Perceptual Classes Concepts Categories

1
Perceptual Classes(Concepts Categories)

• Ps625 Concept Formation
• Dr. Ken Reeve
• Caldwell College Grad ABA Programs

2
Other Names of Perceptual Classes

• similarity-based categories (Wasserman
  DeVolder, 1993)
• open-ended categories (Herrnstein, 1990)
• generalization classes (Stemmer, 1980)
• feature classes (McIlvane, Dube, Green, Serna,
  1993 Stromer, Mackay, 1997).

3
Definition

• set of perceptually similar, but discriminable,
  stimuli that controls a common response
• Responding comes under the control of the entire
  set of stimuli even though the organism was
  directly trained to respond to a smaller subset
  of the stimuli in the class

4
Types of Perceptual Classes

• types of perceptual classes can be differentiated
  by the stimulus domain, or set of limiting
  properties, that characterize the stimuli in the
  classes
• Typically this involves looking at dimensions and
  features

5
Dimensional Perceptual Classes

• dimensional class is, is defined by a range of
  discriminable values along some stimulus
  dimension (Fields et al., 1997 Herrnstein, 1984
  Zentall, Jackson-Smith, Jagielo, 1990)
• stimulus dimension refers to some continuous
  range of measurable values resulting from changes
  in a particular stimulus property, feature, or
  characteristic (Guttman Kalish, 1956 Jenkins
  Harrison, 1960, 1962

6
Dimensional Perceptual Classes

• Within a specific range along the dimension, each
  stimulus occasions a particular response
• This occurs even though the emission of the
  response was reinforced only in the presence of
  some small subset of the stimuli in the range
• Other stimuli outside of this range along the
  dimension do not occasion the response.

7
Types of Dimensional Classes

• dimension along which such a class is defined can
  be a simple physical one such as
• length (Fields et al., 1997)
• Color (visible light wavelength) (Zentall,
  Jackson-Smith, Jagielo, 1990)
• sound frequency (Njegovan, Ito, Mewhort,
  Weisman, 1995 Risley, 1964)
• object rotation (Wasserman, Gagliardi, Cook,
  Kirkpatrick-Steger, Astley, Biederman, 1996)

8
Example of Dimensional Classes

• consider a tabletop containing long and short
  sticks
• child is taught to say short in the presence of
  a 5 cm stick and long in the presence of a 15
  cm stick (see next)

9
Example of Dimensional Classes

• Say long
• Prompt if necessary
• Reinforce if child is correct

10
Example of Dimensional Classes

• Say short
• Prompt if necessary
• Reinforce if child is correct

11
Example of Dimensional Classes

• Continue to alternate between these two trials
  until come defined criterion is met
• Ex. Correct on 6 trials in a row
• Lean out reinforcement

12
Example of Dimensional Classes

• we can then test (with no reinforcement provided)
  the extent to which other sticks along the
  dimension of length will occasion either of these
  verbal responses
• Results Maybe 3 or 4 of the different shorter
  sticks will occasion the verbal response short
  while another 4-5 of the different longer sticks
  will occasion the verbal response of long

13
Example of Dimensional Classes

• The frequency of responding short or long as
  a function of changes along length can then be
  plotted as a generalization gradient (Guttman
  Kalish, 1956 Jenkins Harrison, 1960, 1962
  Honig Urcuioli, 1981).

14
Example of Dimensional Classes

• The gradient will show that all sticks within a
  particular range occasion the verbal response
  short every time they are presented while
  another range along the dimension will occasion
  the verbal response long.
• Thus, these two ranges of stimuli may be
  functioning as separate dimensional classes

15
Example of Dimensional Classes

• some sticks that are intermediate in length may
  occasion the response short sometimes and
  long other times
• These stimuli are not functioning as members of
  either dimensional class

16
Example of Dimensional Classes

• Thus, in a dimensional class, only stimuli within
  a specific range of values along the dimension
  occasion a particular response even though the
  physical dimension itself is continuous (Fields
  et al., 1997 Risley, 1964 Zentall et al., 1990)

