Chapter 9: Concepts & Knowledge Flashcards
Schematic knowledge
our general knowledge about something in the world that we gain through experience
key components of knowledge
- There is regularity in the way different people organize knowledge and we use this to make inferences
- Much of our knowledge is implicit
- All knowledge is interconnected
- Knowledge depends on semantic memory
semantic knowledge
includes facts we know about the world and ourselves
category
a group of items that are grouped together because they are similar in some way
exemplars
individual members of a category
concept
a mental representation of a category
Medin on concepts
concepts depend on the structure of the world and how we assign meaning to it
Commonsense knowledge problem
a problem in classical AI in which computers don’t possess the same commonsense knowledge as humans because knowledge has to be explicit in classical AI compared to humans’ ability to infer implicit knowledge
classical approach to categorization
- Categories are defined by sets of defining features
- Categorization is a matter of considering defining features of individual exemplars
- Categories are all-or-none
defining features
a feature that is both necessary and sufficient for category membership according to the classical approach
evidence against the classical approach to categorization
- It’s nearly impossible to identify defining category features for most categories
- People recognize that some exemplars are better category members than others
Wittgenstein on the classical appraoch to categorization
Concepts are not based on defining features, but rather are defined by the resemblance to a collection of features.An item can be categorized into more than one category
Rosch, 1975 typicality experiment
asked participants to rate items based on how good of an example of a category an item was. She found that participants could easily give typicality ratings and they agreed with them.
typicality ratings
an experiment in which participants rate how good an exemplar is of a category
typicality effects
experimental effects in which participants behave differently toward typical category members than atypical members, suggesting that typical items have a privileged place in that category
lexical decision task
an experiment task in which participants must decide whether a string of letters is a word or non-word as fast as possible
semantic priming
the observation that a response is facilitated if it is preceded by a semantically relevant stimulus
examples of typicality effects
- we name typical category members before atypical ones
- we are faster to put typical members into categories than atypical ones
- typical exemplars show larger priming effects than atypical ones
- infants learn typical category members first
- when producing sentences, we list typical category members before atypical ones
are typicality effects exclusive to knowledge tasks?
no, typicality effects are also present in sentence production and other cognitive tasks
who proposed the prototype theory of categorization?
Eleanor Rosch
prototype theory of categorization
Proposes that we consider which features are most likely among category members
characteristic features
a feature that category members likely possess but is not required for category membership
prototype
a mental average of all category members; the most typical example of a category member
do prototypes have to exist in reality?
no
what constitutes a typical category member according to the prototype theory?
The more similar something is to a prototype the more typical a category member it is
family ressemblance
the proposal that all category members can belong to a category without all sharing a single common feature. What is needed for category membership is for each category member to share a feature with at least one other member
context effect
typicality depends on the context
criticism of prototype thoery
it fails to account for atypical category members and the effect of context
exemplar theory of categorization
Proposes that we store examples of items we have encountered in the past
how does categorization occur according to the exemplar theory?
Categorization occurs by comparing new items to the ones we have in memory at retrieval and looking for similarities between their features
advantages of the exemplar theory
it accounts for typicality and context effects
criticisms of the exemplar theory
it seems unlikely that we store copies of all the items we have previously encountered in memory. both exemplar and prototype theories are based on similarities and these are applied even to categories that are rule-based
Dopkins & Gleason, 1997 computer categorization experiment
had participants learn to categorize rectangles on a computer screen into two different categories. They weren’t given any rules to help them learn to sort the rectangles and were simply told if they had classified each rectangle correctly. The rectangles could be classified according to their length or position on the screen. Found that when participants had to categorize new, ambiguous exemplars that could theoretically belong to either category they chose to categorize them based on previously seen exemplars rather than a prototype
what theory do most researchers agree with?
most researchers agree that we are flexible in our categorization strategies
Armstrong et al., 1983 typicality ratings experiment
had participants give typicality ratings to naturally occurring categories but also to categories that have well-defined membership rules. Found that participants could give consistent typicality ratings to well-defined categories
Murphy & Medin’s similarity paradigm
there is an infinite number of ways that any two items can be similar
knowledge-based theories of categorization
We rely on our broad knowledge to explain the reasons for category membership
psychological essentialism
the idea that categories have a natural underlying true nature that cannot be stated explicitly
Murphy & Allopena, 1994 knowledge categorization experiment
discovered that participants have difficulty learning about things that don’t make sense. Demonstrates that we use our prior knowledge to explain particular combinations of features
downside of categorizing people based on psychological essentialism
it can lead to stereotyping
who is more likely to stereotype?
