PL3103 Cognitive Psychology - Semantic Memory PDF

PL3103 Cognitive Psychology Dr. Cynthia Siew Today’s Topic: Semantic Memory (Week 6) After today’s lecture, you will learn about: 1. Key concepts in semantic memory (concept, category, propositions, schema, stereotype). 2. Theories of categorizing concepts (common...

PL3103 Cognitive Psychology Dr. Cynthia Siew Today’s Topic: Semantic Memory (Week 6) After today’s lecture, you will learn about: 1. Key concepts in semantic memory (concept, category, propositions, schema, stereotype). 2. Theories of categorizing concepts (common features, pro- totype, exemplar, knowledge-based). 3. Theories of semantic organization (hierarchical, network- based) and the mechanism of spreading activation. What is semantic memory? The part of declarative (explicit) long-term memory that stores your knowledge of general facts and concepts of the world. It also includes your knowledge of the meanings of thousands of words that you know in a particular language. What are the building blocks of semantic memory? Individual elements Concept: A mental representation of a category of ob- jects. Category: A set of class of objects that belong together. Combinations of concepts 1  Propositions Schemas Stereotypes Concept Category Schema When is a concept a concept? https://edition.cnn.com/2024/09/ 12/business/china-pets-over- Theoretical approaches to defining concepts: toddlers-aging-population-intl- hnk/index.html common-features prototype exemplar knowledge-based Common-feature approach Concepts are defined by essential, common features that are shared across all concepts of the same category. To decide if something belongs to a category, you check if it has the defining features of a particular category. Example: The concept of a fruit. Defining features: The ripe, reproductive body of a seed plant. Common features in most examples but not defining: Sweet, juicy, round, brightly colored. 2 3 4 Issues with the common-feature approach 1. It is very difficult to discover the set of defining features for any concept. 2. Despite having defining features, many concepts are still not easily classified, as they do not align with human intuitions. Demonstration of “fuzzy boundaries” [Everyone] Answer the questions on PollEverywhere! pollev.com/cynthia Fuzzy boundaries refer to the grey zone of transition that exist between categories, making it very difficult to define concepts on the basis of essential features. survey on concepts with fuzzy boundaries - fruits, veggies, sports Prototype approach Each category has a prototype. A prototype is a central description or conceptual store repre- senting the category, stored in your long-term memory. To decide if something belongs to a category, it is compared to the “prototype” to see if there is a good match of features. A high level of match is known as family resemblance. Demonstration of “typical members” [Everyone] Answer the questions on PollEverywhere! pollev.com/cynthia An item that is most representative of a category because it shares many features with the prototype is a highly typical member of that category. 5 Typicality effect: people are faster to decide that a typical mem- ber belongs to a particular category as compared to a non- typical member (robin vs. ostrich). survey to click on the image with the most prototypical example of a category to illustrate typical and non-typical members Limitations of prototype approaches not all concepts have clear prototypes: abstract concepts, or newly formed “ad-hoc” categories does not account for expert evaluations of concepts: e.g., their typicality decisions are based on a more complex evaluation, not just matching of features to the prototype Exemplar approach Instead of storing a single abstract “prototype” of a category, we store examples of the category in long-term memory. We also store “frequency” information about specific exemplars, i.e. how often it was encountered. To decide if something belongs to a category, it is compared to the examples (instances or exemplars) of various categories. Comparing exemplar and prototype approaches Each item has: 1. A prototype similarity score - computed by counting the match of its features to the prototype. 2. An exemplar similarity score - computed by averaging the similarity of the item to other items in the same category. DV: Category naming task - how long does it take to list the category that an “eagle” belongs to? Findings 1. Exemplar scores were better correlated with RTs than prototype scores. 6  2. Both exemplar and prototype scores moderately corre- lated with each other. Storms, G., De Boeck, P., & Ruts, W. (2000). Prototype and Evaluation exemplar-based information in natural language categories. Journal exemplar theories assume no abstraction of concepts; pro- of Memory and Language, 42(1), totype theories assume total abstraction 51-73. is it memory-efficient to store every single instance of a category? can there be any “general knowledge” of the world with- out any abstraction? best of both worlds: concepts and categories may vary on the amount of abstraction, depending on factors like experience/expertise and complexity/simplicity Knowledge-based approach Categories are based on knowledge about causal, functional, or structural properties of things in the world. To decide if something belongs to a category, it is compared to our knowledge about causal patterns across features that define a particular category. (Not sufficient to use number of shared features) Thought experiment Imagine if someone designed an airplane that is the same size as a medium sized bird, has feathers pasted on its exterior, has a nose shaped like a beak, and has specially designed wings that can flap. Would you categorize this object as a “bird”? [Vote on PollEv] 7 Evidence for knowledge-based categorization in children Which animal lives in a nest? dodo is more likely than the pterodactyl dinosaur to live in the nest, even though it is perceptually dissimilar than the target