Studies on Concept Formation

1
Studies on Concept Formation
Psych 1090 Lecture 2
2
The definition of concept, like that of
cognition
is rather flexible and involves many different
types of behavior
3
So we are going to examine concept formation from
the simplest to the most complex cases
And do it for a small (for the moment) variety of
species
4
Concept formation, of at least some sort, is a
very basic phenomenon
Whether consciously or unconsciously, and
organism must distinguish, e.g.,

• predator from prey/ friend-foe

• food from not-food

• possible mates from non-mates

5
And, in any case, creatures must organize the
big, confusing world into some form of workable
hierarchy
so that attention can be paid, shifted, directed,
etc. appropriately for survival
6
One big issue, that we mentioned last time,
is whether animals, lacking language, can form
concepts
and, I said that they canbut now Ill give some
real data
7
We can split concept formation into three major
categories.

• object recognition

what is or is not a key or a pencil

• relational/abstract

bigger/smaller same/different

• equivalence/transitive inference

AgtB, BgtC, CgtD, DgtEB?D
8
Well hold off on concepts like same/different
and transitive inference for now
Theres enough just on other areas of concept
formation for todays lecture
9
But even the simplest level, object recognition
isnt really so simple.
Think about the prototypical bird
beak, feathers, wings, flight
10
But what about
11
Some of these creatures have some bird-like
characteristics and are clearly not birds
Others lack some of these bird-like
characteristics and clearly are birds
12
And, of course, what might be one important
category to a bird
Might be two to a human
13
For a bird, the snakes are functionally
equivalent
in that both are predators
But for humans one is dangerous and one is not
14
And although humans can, for example, detect
many variations in neighboring white-crowned
sparrow dialects,
To a bird in a given dialect, all the others are
lumped into the foreign category,
15
And in different areas,
different parts of the songs (e.g., trills versus
introductory notes)
may be critical for recognition
16
So, one can train an animal like a pigeon to, for
example, categorize all trees
by pecking to picture of a few trees and not to a
picture of a bird
But pigeons were less easily trained to form the
category car
Something unnatural for it
17
And pigeons completely failed when asked to
categorize
one particular oak leaf from among a number of
oak leaves
18
A discrimination not at all important to a pigeon
19
More recently, a number of researchers (e.g.,
Cerella, Watanabe, etc.) showed that pigeons
emphasize local cues
So they had considerable trouble distinguishing
some, but not others of the following.
20
Pigeons couldnt distinguish intact versus
distorted cubes
21
TTTTTT T T T
T T T
TTTTTTT T
T T T
T T
Pigeons could find the T in either picture, but
not the H (Anim. Cogn. 2002)
22
And if given a regular triangle as the positively
rewarded choice, and then shown
23
Pigeons chose the trapezoid rather than the
occluded item
Suggesting that they chose on the basis of just
the line figure
24
More recently, Watanabes group showed that
pigeons cant really discriminate scrambled from
unscrambled cartoon figure heads
At least until they got really scrambled
25
1x1 3x3
6x6 12x12
26
 
