3 points for this lecture

1
3 points for this lecture

• 1. Knowledge influences perception.
• 2. Knowledge can influence perception because
  learning speeds up access to LTM.
• 3. Access to LTM is speeded up by development of
  a virtual short term memory, called Long Term
  Working Memory (LT-WM)

2
3 points for this lecture

• Knowledge influences perception.
• Knowledge can influence perception because
  learning speeds up access to LTM.
• Access to LTM is speeded up by development of a
  virtual short term memory, called Long Term
  Working Memory (LT-WM

3
1. Knowledge influences perception

• Biederman, Rabinowitz, Glass, Stacy (1974)
• Subjects better at identifying briefly-presented
  objects that were expected in a context.
• Its easier to see things you expect to see.
• All of us use expectancies in seeing the world.
• But what do we expect to see?

4
What do we expect to see?

• Chase Simon (1972)
• Compared chess master, intermediate and novice
  players.
• Subjects viewed chessboard in midgame, then
  reconstructed it from memory (0 delay).
• Grandmaster saw and remembered more than the
  other two.

5
How did the GMs knowledge help him?

• The GM looked for patterns in the display.
• Two pieces classed as in the same chunk if set
  down less than 2 seconds apart.
• GM had more and larger chunks.
• Pieces put down in succession by GM shared more
  relations (e.g., type, colour, defence).

6
How do patterns help experts?

• Chi, Feltovich, Glaser (1981)
• Compared 1st yr. undergrads (novices) and senior
  Ph.D. students in physics (experts).
• Subjects grouped physics problems.
• Novices classified on basis of surface experts
  used underlying structure (e.g., Newtons Second
  Law), ignoring surface differences.

7
Review

• We all use expectancies in ordinary perception.
  Its easier to see things we expect to see
  (Biederman).
• Experts show a pronounced form of this effect
  they develop precise expectations for their skill
  domain. (Chase Simon)
• Those expectations allow experts to recover the
  underlying structure of their domain. (Chi et al.)

8
3 points for this lecture

• Knowledge influences perception.
• Knowledge can influence perception because
  learning speeds up access to LTM.
• Access to LTM is speeded up by development of a
  virtual short term memory, called Long Term
  Working Memory (LT-WM

9
How can knowledge influence perception?

• Perception happens fast.
• How can we retrieve knowledge fast enough to
  influence rapid perception? Two theories
• Superior performance based on innate ability.
• Superior performance based on learning.

10
Superior performance based on talent

• The first possibility is that experts are good
  because of some trait, something they were born
  with.
• This view has three implications
• 1. People with basic training should be capable
  of excellent performance because they have talent.

11
Superior performance based on talent

• 3 implications of the talent hypothesis
• 2. Aptitude tests should be good predictors of
  performance even after years of experience.
• 3. Should be an upper limit to how good a
  persons performance can be (specified by their
  talent).

12
3 implications of the talent hypothesis

• All three claims are false.
• 1. With only basic training, no-one does well.
• 2. Aptitude tests are poor predictors of
  performance after several years of experience.
• 3. If there is an upper limit to performance, we
  havent found it yet. (Consider Olympic athletes,
  difficult violin pieces, of 100 years ago.)

13
Superior performance based on learning

• Some examples of superior performance
• Blindfolded chess master, George Koltanowski
  could play 30 opponents at once, winning most
  games, drawing the others. (Koltanowski, 1985)
• An expert waiter, J.C., rapidly takes orders
  from up to 20 customers at one table. Never mixes
  them up. Always delivers right meal to each
  person. (Ericsson Polson, 1988).

14
Superior performance based on learning

• How do these experts do this? In playing chess or
  taking food orders, you need
• Fast access to a memory store.
• Large capacity in that memory store.
• But humans have two stores one for fast access
  (STM) and one for large capacity (LTM).

15
3 points for this lecture

• Knowledge influences perception.
• Knowledge can influence perception because
  learning speeds up access to LTM.
• Access to LTM is speeded up by development of a
  virtual short term memory, called Long Term
  Working Memory (LT-WM

16
Virtual short-term memory

• Ericsson Kintsch (1995) argue that experts have
  a virtual short-term memory. E K call it, Long
  Term Working Memory (LT-WM).
• Through experience, you set up a virtual STM
  inside LTM a rapid access store without the
  capacity limit.
• Based on Chase Ericsson (1982).

17
Chase Ericcson (1982)

• Used the digit-span task
• Subject hears a sequence of digits, like 7 4
  9 5 1 3, and repeats them back.
• Score number repeated back without error.
• Subject S.F., a long distance runner, had a digit
  span of over 80 digits.

18
How did S.F. do that?

• S.F. began with groups of four or five numbers,
  which he coded as times for distances (e.g., 3
  5 9 6 3 min. 59.6 seconds, for 1 mile
  race).
• He then grouped the groups into supergroups, then
  grouped the supergroups, producing a hierarchical
  network structure.
• At the top of the hierarchy was a node. That
  node went into STM.

19
Top-level node
3 5 6 9 4 2 8 1 7 6 1 3 5 2 9 8 6 1 5
3 4 7 3 6 8 5 2 9 1 4 7 3 2
20
How did S.F. do that?

• At the end of a long session of hearing, storing,
  and recalling lists of digits, he could
  accurately retrieve all of the lists.
• C E could specify a location in the network for
  a given list, and S.F. could tell them the digits
  in that location in the network in his memory.
• He must have been capable of very rapid storage
  in a long-term store.

21
Extending the model to expertise in general

• Ericsson Kintsch expanded Chase Ericssons
  idea into a general model of expert behaviour
• Network retrieval structures are rapidly created
  and stored by experts.
• With top-level node in STM, the whole structure
  becomes rapidly available.

22
Review

• Information is stored in LTM. Each item is
  associated with a cue.
• All cues are related in a hierarchical retrieval
  structure, under a top-level node.
• With top-level node in STM, any item under node
  can be retrieved.
• Fast access large capacity LT-WM