VanRullen & Thorpe (2001) studied the time course of visual processing using EEG. Their results identified two mechanisms: 1) An early perceptual process starting at 75-80ms that is task-independent and category-dependent. 2) A later task-related process starting after 150ms that is category-independent and involves decision making.
Rogers & Patterson (2007) sought to explain contradictions in categorization research. Their experiments using reaction times and patients with semantic dementia supported a parallel distributed processing model where more general categories are processed first, though full activation of basic categories occurs faster.
Mack & Palmeri (2015) explored factors influencing category advantages. Their 5 experiments showed that both brief exposure and
2. Rosch’s prototype theory argues
that basic categories should be
those which “yield the most
information for the least
cognitive load” (Rosch, 1976,
p.428), i.e. basic objects are the
categories for which the cue
validity is the highest.
It stands to reason, then, that in
terms of primacy, these basic
level categories should be
processed first.
Or do they?
4. WHAT THEY SET OUT TO DO
Current theories of visual processing suggest two mechanisms:
1. A perceptual process at the low-end level, and
2. A decision process at the high-end level.
How these mechanisms can be dissociated in time and space was the goal of
this study.
In addition, VanRullen & Thorpe made the following objections:
• The use of reaction times as a DV make it difficult to separate the
respective durations of perception, decision, and motor responses.
• They also noted that the fact that neural activity varies with respect
to the properties of the visual input is not sufficient to conclude that
a person is actively recognizing the identity of the category involved.
5. WHAT WAS THEIR STRATEGY
VanRullen & Tharpe used an alternating dual-task
paradigm of event-related potentials (ERPs) to compare
the processing of:
(1) the same visual category having different task-related
(behavioral) status, and
(2) different visual categories having the same behavioral
status.
This design allowed them to dissociate low-level
sensory activity from high-level task-related
mechanism.
6. WHAT WERE THEIR METHODS
Sample size:
16 subjects
8 men
8 women
Ages 21-50 yrs old.
Stimuli:
Two categorization, release-
button tasks (go/no-go) in
alternation.
Each task consisted of 10 series
of 96 images, half targets, half
non-targets.
Images were flashed for only 20
msec; these include pictures of
animals (birds, fish, insects, etc.),
or of transport (cars, trains,
trucks, etc.)
Distractor images were street
buildings, fruits, etc.
Data:
Using EEGs, event-related
potentials were recorded
from 32 electrode sites.
Electrodes were grouped
into frontal, central,
parietal, and occipital
groups.
Intersubject t-tests were
conducted at p < .01 for
extra precision.
There was no difference in
performance between
tasks.
7.
8. Differential activity between
targets and distractors more
marked in the frontal lobe after
150 msec.
Despite “transport” being an
artificial category, it produces
similar electrical responses.
Small activity found around
75 msec., but dismissed as
changes of the experimental
protocol.
9. Here, the differences in the two visual categories seem to occur after 75-80 msec, and
are totally independent of the task and of the status of the images in each task (target
and non-target represented equally in each category-specific waveform).
Parietal sides
10. Grouping of all waveforms from different categories when they were targets vs. all
waveforms for same categories when they were non-targets. Here, no differences
occur before 150 msec, and are totally independent of the visual category.
Frontal sides
11. WHAT WERE THEIR RESULTS
VanRullen & Thorpe identified two mechanisms:
1. An early perceptual, task-independent, and category-dependent process
starting at 75-80 msec after stimulus.
2. A later, task-related, category-independent mechanism starting after 150
msec involving subject’s decision making (target present or not present?)
The first is supposedly due to the extraction of visual features happening
in extrastriate areas such as V2 or V4.
The second takes place after visual recognition, and the authors suggest
that this decision-related activity could be located in occipito-temporal
regions.
12. WHAT WERE THEIR CONCLUSIONS
Categorization appears to unfold in two
stages:
The behavioral stage,
The perceptual stage, which most likely involves simple
visual encoding mechanisms,
including the extraction of basic
visual properties.
which most likely involves the
decision made by the subject
concerning the object, regardless
of the visual category.
Unaware
Aware
14. WHAT THEY SET OUT TO DO
A paradox exist in literature regarding the process of categorization:
1. There’s substantial evidence for Rosch’s idea of basic level categories,
which are intermediate level categories processed faster, and more often,
than more general and more specific categories.
2. On the other hand, patients with semantic dementia seem to be better at
processing general categories than basic-level or more specific categories,
even as the their cognitive system deteriorates.