17
MoreDimensional Classes

• quantifiable measures of the stimuli (such as
  length) are easily obtained
• These measures can then be used to determine how
  they correspond to the control of responding by
  specific ranges of stimuli along the dimension
  (Blough, 1990)
• Thus, stimuli comprising dimensional classes
  differ quantitatively from each other

18
Another Type of Perceptual Class Fuzzy Classes

• In these classes, properties of the stimuli are
  more difficult to objectively measure because the
  stimuli differ qualitatively from one another
  (Adams, Fields, Verhave, 1993 Blough, 1990
  Herrnstein, 1990 Lea, 1984 Rosch Mervis,
  1975 Von Fersen Lea, 1990 Wittgenstein, 1968)
• stimuli in fuzzy classes contain different
  combinations of discrete physical
  characteristics, or features, that occur in
  different combinations across the class members
  (Cerella, 1979 Cook et al., 1990 Jitsumori,
  1993, 1996 Herrnstein, 1984)
• Thus, stimuli in fuzzy classes differ from one
  another across a multiplicity of features and/or
  along different dimensions

19
Other Names of Fuzzy Classes

• Fuzzy categories have been referred to by a
  variety of names including
• basic level classes (Rosch, 1978 Rosch Mervis,
  1975 Rosch, Mervis, Gray, Johnson,
  Boyes-Bream, 1976)
• natural classes (Herrnstein, 1990 Lea Ryan,
  1984 Wasserman DeVolder, 1993 Wittgenstein,
  1968)
• ill-defined classes (Bourne, Dominowski,
  Loftus, 1979 Homa Little, 1985 Omohundro,
  1981 Neisser, 1967)
• probabilistic categories (Medin Smith, 1984)

20
Examples Fuzzy Classes

• researchers have investigated fuzzy categories
  comprised of
• leaves (Cerella, 1979)
• fish (Herrnstein de Villiers, 1980)
• people (Malott Siddall, 1972)
• cars, cats, and flowers (Wasserman, Kiedinger,
  Bhatt, 1988)
• locations on a college campus (Honig Stewart,
  1988)
• trees and water (Herrnstein, Loveland, Cable,
  1976)
• different musical pieces (Porter Neuringer,
  1985)

21
Fuzzy Classes Control of Responding

• responding to the members of a fuzzy category
  comes under the control of some unspecified
  combination of many of the potential features
  found across the class members (Herrnstein, 1990
  Jitsumori, 1993, 1996 Lea Ryan, 1984 Medin
  Smith, 1984 Rosch Mervis, 1975 Smith Medin,
  1981 Wasserman, Kiedinger, Bhatt, 1988)
• That is, of the many possible stimulus features
  that occur across class members, a class exemplar
  must contain some non-specified number of these
  features to evoke the same responding as other
  class members
• Thus, the phrase some of many characterizes the
  stimulus domains of fuzzy classes because the
  specific features are not easily identified or
  quantified
• In addition, no one characteristic is either
  necessary nor sufficient to define membership in
  a fuzzy class.

22
Fuzzy Classes

• unlike stimuli that comprise dimensional classes,
  stimuli in fuzzy classes cannot be arrayed along
  simple dimensions such as length, temperature,
  brightness, size, etc.
• Stimuli in fuzzy classes, however, can be scaled
  along psychometrically defined dimensions.
• One example that illustrates the creation of such
  a dimension involves the sorting of stimuli
  according to their typicality in a particular
  category (Bourne, Dominowski, Loftus, 1979
  Cook et al., 1990 Lea Ryan, 1990 Rosch
  Mervis, 1975 Smith, 1989).
• More