27
Could discriminate scrambled from unscrambled
pictures of pigeon heads
28
And Wassermans group argued that scrambling a
line figure makes the task more difficult for the
pigeon
29
But pigeons seem to have a hierarchy of attention
So they may first look for curves versus straight
edges
And then look for specific types of angles
30
And other studies suggest that motion is a
critical factor
31
Some researchers nevertheless argue that pigeons
can be trained to do any kind of discrimination,
and that may be correct
But the point is that some discriminations are
more ethologically relevant
and thus are trained more easily
32
Other more subtle issues come up as well.
If you use video for birds, you have to use LCD
projectors
Because their flicker-fusion rates differ from
those of humans,
CRT screens are difficult to view
33
And what is true for a pigeon may not be true for
another type of bird
Parrots go anorexic in an operant paradigm and
stop responding
And some hummingbirds, who are win/shift
foragers, have real trouble w/ a win/stay operant
paradigm
34
Theres the issue, too, of neighbor/stranger
discrimination in song sparrows
And how at first some data showed it existed and
other data didnt.
But the real issue was knowing the exact
territorial boundary
35
And, of course, most studies on categorization
are done with animals that categorize the world
visually,
much like humans
But pigeons and people would fail if the criteria
were scents.
or subtle auditory cues
36
And studies of polymorphous stimuli--
where a combination of attributes must be present
for an item to be judged as correct
Suggest that in pigeons the additive features
should be ecologically related for success
37
Remember our prototypic birdif we assume wings
AND beaks,
X
X
38
But if we assume wings AND flight,
X
X
39
So, how do we go about dealing with such issues?
40
Lets look at the study on monkeys and humans,
Dépy et al. on the conjoined search
41
Training was on marked items
yellow vs. blue was color discrimination, not a
bar
42
The idea was to determine which 2 of 3 qualities,
color/shape/location, put a stimulus in C1 or C2
43
I did it by marking C1 as blue, except if blue,
round, and down C2 as yellow, except for yellow,
square up
So I didnt match the human subjects in the
study, nor did I use the experimenters prototype
44
I did it more like the baboons than the humans
Of course, I didnt do it by seeing slides for
800 ms
and being told if I were correct or not, either
45
But I would have gotten the prototype right 100,
like the baboons
And would matched them on Csl, too
46
So Im not at all sure one can draw conclusions
as to how a particular species does a particular
task
47
And, of course, animals will try to succeed
however they can
Studies on homing pigeons suggest that theyll
use sight for local features or global features,
star navigation, smell, or magnetic currents
Depending upon what the experimenters knock out!
48
Thus issues of external validity
are extremely important when determining whether
an animal is competent or not in any task,
and particularly in a categorization task
49
My parrot, Alex, learned to identify a number of
objects with English labels
He could also generalize from a piece of an index
card to a large 12x18 piece of paper
50
So we trained him on a set of items,
then tested him on variations of these things
Differently sized pieces of paper, wood, rawhide,
etc.
51
(No Transcript)
52
But a separate issue is whether animals
understand categorical classes..
Not just what is or is not wood or green or
square
But that each of these attributes represented a
different category for the same object
53
That is, could an animal respond to What color?
versus What shape? versus What matter?
54
That is, could an animal understand a
hierarchical organization
in which various symbol represented different
class labels
under which other classes could be categorized
55
Note that this task is not the same as the
conditional discrimination we saw last time
Remember.pick odd color if backed in white, odd
shape if backed in black.
56
Here an animal must

• attend to a multivariant item

• attend to the particular question

• determine what attribute is targeted

• determine which instance of many possible
  choices is correct (e.g., which of 7 colors or of
  5 shapes)

• and then encode that info into a vocal label

57
This is the kind of task that we gave to Alex
It differed from tasks given most chimps, for
example, in that

• it required reclassification of the same items
  at different times wrt different queries

