Presentation is loading. Please wait.

Presentation is loading. Please wait.

Psy1302 Psychology of Language Lecture 16 Words and Meanings II.

Similar presentations


Presentation on theme: "Psy1302 Psychology of Language Lecture 16 Words and Meanings II."— Presentation transcript:

1 Psy1302 Psychology of Language Lecture 16 Words and Meanings II

2 Review Classical Theory of Concepts Classical Theory of Concepts –Other Names: Defining Features Theory Defining Features Theory Definitional Theory Definitional Theory –Pros and Cons Review: Classical Theory of Concepts

3 Pros: Explanatory Power 1. Explains category membership in terms of necessary and sufficient features 2. Allows identification of new candidates 3. Explains how you learned the meanings of the words. 4. Provides descriptions that can support semantic compositionality. 5. Semantic feature can explain relationships between words. Review: Classical Theory of Concepts

4 Cons: Problems for Classical Theory Difficulty in coming up with a set of necessary and sufficient features Difficulty in coming up with a set of necessary and sufficient features –E.g. game, bachelor –Feature Naming: list features of “apple”, “lemon”, “fig”. Necessary and sufficient features DO NOT emerge Necessary and sufficient features DO NOT emerge Category membership may not be ALL-OR-NONE Category membership may not be ALL-OR-NONE –Category boundaries are fuzzy. Cup vs. Bowl Cup vs. Bowl Categories have graded membership Categories have graded membership –Typicality effects: some members are better than others Review: Classical Theory of Concepts

5 Pro Prototype Theory (Experiments showing Graded Membership) Prototypicality Ratings correlates with Production Task Prototypicality Ratings correlates with Production Task (Rate 1-7 Good/Bad)(List Members) (Rate 1-7 Good/Bad)(List Members) –Prototypical members are listed early. Prototypicality Ratings correlates with Verification Task Prototypicality Ratings correlates with Verification Task (Rating 1-7 Good/Bad)(RT of X is a ___.) (Rating 1-7 Good/Bad)(RT of X is a ___.) –Prototypical members are responded to faster. Prototypicality Rating correlates with Feature Naming Prototypicality Rating correlates with Feature Naming (Rate 1-7 Good/Bad) (List features of…) –Prototypical members share more features with other members. Review: Prototype Theory of Concepts

6 Prototype Theory Concepts are made of: Concepts are made of: –features –perceptually grounded (like the classical theory) How are features combined: How are features combined: –family resemblance: no single feature necessary –more shared features = better category member (different than the classical theory) Prototype Theory (Probabilistic Features)

7 One formalization of Family Resemblance Structure Smith Family Smith Family Prototype Theory (Probabilistic Features) – Family Resemblance Structure

8 Degree of Category Membership (“Smithness”) depends on Degree of Category Membership (“Smithness”) depends on –the number of features and –how central they are to “Smithness” Family Resemblance “Smith” family Prototype Theory (Probabilistic Features) – Family Resemblance Structure

9 Smith Features Smith Features –Beard 8/8 = 1 –Brown hair6/8 =.75 –Big nose6/8 =.75 –Big ears6/8 =.75 –Mustache5/8 =.625 Smith Features Smith Features –Beard 8/8 –Brown hair6/8 –Big nose6/8 –Big ears6/8 –Mustache5/8 Smith Features Smith Features –Beard –Brown hair –Big nose –Big ears –Mustache Prototype Theory (Probabilistic Features) – Family Resemblance Structure Family Resemblance “Smith” family

10 Middle Smith has all features Middle Smith has all features –beard 1 * 1.0 –brown hair 1 *.75 –big nose 1 *.75 –big ears 1 *.75 –mustache 1*.625 --------------------------- --------------------------- –Total 3.8 Prototype Theory (Probabilistic Features) – Family Resemblance Structure

11 Smith #3 a few features Smith #3 a few features –beard 1* 1.0 –brown hair 1*.75 –big nose 0 *.75 –big ears 1 *.75 –mustache 0 *.625 -------------------------- -------------------------- –Total 2.5 –poorer instance than middle Smith Prototype Theory (Probabilistic Features) – Family Resemblance Structure

12 Item with too few features is not a member of the category Item with too few features is not a member of the category –beard 0* 1 –brown hair 0 *.75 –big nose 1 *.75 –big ears 0 *.75 –mustache 0 *.625 ----------------------- ----------------------- –Total.75 –not a Smith Prototype Theory (Probabilistic Features) – Family Resemblance Structure