23
Fuzzy Classes Rosch Mervis 1975

• Human participants were presented with six
  different sets of visual stimuli (car, truck,
  airplane, chair, table, and lamp) with 15 members
  in each.
• Participants were asked to rate each stimulus in
  the six sets on a 7-point scale corresponding to
  the degree to which that stimulus was
  representative (or typical) of the category
  (thus, 7 was the most typical)
• these best exemplars, or most representative
  members of each class, were called prototypes
  for each class by Rosch Mervis
• Other stimuli could thus be ranked according to
  how prototypical they were of each category and
  then plotted along a dimension of prototypicality
• Lets try to rate the followingMoregtgtgt

24
Fuzzy Classes Rosch Mervis 1975
25
Dimensional Fuzzy Classes

• Both dimensional classes and fuzzy categories
  have been referred to as open-ended categories
  (Herrnstein, 1990) because the number of
  potential class members in both is theoretically
  unlimited
• With regard to dimensional classes, because the
  members can be arrayed along some continuous
  dimension there are an infinite number of
  exemplars for each dimensional class
• With regard to fuzzy classes, because there is
  presumably no limit to the number of combinations
  of some of many features that members of a
  basic level category can contain, there is also a
  potentially infinite number of exemplars

26
Polymorphous Classes

• Unlike dimensional and fuzzy classes, a third
  type of perceptual category, called a
  polymorphous class, contains a closed or finite
  number of potential class members (Herrnstein,
  1990)
• Polymorphous categories are frequently
  investigated as artificial laboratory analogs
  of basic level or fuzzy categories
• Unlike the unspecified some of many features
  that define fuzzy class membership, the features
  found in members of polymorphous classes are
  generally experimenter-defined and, thus, more
  easily quantified

27
Polymorphous Classes

• use of experimenter-defined polymorphous classes
  came from difficulty identifying the some of
  many stimulus features that come to control
  behavior in naturally occurring fuzzy categories
  such as trees, fish, etc. (Blough, 1990
  Jitsumori, 1993, 1996 Lea Ryan, 1990.
• Membership of stimuli in polymorphous classes is
  defined by a mathematical rule in which at least
  m of some n experimenter-defined features must be
  present in a stimulus (Aydin Pearce, 1994
  Fersen Lea, 1990 Jitsumori, 1993, 1996 Lea,
  1984
• These m features are a subset of a finite number
  of n features that the stimuli in polymorphous
  classes can contain
• Thus, polymorphous classes may contain a large
  number of exemplars, but they are limited in size
  by the number of possible n features each
  exemplar may possess.

28
Example of Polymorphous Class

• Diagnostic Criteria for Autistic Disorder
• A total of six (or more) items from (A), (B), and
  (C), with at least two from (A), and one each
  from (B) and (C)
• A qualitative impairment in social interaction
• B qualitative impairments in communication
• C restricted repetitive and stereotyped
  patterns of behavior

29
Example of Polymorphous Class

• Stimulus must have at least 3 of 4 (m of n) of
  the following features to be a member
• Star shaped
• Black dot in interior
• Circle touching it
• Plus sign outside it

30
(No Transcript)
31

32

33
Required Functions of PC

• Regardless of stimulus domains used, specific
  behavioral properties must be demonstrated in the
  presence of the stimulus members to infer that
  they are functioning as a class
• each stimulus in the set occasions a particular
  response stimuli in other sets do not occasion
  that same response (Adams et al., 1993 Fields,
  Adams, Buffington, Yang, Verhave, 1996
  Wasserman DeVolder, 1993).
• This emerges after training has occurred with
  only a subset of the stimuli in the potential
  class (Lea, 1984 Herrnstein, 1990).

34
How is this conducted?

• first a response is trained to occur in the
  presence of only one or some of the stimuli in
  the set
• Next, novel stimuli are presented under
  extinction conditions in a transfer test to
  determine the extent to which control of
  responding by the training stimuli transfers to
  the novel stimuli
• When novel stimuli occasion responding to the
  same extent as those stimuli used in training,
  then the entire set of stimuli presented during
  training and during the transfer test may be
  functioning as a class (see next)

35
Why may?

• Although class-consistent responding during the
  transfer test is necessary to infer that
  perceptual classes have been established, it is
  not sufficient to do so
• Such performances can be explained in two ways
• One explanation is that the stimuli in such sets,
  while different physically, may not be
  discriminable from each other (psychophysical
  limits of the sensory system of the organism).
  Thus, the different stimuli in the set may
  occasion the same response simply because an
  organism is unable to discriminate among the
  stimuli within the set, or between the stimuli
  used during training and transfer testing.

