The nature of working memory capacity in sentence comprehension: Evidence against domain-specific working memory resources

1
The nature of working memory capacity in sentence
comprehension Evidence against domain-specific
working memory resources

• Federenko, Gibson, Rohde
• Journal of Memory and Language, 2006

Kacey Wochna Psycholinguistics November 3rd, 2010
2
Background

• What is the nature and functional organization of
  working memory?
• Domain-general or domain-specific?
• Is verbal working memory general or specific?
• All verbally mediated tasks use the same pool of
  VWM resources (King Just, 1991 Just
  Carpenter, 1992)
• Linguistic and non-linguistic verbal tasks use
  different VWM pools (Caplan Waters, 1999)

3
Background

• Dual-Task Approach On-line sentence processing
  and a non-linguistic verbal task are performed
  simultaneously
• If VWM is domain-specific...
• Memory load in a non-linguistic verbal task
  should not interact with syntactic processing (as
  in Caplan Waters, 1999)
• If VWM is domain-general...
• The two tasks should interact
• Interaction found using the individual
  differences approach (Just and colleagues)
• Caplan and Waters found off-line interactions,
  but argued that off-line processing goes beyond
  linguistic processing

4
Current Study

• Based on Gordon, Hendrick, and Levine (2002)
• In previous research, memory load was defined as
  the number of items kept active in memory
• Proposed that load should be measured by the
  amount of interference produced by the items kept
  active in memory
• Similarity-based interference probably affects
  retrieval
• Low syntactic complexity (subject-extracted
  cleft)
• It was the dancer that liked the fireman before
  the argument began.
• High syntactic complexity (object-extracted
  cleft)
• It was the dancer that the fireman liked before
  the argument began.
• Match (high similarity) poet cartoonist
  voter
• Non-match (low similarity) Jim Greg Andy

5
Current Study

• Difference in rate of comprehension errors
  between low and high complexity was larger when
  memory load was matched
• BUT on-line reading times only showed a trend
  towards this interaction
• Federenko et al.
• Moving-window instead of center-screen
  presentation
• Amount of interference is possibly a function of
  both the similarity of the items and the number
  of items

6
Method

• 32 experimental items, self-paced moving-window
  presentation
• Low syntactic complexity (subject-extracted
  relative clause)
• The physician who consulted the cardiologist
  checked the files in the office.
• High syntactic complexity (object-extracted
  relative clause)
• The physician who the cardiologist consulted
  checked the files in the office.
• Match (high similarity) poet cartoonist
  voter
• Non-match (low similarity) Joel Greg Andy
• Easy load (one noun) Hard load (three nouns)
• Procedure
• memory nouns -gt sentence -gt recall -gt 2
  comprehension questions

7
Results

• Comprehension questions
• Better accuracy in the easy load condition than
  the hard load condition
• Marginal three-way interaction of load,
  similarity, and complexity trend for load to
  affect high complexity sentences more was more
  pronounced in the match condition
• Didnt replicate Gordon et al.s finding,
  probably due to procedural differences

8
Results

• Reading times
• Three-way interaction
• in critical region
• Easy load
• Longer reading times
• for high complexity
• No effect of match
• Hard load
• Match only caused
• longer reading times for
• high complexity

9
Results

• When the number of items that had to be kept in
  memory was greater, people processed
  syntactically complex sentences more slowly when
  the items to be kept in memory were more similar
  to the nouns in the sentences.
• Non-linguistic verbal memory loads interact with
  syntactic processing
• Both the similarity of items and the number of
  items contribute to interference
• Support for domain-general VWM system