Transcript Raferty

False Memory Phenomena
James Raftery
Psy/Orf 322
Spring 2004
Can we trust our memory?
• Victim Testimony
• “Repressed” abuse in therapy
What is the issue?
• Memory as constructive process
• Coherent memory
– Bits of actual memory
– Context
– Repetition
False Memory Experiment
• Quantitative Measurements
– Inherent difficulties
• Simple measurable tasks = little error
• Reproductive VS Reconstructive?
– Henry Roediger/Kathleen McDermott’s
experiments
Method
•
•
•
•
•
Lists of Words
“Critical Lure”
Bidirectional associations
42-item lists read aloud
Subject wrote down as
many as could be
remembered without
‘guessing’
• At the end asked to judge
if they had seen certain
words on the list
Experiment Results
• Critical Lure 55% (higher than in exp1)
• Recall Condition Enhanced Effect
• Had physical memories of the lures
Implications
• False Memory is quantifiable
• All memory is constructive, not
reproductive – even rote recall
• Effects were substantial in a lab setting, in
a traumatic situation effects are likely even
more powerful
Parallel Distributed Process
• Experiences are represented in the mind
by their several aspects (within descriptive
categories)
• Different experiences share aspects
• Connections between two aspects are
strengthened when both are activated
Dynamic Nature Of PDP Memory
• Later events alter connections
• When one aspect is used as trigger for
memory, it activates all others that it is
connected with
• A lack of a strong connection will allow for
several weak connections to be asserted,
giving rise to confusion or generalization
False Memory Simulator
• Purpose: To replicate Roediger and
McDermott’s results with a neural network
• All words in list related to each other
• Primarily activate themselves, but to some
extent activate similar words
• If a word (critical lure) reaches a certain
threshold of excitement it will register as
having been on the list
Implementation
• Each word has a weight with all other
words on list
– Weights stored in a matrix
– Strong weight associated between a word and
itself
– Vector of input words put through matrix
determines net excitation of all words
– If a specific threshold is reached, the network
will output it as a recalled word