How to explain this seemingly contradiction was the goal of this study.
15. WHAT THEY SET OUT TO DO
Rogers & Patterson also examine how spreading-activation models
have been used to in the past to account for these differences:
bird node
robin node
animal node
Point of entry
Jolicoeur’s Traditional Model Warrington’s Modified Model
animal node
bird node
robin node
Point of entry
semantic dementia
16. WHAT THEY SET OUT TO DO
They also argue that Murphy’s differentiation theory falls short
because it doesn’t explain why basic-level categories are more
susceptible to SD:
bird category robin categoryanimal category
Optimal
semantic dementia ??
Too low Too high
17. WHAT WAS THEIR STRATEGY
(1) There really is a paradox,
(2) That a parallel distributed processing (PDP) model is
better at explaining the phenomena than either
spreading- activation and differentiation models, and
(3) That PDP’s predictions can be empirically tested.
VanRullen & Tharpe used three experiments to demonstrate that:
18. WHAT WERE THEIR METHODS
EXPERIMENT 1
Sample size:
28 subjects
14 men
14 women
Ages 55-75 yrs old.
Stimuli:
72 color pictures consisting of:
Animals and vehicles at the
general level; dogs, birds, cars,
and boats at the basic level; and
Pekinese, Labrador, kingfisher,
robin, yacht, ferry, BMW, and
Morris at the specific level.
Each pic appears once as a
target and once as a distractor,
for a total of 144 trials.
Distractor images were taking
from the same 72 pictures.
Data:
On each trial, a name
would appear, followed by
blank screen, and then an
onscreen picture until
response was detected.
Trials were ordered
randomly. Reaction times
and accuracy were
recorded using DMDX
software.
Repeated-measures
ANOVAs and planned
comparisons t-tests were
used to test for
significance.
20. WHAT WERE THEIR METHODS
EXPERIMENT II
Sample size:
8 patients
suffering from
semantic
dementia (SD).
6 men
2 women
Ages 50-72 yrs
old.
Stimuli:
Same 72 color pictures used in
experiment 1.
Each pic was printed, and
presented along with the
corresponding word, for a
total of 144 trials.
Distractor images were taking
from the same 72 pictures.
Data:
On each trial, word + picture
would be shown, and the
experimenter reads the word
aloud. Participants were then
asked if the picture matched
the word.
Pictures were presented semi-
randomly. Accuracy was
recorded by experimenters.
Reaction times were not
recorded.
Repeated-measures ANOVAs
and planned comparisons t-
tests were used to test for
significance.
21. WHAT WERE THEIR RESULTS
severemild
There was significant interaction between severity and category
level,
F(2, 12)= 8.2, p<.006, = .58
22. A PARALLEL DISTRIBUTED PROCESSING (PDP)
THEORY
Semantic representations are instantiated as distributed
patters of neural activity, with different patters
corresponding to different concepts.
Surface representations capture modality-specific similarity
structure, whereas semantic representations capture
conceptual similarity structure.
Thus, items that are “the same kind of thing” will be
represented as similar even if they differ in particulars;
items that are “different kinds of things” will be
represented as dissimilar even if the share the same
characteristics.
23.
24. Basic-level categories
are distinct and
informative, located in
tight, widely
separated clusters—
they are “just right.”
Not distinctive enough
Not informative enough
Not distinctive enough
Not informative enough
25. Semantic dementia
disturbs the activity
patters of basic-level
and specific categories
first before disturbing
patterns of general
categories.
semantic
dementia
26. Basic-level effects
arise as a result of
similarity structures
coded in the hub, but
if conditions are
changed (e.g. by SD,
or by time pressure),
the basic-level
advantage would
turn into a
disadvantage.
In parallel
In normal
conditions, animal
activates first, but
bird reaches full
activation before
animal or canary
does.
27. WHAT WERE THEIR METHODS
EXPERIMENT III
Sample size:
28 subjects
14 men
14 women
Ages 55-75 yrs old.
Stimuli:
72 color pictures consisting
same as experiment 1.
Each pic appears once as a
target and once as a distractor,
for a total of 144 trials.
Distractor images were taking
from the same 72 pictures.
Data:
Same as experiment I, but
responses were timed with a
deadline.
All stimuli and conditions
from experiment I were
repeated at three different
deadlines: slow, medium,
and fast.
