The neuroscience of memory

The neuroscience of memory
The standard model of memory
• Sensory store (iconic memory)
• Short-term memory
• Long-term memory
Problems with the standard
model
• Is the sensory store really “memory”?
• Is the short-term store really so passive?
• What defines an “item?”
• Levels of Processing
Working memory
• There is clear evidence that the short-term
store is more active than just a temporary
warehouse. It is where we actually process
the information we’re using.
• Information enters working memory from
both perception and from LTM.
• Baddeley:
• Central Executive
• Phonological loop
• Visuo-spatial scratchpad
Neurological evidence of WM
Information encoding
• Propositional Hypothesis: All information is
stored in an abstract, amodal format called
“propositions”
• Based on the mathematics of propositional logic
• Propositions represent facts about the world, and can capture
properties of objects or relationships between two objects.
• Dual-Code Theory: We store information in two
types of codes:
• Verbal codes are abstract and amodal. Could be similar to
propositions
• Analog/Imaginal codes are tied to the particular sensory
modalities. For example, when we imagine something we are
actually loading a picture-based representation into working
memory.
Mental Rotation
• Shepard & Metzler (1967) did the classic
mental rotation experiment.
– Show people two arbitrary, 3-D objects at
different orientations.
– Ask people to judge whether the objects are
mirror images of each other or not.
– The amount of time it takes people to answer is
directly proportional to the degrees of rotation
by which the objects differ.
Relative size & Image scaling
• When asked to imagine two objects, it takes
longer to make judgments about the smaller
object
• It also takes longer to make judgments
about smaller features of objects
Image scanning
• Give people a picture of something, such as
a map of a simplistic island and have them
memorize it.
• Ask people to scan the image from one
point on the map to another.
• The time it takes to do this is proportional to
how far apart the two points are.
The Mind’s Eye hypothesis
• Kosslyn proposed that mental imagery is
functionally equivalent to vision.
• What he means by this is that mental
imagery uses the same internal
representations and processes that vision
does, but without visual input.
Neurological evidence for mental
imagery
• fMRI studies have shown that when
subjects perform imagery tasks, early visual
areas of the brain become active.
• It is also the case that when subjects are
asked to think about particular properties,
such as color or action, the parts of the brain
that process those properties (V4 and MT)
become active.
Unitary Content Hypotheses
(UCH)
(Caramazza et al., 1990; Caramazza & Shelton, 1998; Riddoch et al., 1988; Pylyshyn, 1973)
vision
naming
touch
categorization
hearing
etc.
SEMANTIC
MEMORY
usage
etc.
Multiple Semantics Hypotheses
(MSH)
(Paivio, 1971; Beauvois, 1982; Shallice, 1987, 1988; McCarthy & Warrington, 1988)
SEMANTIC MEMORY
vision
Visual
semantics
verbal
Verbal
semantics
touch
tactile
semantics
etc.
categorization
naming
usage
etc.
The Evidence
• Category-specific deficits (e.g., Warrington & McCarthy, 1983, 1987; Warrington &
Shallice, 1984; Gainotti & Silveri, 1996)
– Patients show impairments in processing living things vs. manmade objects and vice versa.
– MSH explanation: the Sensory-Functional Theory (Warrington & Shallice,
1984; Farah & McClelland, 1991)
• Modality-specific deficits
–Patients are unable to name visually presented
objects, but can name them from other modalities and
can access other semantic information about visually
presented stimuli (Beauvois, 1982)
–UCH explanation: priviliged access and privileged
relationships.
More evidence
• Brain-imaging (Martin et al., 1995; Martin et al., 1996)
– Brain areas associated with visual processing more
active when animals are being named;
– Areas associated with generating action words or
imagining actions active when naming man-made
objects
– Areas associated with color perception are active when
color words are generated, even for man-made objects
– Areas associated with motion perception are active
when action words are generated for man-made objects
• UCH: Are they just tapping into color or action
concepts?
Evidence (cont.)
• Naming vs. Categorization (Potter & Faulconer, 1976; Guenther et al.,
1980, Seifert, 1997)
– Faster to name words vs. pictures
– Faster to categorize pictures vs. words
• Stroop-like effects (Glaser, 1992; Glaser & Glaser, 1989)
– Incongruent words interfere with pictures more than
pictures interfere with words in a naming task
– Incongruent pictures interfere with words more than
words interfere with pictures in a categorization task
Conclusions
While both the UCH and MSH can be
modified to account for any one piece of
evidence, the preponderance of evidence
seems to favor a system organized more
closely to the MSH.
Computational models:
Graded specialization
• Multiple input and output
dimensions
• Encodes the notion of privileged
access and relationships in the
systematicity of the environment
• Further presumes that neurons like
short connections
• As a result, neurons close to
modality-specific inputs and
outputs respond better to those
inputs and outputs.
(From Plaut, 2002)
Future Directions
• Most existing models were designed and
tailored to account for only one of the many
phenomena mentioned
– e.g., Farah & McClelland, 1991 - SFT;
Caramazza et al., 1990 - optic aphasia, etc.
Graded specialization:
Accounting for evidence
• Optic aphasia - damage to the pathways between visual inputs and
naming outputs
• Category specific deficits - systematic relationship between
information from a given category leads to vertical organization of
category information that is orthogonal to modality-based organization.
• Experiment 1 explained in terms of different inputs accessing different
category representations
– Pictures have systematic relationships to each other, and thus to
categories.
– Words don’t.
• Experiment 3 simply implies that words have more systematic
relationships to their functions than to their categories.