36
Why may?

• Such performances can be explained in two ways
• The other explanation (the one we want to be
  true!) is that the stimuli within the set, while
  physically similar, are discriminable from each
  other and still occasion the same response.
• To demonstrate this, need an independent measure
  of discriminability
• Demonstrating this within-set discriminability
  would support the conclusion that perceptual
  classes had been established (because responding
  would have transferred to discriminable, novel
  stimuli). (this test will be explained in a
  subsequent slide dealing with training/testing
  procedures)

37
Training Testing Procedures Simple
Discrimination

• During training, participants are exposed to at
  least two different stimuli, each of which is a
  member of a potential perceptual class
• If only one stimulus is used for each potential
  class during discrimination training, the
  procedure is referred to as single exemplar
  training (Bhatt Wright, 1992 Cerella, 1972).
• Thus, if two potential classes were being
  established, two stimuli would be used during
  training if three potential classes were being
  established, three stimuli would be used, etc.

38
Training Testing Procedures Simple
Discrimination

• training stimuli can be presented either one at a
  time (successive discrimination), or concurrently
  (simultaneous discrimination).
• Each stimulus in a set serves as a discriminative
  stimulus (S) for a particular response.
• This response may entail physically contacting
  the stimulus or emitting some other behavior such
  as a lever press, key peck, or verbal response.
• response specific to each stimulus is reinforced
  only when it occurs in presence of that
  particular stimulus (emitting the response at any
  other time is not followed by reinforcement).

39
Risley (1964)

• Used single exemplar successive discrimination
  procedure to train adults with mental retardation
  to press a bar specific to a particular sound
  frequency
• 1900 cps sound right bar presses
• 600 cps sound left bar presses
• Once the participants were reliably responding
  correctly, 16 different tones (range 160-12,000
  cps) were presented under extinction conditions.
• generalization test measured extent to which left
  or right bar presses were maintained in the
  presence of the multiple probe stimuli.
• resulting generalization gradients showed that
  dimension of sound frequency was divided by the
  participants into two potential dimensional
  classes.

40
Training Testing Procedures Simple
Discrimination

• perceptual classes can also be established using
  multiple stimuli from each potential class during
  training (called multiple exemplar training
  (Becker, 1971 Cook et al., 1990 Haring, Breen,
  Laitenen, 1989 Homa Little, 1985) or
  multiple stimulus discrimination training (Lea
  Ryan, 1990).
• For example, rather than training a
  discrimination between only one picture of a dog
  or cat, many pictures of cats and dogs could be
  used.

41
Herrnstein Loveland (1964)

• successive discrimination procedure with multiple
  exemplar training was used to investigate the
  demonstration of fuzzy categories with pigeons.
• In training, presented over 1200 slides
  containing natural settings such as trees,
  meadows, towns, and the countryside.
• 50 of slides contained a person or part of a
  person (in other slides, no person was present).
  More

42
Herrnstein Loveland (1964)

• Pigeons were trained to peck a key to gain access
  to food when instances of person (S) were
  presented on a slide viewing screen
• When slides depicting no people (S-) were
  presented, pecks were under extinction
• Discrimination training continued until the
  pigeons likelihood of pecking was systematically
  higher in the presence of the person slide
  exemplars than the non-person slide exemplars.
  More

43
Herrnstein Loveland (1964)

• Although the pigeons learned the discrimination
  between the sets of stimuli, it cannot be
  inferred that perceptual classes had emerged
  birds may have simply learned many independent
  discriminations in which specific slides were
  correlated with reinforcement for responding
  while others were not
• To assess whether perceptual classes had been
  established, need to present novel person and
  non-person slides.
• this transfer test showed that the birds
  systematically responded to the novel person
  slides to a greater degree than non-person slides
  even though they had never been observed before.

