Three points for today

1
Three points for today

• Sensory memory (SM) contains highly transient
  information about the dynamic sensory array.
• Stabilizing the contents of SM allows
  identification of external stimuli, which in turn
  permits access to knowledge.
• Stabilization occurs when the contents of SM are
  extracted and stored in Short Term Memory (STM)

2
The world and our representation of it

• A. External world
• Dynamic (in motion with respect to us).
• That is, something is always happening.
• Representations must be created fast.
• B. Internal representation of the world
• Begins with receptor surface.
• Receptor surface codes information about what is
  out there in the world.

3
Question

• Can the receptor surface drive all responses?
• That is, can all responses be prompted and guided
  only by information available at the receptor
  surface?
• E.g., shadow passing over squirrel causes
  squirrel to run for shelter without thinking.

4
(No Transcript)
5
Which behaviours could receptors control?

• Behaviours based on distinctions available at
  receptors - e.g., for vision
• light intensity
• wavelength (hue)
• contrast (light vs. dark)
• Nothing involving learning.

6
Which behaviours can receptors not control?

• Behaviours based on abstract properties for
  vision, these properties include
• size
• shape
• category
• name
• These properties are computed based on
  information extracted from SM.

7
Behaviours receptor surface cannot control

• Most human behaviours of interest require
  information about those abstract properties. But
  because the world is dynamic, SM is continually
  being refreshed.
• Representations in SM dont last long enough for
  analysis of abstract properties.
• We need to extract and stabilize these
  representations, in order to find out what is out
  there. How do we do that?

8
How do we extract information from SM?

• Two issues
• 1. How does the extraction process work?
• 2. What is the nature of the stimulus
  representation that results?

9
How does the extraction process work?

• Well begin by looking at some of the functional
  characteristics of SM that will influence our
  model of the extraction process. Well ask
• How long do representations in SM last?
• How many representations can SM contain?
• Then, well look at Rumelharts model .

10
Functional characteristics of sensory memory

• Sperling (1960)
• tachistoscopic procedure very brief exposure of
  a matrix of letters (because letters are a
  convenient stimulus)
• subjects task is to report whats in the
  matrix.
• until Sperling, whole report technique used.

11
80 years of Whole Report results

• Consider a typical display of 3 rows of 4
  letters per row.
• Subjects in the whole-report task typically
  reported 4 or 5 out of 12 letters.
• Did that mean that SM could contain only 4 or 5
  letters at one time?
• Or did the rest of the display decay during
  readout of the first 4 or 5?

12
Theory 1 thats all we can get into SM
J V E C M B S A Y L K R
J V E C M
J V E C M
13
Theory 2 rest decays during read-out
J V E C M B S A Y L K R
J V E C M B S A Y L K R
J V E C M Thats all I remember.
14
Sperlings (1960) Partial Report Technique

• After display disappears, a tone signals which
  row to report.
• High tone, top row. Medium tone, middle row,
  etc.
• Logic if you can report a randomly selected
  row, you could have reported the other rows, too
  if a different tone had sounded.

15
Sperlings results

• 76 correct in partial report condition.
• Suggests that most of a complex display is
  available in SM immediately after offset.
• 76 of 4 letters 3 letters. Why not 4 or 5 as
  in whole report condition?
• After 5 seconds, most of array has decayed.

16
Characteristics of Sensory Memory

• We asked about the duration and capacity of
  Sensory Memory. Sperlings data tell us
• Duration under ideal circumstances (display
  followed by darkness), a second or two.
• Capacity up to 75 of a complex visual display.
• Now, lets look at Rumelharts model

17
Rumelharts model of the extraction process

• Extraction process takes information from SM to
  STM.
• Process operates in parallel on all contents of
  SM (unless task instructions say focus).
• Process influenced by stimulus quality better
  quality, faster process.
• Explains Sperlings results.

18
Rumelharts model
J V E C M B S A Y L K R
J V E C M B S A Y L K R
M B S
19
The nature of the stimulus representation

• In SM, the representation is unprocessed it
  contains information about environmental energy
  levels (e.g., where the image is light or dark).
  But the representation in STM is abstract. That
  means it has some form. What is that form?
• Think of the difference between muffin mix in the
  mixing bowl and muffins. The mixing bowl is SM.
  The muffins are the contents of STM. How do we
  get from mix to muffins?

20
The nature of the stimulus representation

• Three suggestions, in historical order
• Template theory
• Feature theory
• Structural description theory

21
Template theory

• information extracted from SM processed to give
  stimulus representation in form of a picture of
  the object being viewed.
• memory also contains pictures of objects.
• recognition involves searching memory for the
  picture that matches current stimulus
  representation.

22
(No Transcript)
23
(No Transcript)
24
A problem with template theory

• How do you recognize objects at an unfamiliar
  orientation?

25
Feature theory

• all visual objects are composed of features
• a stimulus representation is created by
  extracting features from the visual object
• memory contains representations in the form of
  lists of features of recognizable objects
• recognition consists of matching feature sets

26
Problems with feature theory

• 1. What are the features (that is, what are the
  visual features of objects that we supposedly
  extract from visual SM)?
• nobody knows
• 2. What about spatial relations among features?
• not specified in the theory

27
Chair?
28
Template and Feature theories compared

• Template theory deals with whole objects.
  Templates do not have component parts. A template
  is like a photograph of an object. If you have a
  photo of a Lego building, you cannot separate the
  individual bricks out. The photo is all one
  piece, though the actual object represented is
  made of many pieces.
• Feature theory deals with representations that
  have internal structure the whole is made up
  of many parts. Pairs of objects may share some
  features and not others.

29
Structural descriptions
A feature theory with spatial relations among
features specified in description. Biedermans
Recognition by Components theory Geons
geometric ions (primitives)
30
(No Transcript)
31
In column B, the geons can be recovered, so the
objects can be identified. Not so in column C.
32
Problems with SDT

• Difficult to specify geons that compose some
  objects.
• what are the geons for a loaf of bread?

33
Selection is the responsibility of attention
see next chapter. But note William
James Selection is the function of
consciousness. Consciousness refers to
experiencing the world.
34
Which comes first? Do we have consciousness
because we have stable representations? Or do we
have stable representations because we have
consciousness (so are able to choose objects and
responses)?