• it went further than testing transfer of symbol
  labels to new items

58
We trained him on a small subset of items in the
lab
Then tested him on everything else
59
COLOR QUERIES
LABEL USED BY PARROT
K W H K W H
K W H SHAPE OTHER
K
6
K,K
4
W
BCW
GH
5
H
4
K
5
RW
RW
RW
W
H
6
H
2
K
5
K
U,W
U,W
W
6
OBJECTS PRESENTED TO THE PARROT
4
H
U
K
4
BPW
8
BCW
1
W
H
5
4
K
W
5
1
H
6
K
5
4K
W
7
5
U,H
H
60
SHAPE QUERIES
LABEL USED BY PARROT
K W H K W H
COLOR OTHER
K
7
3W
W
8
1
3H,3BCW
H
4
K
4
W
7
1
3W
H
4
H,H
5
K
3K
5
1
W
CW
H
4
OBJECTS PRESENTED TO THE PARROT
2CH
K
6
10
W
W
H
4
UCH
4
K
1
W
5
4CGW
H
6
K
4
UK,CK
W
4
4W
6
H
61
We did, of course, look at transfer as well
62
NOVEL COLOR/SHAPE COMBINATIONS
LABEL USED BY PARROT
WHAT COLOR
WHAT SHAPE
ERRORS
ERRORS
W W H W W K W W W W H W W
BW(2) GW(2) GH RW GyW GyK GyW YW BW RW YW
2CW(2) 2CW(2) 2CW 3CK 3CW(2) 4CW 5CW(2) 5CH 5CW
5CW(2)
2CH(1)
W(1)
W(1)
2W(1)
OBJECTS PRESENTED TO THE PARROT
UW(1)
UW(1)
63
Now, this particular set of experiments did seem
to rely on language,
but in reality it merely relied on the ability to
understand symbols
and in principle could be done in an operant
setting
64
First train an animal with a number of different
symbols
Each of which represent a particular color or
shape or matter
Then train another set of symbols to represent
color, shape, matter, maybe via a sorting
system
65
Then train the animal to pick one color symbol in
the presence of an object and the symbol for
color
And likewise for shape
And then do transfer tests on novel items
Language just is easier
66
Now, one of the big issues that may not seem
obvious so far
The pigeon work is almost exclusively done with
slides or pictures
And the work with Alex was done exclusively with
3D objects
67
But does an animal understand that a photo is a
representation of a real-world item?
Even primitive tribes dont understand that
relationship
why would animals?
68
But that brings into question what kind of
categories were the pigeons forming
Did the categories have anything to do with real
life?
69
Although Watanabe showed that pigeons could
transfer food/not food discrimination
from objects to pictures,
his work consistently used items that were
familiar to the subjects
70
Some work by Spetch, that we wont discuss in
detail,
suggests that pigeons could associate pictures
with real-world situations and locations
But they have a difficult time connecting 2D w/
3D
And so does Alex
71
He initially called all pictures four-corner
paper
And weve spent a lot of time working to try to
train him on a separate category of picture-toy
72
But what about nonhuman primates?
Are they easier to train in this respect than
birds?
73
Savage-Rumbaugh and colleagues had to train their
chimps to associate objects and photographs
actually taping the photos to the objects
until the apes made the associations themselves
74
Note that in the baboon study on food vs nonfood
discriminations,
Bovet and Vauclair report first trial responses
to novel items.
Such data are critical and often are omitted
75
On anything other than first trial responses,
learning can be involved,
and in many studies, like this,
we want to see what the animal knows before
learning occurs
76
Note that the baboons did need significant
training to discriminate the pictures of the
apple versus the padlock,
even after being able to work almost immediately
with real 3D objects.
77
Such is not surprising,
given that young children without photo
experience respond similarly
And in terms of real correspondence,
children need to be at least 2.5 yrs old to see
pictures as representations of situations
78
Logicallya photo isnt edible,
so why would one want to respond to it as an
edible item????
79
Why did the baboons do less well on whole photos
rather than cut-outs?
Cut-outs look more like the objects
whereas photos look more unreal
80
So, again, there are real issues of external
validity in determining animal abilities
81
But so far we have looked only at concrete
concepts.
What about more abstract concepts?
Bigger/smaller?
And number?
82
These concepts are more difficult because they
have to do not with individual items
But with the relation between or among items
And how that relationship might vary
83
For relative concepts, such as bigger or smaller,
or lighter or darker,
what is correct on one trial may be incorrect on
another
Such was one of the problems when researchers
first looked at relational concepts in nonhumans
84
An animal in an operant paradigm is, for example,