Participants with higher essentialist beliefs were more likely to endorse a variety of stereotypes about different groups of people
basic level categories
the level of categorization that people find most natural. It is the most cognitively efficient level that is both informative and distinctive (ex. dog)
subordinate categories
the level of category that is below the basic level. This level is more informative than the basic level but less distinctive (ex. poodle)
superordinate categories
the category level that is above the basic level. This level is more distinctive than the basic level but less informative
semantic network models
A model for how to store semantic information in computer memory became one of the most influential ways in shaping the way cognitive psychologists study semantic knowledge
Collins & Quillian’s semantic model
knowledge is stored as concepts within individual units called nodes, which are activated as a result of input from the environment. Semantic networks are organized hierarchically with superordinate categories occupying the uppermost level of the network and subordinate categories occupying lower levels. Knowledge is represented as a series of interconnected nodes.
cognitive economy
the tendency to conserve cognitive effort and resources
how do semantic networks demonstrate cognitive economy?
Semantic networks demonstrate this because a property is stored only once at the highest level in the hierarchy. We use the simplest terms that is meaningful for the situation
property inheritance
a characteristic of semantic network models in which nodes inherit the properties of nodes higher in the hierarchy to which they are connected.
how do semantic networks demonstrate property inheritance?
Semantic networks demonstrate this because subordinate categories inherit the properties of the superordinate categories they are connected to
what nodes take longer to activate according to semantic network models?
Nodes farther away from each other in the hierarchy will take longer to activate
semantic network verification task
an experimental task in which participants have to judge whether a sentence is true or false as fast as possible
Collins & Quillian sentence verification experiment
participants responded fastest in a sentence verification task to sentences that could be answered by searching between the fewest levels in the hierarchy
criticisms of semantic network models
a hierarchical model of semantic knowledge failed to account for typicality effects
spreading network activation model
a semantic network model proposed by Collins & Loftus, 1975 in which concepts are organized based on their semantic similarity to each other (non-hierarchical)
Meyer & Schaveneveldt lexical decision task
used a lexical decision task to demonstrate semantic priming between two related words.
schemata
a cognitive structure representing a person’s knowledge about an item or situation
method of repeated reproduction
an experimental technique first described by Barlett in which participants must reproduce an item directly from memory
Barlett’s view of memory
memory is reconstructive; we rely on our past knowledge and experience to help us reconstruct memory the best we can
which people have similar schemata?
People with similar experiences
Bernsten & Rubin, 2004 babies & schemata experiment
asked undergraduate students to imagine a newborn baby and make predictions about what would happen in its life as the baby got older. Found that there was a lot of overlap in people’s predictions and these predictions included expectations about what is expected in a typical-life course within a culture
is the schematic view of knowledge mutually exclusive?
no, there is some overlap between schema and other more detailed theories of knowledge
symbolic grounding problem
a problem first described by Stephen Harnad which relates how symbols get their meaning in the real world. Symbol systems need a way to connect to the real world to avoid an endless cycle of symbolic representation
how has AI addressed the symbol grouping problem?
AI has addressed this problem with robotics (can get direct access to stimuli from the environment)
Artificial neural network (ANN) connectionist models
machines based on the structure of the human brain with neuron-like nodes that are highly connected to each other. Knowledge is contained in the distribution of weights between the connected nodes and these weights determine which patterns will ultimately produce specific outputs.
pro of neural network models
They can explain some phenomena that traditional cognitive models can’t
con of neural networks
It is difficult to explain or interpret neural networks once they have been trained
graceful degradation
a property of neural networks in which damage to part of the network results in relatively few deficits because information is distributed across the network and no single node contains information
what’s an example of graceful degradation
category-specific deficits
category-specific deficits
a specific loss of semantic knowledge from one category but not another as a result of brain damage. One of the most common category-specific deficits is when a patient loses knowledge of living things but retains knowledge of non-living things.
Ralph et al., 2007 ANN & category-specific deficits
used an ANN to stimulate the pattern of deficits of a group of brain-damaged patients with a living things category-specific deficit. Found that neural network models that previously were able to name a picture of a fox were no longer able to do so, but could still name a picture of the car, supporting connectionist models.
embodied cognition
Attempts to capture grounding by considering the interaction between the brain, body and the environment.
is there a general theory of embodied cognition?
no
dentist experiment
demonstrates that participants were better at remembering their trip to the dentist when reclining in a chair compared to standing, supporting embodied cognition
hot vs. iced coffee experiment
found that people are more likely to rate descriptions of people as warm after holding a hot cup of coffee compared to holding a cup of iced coffee, supporting embodied cognition
Barsalou, 2008 on the basis of cognition
proposes that cognition is grounded in sensory experiences and that these sensorimotor experiences are used to understand abstract cognitive processes
Hauk et al., 2004 motor areas experiment
found that similar brain areas were active when participants read the word kick to when they actually moved their foot, supporting the claim that knowledge is stored as modality-specific neural activity
Zwaan et al., 2002 priming sentences experiment
found that participants were faster to respond to images that matched the typical context-consistent shape of a prime sentence compared to shapes that didn’t match the prime sentence, providing evidence for context-dependent processing
Embodied vs. Classical Theories
- Embodied theories: knowledge is goal-driven, flexible, and context-dependent
- Classic theories: knowledge is abstract and independent of context
cognitive neuroscience
concerned with investigating how the brain controls thought
semantic dementia
a type of dementia characterised by a loss in semantic memory, leading to deficits in naming ability, language comprehension, and object use
damage to what brain area is affected by semantic dementia?