13 One formalization of Family Resemblance Structure Features have associated probability Features have associated probability These probabilities may be thought of as weights on the features for membership/identification purposes These probabilities may be thought of as weights on the features for membership/identification purposes Category membership is based on a weighted sum of the features. Category membership is based on a weighted sum of the features. Prototype Theory (Probabilistic Features) – Family Resemblance Structure

14 Press a button to answer TRUE or FALSE to the following statements Press a button to answer TRUE or FALSE to the following statements QuestionResponse QuestionResponse A canary is a bird A ostrich is a bird TRUETRUE fast slow Typicality effect Verification Task Prototype Theory (Probabilistic Features) – Applied to Experimental Data

15 Prototype Theory (Rosch & Mervis, 1975) Conceptual category (e.g. birds) is represented by a prototype Conceptual category (e.g. birds) is represented by a prototype –an average of all the exemplars in the category –not a real instance, just an abstraction –the ‘average bird’ will be more like a canary than an ostrich Verification task: compare the exemplar to the prototype Verification task: compare the exemplar to the prototype –A canary is a bird big overlap = FAST big overlap = FAST –An ostrich is a bird small overlap = SLOW small overlap = SLOW Typicality effect! =? = ? Prototype Theory (Probabilistic Features) – Applied to Experimental Data

16 Prototype Theory Summary 1. Certain members of a category are prototypical – or instantiate the prototype 2. Categories form around prototypes; new members added on basis of resemblance to prototype 3. No requirement that a property or set of properties be shared by all members 4. Features/attributes generally gradable 5. Category membership a matter of degree 6. Categories do not have clear boundaries Prototype Theory (Probabilistic Features)

17 Pros: Explanatory power of Prototype Theory Explains category membership ratings Explains category membership ratings – membership is graded not absolute Explains sentence verification results: Explains sentence verification results: –more typical instances quickly identified because they have more of the category features. Consistent with feature listing results Consistent with feature listing results – more typical instances have more features in common with other members Prototype Theory (Probabilistic Features)

18 Hypothesis 4b: Exemplar Theory Hypothesis 4b: Exemplar Theory –Alternative to prototype theory

19 Conceptual category (e.g. birds) is represented by an instance of every exemplar Conceptual category (e.g. birds) is represented by an instance of every exemplar –No single concept ‘bird’ Exemplar Theory

20 Armstrong, Gleitman & Gleitman Structure of the Argument Prototype Claim: If a category shows typicality effects, then it must have a prototype structure Contrapositive: Whenever “if A then B” is true. “If not B, then not A” must also be true. A B Prototype Theory (Probabilistic Features) – PROBLEM

21 Armstrong, Gleitman & Gleitman Structure of the Argument Prototype Claim: If a category shows typicality effects, then it must have a prototype structure If a category shows typicality effects, then it must have a prototype structure Contrapositive of Prototype Claim: If a category does not have a prototype structure then it will not show typicality effects. If a category does not have a prototype structure then it will not show typicality effects. A B Prototype Theory (Probabilistic Features) – PROBLEM

22 Armstrong, Gleitman & Gleitman Structure of the Argument If you can disprove the contrapositive then the original claim must be false. If you can disprove the contrapositive then the original claim must be false. Contrapositive of Prototype Claim: If a category does not have a prototype structure then it will not show typicality effects. If a category does not have a prototype structure then it will not show typicality effects. A B Prototype Theory (Probabilistic Features) – PROBLEM

23 Armstrong, Gleitman & Gleitman Structure of the Argument AGG set out to disprove the contrapositive by conducting typicality tests on well- defined categories: AGG set out to disprove the contrapositive by conducting typicality tests on well- defined categories: –mathematical categories: even number, triangle, plane –gender categories: female, male –kinship terms: uncle, grandmother Contrapositive: Well-defined categories (those that do not have prototypical structure) should not show typicality effect. Contrapositive: Well-defined categories (those that do not have prototypical structure) should not show typicality effect. Prototype Theory (Probabilistic Features) – PROBLEM

24 Exp. 1: Exemplar Ratings for ordinary categories FRUIT apple 1.3 apple 1.3 strawberry 2.1 strawberry 2.1 pineapple 2.7 pineapple 2.7 fig 5.2 fig 5.2 olive 6.4 olive 6.4SPORT football 1.4 football 1.4 hockey 1.8 hockey 1.8 gymnastics 2.8 gymnastics 2.8 archery 4.8 archery 4.8 weight-lifting 5.1 weight-lifting 5.1 Prototype Theory (Probabilistic Features) – PROBLEM