44
Bhatt, Wasserman, Reynolds, Knauss (1988)

• used a multiple exemplar successive
  discrimination procedure to see if pigeons could
  learn four fuzzy categories
• pigeons were randomly exposed to 10 exemplars of
  four categories consisting of cats, flowers,
  people, and cars on a slide viewing screen
• For a particular category, each exemplar served
  as an S for pecks to a key located in one corner
  of a rectangular array
• discrimination training continued until the
  pigeons reached an asymptote of responding
  correctly on about 75 of trials (well above
  chance!)

45
Bhatt, Wasserman, Reynolds, Knauss (1988)

• Based on the performances during discrimination
  training alone, two possible inferences can be
  made the pigeons had simply learned many
  unrelated discriminations in which particular
  slides were correlated with reinforcement for
  pecking at a particular location or
• pecking had come under the control of the members
  of a perceptual class.
• Transfer tests were next used to determine which
  of these explanations accounted for pigeon
  performances.

46
Bhatt, Wasserman, Reynolds, Knauss (1988)

• In testing, 10 new slides depicting novel
  exemplars were presented for each of the four
  stimulus sets
• If learning many unrelated discriminations had
  occurred, correct responding to the novel
  exemplars would not be expected to exceed chance
  responding of 25 correct.
• transfer test, however, showed that the pigeons
  responded correctly to the novel stimuli at
  levels significantly greater than chance (but not
  as high as the 75 correct during training)
• this provides minimal requirements to infer that
  fuzzy classes had been established.

47
Lea Harrison (1978)

• studied emergence of polymorphous categories with
  pigeons
• used multiple exemplar successive discrimination
  procedure
• six potential features could be present in the
  stimuli used in training and testing
• features were
• 2 different shapes
• 2 brightness levels
• 2 colors
• Three features were used in training for the S
  or positive set (green background, circular
  figure, or black figure). Another three features
  were used for the S- or negative set.

48
Lea Harrison (1978)

• features were
• 2 different shapes (1 positive and 1 negative)
• 2 brightness levels (1 positive and 1 negative)
• 2 colors (1 positive and 1 negative)
• Each stimulus contained 3 features
• Positive stimuli (reinforced for pecking) had 2
  positive features and 1 negative
• Negative stimuli (extinction for pecking) had 2
  negative features and 1 positive
• Thus, polymorphous rule was 2 of 3 features
  (using our m of n notation)

49
Lea Harrison (1978)

• During training, pigeons learned discrimination
  between the two sets
• Because we cannot infer perceptual class has
  emerged after training alone, stimuli containing
  either all three positive features or all three
  negative features were used during testing
• Birds were successful during testing

50
Training Testing Procedures Conditional
Discrimination

• establishment of perceptual classes is not
  restricted to simple discrimination training
  procedures
• also used conditional discrimination procedures,
  such as matching to sample (Cumming Berryman,
  1965) (although used less frequently than simple
  discrimination)
• Studies of arbitrary class formation (e.g.,
  equivalence classes), however, have used
  conditional discrimination almost exclusively!

51
Training Testing Procedures Conditional
Discrimination

• To establish two potential perceptual classes
  using a matching to sample procedure,
  participants are exposed to at least three
  stimuli during each trial
• first stimulus presented in a trial (called the
  sample), is drawn from a potential perceptual
  class (For two potential classes, at least two
  different sample stimuli would be randomly and
  successively presented).

