Cognitive Psychology C 81 COG 6 Memory Representing

  • Slides: 28
Download presentation
Cognitive Psychology C 81 COG 6. Memory – Representing Meaning Dr Jonathan Stirk

Cognitive Psychology C 81 COG 6. Memory – Representing Meaning Dr Jonathan Stirk

Overview l l Tip-of-tongue phenomenon A network theory of semantic memory (Collins & Quillian,

Overview l l Tip-of-tongue phenomenon A network theory of semantic memory (Collins & Quillian, 1969) – Hierarchical structure of concepts – Recover information by traveling through the network via the links in the hierarchy Problems with the Collins & Quillian model – Hierarchies are not based on human understanding – Typicality effects are not explained – Hierarchical distance is not predictive A feature list model of semantic memory (Smith et al. , 1974) – Defining features vs. Characteristic features – Recover information by comparing the two feature lists – Explains linguistic hedges

 Tip-of-the-tongue Phenomenon State of knowing a fact or name, but not being able

Tip-of-the-tongue Phenomenon State of knowing a fact or name, but not being able to recall or retrieve the information from memory right away l Brown & Mc. Neill (1966) used dictionary definitions to produce TOT states: A navigation instrument used in measuring angular distances, especially the latitude of the sun, moon, and stars. What is it? l

Tip-of-the-tongue Phenomenon l l In TOT state, people can often recall partial information about

Tip-of-the-tongue Phenomenon l l In TOT state, people can often recall partial information about the word (e. g. , Length, number of syllables), or letters of sounds within the word, and related words People guess the first letter about 50% of the time – better than chance People use partial information to cue memory for the word or fact E. g. “I can’t remember the name…oh its green and a vegetable………………. . ah a cucumber, that’s what I mean. ”

Collins & Quillian (1969) l Developed network model of memory – l Developed from

Collins & Quillian (1969) l Developed network model of memory – l Developed from the teachable language Comprehender – l l Computational model of how memory works A computer program intended to “extract and somehow retain meaning from natural language text” (Quillian, 1968) Nodes – represents a concept Links – represent relationships/related concepts

Collins & Quillian (1969)- A Network Model Of Semantic Memory (Semantic Network) Breathes -NODE

Collins & Quillian (1969)- A Network Model Of Semantic Memory (Semantic Network) Breathes -NODE Animal LINK FEATURE Eats Is a Has skin Has wings Bird Can fly Has fins Fish Can swim Has feathers Is a Canary Can sing Is yellow Has gills Is a Ostrich Is a Is tall Can’t fly Salmon Is pink Is edible Swims upstream

Collins & Quillian (1969)- A Network Model Of Semantic Memory l l l Concepts

Collins & Quillian (1969)- A Network Model Of Semantic Memory l l l Concepts are related by links The links impose a hierarchical organization of concepts in a subordinate system Assumption of cognitive economy – l Features or properties are represented only once at the highest level of the hierarchy Questions about knowledge are answered by computing the overlap between features stored at levels of the hierarchy e. g. answering the question “are canaries yellow? ” will take less time than the question “do canaries breathe? ” WHY? ? ?

Are Canaries Yellow? Breathes Animal Eats Has skin Has wings Bird Can fly Has

Are Canaries Yellow? Breathes Animal Eats Has skin Has wings Bird Can fly Has fins Fish Has feathers Canary Can sing Is yellow Ostrich Can swim Has gills Is tall Can’t fly Is pink Salmon Is edible Swims upstream

Do Canaries Breathe? Breathes Animal Eats Has skin Has wings Bird Can fly Has

Do Canaries Breathe? Breathes Animal Eats Has skin Has wings Bird Can fly Has fins Fish Has feathers Canary Can sing Is yellow Ostrich Can swim Has gills Is tall Can’t fly Is pink Salmon Is edible Swims upstream