25 FEMALE mother mother housewife housewife princess princess policewoman policewoman comedienne comedienne EVEN NUMBER 4 8 18 18 34 34 106 106 Exp. 1: Exemplar Ratings for well-defined categories EVEN NUMBER 4 1.1 4 1.1 8 1.5 8 1.5 18 2.6 18 2.6 34 3.4 34 3.4 106 3.9 106 3.9FEMALE mother 1.7 mother 1.7 housewife 2.4 housewife 2.4 princess 3.0 princess 3.0 policewoman 3.9 policewoman 3.9 comedienne 4.5 comedienne 4.5 Prototype Theory (Probabilistic Features) – PROBLEM

26 Verification Task OrdinaryWell-defined +Proto(orange-fruit)(8-even) -Proto(fig-fruit)(56-even) Fast Slow Fast Slow > > Prototype Theory (Probabilistic Features) – PROBLEM

27 Armstrong, Gleitman & Gleitman Conclusion Since well-defined categories also show typicality effects Since well-defined categories also show typicality effects then the presence of these effects, does not prove that ordinary categories have a prototype structure then the presence of these effects, does not prove that ordinary categories have a prototype structure Prototype Theory (Probabilistic Features) – PROBLEM

28 Hypothesis 5: Dual Theory Hypothesis 5: Dual Theory –Dual Route: Classical and Prototype Dual Theory

29 Grandmothers

30 Definitional Features for reasoning with words and determining category membership Definitional Features for reasoning with words and determining category membership Who is a grandmother? Who is a grandmother? –The mother of a parent Prototypes for quick identification Prototypes for quick identification How do you find a grandmother is a crowd? How do you find a grandmother is a crowd? look for the prototypical features look for the prototypical features –(kindly, gray-haired) Dual Theory

31 Mango Pat Tina Fey J. Reno SNL Women? Dual Theory

32 Challenges for any feature theory 1. What are the features which determine category membership? 2. What are the rules which describe how meanings combine. General Challenges

33 Classical Theory (definitional) PET [animal] [animal] [kept for amusement] [kept for amusement] General Challenges

34 Classical Theory (definitional) FISH [aquatic] [aquatic] [water-breathing] [water-breathing] [cold blooded] [cold blooded] [chambered heart] [chambered heart] [animal] [animal] General Challenges

35 Classical Theory (definitional) PET + FISH [aquatic] [aquatic] [water-breathing] [water-breathing] [cold blooded] [cold blooded] [chambered heart] [chambered heart] [animal] [animal] [kept for amusement] [kept for amusement] General Challenges

36 Prototype Theory PET [animal] [animal] [kept for amusement] [kept for amusement] [cute] [cute] [friendly] [friendly] [mammal] [mammal] [furry] [furry] [smallish] [smallish] General Challenges

37 Prototype Theory FISH [aquatic] [aquatic] [water-breathing] [water-breathing] [cold blooded] [cold blooded] [chambered heart] [chambered heart] [animal] [animal] [elongated] [elongated] [spindle shaped] [spindle shaped] [broad caudal fin] [broad caudal fin] General Challenges

38 Prototype Theory Predicted Prototype for Pet Fish: Predicted Prototype for Pet Fish: –a catlike trout? –a fuzzy salmon? General Challenges

39 Rating Task –Is guppy a good PET? –Is guppy a good FISH? –Is guppy a good PET FISH? Prototype Theory Rating Task – –Is guppy a good PET? Bad Member – –Is guppy a good FISH? Bad Member – –Is guppy a good PET FISH? Good Member General Challenges

40 A theory of concepts should explain how words combine to yield the meanings of phrases. A theory of concepts should explain how words combine to yield the meanings of phrases. Prototype Theory fails to do this Prototype Theory fails to do this General Challenges

41 Differing inferences: Differing inferences: –SKILLFUL + SURGEON –SKILLFUL + CARPENTER (is skillful adding the same features in each case?) Feature eating modifiers Feature eating modifiers –STONE LION: –COUNTERFEIT DOLLAR: –FORMER SENATOR: More challenges… (for any feature theory) General Challenges [animate] [legal tender] [member of congress]

42 Challenges for any feature theory 1. What are the features which determine category membership? 2. What are the rules which describe how meanings combine. General Challenges

43 Are these any of these features necessary? TIGER [JUNGLE-DWELLING] [JUNGLE-DWELLING] [4-LEGGED] [4-LEGGED] [FUR-COVERED] [FUR-COVERED] [GROWLY] [GROWLY] [FIERCE] [FIERCE] [STRIPPED] [STRIPPED] [ANIMAL] [ANIMAL] General Challenges What if it lives in a zoo? What if it lost a leg in an accident? What if I shaved it? What if it lost its voice? What if it is a scaredy cat? What if I dye its hair?