Gelman, S. A., & Coley, J. D. (1990). The importance of knowing bird. a dodo is a bird: Categories and inferences in 2-year-old children. Developmental Psychology, 26(5), Hub-and-spoke model 796–804. How can concepts be based on different types of information, and are both stable and context-dependent at the same time? 8  Each spoke is a modality specific region (e.g., sensory, mo- tor [bodily experiences; embodied approach], emotional, verbal) -> provides contextual variability Each concept has a hub which is a modality independent unified conceptual representation that integrates all the information stored in the “spokes” -> provides *stabil- ity** Concluding remarks Concept formation and categorization are complex cognitive processes involved in our semantic memory representation! There is no approach that is entirely correct. Each offers a dif- ferent perspective on various aspects associated with conceptual representation and processing. How is semantic information organized in memory? Theoretical approaches to semantic organization: hierarchical network-based 9 Hierarchies of concepts (Rosch et al., 1976) Concepts can be defined at various levels of generality. 1. Top-level, superordinate category (most abstract) - a type of furniture 2. Middle, basic-level category (best balance of informa- tiveness and distinctiveness) - chair 3. Bottom, subordinate category (most specific) - bar stool Limitation: The three-level hierarchy is too inflexible and arbitrary Thought experiment: Can you list examples of basic and sub- ordinate categories for the superordinate category of “Poke- mon”? Discuss with a friend. Selected students to share your answers on PollEv. potential discussion points: many ways of categorizing poke- mon, generation, type (but some have two types), legendary vs normal, vary depending on how much of an expert you are on pokemon, role of expertise and familiarity Networks of concepts Collin and Loftus (1975) argued that semantic memory was not organized in a hierarchy. Semantic information is organized based on semantic related- ness in a form of a semantic network. 10 Nodes = concepts; Links = connect related concepts How can we find out which concepts are related to each other? 1. Word association task 2. Feature listing task 3. Direct ratings of similarity 4. Computational approaches 1. Demonstration of the word association task https://singlishwords.nus.edu.sg/ 11 Explore associations to common Singlish concepts https://singlishwords.nus.edu.sg/visualise Choose a few “common” Singlish concepts from this list: https://en.wikipedia.org/wiki/Singlish_vocabulary#List_of_ Singlish_words If you would like to learn more about my lab’s research, do check 2. Sample data from feature listing task out our lab website: https:// langcomplab.github.io/singlish.html Data from: https://doomlab. shinyapps.io/double_words/#cue- 3. Sample data from direct judgments of semantic feature_variable_table similarity 4. Semantic vectors Data from: https://fh295.github.io/simlex.html. the meaning of words can be inferred from the other words that co-occur with it in a large corpora of text (usually Simlex-999 is a human-generated written and from the Internet) dataset commonly used to evaluate computational models of language and meaning. 12  latent semantic analysis is a connectionist model that computes co-occurrence patterns among words: the con- nection weights learned by the model correspond to the “semantic vectors” that represent the meaning of words we can quantify the similarity of two words’ meanings by analyzing their vectors Günther, F., Dudschig, C., & Kaup, B. (2015). LSAfun—An R package Spreading activation for computations based on Latent Semantic Analysis. Behavior a key concept in semantic memory Research Methods, 47 (4), 930–944. when a node in the semantic network is activated, acti- vation spreads to other connected nodes which themselves become activated -> contributing to a richer meaning of the original concept 13  Siew, C. S. Q. (2019). spreadr: An R package to simulate spreading Beyond concepts and categories activation in a network. Behavior Research Methods, 51(2), 910–929. https://doi.org/10.3758/s13428-018- Combining concepts 1186-5 Proposition: Combine concepts based on rules to result in a true or false statement. “Dr. Cynthia is a professor in the psychology depart- ment.” “Dr. Cynthia likes to eat sea cucumber.” Schema: A set of related propositions that forms an integrated packet of information about the world, events, or people. How to take public transport in Singapore Young families that have many pets Stereotype: A type of schema containing negative overgeneral- izations of a certain group of people. The importance of schemas facilitate perception and memory – easily recognize the card reader on the bus to tap your card – remembering to tap out when you alight the bus enable predictions, expectations, inferences – such families are likely to not be planning on chil- dren, and feel pressure from their older parents – I expect to see them at the next pet expo 14 War of the ghosts (Bartlett) How much are you able to recall from the story? Reflection questions: How was your memory for the story influenced by your own schemas? How might your interpretation of social situations be influenced by stereotypes? Today’s Topic: Semantic Memory (Week 6) After today’s lecture, you have learned about: 1. Key concepts in semantic memory (concept, category, propositions, schema, stereotype). 15  2. Theories of categorizing concepts (common features, pro- totype, exemplar, knowledge-based). 3. Theories of semantic organization (hierarchical, network- based) and the mechanism of spreading activation. Important info about mid-term test: 10 MCQs worth 1 point each 2 short essay questions worth 5 points each – for each question you have two choices, pick one – max. no of words = 300 (about half a single spaced page) bring a fully charged laptop with Examplify installed duration: 1 hour; usual time and location; closed book See you next week! 16