Transcript n+1

Working Memory:
The Feature Model
Presented by: Umer Fareed
Outline








Basic Concept of Working Memory (WM)
Three Component Model of WM
Working Memory Capacity
The Feature Model
Feature Model Operation
Simulations using Feature Model
Pros and Cons of Feature Model
Conclusion
Basic Concept of Working Memory




A dedicated system that maintains and stores
information in the short term
Supports human thought processes by providing
an interface between perception, long-term
memory and action
Information decays quickly unless actively
rehearsed
Sometimes referred to as short term memory
Three Component Model of WM





Proposed by Baddeley and Hitch
Comprises a control (supervisory) system and
two storage (slave) systems;
The Central Executive
The Visuospatial Sketchpad
The Phonological Loop
Working Memory Capacity




Measures the extent to which a person can
control and sustain attention in face of
interference and distraction
Ability to activate items in memory and to
ignore or disregard interfering items
Correlate highly with many cognitive tasks
including reading comprehension, spelling,
vocabulary learning, writing and reasoning
Not about storage and processing but is
about retention over a period in which there
is distraction from stored information
Working Memory Capacity




Measured by a variety of tasks, commonly
used are Reading Span and Operation Span
Reading span includes reading of a number of
sentences and trying to recall the last word of
each sentence in their correct order
In operation span, subject performs
mathematical calculations and word reading
and then asked to recall the words in order
Larger WM capacity means better ignorance
to irrelevant or distracting information
The Feature Model




Proposed by Nairne in 1988
Computational model of Serial Recall guided by
a set of “Primary Memory” (PM) cues of
varying effectiveness in identifying the target
item from a search set defined within
“Secondary Memory” (SM)
Items are assumed to be represented as a set
of features
A simplifying assumption is that only
immediately adjacent items interfere
The Feature Model



Items in PM and SM comprise sets of internallygenerated modality-independent (MI) and
externally generated modality-dependent (MD)
features organized as row vectors
MD features represent the conditions of
presentation whereas MI features represent the
nature of the item itself
For each partially degraded trace in PM, the
subject tries to select an appropriate recall
candidate by comparing degraded traces with
intact traces in SM search set
The Feature Model
Serial Position Function: Effect of change in MD features
while keeping MI features constant
The Feature Model
Item Presented
1
[ +1 -1 +1 +1, +1 -1 +1 -1]
2
[ -1 -1 +1 -1, -1 -1 -1 -1]
3
[ +1 -1 -1 -1, +1 +1 -1 +1]
4
[ -1 +1 -1 -1, -1 +1 +1 +1]
Primary Memory
[ +1 -1 +1 +1, +1 -1 +1 -1]
[ +1 0 0 +1, +1 0 +1 0]
[ -1 -1 +1 -1, -1 -1 -1 -1]
[ +1 0 0 +1, +1 0 +1 0]
[ -1 0 +1 0, -1 -1 0 -1]
[ +1 -1 +1 +1, +1 -1 +1 -1]
[+1 0 0 +1, +1 0 +1 0]
[ -1 0 +1 0, -1 -1 0 -1]
[+1 -1 0 0, +1 0 -1 0]
[ -1 +1 -1 -1, -1 +1 +1 +1]
Example : Four Items are presented
Feature Model Operation




Each to-be-remembered item is made up of
features
Assume there are 20 MI features and 20 MD
features, each of which is randomly set to a
value of 1 or -1
These 40 features represent the first item
presented, second item presented also
contains 40 features
Item2 features overwrite Item1 features due
to retroactive interference
Feature Model Operation



If MI feature no 5 of item2 has the same value as
MI feature no 5 of item1, then original value of
item1’s feature 5 is overwritten with a value of 0
Then item3 and item4 are presented, the final
item presented is not followed by any external
information but by rehearsal
At end of list presentation, primary memory
contains trace of each item presented but these
traces are degraded due to certain features being
overwritten
Feature Model Operation



Only the modality-independent features of final
item are overwritten whereas its modality
dependent features remain intact
The subject then tries to match each primary
item with an intact secondary memory trace
Beginning with first item, each PM item is
compared with SM items in the comparison set,
SM item with fewest mismatching features will
be selected as a candidate for recall
Simulations using Feature Model




The Serial Position Function
Recall generally declines over serial positions due
to output interference
If feature x of item n+1 is identical to feature x
of item n, then the value representing feature x
of item n is lost and cannot be used as a recall
cue
The recency effect seen in serial recall of
auditory items arises because the MD features of
the last list item are not overwritten whereas
recency effect is not seen in visual presentation
Simulations using Feature Model




The Suffix Effect
Occurs whenever modality-dependent
features of the final item are overwritten
Speech suffix significantly reduces serial recall
of auditory presented lists overwriting
modality-dependent features of list of speech
items
Visual suffixes have little or no effect on
auditory list items
Simulations using Feature Model




Effects of Irrelevant Speech and Articulatory
Suppression
Immediate serial recall of verbal material is
reduced significantly if participants are exposed
to irrelevant speech while studying the list items
If the target item is presented auditorily then
irrelevant speech reduce serial recall
performance under articulatory suppression
Feature adoption occurs when some of the MI
features in PM are replace by features of word
that is articulated
Simulations using Feature Model




Effects of Irrelevant Speech, Phonological
Similarity and Word Length
With visual presentation of list items, the feature
model predicts that there will be no effect of
phonological similarity under conditions of
irrelevant speech
Recall of visually presented material depends
heavily on MI features, therefore, phonological
similarity will produce no further damage to them
World length effect is eliminated by articulating
suppression
Simulations using Feature Model
Suffix Effect and Articulatory Suppression Effect
Pros & Cons of Feature Model



Can address effects at different serial
positions, including modality and suffix effects
observable with auditory stimuli
Readily explains the absence of time-based
word-length effects whereas working memory
(Baddeley) has to predict that they will occur
Provides precise and unambiguous
predictions as numbers used are easy to
compare
Pros & Cons of Feature Model



No role of time in the Feature Model
Only addresses serial recall not free
recall or recognition
Recall of single list implemented in the
model whereas most experiments ask
subjects to recall multiple lists
Conclusion



Differs from other working memory
models as it does not use the concept of
decay, rather interference degrades
memory performance
Accounts for suffix effect remains even
when subjects engage in articulatory
suppression
Directly addresses modality effects
Questions
1.
2.
3.
How has working memory capacity
been found to affect dichotic listening?
Identify two shortcomings of Nairne’s
feature model?
Identify two advantages that Nairne’s
feature model enjoys over Baddeley’s
working memory model?