44 What is necessary? Has tiger DNA? Has tiger DNA? Has tiger essence? Has tiger essence? What the experts call a tiger? What the experts call a tiger? But how can these be reduced to sensory primitives? But how can these be reduced to sensory primitives? General Challenges

45 Are these any of these features sufficient? What if I transplanted a mean striped tomcat to the jungle? [JUNGLE-DWELLING] [JUNGLE-DWELLING] [4-LEGGED] [4-LEGGED] [FUR-COVERED] [FUR-COVERED] [GROWLY] [GROWLY] [FIERCE] [FIERCE] [STRIPPED] [STRIPPED] [ANIMAL] [ANIMAL] General Challenges

46 Doctors took a raccoon and shaved away some of its fur. They dyed what was left all black. Then they bleached a single stripe all white down the center of tits back. Then, with surgery, they put in its body a sac of super smelly yucky stuff, just like a skunk has. When they were all done, the animal looked like this. After the operation, was this a skunk or a raccoon?

47 Doctors took a coffeepot that looked like this. They sawed off the handle, sealed the top, took off the top knob, closed the spout, and sawed it off. They also sawed off the base and attached a flat piece of metal. They attached a little stick, cut a window in it, and filled the metal container with bird food. When they were done, it looked like this After the operation, was this a coffeepot or a birdfeeder?

48 Results

49 Results Children know that appearances aren't everything. Children know that appearances aren't everything. Animal generalizations should be based on membership in a category, not on appearance. Animal generalizations should be based on membership in a category, not on appearance. This ability increases with age. This ability increases with age. Children also shift from using characteristic properties to categorize to using defining ones. Children also shift from using characteristic properties to categorize to using defining ones.

50 (Theory theory) Hypothesis 6: Theory-Based Theory of Concepts Concepts are representations whose structure consists in their relations to other concepts as specified by a mental theory Theory-Based Theory (Theory Theory)

51 Hypothesis 6: Theory-Based Theory Theory-Based Theory (Theory Theory) (Theory theory) Causal theory of category membership.

52 Theory Theory Causal knowledge is critical to concept learning in at least three ways: Causal knowledge is critical to concept learning in at least three ways: –Causal knowledge helps us decide which features are relevant to category membership –Causal knowledge helps us decide which features are central and which peripheral. –Causal knowledge affects our intuitions about when category members will retain their identity and when they will be transformed. Theory-Based Theory (Theory Theory)

53 Some Challenges for Theory Theory What qualifies as a theory? It is possible to have a concept, even if your theory is deficient or wrong? The content of a concept changes when the theory changes How do theories change? Theory-Based Theory (Theory Theory)

54 A Child’s Theory Example from Callahan & Oakes (1992) Situation: It was bedtime & Child is 4 yrs 2 mon. (4;2) Child: Why does Daddy, James (big brother), and me have blue eyes and you have green eyes? Mother: (Told her she got her eyes from Daddy. Then said goodnight and left the room.) Child: (child calls mother back 5 minutes later) I like Pee Wee Herman and I have blue eyes. Daddy likes Pee Wee Herman and he has blue eyes. James likes Pee Wee Herman and he has blue eyes. If you liked Pee Wee Herman you could get blue eyes too. Theory-Based Theory (Theory Theory)

55 A Child’s Theory Example from Callahan & Oakes (1992) Mother: (I told her it would take more than liking Pee Wee Herman to make my eyes blue. I realized that she didn’t understand me, so I explained that God gave me this color and that they couldn’t be changed.) Child: Could you try to like Pee Wee Herman so we could see if your eyes turn blue? Mother: (I said I would think about it, but if my eyes stayed green it was ok.) Theory-Based Theory (Theory Theory)

56 Concepts A word denotes a concept Word meaning (lexical concepts) Combination of words could get you more complex concepts Categorization = psychological application of a concept Recapitulation: Where we started

57 Theories of Concepts Meaning as Reference Meaning as Ideas Classical Theory of Concepts – –(Defining Features Representation) Prototype Theory – –(Probabilistic Feature Representation) Dual Theory – –(Both Classical and Prototype) Theory Theory of Concepts – –(Theory-based representations) These theories differ mostly in what they consider the structure of concepts to be Recapitulation: Where we started and what we discussed


Download ppt "Psy1302 Psychology of Language Lecture 16 Words and Meanings II."

Similar presentations


Ads by Google