Conceptual Hierarchies Arise from the Dynamics of Learning and Processing: Insights from a Flat Attractor Network Christopher M. O’ConnorKen McRaeGeorge.

Slides:



Advertisements
Similar presentations
Cross Cultural Research
Advertisements

Figure 2. L2 Cognates vs. L2 Non-cognates in both language groups at the anterior electrode site Fz (finding A). Figure 3. L2 Cognates vs. L2 Non-cognates.
18 and 24-month-olds use syntactic knowledge of functional categories for determining meaning and reference Yarden Kedar Marianella Casasola Barbara Lust.
DEPARTMENT OF PSYCHOLOGY U N I V E R S I T Y O F C O P E N H A G E N Suppression of neutral but not emotional words Background Anderson & Green (2001)
General Knowledge Structure of Semantic Memory Schemas & Scripts
Cognitive - knowledge.ppt © 2001 Laura Snodgrass, Ph.D.1 Knowledge Structure of semantic memory –relationships among concepts –organization of memory –memory.
Chapter 9 Knowledge.
Chapter 7 Knowledge Terms: concept, categorization, prototype, typicality effect, object concepts, rule-governed, exemplars, hierarchical organization,
Representation/organization in LTM Typical empirical testing paradigm: propositional verification task – rt to rose is flower, vs. rose is plant. Set Theoretical.
›SIFT features [3] were computed for 100 images (from ImageNet [4]) for each of our 48 subordinate-level categories. ›The visual features of an image were.
Cognitive Linguistics Croft&Cruse 4: Categories,concepts, and meanings, pt. 1.
Knowledge ß How do we organize our knowledge? ß How do we access our knowledge? ß Do we really use categories?
PSY 369: Psycholinguistics Mental representations II.
Concepts and Categories. Functions of Concepts By dividing the world into classes of things to decrease the amount of information we need to learn, perceive,
Knowing Semantic memory.
Natural Categories Hierarchical organization of categories –Superordinate (e.g., furniture) –Basic-level (e.g., chair) –Subordinate (e.g., armchair) Rosch.
Scientific Creativity, Logic, and Chance: The Integration of Product, Person, and Process Research Traditions.
Chapter Seven The Network Approach: Mind as a Web.
Second Language Proficiency Places Cognitive Constraints on Sentence Processing Noriko Hoshino Department of Psychology The Pennsylvania State University.
Reading. Reading Research Processes involved in reading –Orthography (the spelling of words) –Phonology (the sound of words) –Word meaning –Syntax –Higher-level.
1 Validation and Verification of Simulation Models.
Knowledge information that is gained and retained what someone has acquired and learned organized in some way into our memory.
Categorization  How do we organize our knowledge?  How do we retrieve knowledge when we need it?
How do we store information for the long-term in memory?
Cognitive Psychology, 2 nd Ed. Chapter 8 Semantic Memory.
An Electrophysiological study of translation priming in French/English bilinguals Katherine J. Midgley 1,2, Jonathan Grainger 2 & Phillip J. Holcomb 1.
Lecture 10 – Semantic Networks 1 Two questions about knowledge of the world 1.How are concepts stored? We have already considered prototype and other models.
General Knowledge Dr. Claudia J. Stanny EXP 4507 Memory & Cognition Spring 2009.
Electrophysiological Correlates of Repetition and Translation Priming in Different Script Bilinguals Noriko Hoshino 1, Katherine J. Midgley 1,2, Phillip.
Semantic Memory Memory for meaning
Representation, Development and Disintegration of Conceptual Knowledge: A Parallel-Distributed Processing Approach James L. McClelland Department of Psychology.
Knowledge and Memory: How we conceptualize information.
Semantic Memory Psychology Introduction This is our memory for facts about the world This is our memory for facts about the world How do we know.
Where is the semantic network?. What is the semantic network?  Knowledge of objects, people, concepts and word meanings  Spreading-activation theory.
Semantic Memory Knowledge memory Main questions How do we gain knowledge? How is our knowledge represented and organised in the mind-brain? What happens.
Semantic Memory Knowledge memory Main questions How do we gain knowledge? How is our knowledge represented and organised in the mind-brain? What happens.
Emergence of Semantic Knowledge from Experience Jay McClelland Stanford University.
The Impact of Exposure to MSA on the Acquisition of Basic Language and Literacy Skills in Arabic Elinor Saiegh-Haddad Bar-Ilan University
PSY 323 – COGNITION Chapter 9: Knowledge.  Categorization ◦ Process by which things are placed into groups  Concept ◦ Mental groupings of similar objects,
The Influence of Feature Type, Feature Structure and Psycholinguistic Parameters on the Naming Performance of Semantic Dementia and Alzheimer’s Patients.
Access Into Memory: Does Associative Memory Come First? Erin Buchanan, Ph.D., University of Mississippi Abstract Two experiments measuring the reaction.
Category Structure Psychology 355: Cognitive Psychology Instructor: John Miyamoto 05/20 /2015: Lecture 08-2 This Powerpoint presentation may contain macros.