52
Training Testing Procedures Conditional
Discrimination
sample
53
Training Testing Procedures Conditional
Discrimination

• Following presentation of the sample, some
  observing response (Wyckoff, 1952) such as
  touching the sample, must be emitted. Why?
• Next, at least two additional stimuli, called
  comparisons, are presented.
• One is drawn from the same potential perceptual
  class as the sample and is called the positive
  comparison
• other comparison comes from a different class
  and is called the negative comparison

54
Training Testing Procedures Conditional
Discrimination
Sample 1
- comparison
comparison
55
Training Testing Procedures Conditional
Discrimination
Sample 2
- comparison
comparison
56
Training Testing Procedures Conditional
Discrimination

• Note that the behavioral functions of the
  comparison stimuli change as a condition of what
  the sample is
• That is, the negative comparison and positive
  comparison swap functions depending on the sample

57
Training Testing Procedures Conditional
Discrimination

• forming perceptual classes using a conditional
  discrimination procedure requires that only some
  stimuli in a potential class occasion selection
  of other stimulus members as a result of direct
  training
• remaining stimuli must occasion selection of
  other set members even though they had never been
  paired together before in a conditional
  discrimination
• To assess this, a match to sample transfer test
  containing multiple novel stimuli not used in
  training must be presented to assess the
  emergence of perceptual classes

58
Training Testing Procedures Conditional
Discrimination

• As in simple discrimination training, conditional
  discrimination training may involve either single
  or multiple exemplar training
• In single exemplar training, only one stimulus is
  drawn from each potential class (such as in our
  circle example).
• This stimulus serves as both a sample and a
  comparison during conditional discrimination
  trials.
• When multiple exemplars are used during
  conditional discrimination training, many stimuli
  drawn from the potential classes serve as samples
  and comparisons

59
Fields, Reeve, Adams, Brown, Verhave (1997)

• Used conditional discrimination training
  procedure to establish two dimensional classes
  with college students
• Training was conducted with single exemplars
  drawn from each of two potential dimensional
  classes
• Sample stimuli were short or long lines (1- or
  25-units long)
• two comparison stimuli were the same 1- and
  25-unit lines.

60
Fields, Reeve, Adams, Brown, Verhave (1997)
Sample 1
comparison
- comparison
61
Fields, Reeve, Adams, Brown, Verhave (1997)

• Once training reached 100, students were
  presented with samples of intermediate length
  lines (varied from 2 to 24 units)
• comparisons were identical to those used in
  training (1 or 25 24 units).
• transfer test determined extent to which
  selection responding generalized to the test
  stimuli
• resulting generalization gradients showed that a
  range of short test lines always occasioned the
  selection of the short comparison line (likewise,
  a range of long test lines always occasioned the
  selection of the long comparison)

62
Fields, Reeve, Adams, Brown, Verhave (1997)
Test sample
Test comparison
test comparison
63
Discriminabilty Measures

• Most researchers investigating the demonstration
  of perceptual classes rely on performances
  observed during transfer testing to infer that
  classes have emerged
• While class-consistent performances during
  transfer testing may be indicative of the
  emergence of perceptual classes, an additional
  requirement must be satisfied to make such an
  inference
• Specifically, some stimuli in the potential
  perceptual class must be discriminable from the
  other stimuli within the set
• If within-set discriminability is demonstrated
  following class-consistent performances during a
  transfer test, such an extension of class
  membership represents the demonstration of a
  perceptual class (Cook et al., 1990 Fields et
  al., 1997 Lea, 1984 Wasserman, Kiedinger,
  Bhatt, 1988).

64
Discriminabilty Measures

• Two general procedures have been used to provide
  discriminability measures of stimuli that
  constitute potential perceptual classes
• In one study (Fields et al., 1997), a conditional
  discrimination format was used
• Recall that in their study a range of short test
  lines always occasioned the selection of the
  short line used in training. In addition, a range
  of long test lines always occasioned the
  selection of the long line
• If participants could not discriminate the test
  lines from those used in training, however,
  transfer of responding to the test stimuli would
  not be indicative of class formation (Cook et
  al., 1990 Fields et al., 1997 Wasserman,
  Kiedinger, Bhatt, 1988).
• Rather, during testing, control of responding
  would simply have been maintained by stimuli that
  were not discriminable from the training stimuli.
  