Do Canaries Breathe? Breathes L 3 Animal Eats Has skin Has wings L 2

Do Canaries Breathe? Breathes L 3 Animal Eats Has skin Has wings L 2 Bird Can fly Has fins Fish Has feathers L 1 Canary Can sing Is yellow Ostrich Can swim Has gills Is tall Can’t fly Is pink Salmon Is edible Swims upstream

Collins & Quillian (1969) - Assumptions Of Model It takes time to move from

Collins & Quillian (1969) - Assumptions Of Model It takes time to move from one level of the hierarchy to a different level l It takes additional time to retrieve features (properties) stored at a level l – It should be faster to answer questions about category membership than about properties when one has to move through the levels of the hierarchy

Collins & Quillian (1969) – Sentence Verification Task EXPT. 1: comparing questions requiring the

Collins & Quillian (1969) – Sentence Verification Task EXPT. 1: comparing questions requiring the comparison of features/properties stored at different levels l l l Level 1 statements e. g. Canaries can sing Level 2 statements e. g. Canaries can fly Level 3 statements e. g. Canaries have skin In terms of speed of response Level 1 > level 2 > level 3 (Fastest) (slowest) More steps through the hierarchy produces slower answers

Findings On SVT Mean Reaction Time (msecs) 1500 (P 2) A canary has skin

Findings On SVT Mean Reaction Time (msecs) 1500 (P 2) A canary has skin (P 1) A canary can fly (P 0) A canary can sing (S 2) A canary is an animal (S 1) A canary is a bird 1000 (S 0) A canary is a canary 0 1 Number of links 2

Collins & Quillian (1969) - A Network Model Of Semantic Memory EXPT. 2: Does

Collins & Quillian (1969) - A Network Model Of Semantic Memory EXPT. 2: Does use of a retrieval path help subsequent retrievals? l e. g. Is a canary a bird? follows (i) Can a canary fly? or (ii) Can a canary sing? l Result – (i) gives more help than (ii) because the same retrieval path is used

Activation Of Retrieval Paths Breathes Animal Eats Has skin Has wings Bird Can fly

Activation Of Retrieval Paths Breathes Animal Eats Has skin Has wings Bird Can fly Has feathers Canary Can sing Can a canary fly? Is yellow Is a canary a bird?

Activation Of Retrieval Paths Breathes Animal Eats Has skin Has wings Bird Can fly

Activation Of Retrieval Paths Breathes Animal Eats Has skin Has wings Bird Can fly Has feathers Canary Can sing Can a canary sing? Is yellow Is a canary a bird?

Summary Of Evidence In Support Of Model l The results from the sentence verification

Summary Of Evidence In Support Of Model l The results from the sentence verification task support the network model – – – Verification time depends on number of levels must go through in network Property judgments are slower than category membership judgments Priming occurs when previous question ‘activates’ a pathway within the network required to answer a next question Also l Category size effect – – Members of smaller categories are classified faster than members of larger categories Smaller categories lower down in hierarchy therefore reached faster

Collins & Quillian (1969) - Problems For The Network Model Of Semantic Memory l

Collins & Quillian (1969) - Problems For The Network Model Of Semantic Memory l The model based upon hierarchies just does not work Is a chimp a mammal? vs. Is a chimp an animal? Animals Mammals Chimp Birds Fish Gorilla Human Reversal of category size effect! Takes longer for “Is a chimp a mammal” This can’t be explained by C&Q model.

Collins & Quillian (1969) - Problems For The Network Model Of Semantic Memory l

Collins & Quillian (1969) - Problems For The Network Model Of Semantic Memory l l Typical instances gain faster decisions (cf. next slide) Is a canary a bird? vs. Is a chicken a bird? Hierarchical distance does not always predict the decision time Can a shark move? vs. Can an animal move? – Hierarchical organization may be no more than a re-ordering of associative strength between features – Q: What features are associated with the items? A: Those features which Collins & Quillian assumed to be stored at that node So: Associative strength may be influencing retrieval time, rather than hierarchical distance – –

