Working memory and categorisation Exploring the relationship between these facets of cognitive funct

1
Working memory and categorisationExploring the
relationship between these facets of cognitive
functioning

• Stewart Craig
• Stephan Lewandowsky
• University of Western Australia

2
Individual variation in categorisation

• Different transfer strategy use
• Individuals categorise in various ways
• Not all doctors will diagnose the same
• Also occurs in the lab
• Participants can show strikingly different
  strategy use
• Importance of individual differences has recently
  been recognised

3
1 1 1 2
2 2 1 1
1 2 2 2
2 1 2 1
4
Individual variation
5

• Has been recognised that these strategies may
  hold important information about the way people
  categorise
• We want to explore what leads people to choose
  particular strategies?

6
Relationship to working memory capacity

• Evidence for relationship between WMC and
  categorisation
• Lewandowsky (2009)
• High WMC learnt categories Shepard, Hovland, and
  Jenkins (1961) categories quicker
• Yang, Lewandowsky, and Jheng (2006)
• WMC was only variable that predicted strategy
  choice in categorisation task

7
Aims

• Explore relationship between WMC and
  categorisation
• Explore strategy selection in categorisation

8
Experiment

• 143 participants
• - Sufficiently large for SEM
• 2 Sessions
• 1. Categorisation task
• 2. WMC tasks

9
Working memory capacity tasks

• Presented with series of working memory tasks
• Memory updating task (Salthouse, 1991 Oberauer,
  2000)
• Operation span task (Turner, 1989)
• Sentence span task (Daneman, 1980)
• Spatial short-term memory task (Oberauer, 1983)

10
Categorisation task

• Presented with alien cells
• The stimuli varied on four binary dimensions
• Alien cells
• Size, wall colour, centre colour, no cell walls

11
Category structure

• 5 4 category pattern (Medin Schaffer, 1978)
• Wanted to explore WMC in relation to popular
  category structure
• People show different strategy use

12

• Training items
• (receive feedback)

Transfer items (no feedback)
32 blocks of training Transfer block at end
13
Learning data
14
Consistent with previous data(e.g. Johansen
Palmeri, 2002 Medin Schaffer, 1978 Nosofsky,
Palmeri, Mckinley, 1994)
15
Generalisation profiles
Trans item 1234567

• R1

X
R3
16

• Lack of relationship between WMC and frequency of
  use

17
(No Transcript)
18
Final SEM

• Tested various models
• Latent categorisation and WMC variables
• Explore each part of model first

19

• Single WMC variable
• All four tasks
• ?2(15) 2.4
• CFI .996
• RMSEA .044
• SRMR .26

.71
.39
.88
.65
20
(No Transcript)
21

• Latent categorisation variable
• Three measures
• Training behaviour
• 1. a
• Transfer behaviour
• 2. Simplicity
• 3. Hamming distance
• Will explore in turn

22

• 1. Learning

• Rate at which people learn
• Best captured by exponential learning curve
• Exponential fit better than power law in 61.03
  cases
• a is slope of curve

Example participant
23
2. Simplicity

• People may categorise items based on simplicity
  (e.g. Chater, 1996, 1999 Feldman. 2003, Pothos
  Chater, 2002)
• People may prefer simplest way to categorise
  objects
• Captured by Bayesian models of categorisation
• Feldman (2003) found learning of Shepard et al.
  (1961) related to minimum bit length

24
Exploring simplicity

• Looked at simplicity in the present category
  space
• Model designed by Pothos and Chater (2002)
• Model looks at pairwise inequalities between
  distance of objects in category space
• Forms grouping which minimises within group
  differences, maximises between
• Simplicity (in bits) given by sum of
• Data left unspecified cost to correct errors
  cost to specify clusters

25
Simplicity in task
26
Simplicity in task
X strategy
R3 rule
R1 rule
27

• So
• Seems that people tend towards simpler patterns
  of responding, consistent with idea of simplicity
• Simplicity can be used to capture strategy choice
• Not all simple patterns are used
• Simplicity doesnt explain everything

28
3. Hamming distance

• To obtain indication of different strategy use
• Focusing on popular response patterns
• R1 rule
• R3 rule
• Exemplar strategy
• Calculate Hamming distance from each of these
  strategies
• Includes people who just missed strategies

29
Hamming distance
30
People tend towards popular strategies
31
Categorisation

• 1. a
• 2. Simplicity
• 3. Hamming distance
• All load on latent variable
• Best model includes both learning behaviour and
  strategy choice

.87
.87
.57
-.23
?2(2) 0.6 RMSEA 0 CFI 1 SRMR 0.0109

• Link between training and transfer behaviour

32
Back to model
33
Back to model
?2(15) 13.2 CFI 1 RMSEA 0 SRMR 0.0352
34
?2(15) 13.2 CFI 1 RMSEA 0 SRMR 0.0352
35
Summary

• WMC predicts categorisation behaviour
• Relationship between WMC and training replicates
  Lewandowsky (2009)
• Could not predict which strategy people would use
• WMC related to closer to popular strategy use
• WMC related to simplicity of transfer behaviour

36
Thank you