trained to choose between two samples
Grey is rewarded, and the animal learns to choose
it after a number of trials
85
The subject is then given a transfer test of the
following two choices
Generally, the subject will initially hit the
grey item, because that was what was rewarded
86
If response to grey is extinguished, the subject
will learn to hit black
And if switched back to white and grey, may have
a bit of trouble going back to grey,
because that action was indeed extinguished
87
Eventually, it will learn the condition of when
to hit and when to avoid grey,
But the subject hasnt really learned darker
than as a concept
88
Ethological studies, although not specifically
designed to contrast relative versus absolute
concept learning in animals,
nevertheless suggest that some avian species can
respond to natural stimuli on a relative basis
89
In the wild, eastern woodpewees (Contopus virens)
may use the relative number of repetitions of a
song type sung by a conspecific
to assess the ways in which the singer will
engage in social behavior (Smith, 1988)
more reps more aggression
90
Some species seem able to judge the relative size
and motivational state of competitors
by assessing the relative frequency (as measured
in kHz) of their vocalizations
and whether the ending frequency of a
vocalization is relatively higher or lower than
the starting frequency (Morton, 1977, 1982)
91
Deeper pitch usually signals a larger and an
angrier animal,
And a lower ending frequency also generally
signals aggression
92
In the laboratory, great tits (Parus major),
after being trained to respond differentially to
songs with either a large or a small number of
notes,
respond to songs with intermediate numbers of
notes on a relative basis (Weary, 1989)
93
Studies of optimal foraging also suggest
that birds base their feeding strategies on the
relative amount of food they can obtain per unit
of time (see Kamil, 1988 Kamil Roitblat, 1985)
94
But most of these studies cannot eliminate the
possibility that the animal is somehow learning
about
the proper choice rather than
making relative judgments
95
One problem is that many animal subjects appear
to respond preferentially on an absolute basis
Or the task is set up so that the ecological
relevance only makes sense on an absolute basis
96
Hulses lab has shown that starlings (Sturnus
vulgaris) can be trained to discriminate
ascending from descending strings of notes
But the birds fall apart if the strings are
transposed to a different overall pitch
97
As we just saw, the pitch of a signal for a bird
likely tells something about the overall state of
the signaller,
so shifting the absolute pitch changes the
overall import of the signal to such an extent
that the likelihood of responding to the
fine-tuning may be overwhelmed
98
And even if an animal is trained and tested
appropriately,
The animal might not be shifted to anything other
than exemplars that differ only slightly from
those used in training
99
If an animal is trained on a set of balls and
tested even on a new set of differently sized
balls,
we still dont know if the concept of
bigger/smaller has generalized to dumbbells and
blocks.
100
One of the advantages of working with a talking
parrot is that one can train it to tell you, for
example, What color bigger/smaller? for
familiar items
So that it can be tested on objects it has never
before seen
101
We could use objects larger or smaller, novel
objects, objects of the same size
and see if he could respond with none, and also
ask about what matter was bigger or smaller, all
without additional training
102
(No Transcript)
103
Another issue about abstract concepts is that a
subject has to learn to reclassify items
So it may know that these things are all blocks,
But now it has to distinguish sets of three from
six
if it wants to show number concepts
104
What constitutes numerical competence?
Number can be a descriptive category

• chose one set from competing arrays

• chose with respect to more/less

• match-to-sample with respect to quantity

• respond to one sequential series of events

Most of these do not involve exact number.
105
Even if subjects discriminate a specific amount,
such as threeness when given a variety of
choices
106
The subjects

• may have only a recognition of pattern

• may be responding to mass or contour

• may be responding to brightness or density

• may actually be subitizing.

107
Subitizing

• is a perceptual mechanism

• generally depends on canonical arrays

-- think dice, dominoes
-- but can be sequential

• is used when time constraints exist

• is usually approximate for larger quantities

108
Counting is a different matter

• Produce a standard sequence of number tags

• Apply a unique tag to each item to be counted

• know that the last number tag used tells the
  quantity of interest

If arrays are not canonical, subitizing is
thought to occur for quantities ?4 and counting
for quantity gt 4
109
Numerical research in animals has a long history.
Current thinking suggests that humans and animals
share processes that involve subitizing.
Various proposals exist for the mechanisms
involved.e.g., object files, accumulators.
110
Several studies have begun to approach number
competence in, e.g., monkeys and pigeons by

• examining ordering of quantities

• matching symbols with numbers of actions
  (yielding approximate values)

• examining more vs. less after adding or
  subtracting items (yielding approximate values)

111
Most researchers argue that true counting can
exist only with language .even for humans
And, of course, most animals do not have
language.
However, a few apes, dolphins, and parrots have
acquired elements of human communication systems,
including number labels..
112
So, lets talk about numerical competence in
avian species
113
Look at the work historically.
Some of the most elegant studies with avian
species was done by Koehler..
114
(No Transcript)
115
(No Transcript)
116
Koehler and his colleagues transferred this
behavior across modalities.
Flashes of light
Notes on a flute
117
Related work on humans..
What happens when there isnt time to count?
118
(No Transcript)
119
(No Transcript)
120
(No Transcript)
121
(No Transcript)
122
Researchers found that humans bottomed out at
about the same level as the pigeons-
about 4
123
So, how might these studies relate to counting,
or at least precise number recognition?
124

