Categorization and Concepts Basic cognitive function is to categorize
Use experience to aid in future behavior and decision-making Cognitive economy
Concepts Mental representation of a category serving multiple
functions We can use associations to organize the environment
and our behavior Distill our experience (knowledge) by utilizing
functional relations
Functions of Concepts Classification
Determine category membership Understanding, making predictions, inference
Once classified one can then understand its relevant parts, know how to interact with it, infer other properties
Explanation and Reasoning For example, of others’ behavior
Learning New entities compared to and understood in terms of old and provide
feedback for modification Communication
Shared concepts and categorization allow for easier expression of ideas to others
Categories Categories
Collection of objects, attributes, or actions, etc. List of concepts Hierarchy
Set of entities or examples picked out by the concept How is experience distilled? How are functional relations established?
Category learning How is knowledge represented in a category?
Structure Schema
General knowledge structure that integrates objects, attributes, and actions into a cohesive representation Script Sequence
How do we use categorical knowledge?
Classification Determining the category membership of
various things (objects, properties, abstractions etc.)
Allows for treating otherwise discriminable entities as similar Similarity as the organizing principle for
categories and categorization
Structure of Categories Classical View Natural categories were structured in terms of
necessary and sufficient features If some entity has the set of necessary and
sufficient features, it belongs to that category, otherwise it does not
Rigid category boundaries
Classical view Problems
Duck-billed platypus and brown dwarf There simply do not seem to be defining features
for many categories
Perhaps features are not available to consciousness? Uncertain as to whether the necessary feature is present? Unlikely as folks are in disagreement as to what
would constitute category membership (even with themselves at different times)
Even when certain, some examples are obviously better than others
Bye-bye classical view
Probabilistic View Certain features may be necessary, and so weighted
heavily in categorization Probabilistic features, which are usually present but
not always, will also influence categorization E.g. Flies, for birds
How might we classify and represent structured knowledge?
Features/Typicality Theories
Prototype Exemplar
Features and typicality Some instances may have more features than others The more frequently a category member’s properties appear
within a category the more typical a member it is Robins vs. Penguins
Arrange objects based on some attribution. Comparison to average member (central tendency) Based on experience with category which may be different for different
folks
Great DaneChihuahua
Labrador
“Dogs”
Prototype Categorization instead may
reflect typicality judgments based on comparison to an ideal Concepts as abstractions
People abstract common elements of a formed category and use a common representation to stand for that category
How is the category updated? Family Resemblance
Overlap of common attributes Classification is made based on
overlap between prototype and exemplar
Prototype The prototype view can explain both typicality effects and the
fact that prototypes that had not been previously presented are correctly classified (even more accurately)
Problems with prototype explanation Doesn’t take into account category size or variability in examples Context
What may be more typical in one setting may not be elsewhere Correlations among attributes
E.g. smaller birds more likely to sing Implies linear separability among categories
Categorization is perfect by adding up and weighing the evidence from features present
If this is not the case for separating categories, one would be hard pressed to come up with worthwhile prototypes
Exemplar theory Exemplar theory
Sort of a bottom up approach to categorization Each instance is compared to others from past experience Category arises by the lumping together of similar exemplars
Similarity based retrieval Since the exemplar approach retains more information about
the category itself it gets around some of the problems faced by the prototype theory (e.g. context effects), but also how a prototype could be recognized at test when wasn’t presented previously
Has similarity to previous examples and activates those stored representations
Exemplar/Prototype theory Hybrid view
Perhaps a little of both* It may be that concepts
rarely consist of only prototype or exemplar representation Once rule is learned
categorize according to it. When exceptions arise, use an exemplar approach
E.g. grammatical rules
MC’s thought for the day: metacategorizationHow do we classify the empirical evidence as supporting (belonging to) one theory or another?
Between Category structure Up to this point the discussion has focused on
classifying items within one category or another i.e. how a particular category is represented Within category structure
But how are categories themselves organized? Between category structure
Types of Categories Examples
Abstract vs. Concrete Love vs. Mammal
Hierarchical vs. Non Mammal vs. woman
Different processes required? Hard to determine difference in kind
Hierarchical Membership assumes a hierarchy such that
classification in a subordinate category means an exemplar belongs to the superordinate category Poodle Animal
Basic level The default category classification
How will an item be typically classified? Poodle as dog rather than animal
The basic level is found at a middle level of abstraction (e.g. between type of dog and more abstract categories like Living)
Typically learned first, the natural level at which objects are named and the level at which exemplars are likely to share the most features
With expertise, the basic level may move to a subordinate level Child: Dog vs. Cat Adult: Poodle vs. Irish setter Expert: Minature vs. Toy
Structure of Categories Rosch
Hierarchal structure of concepts
Vehicles
CAR TRUCK BOAT
Sedan Sports SUV Garbage Row Yacht
-Corvette
-Mustang
Structure of Categories Vertical = Level of abstraction Horizontal = variability within category
Vehicles
CAR TRUCK BOAT
Sports SUV Garbage Row YachtSedan
Vertical StructureSuperordinateVehicles
CAR TRUCK BOAT
Sports SUV Garbage Row YachtSedan
Basic
Subordinate
Superordinate = defines category
Basic = overlap of common features
Subordinate = examplars
Properties of Hierarchy Each level gives a similar degree of information Converging operations for Basic Level
Common attributes Shape similarity Ease of labeling Similar verification time
SuperordinateVehicles
CAR TRUCK BOAT
Sports SUV Garbage Row YachtSedan
Basic
Subordinate
Non-hierarchical No clear structure
How would you classify yourself? No clear hierarchy, no basic level
E.g. socially relevant categories to which a member may belong to several
The various applicable categories can be seen as competing for classification rights Those used more frequently and recently will be more
likely applied for classifying a new instance E.g. gender, race
What processes are involved in categorization?
