Working Memory: The Feature Model

1
Working MemoryThe Feature Model

• Presented by Umer Fareed

2
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

3
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

4
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

5
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

6
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

7
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

8
The Feature Model

• Items in PM and SM comprise sets of
  internally-generated 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

9
The Feature Model

• Serial Position Function Effect of change in MD
  features
• while keeping MI features constant

10
The Feature Model

• Item Presented Primary Memory
• 1 -1 1 1, 1 -1 1 -1 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 -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 -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
• 4 -1 1 -1 -1, -1 1 1 1 -1
  1 -1 -1, -1 1 1 1
• Example Four Items are presented

11
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

12
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
  item1s 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

13
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

14
Simulations using Feature Model

• The Serial Position Function
• Recall generally declines over serial positions
  due to output interference
• If feature x of item n1 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

15
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

16
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

17
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

18
Simulations using Feature Model

• Suffix Effect and Articulatory Suppression Effect

19
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

20
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

21
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

22
Questions

1. How has working memory capacity been found to
   affect dichotic listening?
2. Identify two shortcomings of Nairnes feature
   model?
3. Identify two advantages that Nairnes feature
   model enjoys over Baddeleys working memory model?