Associated with the degeneration of neurons in the anterior temporal lobe (ATL)
brain activation during semantic processing
In healthy brains, there is widely distributed activity during semantic processing
hub-and-spoke model
a model of knowledge representation in which abstract concept information is stored in a central hub in the ATL and modality-specific information is stored in modality-specific stores throughout the brain
who proposed the hub-and-spoke model?
Karalyn Pattesron & Matthew Lambon Ralph
Inferior parietal lobule (IPL)
a cortical region that is known to be involved in visually guided hand movements
what is the likely neural mechanism behind semantic knowledge?
Semantic knowledge is likely distributed across many brain areas
what happened when ATL was stimulated via TMS
it took participants longer to name all pictures of objects
what happened when IPL was stimulated via TMS?
it took more time for participants to name non-living things that could be manipulated with the hands, but there was no effect on the speed of naming living things.
anterior temporal lobe (ATL)
a cortical area that is associated with semantic knowledge
function of concepts
- Concepts are vital to do the right thing with the right kind of thing
- They are used to predict outcomes, guide behaviour, and for communication
development of concepts in children
Children learn basic, superordinate, then subordinate concepts
loss of concepts in dementia patients
- Early in the disease, basic-level concepts are accessed
- As the disease progresses, use superordinate concepts
graded concept organization
Organizing concepts across a level
function of graded concept organization
Allows us to determine what’s good exemplar of a concept
generalization
the process of serving a concept from specific experiences
when does the classic approach work well?
for simple concepts
the cube rule
- Anything that has starch at the bottom is a sandwich
- Anything that has starch on three sides is a taco
- Anything that has starch on all four sides is sushi
how are concepts represented in the semantic memory network?
by connected nodes
how is the relatedness of concepts represented in the semantic memory network?
pointers
exemplar vs. prototype theory
- Both suggest that there is no requirement of a set of properties to be shared by all members
- There’s no defining features
- Differ in what we hold in memory to represent a category
Posner & Keele, 1968 prototype experiment
- participants learned to classify dot patterns that were variants of a prototype. They classified the studied patterns, new patterns, and the prototype into groups.
- Worse at classifying new compared to old patterns
- Equally able to classify prototype and old patterns
- Suggests the participants formed a prototype in memory when they are learning the related exemplars
priming
when information you have previously been exposed to unconsciously affects processing
multiple forms of representation theory
Different forms of representation are used to meet different task demands:
Specificity task= exemplar
Generalized task= prototype
advantage to multiple forms of representation theory
supports flexible thought
ad-hoc categories
A category concept that is invented for a specific purpose or goal
conceptual expansion
a process of thinking outside traditional conceptual boundaries
conceptual expansion is linked to
creativity
White et al., 2018 ADHD & creativity study
Had students with and without ADHD to imagine what fruit from another planet might look like and draw them. Found that individuals with ADHD had more creative drawings
how is knowledge stored in the brain?
- Knowledge is stored as sensorimotor neural representations
- The particular representation that is accessed as a function of what sensorimotor domain is required
perceptual symbols system
- Perception and concepts knowledge are linked as perceptual symbols
- Activating a concept will engage certain sensory perceptions to engage mental stimulation as a function of the goals of the current task
- Rejects the view that concepts are abstract representations
property verification tasks findings
- people are faster to respond if a previous trial asked about a feature from the same percept
- Demonstrates that we recruit concepts based on sense/perceptions
brain representation and action words study
In an MRI scanner, participants passively read action words. The brain regions that process movements associated with those words were active during passive reading
the loss of concept depends on ___
where the brain injury is
types of category-specific deficits in brain-injury patients
- Some have selective impairment in naming living things
- Some have selective impairments in non-living things
Caramazza et al., 2003 brain deficits and naming concept information experiment
participants with a selective deficit in naming living things would name them incorrectly, but they were able to access concept information for other categories
picture matching task
point to the picture in a display that corresponds to a spoken word
Warrington & McCarthy, 1987 brain deficits and picture matching task experiment
patients with selective impairment in naming non-living things were shown a picture-matching task. They were unable to complete the task for the category that they had impairment in.
how are living things defined?
by visual features
how are inanimate objects defined?
by functional features