Similarity and Attribution Contrasting Approaches To Semantic Knowledge Representation and Inference Jay McClelland Stanford University.
Rapid integration of new schema- consistent information in the Complementary Learning Systems Theory Jay McClelland, Stanford University.
Long Term Memory: Semantic Kimberley Clow
1 How is knowledge stored? Human knowledge comes in 2 varieties: Concepts Concepts Relations among concepts Relations among concepts So any theory of how.
REFERENCES Bargh, J. A., Gollwitzer, P. M., Lee-Chai, A., Barndollar, K., & Troetschel, R. (2001). The automated will: Nonconscious activation and pursuit.
LEXICAL INTERFACE 4 OCT 30, 2015 – DAY 27 Brain & Language LING NSCI Fall 2015.
The Emergent Structure of Semantic Knowledge
Organization of Semantic Memory Typical empirical testing paradigm: propositional verification task – rt to car has four wheels vs. car is a status symbol.
Semantic Memory Psychology Introduction This is our memory for facts about the world How do we know that the capital of Viet Nam is Hanoi How is.
Verbal Representation of Knowledge
Theories of Priming II : Types of Primes Timothy McNamara Journal of Experimental Psychology,1994 조 성 식조 성 식.
Emergent Semantics: Meaning and Metaphor Jay McClelland Department of Psychology and Center for Mind, Brain, and Computation Stanford University.
Development and Disintegration of Conceptual Knowledge: A Parallel-Distributed Processing Approach James L. McClelland Department of Psychology and Center.
VISUAL WORD RECOGNITION. What is Word Recognition? Features, letters & word interactions Interactive Activation Model Lexical and Sublexical Approach.
Chapter 9 Knowledge. Some Questions to Consider Why is it difficult to decide if a particular object belongs to a particular category, such as “chair,”
Representational Hypotheses in Cognitive Psychology
Semantic Priming Effects in a Bilingual Gujarati Speaker
Psychology 209 – Winter 2017 January 31, 2017
PSY 323 – Cognition Chapter 9: Knowledge.
Experiment 2 – Discussion Experiment 1 – Discussion
Phonological Priming and Lexical Access in Spoken Word Recognition
Emergence of Semantics from Experience
Class Schedule In-text Citations Long-term Memory: Organization
[Human Memory] 10.Knowledge
Phonological Priming and Lexical Access in Spoken Word Recognition
Noriko Hoshino Department of Psychology
How is knowledge stored?
The Network Approach: Mind as a Web
Presentation transcript:

Conceptual Hierarchies Arise from the Dynamics of Learning and Processing: Insights from a Flat Attractor Network Christopher M. O’ConnorKen McRaeGeorge S. Cree University of Western OntarioUniversity of Western OntarioUniversity of Toronto at Scarborough London, Ontario, CanadaLondon, Ontario, CanadaToronto, Ontario, Canada Superordinate & Basic-level Representations Superordinate Priming & Temporal Dynamics of Similarity  both spreading activation (Collins & Loftus, 1975) and attractor networks predict that magnitude of semantic priming is determined by degree of semantic similarity  supported experimentally using basic-level concepts (McRae & Boisvert, 1998)  simulated using feature-based attractor nets (Cree, McRae, & McNorgan, 1999)  therefore, the degree that an exemplar target is primed by its superordinate should vary as a function of typicality  high typicality > medium typicality > low typicality  however, Schwanenflugel and Rey (1986) found that short SOA superordinate priming does not vary as a function of target exemplar typicality  replicated and simulated their experiment Experiment  72 superordinate-exemplar pairs, e.g., vegetable paired with peas, turnip, garlic  12 superordinate primes with 2 exemplars each of low, medium, and high typicality  200ms superordinate prime, 50ms ISI, exemplar target until response (concrete object?) Results: replicated Schwanenflugel and Rey (1986)  main effect of relatedness, F1(1, 42) = 8.09, p <.01,F2(1, 66) = 3.52, p <.07  main effect of relatedness, F1(1, 42) = 8.09, p <.01, F2(1, 66) = 3.52, p <.07  no interaction between typicality & relatedness, F1 < 1, F2 < 1 Feature Verification Conclusions  semantic memory can be represented as a single layer of semantics  without a transparent hierarchical structure  accounts for graded structure of categories  predicts online superordinate verification latencies; novel result  due to the temporal dynamics of similarity, accounts for counterintuitive and seemingly inconsistent results regarding basic-level vs. superordinate priming  results counter to hierarchical spreading activation theories Introduction  people’s conceptual knowledge structure for concrete nouns traditionally viewed as hierarchical (Collins & Quillian, 1969)  superordinate concepts (vegetable) represented at a different level in hierarchy than basic-level concepts (carrot, or pumpkin)  flat attractor networks – i.e., models with a single layer of semantics – have provided insight to a number of phenomena regarding basic-level concepts  semantic priming  statistically-based feature correlations  concept-feature distributional statistics  unclear how these networks could learn and represent superordinate concepts  can such a network account for established results and provide novel insights? Goals  demonstrate that a flat attractor network can learn superordinate concepts  simulate typicality ratings to show model accounts for graded structure  simulate feature verification latencies to demonstrate superordinate representations