Does judgment of similarity in and of itself explain categorization?
Variable People’s judgments of similarity
change depending on the situation Medin Goldstone & Gentner (1993)
Depending on which pair of objects shown would change what determined a judgment of similarity
Similarity What constraints if any are placed on determinations
of similarity? What constraints does similarity place on what counts as a feature? Rocks and squirrels
Both exist, are bounded, can be run over etc.
Can similarity alone explain classification? Perhaps serves as guideline rather than definitive
delineator Abandoned if additional info suggests it is misleading Gelman & Markman (1986)
Classification by theory Organization of concepts is knowledge-based as
opposed to similarity-based Apply theory to the data
Concepts develop and change with experience/evidence E.g. various mental disorders
Theory and Similarity Theories will affect similarity judgments Similarity constrains theory Psychological essentialism
The way people approach the world Essences of things (e.g. what makes male or female)
Models of Categorization Generalized Context Model Exemplar-Based Random Walk
See Nosofsky link on class webpage ALCOVE Combinations of exemplar and rule-based processing Decision-bound approaches Rational model
Anderson
Categorization and memory What memory system or systems are used during
category learning? Essentially theories of category learning virtually all
assumed a single category learning system E.g. exemplar theory
When a novel stimulus is encountered, its similarity is computed to the memory representation of every previously seen exemplar from each potentially relevant category, and a response is chosen on the basis of these similarity computations
Category learning uses many, or perhaps all of the major memory systems that have been hypothesized by memory researchers.
Working memory Heavily used in reasoning and problem solving Could be the primary mediating memory system in tasks
where the categories are learned quickly. Two possibilities:
The categories contain few enough exemplars that the process of explicitly memorizing their category labels does not exceed the span of working memory Though possible, probably unlikely, however if comparisons are made
to a single ideal or prototype perhaps Working memory could be used if the category structures were
simple enough that they could be discovered quickly via a logical reasoning process. In other words if the means of categorization can be reduced to one or
two dimensions (e.g. some rule)
Working memory Evidence
Single rule-based categorization is interfered with in divided attention tasks where more complex category learning is not
Rule-based category learning is possibly mediated by a conscious process of hypothesis generation and testing. If the feedback indicates response was incorrect, then must decide
whether to try the same rule again, or whether to switch to a new rule If the latter decision is made then a new rule must be selected and
attention must be switched from the old rule to the new. Such operations require attention and working memory.
Episodic and semantic memory Memory for personally experienced events and general world knowledge No empirical evidence from category learning suggests separate
contributions of episodic and semantic memory systems These declarative memory systems are used during explicit memorization,
so category structures that encourage memorization are especially likely to be learned via these systems.
Two conditions: First, memorization is an especially effective strategy if each category
contains a small number of perceptually distinct exemplars. Second, other simpler strategies are ineffective
Indirect evidence from successful exemplar-based models that assume use of stored representations from prior learning
Some direct evidence from amnesiacs that suffer in category learning
Non-declarative memory Procedural knowledge
Memories of skills that are learned through practice Little awareness of details Is slow and incremental and it requires immediate and consistent feedback
Like declarative memory systems, would not be utilized for simple rule-based categorization
Example of radiologists and tumors Many exemplars in the set of X-rays, but identification takes practice and
process is not well-defined by practitioners Evidence
Information integration (more complex multi-dimensional categorization) tasks affected similarly as serial reaction time tasks Changing the way in which one responds (key press) leads to poorer performance
that is not seen in simple rule-based categorization tasks As with procedural tasks, complex category learning can be hindered without
appropriately timed feedback
Perceptual learning The specific and relatively permanent
modification of perception and behavior following sensory experience
No behavioral evidence implicating the perceptual representation system, jury out on neuropsych evidence
Use of categories in reasoning Ad hoc categories
Spontaneously constructed for the purposes of some goal Constructed differently from other categories?
Show similar results e.g. typicality effects, however, more of a comparison to an ideal rather than prototype
Gist: goals can affect category structure Conceptual combination
Construction of new concepts by combining the previous representations Recall structural alignment
Typicality may not be predictable from previous concepts Properties of new concepts may not be present in old.
Use of Categories Classification
Process of assigning objects to categories Treat (use) different “things” as the same
Explanation Bringing knowledge to bear in novel situation By classifying a novel event into an existing
category, an explanation is provided.
Use of Categories Prediction
Understanding of an event guides reactions and behaviors Allows us to expect certain outcomes or properties
Reasoning Categories are the basis for inferences
Allow categorical knowledge to stand for an event Allows for “filling-in” of ambiguous information
Allow for conceptual combinations Paper Bee Wooden Spoon