of stimuli E.g. letters
Two tasks required of participant Name the white letter
Target Was there an X?
Probe When the time between
target and probe is short, participants are more likely to miss the probe
PRP and Attentional Blink summary of results
Task 2 RT
0
100
200
300
400
500
600
700
800
0 200 400 600 800 1000 1200
Interstimulus Interval (msec)
Tas
k 2
RT
Dual Task
Control
.4
.5
.6
.7
.8
.9
1.0
0 200 400 600 800 1000Stimulus Onset Asynchrony
Control
Experimental
As one can see, with more time between stimulus and probe, attention has ‘returned’and the probe is more readily identified.
% C
orre
ct P
robe
D
etec
tion
PRP and Attentional Blink Suggests appears to be a bottleneck in response selection and
consolidating the perception into a reportable memory Both tasks use the process before response can be made Can’t be used at the same time
However… Shapiro notes in the article there are cases in which no AB is seen
May be related to stimulus similarity Awh et al. (2004)
Digit response (target) Face response (probe) Faces unaffected Perhaps competition among multiple limited-capacity resources rather
than (dis)similarity Dual task costs can be predicted based on the degree to which each task
calls upon overlapping components of a broad range of resources Target hits on resources required for probe processing
Multiple Resource Theories Pashler (1998) Attention
Perceptual component that acts as both a filter and has resource limitations
Bottleneck component corresponding to response selection Some information may be blocked early on, but even that which
is not filtered is subject to available resources Complex Tasks
Capacities can be coordinated Response selection must occur for one task before next
can be completed Coordinated by the Central Executive
Resource Theories Limitations
Nature of the limitation is unspecified Not really testable
If two tasks can’t be performed without some impairment shared resource which is limited
If no impairment they don’t require same resource Ambiguous results multiple resources By explaining everything it may not really be able to
provide a true understanding
Decision Noise An alternative explanation of such results is
that, in tasks that require multiple decisions, accuracy will decline just because there is more opportunity to make errors (Shiu & Pashler, 1994) I.e. more noise in terms of signal detection
Evidence from Visual Search Spatial Cuing
Decision Noise Looking for a red T
Resource perspective: If accuracy decreases with
increase in red items but not with increase in green, we might conclude… Some resource allocated to
red but not green Each red item receives
fewer available resources with increase in red items
Decision Noise Decision noise
perspective More noise with
additional red items More opportunity for
decision error Decisions are not made
for green letters, so no performance detriment with increase in green items
Decision Noise We accumulate evidence over time until
criterion reached More time (longer RT) more accurate Less time (quicker), less accurate
Changes in criterion, sensitivity will influence speed and accuracy (no need to refer to resources)
Decision Noise Resource explanation
Accuracy better for valid trials due to resource allocation to cued location
Decision noise explanation Accuracy based on weighted
combination of noise at non-target locations and signal+noise at target location
Valid trials: hi success due to more weight given to cued location
Invalid: still more weight given to cued locations, but this is noise on invalid trials
Decision Noise Consider the following
experiment Exogenous (peripheral)
cue, followed by target, followed by mask (1 or 4 #s)
Valid, invalid, neutral (no cue) trials
Decision Noise According to resource theories there
should be an increase in errors for invalid trials regardless of number of masks because attentional resources are devoted to another area
However, in single mask condition target is unambiguous (no noise to reduce for valid or invalid trials)
More masks introduce more noise and make detection more difficult for invalid trials that do not have the noise reducing benefits of attention A precue allows nontarget
information to be excluded from the decision (noise reducer)
The end of resources? Not likely
Still results, such as those from visual search where targets are defined by relational cues, that SDT can’t explain
ERP evidence in difficult visual search tasks in favor of shifts of attention for difficult searches
Noise reduction or signal enhancement? In separate experiments Shiu & Pashler noticed decreased
accuracy for neutral trials suggesting attention as noise reducer Compare with Lu & Dosher that found evidence later of
attention as signal (+ noise) amplifier
Complex action What controls where attention is allocated? Automatic processing
Strict: obligatory and completes once started (e.g. feature detection pop-out)
Lenient: very reduced cognitive effort involved Cognitive control
Central Executive – coordinates and controls attention and other cognitive activities
Automaticity Neisser
Scan column of letters for a target (e.g. K)
•Both valid and invalid trials
•Measured reaction time (RT) to response
W P D S
J A L Q
A B C D
Neisser
Initially everything in the search set must be scanned… …with practice, less is scanned
Less effort Automaticity (multiple search targets can eventually be found as
quickly as single)
“Practice” with search set
RT
4 2 1
Search Set Size
Automaticity What can be automatized?
To what extent can certain tasks be automatized
Schneider & Shiffrin (1977) Visual Search Task
Search Set Memory Set Vary the number of elements in each (1, 2, or 4)
Schneider & Shiffrin Positive and Negative Trials Consistent and Variable Mapping
Variable: Target for one trial can used as distractor in another E.g. Memory set numbers,
distractors include numbers Consistent: Stimulus is either
always a target or always a distractor E.g. Memory set numbers,
distractors letters Measure RT to “yes/no” response
Schneider & Shiffrin General effect: Variable vs.
