Professor Greg Francis 5/23/081Representation of knowledgeIIE 269: Cognitive PsychologyGreg FrancisLecture 23What is a shoe?Concepts We have knowledge about the world Due to personal experience Or due to language Such information must be in some kind of format,which we call concepts But what are the concepts? what is the concept of “dog,” “walking,” or “free-marketcapitalism”?Concepts We will look at three topics in concepts Definitions Prototypes ExemplarsDefinitions Plato (and Socrates) spent a lot of efforttrying to define terms like virtue andknowledge they were largely unsuccessful the 20th century philosopher Wittgensteinwondered if definitions of even simpleconcepts were possibleDefinitions Consider the concept shoe, you might define it asWebster’s does A covering for the human foot, usually made of leather,having a thick and somewhat stiff sole and a lighter top. Anything resembling a shoe in form, position, or use. Lots of shoes fit this definitionDefinitions Consider the concept shoe, you might define it as Webster’sdoes A covering for the human foot, usually made of leather, having athick and somewhat stiff sole and a lighter top. Anything resembling a shoe in form, position, or use. But now consider some situations and decide if they are reallyshoes A shoe that is intended for display onlyProfessor Greg Francis 5/23/082Definitions Consider the concept shoe, you might define it as Webster’sdoes A covering for the human foot, usually made of leather, having athick and somewhat stiff sole and a lighter top. Anything resembling a shoe in form, position, or use. But now consider some situations and decide if they are reallyshoes a shoe filled with cement, which cannot be worn a covering worn on the hands of a person without legs who walkson his handsDefinitions The difficulty is the same one that Plato and Socrateshad trying to define virtue for any definition you come up with, I can find examples that donot seem to fit the definition But we all know what a shoe is so our knowledge of this concept must not be based on someprecise definition Note, scientists can (sometimes) create precisedefinitions (e.g., a dog is defined by a DNA pattern orby mating abilities) but the definition is somewhat arbitraryPrototypes Perhaps what defines a concept is similarityamong its members there may be no absolutely necessary characteristics there may be no absolutely sufficient characteristics Prototype theory supposes that similarity isjudged relative to a prototype example of theconcept e.g., an ideal, average, or most frequent version of theconceptPrototypes In prototype theory it is possible for an object tobe “more” or “less” a certain concept Consider the concept “coffee cup”Prototypes In prototype theory it is possible for an object tobe “more” or “less” a certain concept Consider the concept “coffee cup” and variations (some are “cup-ier” than others)Prototypes In prototype theory it is possible for an object tobe “more” or “less” a certain concept Consider the concept “coffee cup” and variations (some are “cup-ier” than others)Professor Greg Francis 5/23/083Prototypes In prototype theory it is possible for an object tobe “more” or “less” a certain concept Consider the concept “coffee cup” and variations (some are “cup-ier” than others)Prototypes In prototype theory it is possible for an object tobe “more” or “less” a certain concept Consider the concept “coffee cup” and variations (some are “cup-ier” than others)Prototypes Lots of experiments suggest the role ofprototypes Posner & Keele (1968): learning category names forrandom dot patterns Discriminate two sets of random dot patterns Each pattern is a variation of one of two prototypepatternsPrototypes PrototypesABPrototypes variations are made by moving some of the dotsvariant of Avariant of B subjects learn to classify many different variants they never see the prototypes themselvesPrototypes The key test is done after subjects learn to classify thevariants reaction time for judgment is recorded for stimuli they have neverseen before» new variants» the prototypes reaction time is faster for the prototypes which suggests that the mental representation of the categories(concepts) are built to favor the prototype of the category Look at CogLab dataProfessor Greg Francis 5/23/084Prototypes Results are based on data from 120 participants. Pattern type Reaction time (ms) Prototypes 847 Variants 805 Data contaminated by a few very long RTs to prototypes Filter data to throw out any student data with RTs longer than 5 seconds Results are based on data from 115 participants. Pattern type Reaction time (ms) Prototypes 683 Variants 710 Unanswered by this (and many other) experiments is what a prototype is: a “thing” that resides in memory and contains information about the categoryfeatures the result of processing information A bit of thought suggests it is the result of processing informationPrototypes Consider the types of concepts you can have and how specific they can be things: bird, dog, chair, shoe,… actions: walking, running, sleeping,… goal-derived: “things to eat on a diet”, “things tocarry out of a house in case of a fire”,... ad hoc: “things that could fall on your head”,“things you might see while in Paris”, “gifts to giveone’s former high school friend who has just hadher second baby”,...Prototypes We can generate new concepts from oldconcepts it’s inconceivable that every possible prototype existsready to be used some must just be built as they are needed perhaps even the prototypes for simple concepts like“bird” or “shoe” are also just built when they areneeded A theory that can account for this processingapproach is exemplar theoryExemplars A concept consists of lots of examples of theconcept e.g., a “coffee cup” concept might contain lots ofexamples of coffee cupsExemplars Comparing an object to see if it is a coffee cupinvolves comparing it to each exemplar andseeing if it matches anything well enoughExemplars Even if it is a new object, it may match severalexemplars well enough to generate an