• If the lines within a set were discriminable,
  however, the emergence of dimensional classes can
  be inferred.

65
Discriminabilty Measures

• To assess this, Fields et al. (1997) reinforced
  the selection of a neither response option for
  all lines previously used in testing (lines 2-24
  units long)
• Below is a sample trial with an intermediate
  length sample line. Only the selection of the
  neither comparison is reinforced.

NEITHER
Line 1
Line 25
66
Discriminabilty Measures

• In addition, reinforcement for selecting the
  short line used in training (1 unit ling) was
  provided only when the identical training line
  appeared as a sample. Likewise, reinforcement for
  selecting the long line (25 units long) was
  provided only if the long training line appeared
  as a sample.
• In summary, the reinforced trials were
• Sample 1 unit gtgt comparison 1 unit
• Sample 25 units gtgt comparison 25 units
• Sample 2-24 units gtgt neither comparison
• Once such selections were occurring reliably, all
  lines were presented as samples under extinction
  conditions
• generalization gradients indicated that
  participants systematically selected the short
  line in the presence of the short line, the
  neither comparison in the presence of all the
  other test lines (2-24), and the long line in the
  presence of the long line.
• These performances indicated that the students
  could discriminate the test lines from the
  training lines.
• Thus, the inference that dimensional classes had
  been established could be made because
  selection-based responses transferred to stimuli
  that were novel and discriminable from those used
  during training.

67
Discriminability Measures Pseudoconcepts

• comprised of stimuli that are arbitrary and
  heterogeneous (they do not share any common
  physical characteristics) (McIlvane, Dube, Green,
  Serna, 1993 Stromer, Mackay, 1997)
• Used in the following study to determine whether
  members of different fuzzy classes were
  discriminable from one another

68
Wasserman, Kiedinger, Bhatt (1988)

• Used same 40 slides as those used by Bhatt et al.
  (1988) in a previous study
• Trained pseudoconcepts with pigeons
• Class 1 cat slides 1-10 (peck upper left key)
• Class 2 cat slides 11-20 (peck upper right key)
• Class 3 flower slides 1-10 (peck lower left
  key)
• Class 4 flower slides 11-20 (peck lower right
  key)
• Thus, we are artificially partitioning two fuzzy
  classes into 4

69
Wasserman, Kiedinger, Bhatt (1988)

• Why? If birds cannot form these pseudoconcepts,
  then the members WITHIN each fuzzy class (cats or
  flowers) are NOT discriminable from each other
• Results showed that birds did confuse some
  members with each fuzzy category but they were
  correct more often than chance would predict

70
Effects of Number of Exemplars

• Only a small number of experiments have
  investigated how the number of training exemplars
  influences the establishment of perceptual
  classes
• these studies focused exclusively on fuzzy class
  formation

71
Cook, Wright, Kendrick (1990)

• used pigeons to investigate the demonstration of
  fuzzy classes consisting of slides of line
  drawings of birds or mammals
• Used either 5 or 35 exemplars of each during
  training
• During generalization test, only birds exposed to
  35 exemplars correctly responded to class members
• Birds exposed to 5 exemplars responded at chance
  levels during testing

72
Homa Little (1985)

• Participants were typical 10-year-old children
• effect of number of training exemplars on class
  demonstration was assessed on a within-subject
  basis
• kids first presented with three different
  experimenter-defined prototype stimuli that
  consisted of computer-generated, 9-sided figures
• kids were informed that each belonged to a
  particular category (designated by letters A, B,
  and C)
• Next, successive discrimination training was
  conducted with 18 new stimuli consisting of 3, 6,
  and 9 exemplars of the three potential fuzzy
  categories (exemplars were distortions of the
  prototype of each class, generated by changing
  the locations of the apices of each figure)
• That is, one class used 3 exemplars, one used 6,
  and one used 9

73
Homa Little (1985)
74
Homa Little (1985)

• During discrimination training, children received
  feedback for all category selections
• training continued until 100 correct
• Following training, transfer test determined
  degree of generalization to novel exemplars of
  the 3 potential classes
• Transfer test used 15 new stimuli from each of
  the 3 classes

