Bio-cog PART TWO
13.1; long-term memory II
Perception
What does it mean to recognize an object?
Stimuli → primary sensation > recognition > ? > response selection → response
Semantic memory: consists of many related features, this forms a whole image
Recognition of an object: the experience that the object belongs to a specific category earlier
experienced (seen, heard, felt, smelled) objects. → top-down
➔ Categorization
Categorization:
- Definitions
- Prototypes
- Examples
Definitions
Definition: A list of necessary and sufficient properties
Example: chair
- 4 legs, seat surface, backrest
There are exceptions, when for example a chair doesn’t have 4 legs, but still is a chair. These
borderline objects, make it harder to form definitions.
Example: male vs female
Do definitions mean anything when we view a male or female? Neurons in the inferior temporal
cortex can distinguish male and female faces. – Afraz 2015
Prototype
Prototype: central representation of a category: the average or centre of gravity of all earlier
experienced objects of that category.
Graded representation: membership of an object to a category depends on the similarity (distance) to
the prototype, how much the object looks like the prototype.
,Examine prototype
- Method: rating scale. Participants rate the typicality of a certain
example for a category.
- Method: sentence verification: apple is a fruit (true/ false)
- Method: priming: in same/ different judgement
memory is drawn towards prototype → bartlett’s drawing
experiment. The participant have to redraw a given image, this
is never 100% accurate.
- Subjects saw the previous image, then drew it themselves form
memory
Prototypical environments and events
Schemata: general knowledge about situations, e.g. kitchen, museum or university office. We fill in the
gaps with existing knowledge.
Scripts: general knowledge about a sequence of events, e.g. how to: make tea, take a metro, buy a
bread etc. Our brain fills in some scripts with information that doesn’t belong there. → false memories
13.2; what is being stored?
Idea of a prototype is attractive:
Concepts are intrinsically fuzzy – they have no fixed boundaries, like human categorization. This
explains finding of typicality (rating, RT, priming, naming, etc.)
What is stored?
- The prototype, abstract from earlier experiences?
- Exemplars: representation of each actual experience
Prototype theory: only the prototype is stored: an abstraction of all earlier experiences.
Exemplar approach: all the experiences are stored. When we are asked about the object, we generate
the prototype. When we categorize we DO use all experiences.
,Prototype vs. exemplar approach
Prototype:
- Advantage: very efficient storage.
- Disadvantage: requires separate storage of unique exemplars, e.g., my dog Bella.
Exemplar:
- Advantage: unique exemplars are stored as any other exemplar; no categorization problem.
- Disadvantage: a lot of capacity required for storage.
Organisation of semantic memory
Semantic memory seems to be structured:
- One fact triggers the retrieval of related facts
- Mistakes are often near-misses
What is the structure of semantic memory like?:
- Semantic network: the hierarchical model
- Multiple-trace theory
- Parallel distributed processing
Hierarchical network
Properties
- Nodes: concepts
- Activation: energy is propagated via links
- Links: properties or examples
- Inheritance: lower nodes inherit properties of higher
nodes
Cognitive economy: each property is represented in a single place in the tree.
Empirical problems:
- Cannot deal with typicality effects. Has been shown that RT ( an ostrich is a bird) > RT (a canary
is a bird). The hierarchical network predicts equal RT’s
- Principle of cognitive economy. It has been shown that: RT (a pig is a mammal) > RT (a pig is an
animal). hierarchical network predicts the opposite.
, Multiple trace model
Models of memory consolidation (either standard or multiple trace model) are also relevant to
knowledge representation.
- Standard trace model: memories become independent of the hippocampus after several years
and stay in the neocortex only
- Multiple trace model: memories keep on interacting with the hippocampus and the neocortex
Assumptions multiple trace model:
- A new memory trace is created for each new experience exemplars
- On each experience all corresponding old memory traces are simultaneously retrieved
- Individual traces become weaker as time passes, while memory strengthens through multiple
traces
➔ Recognition (or activation of facts) is the result of the collective activation of old memory
traces
Explains many phenomena:
- Typicality effect: you have more memory traces of frequent experiences → exemplars
- Automatization: enabled by obligatory retrieval of (many) other memory traces
- Semanticization of episodic memory: when many memory traces are activated in concert, any
unique episodic experience is drowned in the multitude of connections.
Criticism on semantic networks:
- Vulnerability: if one node gets damaged, all knowledge related to that node is lost → specificity
coding (1 neuron focusses on 1 concept)
- No clear ideas about learning → how is knowledge acquired
A solution of these problems is provided by parallel distributed processing → distributed coding
Lecture 13.3
Parallel distributed processing
Developed during the 80s
Not vulnerable: if there is damage, the info is still represented
- Representation of data is depended on a wide range of nodes
Other names are: connectionism, distributed networks and neural networks
The input units are connected with hidden units to output units.