Lecture 12: Semantic Memory (Part 1)

Administrative and Course Information

  • Assessment Protocols:

    • Quiz 4 is administered via Brightspace.

    • Password for access: concept.

    • Available starting at 2:00 PM2:00\,\text{PM}.

    • Time allotment: 10 minutes10\,\text{minutes} total completion window.

    • Protocol: strictly closed-book examination; quiet seating required upon completion.

  • Upcoming Exam Preparation:

    • Exam 2 takes place in 2 weeks2\,\text{weeks}.

    • Recommended study methodology: distribute practice across intervals rather than massing study sessions.

    • Complete textbook practice questions, covering both multiple-choice and open-ended formats.

    • Formulate active responses to questions presented in course outlines.

    • Construct flashcards targeting unfamiliar key terms and definitions.

    • Direct overlap notice: exactly 2–32\text{--}3 multiple-choice questions from the textbook will appear directly on Exam 2.

Foundational Principles: Episodic Memory Review

  • Episodic vs. Semantic Memory Systems:

    • Episodic memory encompasses memory for personally experienced events situated in a specific temporal and spatial context.

    • Semantic memory encompasses decontextualized, general world knowledge, facts, and concepts.

    • The functional and neurocognitive dissociation between semantic and episodic memory is evidenced by three primary empirical lines:

    • Neuropsychological patient dissociations (e.g., retrograde/anterograde amnesic patients who preserve semantic vocabulary and factual world knowledge despite profound episodic amnesia, or semantic dementia patients displaying selective semantic erosion with preserved episodic recall of recent personal events).

    • Functional neuroimaging divergences showing differential brain regional recruitment during autobiographical episodic retrieval versus generic semantic fact retrieval.

    • Behavioral and cognitive developmental differences showing differing developmental trajectories and forgetting rates.

  • Three-Stage Framework of Long-Term Memory Processing:

    • Encoding: the initial acquisition and processing of sensory and perceptual inputs into neural representations.

    • Storage / Consolidation: the maintenance and stabilization of encoded memory traces across time.

    • Retrieval: the reactivation, access, and conscious or unconscious utilization of stored representations.

  • Encoding Dynamics within Episodic Memory:

    • According to the levels-of-processing framework, information is encoded as a byproduct of the depth and type of cognitive processing performed on incoming stimuli.

    • Deep processing involves semantic, meaningful, and elaborative analysis of stimulus properties and associations.

    • Shallow processing involves physical, structural, perceptual, or acoustic/phonemic analysis (e.g., detecting typeface capitalization or rhyme).

    • Experimental testing of levels-of-processing utilizes incidental learning paradigms where orienting tasks require shallow judgments (e.g., "Is the word in capital letters?") versus deep judgments (e.g., "Does the word fit meaningfully into the sentence?"), followed by an unannounced memory test demonstrating superior recall for deeply processed items.

  • Episodic Retrieval Processes and Retrieval Failures:

    • Retrieval cues provide external or internal access pathways that prompt reactivation of stored memory traces.

    • Retrieval cues distinguish memory availability (whether an engram is intact in long-term storage) from accessibility (whether the trace can currently be retrieved given present cues).

    • Retrieval inhibition occurs when the selective retrieval of target memories causes competitive suppression of non-target, competing representations (e.g., retrieval-induced forgetting via inhibitory control mechanisms).

Organization of Semantic Memory: Hierarchical Network Models

  • Definition of Semantic Memory:

    • Semantic memory denotes our structured store of general knowledge about the world, including language, facts, concepts, and relationships.

  • The Collins & Quillian (1969) Hierarchical Network Model:

    • Concepts are arranged within a nested, strictly vertical taxonomic hierarchy of nodes.

    • Subordinate concepts inherit all properties and features of superordinate concepts.

    • Principle of Cognitive Economy:

    • Information and descriptive features are stored exclusively at the highest applicable conceptual node to eliminate informational redundancy.

    • Features common to all living entities are stored only once at the highest level rather than replicated at individual exemplar nodes.

    • Search and Retrieval Mechanism:

    • Verification and retrieval operate via formal search processes moving vertically up or down structural links in the hierarchy.

    • Reaction time (RT\text{RT}) is a direct linear function of the distance (number of hierarchical links) traversed between nodes.

