LANGUAGE AND THOUGHT

1
LANGUAGE AND THOUGHT
2
Overview

• Large number of artificial tasks introduced,
  e.g.
• 2-4-6 task
• Selection (4-card) task
• THOG problem
• Subjects shown not to adhere to normative model.
• Experimenters argue for irrationality of subjects
  - especially confirmation bias.
• Search for descriptive models for responses.

3
2-4-6 task (Wason, 1960)

• Artificial task to examine explicit hypothesis
  testing in lab.
• Procedure
• Experimenter thinks of rule about 3 numbers.
• E. tells S. a triple that fits rule (2-4-6).
• S. generates other triples (e.g., 0-2-4) and E.
  says whether each fits rule (Yes").
• Subject has to guess rule (e.g., Ascending
  numbers by twos.)
• Experiment ends when correct rule guessed (or
  subject gives up).

4
2-4-6 Task Results

• Most popular first guess is Ascending by twos.
• Subjects often fail to guess correct rule.
• Subjects tend to generate only triples consistent
  with their current rule.
• Labelled confirmation bias by Wason

5
2-4-6 Task Sample Protocol

• 1 3 5 Add two each time.
• 16 18 20 To test the theory that it's simply a
  progression of two.
• 99 101103 To test the progression of two theory,
  using odd numbers.
• As these numbers can have hardly any other
  connection, unless it is very remote, the rule is
  a progression of adding two.
• 1 5 9 The average of the numbers on the outside
  is the number between them.
• The rule is that central figure is the mean of
  the two external ones.
• 6 10 4 Difference between the first two
  numbers, added to the second number gives the
  third.
• 7 1115 To test this theory
• 2 25 48 To test this theory.
• The rule is that the difference between the first
  two figures added to the second figure gives the
  third.
• 7 9 11, 11 12 13, 12 9 8, 77 75 71
• Subjects gives up (45 minutes)

6
Normative Model of the 2-4-6 Task

• Logic tells us that a rule can never be proved
  true (without searching all possibilities) - can
  only be proved false.
• Example All swans are white.
• Cannot prove this rule true unless we examine all
  swans and find that each is white.
• BUT one black swan will disprove rule.
• Thus should search for falsification not
  confirmation (compare Poppers ideas about
  scientific method).
• Wason argues that subjects seek only confirmation
  and are thus in error.
• Further tests of confirmation bias include the
  4-card task/selection task (Wason, 1966).
• Again subjects appear to ignore the chance to
  search for falsifying evidence.

7
The Wason Selection Task

• OR, 4-card Task (Wason, 1966)
• Four cards, each has a letter on one side and a
  number on the other. You can see one face only.
• Rule If there is an A on one side of a card
  then there is an even number on the other side of
  the card
• Task Which cards do you need to turn over to
  discover whether the rule is true for all four
  cards?

A
K
7
2
8
Selection Task Sample Results

• Experiment by Beattie and Baron, 1988
• Considerable disagreement among subjects
• A and 2 selected 8 people
• A selected 8 people
• A, B, 2, 3 selected 1 person
• 2 selected 1 person
• A, 3 selected (CORRECT) 0 persons

9
Selection Task Revisited

• Do subjects really fail to seek falsifying
  evidence in selection task?
• Almost all subjects turn over A card which is
  potentially falsifying.
• Evans Lynch (1973) negated version.
• Rule If there is an A then there is NOT a 2.
• Results Most subjects turn over A and 2 (correct
  answer).
• Alternative hypothesis matching bias.
• Concrete materials (sometimes) help subjects
  adhere to normative model.
• Cheng Holyoak (1985) - pragmatic reasoning
  schemas.

10
Concrete Materials Effects in the Selection Task

• Issue Does familiar (non-abstract) context
  improve performance. If so, errors may be less
  significant.
• Wason Shapiro (1971)
• Rule Every time I go to Manchester I travel by
  car.
• Cards Manchester, London, car, train.
• Subjects correctly tend to turn over Manchester
  and train cards only.

11
Concrete Materials Effects in the Selection Task
(Cont.)

• But - consistent failures to replicate (e.g.,
  Yachanin, 1980).
• Also - concrete items alone are not sufficient.
• Manktelow Evans (1979) Every time I eat fish I
  drink gin.
• Cards Gin, Whisky, Haddock, Steak.
• Recurrent problem of whether familiarity improves
  performance.

12
Pragmatic Reasoning Schemas Cheng Holyoak (1985)

• Alternative hypothesis different concrete
  materials invoke different sets of rules which
  produce different card selections.
• Pragmatic reasoning schemas (Cheng, et
  al.,1986).

13
Pragmatic Reasoning SchemasExample Schemas

• Permission schema
• Rule If a customer is drinking alcohol, s/he
  must be over 21.
• Prediction beer, 17 (not juice, 21)
• Converse bias
• Rule If two items carry like electrical charges
  then they will repel each other.
• Prediction all cards
• Arbitrary
• Rule If a card has an A on one side it has a 2
  on the other side.
• Prediction confusion.
• Problems are logically equivalent, but
  psychologically different.

14
Evolutionary Psychology and the Search for
Cheaters

• Cosmides, Gigerenzer
• Suggestion The reason why we are good with
  social rules is that in complex primate societies
  it is important (and was important in
  evolutionary terms) to identify cheaters
  (animals who try to take a benefit without paying
  a cost).
• In the selection task you need to check that
  those that have not paid the cost (e.g. have not
  reached the legal drinking age) are not taking
  the benefit.
• Therefore, need to turn over the so-called
  not-Q card.

15
Confirmation Bias Conclusions

• Debate still rages whether subjects suffer from
  confirmation bias.
• Variety of hypotheses to explain subjects
  behaviour on classic tasks
• matching bias (Evans)
• concrete materials hypotheses
• pragmatic reasoning schema (Cheng et al)

16
Confirmation Bias Conclusions (Cont.)

• Points to take away
• Definition of confirmation bias is suspect
  (subjects in 4-card tasks turn over A)
• Subjects do not appear to adhere to rules of
  logic (normative model)
• Search still on for adequate descriptive model
• Subjects responses are radically changed by
  superficial features of task
• Errors are resilient - no training effects
• Considerable individual differences in responses.