COMPARE AND CONTRAST TWO CLASSICAL APPROACHES IN SEMANTICS: APPROACHES TO LEXICAL-
SEMANTIC REPRESENTATION
Prototype approach versus Exemplar approach
Prototype approach Exemplar approach
Rosch, 1975 Medin & Schaffer, 1978
• Category structure is based on Prototype = a con- • categorisation structure based on similarity of
cept presented by a collection of characteristic object to exemplars of the category versus total
features similarity object to nn-exemplars
• Features are cue to category membership but the • Previously encountered exemplars
Boundaries for the prototype definition are not set: • assumes that we can only retrieve memories of
features can also not exist in the prototype specific instances of a category
• no abstraction of prototypes
• members why resemble the prototype strongly =
high-prototypicality. connections to the family re- • learn concept by being exposed to it (e.g. dog)
semblance • new stimuli are classified according to how
closely they resemble these exemplars
• Typicality effect = describes the fact that the the • Similarity influences classification the most
high prototypical members are highly recognised • Semantic decisions = retrieve exemplars one at
as members of category a time from semantic memory until a decision
• Potential members of the category are identified can be made
by how closely they resemble the prototype or cat- • compute total similarity of current instance to
egory average. memories of positive and negative exemplars
• “best example” of a concept ( dogs, non-dogs)
• a special type of schema • Decide that the exemplar is a dog if it is more
similar to the memories of dogs than memories
of relevant non-dogs
Hertley reports Medin & Shoeben (1988)
- “A penguin in a bird.”” vs. “A sparrow is a bird” Participants judged the typicality of sons into two
classes: SPOON & LARGE SPOON
- queries involving a prototypical members (e.g. is
a robin a bird) elicited faster response times than - typicality was mediated by material
for non-prototypical members - Metal = spoon
- Wooden = large spoon
• Sparrow is more prototypical than penguin, thus
decisions were much faster judgements made on stored instances
• When asked an example of bird ire likely to sate a
prototypical category: Sparrow
• high prototypical objects are strongly affected by
priming
Problems Problems
- does not explain why categories cohere - offers no amount about how categories are
not all concepts have prototypes (abstract) formed beyond memorisation
- prototype for “truth” - Vague definition of exemplar
- not all types of concepts appear to have proto- - Abstract processes ?
types (Harley)
Traditional views of the conceptual system
mental representations do not resemble the perceptual states from which they originate
Transduction problems: what exactly is this redescription process that produces amodal symbols
from modality-specific (perception, action) states?
, Knowledge is embodied or grounded in bodily states and in the brain’s modality-specific systems
(Barsalou, 1999).
Semantic networks
Collin & Quallian, 1969
• this approach proposes that concepts of the mind are arranged in networks = knowledge base
that represents semantic relations between concepts in a network
•
• a functional storage-system for the “meanings” of words represented with graphing systems - semantic
representations
• conceptual categorisation used on logical classification rules = conjunction of necessary features
• Properties of a concept could be “stored” (placed) next to a “node” representing the concept
• links between nodes represent the relationship between objects in a dynamical correlation with other
concepts (with prototypically similar characteristics)
• Structural model of human memory in which related categories and hierarchical organisations are
present
• general concepts are at the top until reaching subcategories at the bottom
Sentence verification Task
(Rips et al., 1973)
true or false to a sentence stating simple facts
Reaction Time = index of how difficult the decision was (the further away the info the slowest the
response)
Prediction: concepts far away in the network will take longer to verify: share less defining features
1) A koala is a Koala
2) A koala is a marsupial
3) A koala is an animal
4) A koala is a fish
RT = 1 < 2 < 3 < 4
(5) A robin is a robin.
(6) A robin is a bird.
(7) A robin is an animal
(8) A robin is a fish.
SEMANTIC REPRESENTATION
Prototype approach versus Exemplar approach
Prototype approach Exemplar approach
Rosch, 1975 Medin & Schaffer, 1978
• Category structure is based on Prototype = a con- • categorisation structure based on similarity of
cept presented by a collection of characteristic object to exemplars of the category versus total
features similarity object to nn-exemplars
• Features are cue to category membership but the • Previously encountered exemplars
Boundaries for the prototype definition are not set: • assumes that we can only retrieve memories of
features can also not exist in the prototype specific instances of a category
• no abstraction of prototypes
• members why resemble the prototype strongly =
high-prototypicality. connections to the family re- • learn concept by being exposed to it (e.g. dog)
semblance • new stimuli are classified according to how
closely they resemble these exemplars
• Typicality effect = describes the fact that the the • Similarity influences classification the most
high prototypical members are highly recognised • Semantic decisions = retrieve exemplars one at
as members of category a time from semantic memory until a decision
• Potential members of the category are identified can be made
by how closely they resemble the prototype or cat- • compute total similarity of current instance to
egory average. memories of positive and negative exemplars
• “best example” of a concept ( dogs, non-dogs)
• a special type of schema • Decide that the exemplar is a dog if it is more
similar to the memories of dogs than memories
of relevant non-dogs
Hertley reports Medin & Shoeben (1988)
- “A penguin in a bird.”” vs. “A sparrow is a bird” Participants judged the typicality of sons into two
classes: SPOON & LARGE SPOON
- queries involving a prototypical members (e.g. is
a robin a bird) elicited faster response times than - typicality was mediated by material
for non-prototypical members - Metal = spoon
- Wooden = large spoon
• Sparrow is more prototypical than penguin, thus
decisions were much faster judgements made on stored instances
• When asked an example of bird ire likely to sate a
prototypical category: Sparrow
• high prototypical objects are strongly affected by
priming
Problems Problems
- does not explain why categories cohere - offers no amount about how categories are
not all concepts have prototypes (abstract) formed beyond memorisation
- prototype for “truth” - Vague definition of exemplar
- not all types of concepts appear to have proto- - Abstract processes ?
types (Harley)
Traditional views of the conceptual system
mental representations do not resemble the perceptual states from which they originate
Transduction problems: what exactly is this redescription process that produces amodal symbols
from modality-specific (perception, action) states?
, Knowledge is embodied or grounded in bodily states and in the brain’s modality-specific systems
(Barsalou, 1999).
Semantic networks
Collin & Quallian, 1969
• this approach proposes that concepts of the mind are arranged in networks = knowledge base
that represents semantic relations between concepts in a network
•
• a functional storage-system for the “meanings” of words represented with graphing systems - semantic
representations
• conceptual categorisation used on logical classification rules = conjunction of necessary features
• Properties of a concept could be “stored” (placed) next to a “node” representing the concept
• links between nodes represent the relationship between objects in a dynamical correlation with other
concepts (with prototypically similar characteristics)
• Structural model of human memory in which related categories and hierarchical organisations are
present
• general concepts are at the top until reaching subcategories at the bottom
Sentence verification Task
(Rips et al., 1973)
true or false to a sentence stating simple facts
Reaction Time = index of how difficult the decision was (the further away the info the slowest the
response)
Prediction: concepts far away in the network will take longer to verify: share less defining features
1) A koala is a Koala
2) A koala is a marsupial
3) A koala is an animal
4) A koala is a fish
RT = 1 < 2 < 3 < 4
(5) A robin is a robin.
(6) A robin is a bird.
(7) A robin is an animal
(8) A robin is a fish.