Repeated-measures
ANOVAs and planned
comparisons t-tests were
used to test for significance.
28. Responses that did not meet the deadline of 100 ms were discarded (21% of all
trials).
29. 53% of trials were
further discarded for
this analysis.
30. WHAT WERE THEIR CONCLUSIONS
PDP predicted that, when pressed for time, participants’
responses would resemble SD patients because decisions
about category membership were impaired—by neural
degradation, in SD case, and by not enough activation time, in
the time-pressure case.
In many theories, the goal of the recognition process is the
activation of entry-level (basic) categories, with little attention
paid to general categories, since they’re assumed to be
“semantic” and usually independent.
Rogers & Patterson argue, instead, that more general levels are
processed first, even if full activation does not occur after
basic-level categories are recognized.
Finally, the authors note that the separation of perceptual
categorization and semantic processing is wrong as they’re
both parts of the same interactive system, and that PDP avoids
32. WHAT THEY SET OUT TO DO
Like the previous article, Mack & Palmeri wanted to explain away the
contradiction between basic-level categories having an advantage during
speeded category trials, and superordinate categories having an
advantage during ultrarapid categorization trials.
In particular, Mack & Palmeri wanted to address the following
methodological factors:
• The exposure duration of the stimulus during speeded categorization
(longer) vs. ultrarapid categorization (very short).
• The local context and structure of each type of trial, as speeded
categorization usually employs randomized design, whereas
ultrarapid uses blocked design, in addition to the possibility of
priming after repetition.
33. WHAT WAS THEIR STRATEGY
(1) A conjunction of both brief exposure and blocked target
category context are necessary to see a superordinate
advantage (experiment 1).
(2) Exposure duration has an effect when targets are blocked
(experiments 2-3).
(3) Target category context has an effect when exposures are
brief (experiments 4-5).
Mack & Palmeri used five (!) experiments to show that:
34. WHAT WERE THEIR METHODS
EXPERIMENT 1
Sample size:
56 students
21 men
35 women
Ages 18-23 yrs old.
Stimuli:
Several images of:
Most popular dog breeds (e.g.
beagle, Labrador, etc.), backyard
birds (e.g. blue jay, American
robin, etc.), many species of
flowers, and trees.
No stimulus image was repeated
during an experimental session.
Data:
Exposure duration (25 ms or
250 ms) and target context
(randomized or blocked)
were fully crossed to create 4
conditions.
Participants were randomly
assigned to one of these
conditions. Trials were
presented in 36 trials sets,
for a total of 228 trials
(including 12 practice trials).
One-way ANOVAs and
planned comparisons t-tests
were used to test for
significance.
35.
36. WHAT WERE THEIR RESULTS
With long exposures, a basic-
level advantage in RT,
sensitivity, or both was
observed regardless of
whether the target category
was blocked or randomized
short long short long
This advantage vanishes
with short exposure +
blocked target category,
the two features of
ultrarapid categorization.
Brief exposure, then, is critical
to eliminate basic-level
advantage, but only when
categorizing at a particular
level of abstraction.
37. WHAT WERE THEIR METHODS
EXPERIMENT 1I
Sample size:
24 students
8 men
16 women
Ages 18-22 yrs old.
Stimuli:
Several images of dogs,
animals, and means of
transportation.
Dog images include the ones
use for experiment 1, plus
some more; animals came from
a variety of other species, in
addition to many categories of
transportation.
No stimulus image was
repeated during an
experimental session.
Data:
Participants performed a
category verification task at
either the superordinate or basic
level.
Target category was blocked;
half of the trials were “yes” (half
of these were dogs), the other
half were “no” (means of
transportation were randomly
chosen for the “no” trials).
Exposure varied at six intervals,
from 25 to 250 ms.
Participants completed 624 trials
in total (52 trials in each of the
12 conditions).
A 2x6 ANOVA and planned
comparisons tests were used to
test for significance.
38.
39. WHAT WERE THEIR RESULTS
There was a significant main effect
of exposure duration, F(5, 115)=
48.37, p<.001, = .028, but no
significant main effect of category
level.There was a significant
interaction between the two
factors: with short exposure,
sensitivity was higher in
superordinate categories.
However, with longer exposure
the situation reverses, with
basic-level categories having
higher sensitivity.
(shot out to Rogers & Patterson)
40. WHAT WERE THEIR METHODS
EXPERIMENT III
Sample size:
14 students
6 men
8 women
Ages 18-24 yrs old.