Consistent Variable mapping: increased RT
across search and memory set size Slopes flat for Consistent across
search and memory set size Consistent mapping allows for
automization and parallel process of items in search display
Effects of practice in variable mapping shows same pattern as here, with just a general reduction in RT (i.e. set size effects remain). Consistent mapping key to
automaticity
Schneider & Shiffrin cont’d. Subjects practiced in consistent mapping condition until
search set size was no longer a factor Switched to varied mapping situation where those items
were now distractors Performance much worse when previously consistently
mapped stimuli were distractors in the target set Stimuli were drawing attention away from other items in the
frame The cost of automaticity Hirst, et al. (1980), some varied mapping situations can improve
with practice
Framework for attentional control Two parameters influence attentional control
Bottom-up (stimulus-based) Example: sudden appearance of stimulus, abrupt changes in the
stimulus array Top-down (goal driven)
Example: expectancies regarding stimulus information (where, when)
Biased competition model (Desimone & Duncan, 1995) Bottom-up and top-down sources together bias the
competition among competing stimuli
Framework for attentional control Attentional template
Represents task demands and goals (e.g. searching for a particular shape and location)
Incoming info compared to template for possible match
Attention strengthens neural representation of info that matches
Framework for attentional control Stimuli and tasks compete for neural representation/motor
output Stroop example, both color and word name compete for vocal
response Mutually inhibitory one to eventually win out
Competition strongest where stimuli are activating the same area of cortex
Interactions among neuronal excitation and inhibiting responses are biased by both bottom-up and top-down influences
Processing can be biased on a number of feature dimensions (color, shape, location etc.)
Working memory implicated in top-down biasing
Binding How are features of stimuli integrated into a
perceptible whole?
Feature Integration Theory (Treisman) Attention needed to bind information together
into a single representation Focusing attention
Enhances the perceptual signal of the features involved
Binds the features together Localizes them to some point in space
Conjunction Search
Typical Findings Single Feature Targets
pop out Flat display size
function Automatic, little to no
attention Conjunction Targets
demand serial search Non-zero slope Require attention
0
500
1000
1500
2000
2500
3000
1 5 15 30Display Size
RT
(m
s)
Feature Target
ConjunctionTarget
Feature Integration Theory Treisman & Schmidt
(1982) Are end numbers odd or
even? What letter and their color
did you see?
Divided attention leads to miscombinations of features (illusory conjunctions) Directing attention to the
location of an object decreases ICs
2 8X T O
Feature Integration Theory Simple features are easily distinguished
regardless of set size, but conjunctions are more difficult to detect with increasing set size Attention is required to bind features, while single
features can be detected automatically
Neuropsych Person with bilateral parietal damage and bilateral
attention deficits- when multiple objects are presented can report features but not the objects to which they belong
Feature Integration Theory Mechanisms
Neurons code conjunctions Problem of combinatorial explosion
Example: bar of light, if 100 neurons needed to represent all colors and 100 for all possible orientations then 10,000 neurons are needed to process all combinations 1,000,000 if we add brightness etc.
Synchronous firing Results somewhat inconsistent and still doesn’t
answer how the end result is accomplished
Fun with attention http://viscog.beckman.uiuc.edu/grafs/demos/1
.html http://viscog.beckman.uiuc.edu/grafs/demos/2
.html http://viscog.beckman.uiuc.edu/grafs/demos/1
2.html http://viscog.beckman.uiuc.edu/grafs/demos/1
5.html
Change Blindness Visual information accessible to
consciousness is transient CB is a phenomenon in which people do not
detect large changes in stimulus array for features of lesser importance
Can occur for both dynamic and static scenes Why?
Lack of attention
Change blindness Phenomenon leads some to suggest there are no internal
representations of scenes or that they are incomplete Outside world as an external memory to be probed by our
senses Just like certain memories are not readily available unless
‘looked’ for, elements of the environment may not be perceived without attention How would you check whether you were seeing all elements in a
scene? Although it seems as if we are perceiving the world as is, we
are only consciously receiving info which is attended to The refrigerator light is always on
Change blindness However it may just be that the comparison
process among representations fails or breaks down in some way, or the preserved information may not be in a format that can be used for conscious change perception Some studies find that when told that a change
occurred, Ss can guess where it was even if they weren’t aware of the change initially
Inattentional Blindness Linking perception and attention
What (if anything) do we perceive w/o attention? Mack & Rock (1998) Participants engaged in another task have an element added at
one point Example: judge which line is longer, but add a critical
stimulus on a later trial. After critical trial participants are asked if they noticed anything unusual (very quick experiment)
+ +.
Quick Demo: Pick a card
I have removed your card!
How? You weren’t paying attention!