Hierarchical network model diagram illustrating concept nodes and property inheritance from Living Thing to specific plants and animals
  • Structural Breakdown of the Collins & Quillian Hierarchy:

    • Top Level: Living Thing

    • Associated features: can grow, is living.

    • Intermediate Superordinate Level 1:

    • Plant (is-a Living Thing): associated features: has roots, has leaves.

    • Animal (is-a Living Thing): associated features: has skin, can move.

    • Intermediate Superordinate Level 2 (Botanical):

    • Tree (is-a Plant): associated features: has bark, has branches.

    • Flower (is-a Plant).

    • Intermediate Superordinate Level 2 (Zoological):

    • Bird (is-a Animal): associated features: has wings, has feathers, can fly.

    • Fish (is-a Animal): associated features: has gills, can swim, has scales.

    • Subordinate Terminal Exemplar Nodes:

    • Oak (is-a Tree): feature: is big.

    • Pine (is-a Tree): feature: is green.

    • Rose (is-a Flower): feature: is red.

    • Daisy (is-a Flower): feature: is yellow.

    • Canary (is-a Bird): features: is yellow, can sing.

    • Robin (is-a Bird): feature: is red.

    • Salmon (is-a Fish): feature: is red.

    • Sunfish (is-a Fish): feature: is yellow.

  • Evaluation of the Hierarchical Model:

    • Strengths:

    • Provided the earliest computationally and empirically testable psychological framework of semantic storage.

    • Explicitly formalized cognitive economy and logical inheritance.

    • Empirical Failures and Limitations:

    • Typicality Effects: Members belonging to the exact same taxonomic level do not exhibit identical retrieval latencies. For example, verifying "A robin is a bird" is systematically faster than verifying "A canary is a bird," despite both exemplar nodes residing exactly 1 link1\,\text{link} away from Bird.

    • Violations of Distance Effects: Semantic relatedness often overrides strict logical taxonomic distance. For example, participants verify "A shark is an animal" faster than "A shark is a fish," directly contradicting the prediction that traversing 2 links2\,\text{links} should require longer verification time than traversing 1 link1\,\text{link}.

Organization of Semantic Memory: Associative Network Models

  • The Collins & Loftus (1975) Spreading Activation Model:

    • Developed specifically to resolve the empirical failures of hierarchical models (typicality effects and distance violations).

    • Abandons rigid hierarchical tree structures in favor of an interconnected associative network.

    • Organization is determined by associative strength and semantic relatedness rather than logical nesting.

    • Conceptual connections are parameterized by: psychological distance between nodes, or explicit numerical weights representing associative link strength.

  • Retrieval via Spreading Activation:

    • Processing or perceiving a concept causes activation to build at that specific node.

    • Activation spreads automatically along associative links outward to neighboring nodes.

    • Spread intensity decreases across distance, attenuating over time and intervening links.

    • Intersecting pathways of activation from two nodes facilitate rapid relational verification.

Spreading activation network demonstrating associative links between semantic nodes such as bus, yellow, student, and related concepts
  • Empirical Verification of Associative Models:

    • Semantic Priming:

    • Presenting a prime stimulus (e.g., word) pre-activates conceptually related representations in long-term memory, accelerating subsequent processing of a related target.

    • Mediated Priming:

    • Activation cascades across intermediate links without direct pairwise association (e.g., node AA primes node BB, which subsequently primes node CC, yielding measurable priming between AA and CC).

    • Lexical Decision Task (LDT\text{LDT}):

    • Experimental paradigm requiring participants to determine as rapidly as possible whether a presented string of letters forms a valid English word.

    • Sample experimental items:

      • Real word stimuli: doctor, planet, yellow, rabbit.

      • Pseudoword / nonword distractors: xffxere, thewqr, moduys.

    • Key empirical takeaway: Reaction times to verify words are significantly shorter when preceded by semantically related primes compared to unrelated primes.

    • Sentence Verification Tasks:

    • Experimental paradigm requiring participants to judge statements as true or false as rapidly as possible.

    • Latency is dictated by the magnitude of associative relatedness: statements possessing dense, strong connections are verified faster because spreading activation travels shorter conceptual distances.

    • Illustrative contrast: verifying "Tigers have stripes" is executed significantly faster than "Tigers have lungs" because the visual and semantic association between tiger and stripes is vastly stronger than between tiger and generic mammalian internal anatomy.