• Geshwind showed that humans use different brain
  areas for labeling versus match-to-sample

• Premack found that his chimpanzees had more
  difficulty labeling quantities than performing
  match-to-sample with number

• Alex was already labeling shapes as 3- or
  4-corner was it a general gestalt, or maybe
  something more?

125
What could Alex really do after some training?
126
(No Transcript)
127
(No Transcript)
128
Now, note that Alex is not the only animal that
has number concepts.
Matsuzawa has shown that chimpanzees can label
quantities up to 8
129
(No Transcript)
130
(No Transcript)
131
Number used by Alex
1 2 3 4 5 6
1
7
1
1
8
4
2
1
6
3
2
1
Number of objects
4
9
7
1
1
5
8
6
1
132
(No Transcript)
133
So.we had production
What about comprehension?
Conceivably, Alex might not truly understand what
his number labels represented
134
(No Transcript)
135
Number Alex produces
1 2 3 4 5 6 n ?
1
8
1
1
9
2
1
1
10
3
Number of objects
8
4
1
1
8
5
2
10
6
5
1
none
136
Alexs results demonstrate competence comparable
to that of nonhuman primates and young children
Matsuzawas chimpanzees can comprehend symbols up
to about 8
His spontaneous use of none to represent
absence of quantity is of particular note
137
But what about something like addition?
Boysens chimps could walk around the room, look
at different collections, then point to the
Arabic number that represented the sum
Beran showed that chimps could track more versus
less when marshmallows were added or subtracted
from sets
138
Alex gave informal evidence of being able to add
sequential sounds.
So we began a study to replicate the Boysen
Berntson work
139
Number Alex produces
1 2 3 4 5 6 n ?
1
8
7
2
1
1
8
3
Number of objects
7
4
1
4
5
2
1
1
7
1
6
3
5
none
140
What about none??
141
All the other work with none involved
attributessame/different, relative size, or, in
a sense, the ATTRIBUTE of a collection
Here, Im asking him to comment on objects that
simply do not exist
142
His concept of zero thus seems to be that of
young children,
who have some but not full comprehension of how
the label is to be used
143
Weve done some training, one session, and he
then seemed to begin to understood how to use
none for zero
That he caught on quickly is important, but more
trials must be done
144
And what about 5?
Data were for first trials only, and he was
correct only 50
When he errs, he gets a total of 4 chancesand
almost every time he said 6.
145
Was he subitizing ?4, and seeing 6 as lots?
If we gave him 10 instead of 2 s, he was 100
accurate
Was he engaging in a counting-like strategy for
5??
Doesnt seem to be using an accumulator or object
files or subitizing, because hes too accurate
146
Given that parrots and primates evolutionary
history dates from the dinosaurs.
Number concepts are likely to be relatively
widespread across species
147
Maybe numerical competence involves giving the
subject the appropriate tools to express latent
abilities.
Certainly, enculturation is important, given
evidence from untrained humans in Peru
148
Future directions for number work---

• completing ordinality and subtraction studies

• training larger numbers

• determining if Alex will comprehend new number
  labels more quickly than old ones

• sequential sounds in younger birds and transfer
  to simultaneous visual

149
So, weve found that animals seem to have some
pretty decent understanding of concepts
150
But I hope what came through was that the
critical issue in determining these abilities
has to do with experimental design
151
The more the experimenter looked at the design
through the eyes of the animal,
The more likely was the animal to succeed on the
task
152
Ive emphasized my own work, because thats the
easiest material for me to present
But researchers working with other animals have,
as weve seen, also found striking abilities
153
And I suggest that if you have time,
you do a lit search on Matsuzawa and Boysen, for
example, and see what their apes have
accomplished
154
Sci-Am show on number work in chimps and rhesus