Stimuli:
Same stimuli as experiment
II.
No stimulus image was
repeated during an
experimental session.
Data:
Same procedure as experiment II,
with the exception of:
Stimulus image appear constantly
for 25ms, followed by a dynamic
mask at five varied intervals, from
25 to 125 ms.
Participants completed six blocks,
3 with the superordinate animal,
and 3 with the basic-level dog, for
a total of 600 trials (60 trials in
each of the 10 conditions).
A 2x5 ANOVA and planned
comparisons tests were used to
test for significance.
41.
42. WHAT WERE THEIR RESULTS
The extent to which categorization is
resilient to the onset of the mask reveals
how much category relevant information
is available at that point in time.
Results suggest that, with brief
exposures, the information relevant for
category decisions favors superordinate
over basic category.
That is, this superordinate information
is available quickly, and is of better
quality initially.
(again, shot out to Rogers & Patterson)
43. ULTRARAPID VS SPEEDED TRIALS
Aside from differences in exposure duration, which
we have seen, do these other procedural differences
matter?
Blocked design
Participants aware of target
category
Decision criteria optimized
for fastest responses
Random design
Participants unaware of
target category
Stimulus exposure is
unlimited
Lots of practice trials Very few (or none) practice trials
44. WHAT WERE THEIR METHODS
EXPERIMENT 1V
Sample size:
16 students
6 men
10 women
Ages 18-23 yrs old.
Data:
Same as experiment 1, with the
exception of:
All stimuli was presented briefly
(25 ms); half of the experiment
used blocked target category,
and the other half used
randomized target category.
Stimuli were randomly assigned
to the either blocked or
randomized category contexts.
There was a filler task of aprox.
30 min. in between.
A 2x3 ANOVA and planned
comparisons tests were used to
test for significance.
Stimuli:
Same images as experiment 1:
Most popular dog breeds (e.g.
beagle, Labrador, etc.), backyard
birds (e.g. blue jay, American
robin, etc.), many species of
flowers, and trees.
No stimulus image was repeated
during an experimental session.
45. WHAT WERE THEIR RESULTS
A basic-level advantage is shown
in the randomized context, but the
RT advantage disappears in the
blocked context.
Blocked target context might be
given participants an opportunity
to increase efficiency (but not
accuracy) in all category levels.
46. WHAT WERE THEIR METHODS
EXPERIMENT V
Sample size:
20 students
7 men
13 women
Ages 18-22 yrs old.
Stimuli:
Same as experiment II,
plus birds from
experiment I.
No stimulus image was
repeated during an
experimental session.
Data:
Participants performed the same
category verification task as in previous
ultrarapid trials.
Trails were critical pairs (one of four
types, depending on same/ different
level of abstraction), baseline pairs (an
unrelated parity task, followed by a
superordinate-basic level prime), and
filler pairs (an unrelated parity task
followed by nonmatching categories).
Participants completed 472 (including
12 practice trials).
A 2x2x2 ANOVA and planned
comparisons tests were used to test for
significance.
47. WHAT WERE THEIR RESULTS
On average, basic-level
categorization is faster than
superordinate
categorization, in both
baseline and prime trials. It
is also robust to local
variation in experimental
context.Conversely, superordinate
categorization is
significantly affected by the
type of prime.This is consistent with a
spreading activation
account.
48. WHAT WERE THEIR RESULTS
This graph shows that the basic-
level advantage in RT is
eliminated after only 4 trials of
superordinate categorization.
The increase in superordinate RT
could be due to a transition from
mediated processing through
semantic knowledge to more
direct perceptual retrieval in
episodic memory.
This increase in RT efficiency is
only available for a limited
window of time.
49. WHAT WERE THEIR CONCLUSIONS
Mack & Palmeri make the following observations:
A common theoretical position is that certain levels of
abstraction are faster, better, and first because they’re primary in
some way by access, logic, and/or development.
Neither exposure duration (time course) nor local categorization
context alone is sufficient to explain the speed of categorization
at different levels of abstraction. Rather, it’s the interaction
between these two that determines when categorization at one
level would be faster than at another level.
In a default state, basic-level categories have an advantage
during visual categorization. However, under very specific
conditions, superordinate categorization can be as fast if not
faster than basic-level categorization after a very short time.
50. FINAL THOUGHTS
Why is superordinate
categorization
sensitive to target
category context when
basic-level