Application: Bilingual Semantic Memory Architecture

  • The Storage Architecture of Multiple Languages:

    • Addresses how individuals who speak more than one language organize and store their cross-linguistic conceptual knowledge.

  • Competing Theoretical Formulations:

    • Dual-Store (Separate-Store) View: Proposes that each known language maintains an independent, segregated semantic and conceptual network with dedicated lexical-semantic nodes.

    • Single-Store View: Proposes a single, unified conceptual network that integrates all lexical representations across all known languages under common semantic nodes.

  • Empirical Evidence Supporting the Single-Store View:

    • Supported by cross-language semantic priming effects.

    • Exposure to a prime concept in one language produces automatic semantic facilitation for related target concepts presented in a different language.

    • Real-world experimental exemplar: presenting the Spanish prime "El libro" (book) reliably accelerates lexical processing of the semantically related English word "magazine".

Concepts and Categorization

  • Conceptual Foundations:

    • Concept: the internal, mental representation of an idea, entity, or class.

    • Can be concrete entities with tangible sensory properties (e.g., dogs, writing utensils, cars).

    • Can be abstract constructs lacking physical form (e.g., democracy, peace, love).

    • Category: the grouping or classification of real-world objects, entities, or instances together under a single concept.

    • Adaptive Utility: categorization reduces perceptual complexity, allows inductive reasoning, and enables immediate behavioral action toward novel stimuli.

Hierarchy of categorization showing Superordinate, Basic, and Subordinate levels using cutlery examples
  • Taxonomic Levels of Categorization:

    • Categories relate to one another in hierarchical, vertical tiers:

    • Superordinate Level:

      • Most general, inclusive category tier.

      • Low within-category similarity and few shared perceptual attributes.

      • Example: CUTLERY.

    • Basic Level:

      • Intermediate level of abstraction.

      • Cognitively privileged tier used predominantly in spontaneous communication, rapid object recognition, and everyday thought.

      • Maximizes cue validity: high similarity among category members combined with distinct boundaries from contrast categories.

      • Examples: spoon, knife, fork.

    • Subordinate Level:

      • Highly specific, concrete exemplars.

      • High perceptual overlap with competing contrast categories.

      • Examples: Soup spoon, Paring knife, Fish fork.

  • Category Boundaries and Family Resemblance:

    • Natural categories rarely possess clean-cut, deterministic boundaries.

    • Definitional challenges demonstrated through games: formulating an exhaustive, non-contradictory definition of a "game" that accommodates chess, solitaire, professional basketball, and ring-around-the-rosie fails due to absence of universal commonalities.

    • Category coherence is instead governed by family resemblance: instances share an overlapping cluster of correlated features, such that members exhibit general similarity to one another without possessing an invariant defining core.

Theories of Categorization

  • Major Theoretical Frameworks:

    • Four central models explain mental category representation:

    • Feature Comparison Theory.

    • Prototype Theory.

    • Exemplar Theory.

    • Situated Simulation Theory.

  • Feature Comparison Theory:

    • Concepts are stored as unbundled lists of linguistic and semantic features.

    • Feature taxonomy:

    • Defining Features: essential, indispensable attributes that an entity must possess to claim category membership (e.g., all birds lay eggs; all triangles have three sides).

    • Characteristic Features: typical, common attributes that members often possess but are not strictly required for membership (e.g., most birds fly; penguins and ostriches do not fly but remain birds).

    • Core Theoretical Limitation:

    • The vast majority of naturally occurring, real-world categories lack true defining features; rigid rule-based lists fail to capture human classification behavior.

  • Prototype Theory:

    • Categories are mentally structured around an abstract, summary representation termed a prototype.

    • The prototype reflects the central tendency or mathematical average of all previously encountered category members, rather than a single specific stored instance.

    • Categorization of novel entities is determined by calculating overall perceptual and conceptual similarity against this mental prototype.

Golden retriever representing a central prototype for the category dog
  • Empirical Evidence Supporting Prototype Theory:

    • Speed of Naming / Generation: Prototypical category members (e.g., robin for bird, golden retriever for dog) are named and generated significantly faster and more frequently in free recall tasks than atypical members.

    • Facilitated Priming: Superordinate category primes yield greater semantic facilitation for prototypical members than for atypical members.