Ch. 9 - Knowledge Flashcards
Schematic knowledge
Schema-based general background knowledge about something in the world that we gain through experience
Concepts
knowledge, non detailed info about a thing. Semantic memory
Categories
items within that concept. Grouping of items
Exemplars
Item that belongs to a category
Commonsense knowledge problem
Problem in AI in which computers don’t possess the same commonsense knowledge as humans because knowledge has to be explicit in AI compared to human’s ability to infer implicit knowledge
Symbol grounding problem
Stevan Harnad 1990. Symbols get their meanings in the real world. Symbol systems need a way to connect to the real world to avoid an endless cycle of symbolic representation. Addressed with robotics, granted real sensory input
Graceful degradation
Property of networks in which damage to part of the network results in few deficits because info is distributed across the network and no single node contains info. Common in human brain.
Category specific deficit
Specific loss of semantic knoweldge from one category but not others after brain damage. e.g. Living things are remembered but nonliving things are not
Concept organization
- Superordinate (less informative, more distinctive),
- Basic (cognitively efficient),
- Detailed/subordinate (more detailed, less distinctive)
Concepts in child development
Basic, superordinate, then subordinate
Concepts in semantic dementia patients
Early on, lose subordinate level. Later, lose basic level. Anterior temporal lobe atrophy, as well as other things
Hub-and-spoke model
Model of knowledge representation in which abstract concept information is stored in a central hub in thr anterior temporal lobe, and modality-specific info is stored in distributed spokes throughout the brain
Cognitive economy
Use of simplest terms that are meaningful in the context of the conversation (e.g. look at that bird. the type of bird is a snowy owl.)
Graded concept organization
Some items are more correspondent to their category than others (trout more fishy than shark)
Property inheritance
Characteristic of semantic network models in which nodes (concept units) inherit the properties of the notes higher in the hierarchy to which they are connected. Idea proposed by Collins and Quilian
Sentence verification task
Experimental task in which participants have to judge whether a sentence is true or false as fast as possible
Spreading activation model
Semantic network model in which concepts are organized based on their semantic similarity to each other. One node activates all surrounding nodes strongly. Weakens with higher degrees of separation.
Schema view of knowledge
Bartlett says what we remember is based on our past experiences and knowledge
Method of repeated reproduction
Bartlett’s method. Participants are shown an abstract image and asked to reproduce it from memory. With time, each attempt more closely resembles a familiar object
Classical approach
- Rule based concept learning
- Defining features: necessary for category membership (dog must be mammal, animate) - these change based on the examples we see
- Characteristic features: common but not essential for membership (dog should have four legs and fur)
- Approach works less well for complex concepts (furless dog) or ambiguous concepts (hot dog = sandwich/taco)
Family resemblence
Proposal that category members can all belong to the same family as long as each member shares one feature with at least one other member
Feature list / similarity based networks
- Feature list: Used for newer concepts. List of features identified in target concept.
- Network: Used for more familiar concepts. Web of features or concepts more or less closely related to target concept.
Prototype theory
We determine overlapping features between exemplars and create a prototype. We then compare new things to each prototype to categorize them. Less strictly rule-based.
Prototype theory study
Participants shown dot patterns that were variants of a prototype, without seeing prototype. They could categorize images they had seen before quicker, as well as unseen prototype.
Typicality effect
Preference to items closest to stereotype (robin vs ostrich). When forming a list, typical examples are listed first. Typical examples are recognized as part of a category faster. 12 month olds can recognize them but not atypicals.
Typicality effect study
Lexical decision task showing categories followed by typical examples, non-typical examples, or non-words. Typical examples were placed faster
Semantic priming
Observation that a response is facilitated if it is preceded by a semantically related stimulus (as done in lexical decision tasks)
Prototype theory problem
Many concepts are context dependent. Harmonica is rated as a typical example around a campfire, but not in a concert hall.
Exemplar theory
No abstract prototype for concepts. Instead, every instance of a category is stored in memory. New items are compared to all other items and rated on similarity. Can function in context-dependent situations, as different exemplars can be retrieved to class new exemplar.
Exemplar theory study
Dopkins/Gleason 1997. Participants had to categorize rectangles into two categories, but not told categories, simply told right or wrong. After several trials, they learned the rule. When shown new rectangles that were on average closer to category 2 but closer to individual rectangles in category 1, they picked category 1. Hence, exemplar theory
Problem with prototype & exemplar theories
- Both do not account for similarity effects, which were perhaps caused by the experimental method and not the unclear borders of a category (odd numbers still had typicality, no fuzzy border).
- Both are based on similarities, yet do not determine how we decide what features to compare. (banana and TV are infinitely similar. Bought at a store, smaller than a truck, etc.)
Psychological essentialism
Proposal that categories have a natural underlying true nature that cannot be stated explicitly. Connected to forming stereotypes.
Psychological essentialism study
Murphy and Allopena 1994. Participants told two scenarios, one that could make sense irl, one that could not. They remembered the one that did better. Prove that when we make categories, we we make connections from past knowledge to explain the combination of features we are making.
Ad-hoc categories
Proof of embodied recognition (parts of brain activated pertaining to task at hand): different things remembered based on goals. We form categories that are temporary (things that catch on fire)
Embodied theories of knoweldge
- Barsalou - body has causal role in thought. Sensorimotor experiences explain abstract cognitive processes
- Hauk, Johnsrude, Pulvermuller - modality-specific knowledge representation
- Zwaam, Stanfield, Yaxley - knowledge access depends on context (refrigerated egg - with shell rather than without)
Perceptual symbols system
Perceptual symbols: Linked perception and conceptual knowledge. Concept retrieval will engage related sensory-perceptions related to the goal (taste cortex activated when thinking what to eat, motivating you to get food)
Perceptual symbols system study
Proves concepts are rooted in motor/sensory systems and not just abstracted ideas.
Property verification task: does the percept (sensory experience) match the object? Participants answer faster if the same modality appears twice (two sound-related items) because cortex is already activated.
Neuropsych case studies
Brane injury patients with category-specific deficits. (some cannot name living things)