may be computed similarly to basic-level concepts  simulate superordinate semantic priming to provide insight into the temporal dynamics of similarity ModelStructure  input:30 wordform units representing spelling/sound of a word  output:2349 semantic feature units representing features taken from McRae et al.’s (2005) feature production norms  e.g.,,,,  e.g.,,,,  single layer of semantics; taxonomic features removed; all semantic features were interconnected  thus, no hierarchy built into the model Training  model learned to map random 3-unit wordform for each concept to semantic features for that concept  basic-level concepts trained in 1-to-1 manner:  3-unit wordform paired with same set of semantic features on every learning trial  superordinate concepts trained in 1-to-many manner  wordform paired with semantic features of one of its exemplars on each trial  e.g., wordform for vegetable paired with features of carrot on one trial, spinach on another, etc.  each exemplar was presented equally often  thus, typicality was NOT built into the model Wordform (30 units) SemanticFeatures (2349 units)  activation of features influenced by:  Feature Frequency:  if many exemplars possess a feature, it is strongly activated  Category Cohesion:  degree of featural overlap of exemplars determines activation of superordinate features  more overlap = more activation  Feature Correlations:  activate one another during the computation of meaning  Superordinate representations:  most features have intermediate activations  Basic-level representations:  all features have activations close to 1 (on) FeatureActivation Vegetable FeatureActivation Celery CategoryNCosine/Fam Res/Cosine/ TypicalityTypicalityFam Res furniture17.76**.62**.78** fruit29.71**.69**.91** appliance14.61*.73**.89** weapon39.58**.70**.76** utensil22.57**.52**.68** bird29.57**.49**.69** insect13.52*.69**.77** carnivore19.52*.45*.83** container14.46*.50*.51** vegetable31.45**.50**.90** musical instrument18.44*.54*.94** instrument18.44*.54*.94** clothing39.43**.50**.73** tool34.41**.38*.65** fish ** animal133.18*.12.55** pet ** herbivore ** predator ** mammal ** vehicle ** * p <.05, ** p <.01 Fam Res = Family Resemblance Typicality Ratings  important for any semantic memory model to simulate graded structure Experiment  collected behavioral typicality ratings for all 20 categories (7-point scale) Simulation  superordinate wordform presented & representation recorded  basic-level wordform presented & representation recorded  computed cosine similarity between each superordinate & exemplar  computed correlation between typicality ratings & cosines for each category  correlation between typicality ratings & family resemblance served as baseline Results  models predicts typicality ratings at least as well as family resemblance  therefore, the model was successful in simulating graded structure Feature Verification  similar “flat” attractor networks have simulated basic-level feature verification  model can also simulate verification of superordinate features Experiment  54 superordinate-feature pairs such as: furniture & fruit  54 superordinate-feature pairs such as: furniture & fruit  superordinate name for 400 ms, feature name until participant responded  "Is the feature characteristic of the category?" Simulation  present superordinate wordform and record feature's activation over 20 time ticks  correlated model's feature activation with human verification latency  feature activation in model predicts human verification from ticks References Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal of Verbal Learning and Verbal Behavior, 8, Collins, A. M., & Loftus, E. F. (1975). A spreading activation theory of semantic processing. Psychological Review, 82, Cree, G. S., McRae, K, & McNorgan, C. (1999). An attractor model of lexical conceptual processing: Simulating semantic priming. Cognitive Science, 23, McRae, K. & Boivert, S. (1998). Automatic semantic similarity priming. Journal of Experimental Psychology: Learning, Memory and Cognition, 24, McRae, K., Cree, G. S., Seidenberg, M. S., & McNorgan, C. (2005). Semantic feature production norms for a large set of living and nonliving things. Behavior Research Methods, 37, Schwanenflugel, P. J., & Rey, M. (1986). Interlingual semantic facilitation: Evidence for a common representational system in the bilingual lexicon. Journal of Memory and Language, 25, Simulation  superordinate prime wordform presented to model for 15 ticks  exemplar target presented for 20 ticks  cross entropy error recorded over last 20 ticks Results  typicality & relatedness did not interact, F < 1  main effect of relatedness, F(1, 66) = , p <.001  related lower than unrelated for ticks 1 to 13 Explanation  why is priming from superordinate to exemplar different than priming between basic-level concepts?  superordinate features have intermediate activations, which (due to the sigmoid activation function) require less change in net input to be turned on or off  basic-level priming: features in prime but not in target relatively difficult to turn off  prime & target must have high degree of featural overlap to produce priming  superordinate priming: activation of prime's features more easily changed  priming still results (vs. unrelated superordinate), but less sensitive to similarity  therefore, same amount of facilitation for exemplars of all typicality levels Acknowledgements NSERC grant OGP & NIH grant R01-MH to Ken McRae