Typicality Effect (Rips, Shoben & Smith (1973) Breathes Animal Eats Has skin Has wings

Typicality Effect (Rips, Shoben & Smith (1973) Breathes Animal Eats Has skin Has wings Bird Can fly Is there a model which can? Has feathers Canary Can sing Is yellow Faster (typical) Chicken The hierarchical network model cannot explain the findings of Rips et al (1973) Lays eggs Clucks Slower (atypical)

A Feature List Model Of Semantic Memory - Smith, Shoben & Rips (1974) l

A Feature List Model Of Semantic Memory - Smith, Shoben & Rips (1974) l l l The meaning of an item/instance is represented by a list of its features/attributes Some features are essential to the item, and others are merely characteristic DEFINING FEATURES are those which are essential to the concept e. g. Canaries - have wings, can eat, are yellow CHARACTERISTIC FEATURES are those which are commonly observed, but which are not essential e. g. Canaries - are small, are harmless, are found in cages, Chickens – are food etc. Questions of knowledge are answered by consulting the lists of features, looking for overlap between lists

Feature List Model BIRD CHICKEN CANARY l. Is animate l. Has feathers l. Can

Feature List Model BIRD CHICKEN CANARY l. Is animate l. Has feathers l. Can fly l. Etc. animate l. Has feathers l. Can lay eggs l. Etc. animate l. Has feathers l. Can sing l. Etc. Defining Features (essential) Characteristic Features (non-essential) No cognitive economy in this model!

Smith, Shoben & Rips (1974) Two Stages Of Comparison When Answering Questions About Knowledge

Smith, Shoben & Rips (1974) Two Stages Of Comparison When Answering Questions About Knowledge (e. g. Is a canary a bird? ) Stage A 1. Compare all features of both concepts 2. If high overlap, then respond YES 3. If no/low overlap, then respond NO l But what if feature overlap is only moderate? ?

Stage A Is a Canary a Bird? Stage A Retrieve features for “Canary” and“Bird”,

Stage A Is a Canary a Bird? Stage A Retrieve features for “Canary” and“Bird”, determine overall similarity of both defining and characteristic features LOW How much similarity between two concepts? HIGH

Stage B LOW How much similarity between two concepts? HIGH Stage B Compare only

Stage B LOW How much similarity between two concepts? HIGH Stage B Compare only defining features Respond “False/No” Match? Respond “True/Yes”

Smith, Shoben & Rips (1974) Stage B l Compare only the defining features of

Smith, Shoben & Rips (1974) Stage B l Compare only the defining features of each concept. l Typical instances require only stage A. l Is a canary a bird? l Atypical instances require both stage A and stage B. l Is a chicken a bird? l Collins & Quillian's data are explained: – The features of canary overlap more with the features of bird and so there should be a faster decision, and indeed there is.

Linguistic Hedges As Fuzzy Set Feature Lists In Natural Language (Lakoff, 1972) l l

Linguistic Hedges As Fuzzy Set Feature Lists In Natural Language (Lakoff, 1972) l l l Hedges are modifiers that signal the use of fuzzy sets: – e. g. Strictly speaking; On the whole; Generally speaking Fuzzy sets are violations of those sets that are described by Smith, Shoben & Rips' DEFINING FEATURES A fuzzy set does not have a clear, well-defined boundary “Technically speaking, a penguin is a bird” – means that a penguin has the defining features, but not the characteristic features of a bird “Loosely speaking, a whale is a fish” – means that a whale has the characteristic features, but not the defining features of a fish

Feature Lists In Natural Language: Simile And Metaphor l Using Feature Lists to analyze

Feature Lists In Natural Language: Simile And Metaphor l Using Feature Lists to analyze simile and metaphor: – – – l e. g. Portland Building is like a wart on the face of the campus (simile) e. g. Eric Cantona: “ When the seagulls follow the trawler, it is because they think that the sardines will be thrown into the sea” (metaphor) Construct the feature list for one comparison, and apply it to the other