75
Homa Little (1985)

• Test results showed that when only three
  exemplars were used during discrimination
  training for a particular potential class,
  class-consistent responding during the transfer
  test was near chance levels (33)
• This increased to about 50 for 6 exemplar
  training and 75 for the 9 exemplar training
  condition
• Thus, increases in exemplars increased likelihood
  of class formation
• Similar findings have been found by others
  (Becker, 1971 Bhatt Wright 1992 Engelmann,
  Carnine, 1982 Homa Chambliss, 1975 Homa,
  Cross, Cornell, Goldman, Swartz, 1973 Homa,
  Sterling, Treple, 1981 Omohundro, 1981 Sands,
  Delius, 1988)

76
Cerella (1979)

• Found that fuzzy classes could be established
  with only a single exemplar during training for
  pigeons
• successive discrimination training was used
• For group 1, 40 slides of different oak leaves
  were positive exemplars (Ss for pecking).
  Negative exemplars were 40 slides of leaves from
  20 other species of trees.
• For group 2, a single oak leaf was used as a
  positive exemplar (S for pecking) with the same
  40 negative exemplars used with group 1.

77
Cerella (1979)

• Following discrimination training by pigeons in
  both groups, a transfer test was conducted with
  40 novel oak leaves plus the 40 negative
  exemplars (non-oak) from training
• results showed that the pigeons in both groups
  correctly discriminated between instances of oak
  and non-oak leaves (with no differences across
  groups)
• But, is it possible that the birds couldnt tell
  the difference among the members in the oak
  class?...

78
Cerella (1979)

• Next, used a successive discrimination procedure
• pecks were now reinforced ONLY in the presence of
  the oak leaf slide used in single exemplar
  discrimination training
• Pecks were NOT reinforced for 40 oak leaf slides
  previously used during transfer testing
• Despite over 100 training sessions, pigeons could
  not discriminate between the single exemplar
  slide and the novel oak leaf slides
• Sowe really cannot infer that classes had formed

79
Why is ME Training Superior for Fuzzy Classes?

• Remember in a fuzzy class, no one exemplar
  contains all the necessary and sufficient
  features that define membership
• In addition, no one stimulus feature is either
  necessary or sufficient for membership
• Rather, membership in a potential fuzzy class
  requires presence of a combination of some of
  many features
• Each feature occurs with different probabilities
  across the class members
• For control of responding to be exerted by these
  combinations of features, the participant must,
  by necessity, be exposed to more than one class
  exemplar
• As the number of exemplars used during
  discrimination increases, there is a greater
  probability that the participant will be exposed
  to the some of many stimulus features that define
  class membership

80
Why is ME Training Superior for Fuzzy Classes?

• Thus, with each successive presentation of
  different exemplars drawn from a particular
  potential class, the some of many features that
  define category membership become correlated with
  reinforcement for responding
• responding in the presence of the some of many
  features that define a particular class is
  reinforced with a greater frequency than
  responding in the presence of irrelevant features
• these differential reinforcement histories select
  the combinations of features differentiating
  positive and negative exemplars of a class
  (Balsam, 1988).
• Only the presentation of multiple exemplars of
  the class during training makes it likely that
  class-defining combinations of many features,
  albeit ambiguously specified, will come to
  control responding

81
Why is ME Training Superior for Fuzzy Classes?

• When only a few or a single exemplar are used
  during training, however, there is a decreased
  probability that a sufficient some number of
  class-defining features will be correlated with
  reinforcement
• presentation of a single exemplar provides an
  equal likelihood that both relevant and
  irrelevant features will be correlated with
  reinforcement for responding
• This would be evident in non-differential
  responding to both positive and negative
  instances of a class during a transfer test
• Such performances indicate that responding during
  training had simply come under the control of
  individual stimuli rather than by the some of
  many defining features defining the class

82

• end