Lecture 7 (Knowing) (1).pptx

Knowing Lecture 7 Overview of Major Topics in Chapter 7 • Semantic Memory (network versus feature list models). • Categorization, Classification and Prototypes. • Priming in Semantic Memory. • Context, Connectionism, and the Brain. Semantic Memory • Our permanent memory store of general world kn...

Knowing Lecture 7 Overview of Major Topics in Chapter 7 • Semantic Memory (network versus feature list models). • Categorization, Classification and Prototypes. • Priming in Semantic Memory. • Context, Connectionism, and the Brain. Semantic Memory • Our permanent memory store of general world knowledge. • How is knowledge (meaning) represented in memory? And how do we retrieve that knowledge? Loftus & Palmer (1974) • Follow-up question: – Was there broken glass? • Smashed: ~30% yes • Contacted: ~10% yes Estimate of Vehicle Speed as a Function of Verb Use 45 Speed (MPH) • Asked: – How fast were the cars going when they ______ each other? 40 35 30 Smashed Collided Bumped Hit Contacted The Collins and Quillian Model • Network Model of Semantic Memory • Network: An interrelated set of concepts / body of knowledge. • Node: A point or location in the network representing a single concept. • Pathways: Labeled directional associations between concepts. Spreading Activation • The mental activity of accessing and retrieving information from the network. • Takes passive concepts (those not currently in working memory) and activates them (puts them in working memory). • Activation then spreads to related nodes (e.g., activation to the doctor node would also spread to the nurse node). Semantic Network Propositions • Express a relationship between two concepts. • Examples: A robin has wings. An apple is a fruit. Pathways and Propositions • Pathways connect two nodes together to form propositions. • “ISA” pathways express category membership (e.g., A robin is a bird). • Property pathways express properties that concepts possess (e.g., x has the property of y-- a robin has the property of wings). Intersection Search • True or False: A robin is a bird? • Activation lights the robin node, and then spreads to its neighbors. • Activation also lights the bird node, and then spreads to its neighbors. • The two spreads of activation eventually collide-an intersection-- which lets you answer “True, a robin is a bird.” Semantic Relatedness • The distance between two nodes in a network is determined by semantic relatedness. • Concepts close in meaning / highly related (e.g., doctor, nurse) are stored close together in memory. • Unrelated concepts (doctor, truck) are stored far away. CANADA BEAVER MOOSE Smith’s Feature Comparison Model • Semantic memory is a collection of lists. • Feature Lists: Contain semantic features-- simple, one element characteristics-- of each concept stored in memory. Feature lists have a much simpler structure than network models. That is, they are more parsimonious. Sample Feature Lists Defining Features • Features absolutely essential to the concept. • Robins must be physical objects, have red breasts and feathers... • Defining features appear at the top of each feature list. Characteristic Features • Features that are common but not essential to the meaning of a concept. • Example: A robin perches in trees. • Characteristic features appear at the bottom of each feature list. Feature Comparison • The major process of information retrieval in the feature list model. • True or False: A robin is a bird? • To answer, first access each feature list from memory. • Second, compare each list for common features (feature overlap). Feature Comparison, Continued • Stage 1 comparison is fast and involves a global comparison of how much the features in each list overlap. • Stage 2 comparison is slow (occurring only when the lists have an intermediate amount of overlap). • Stage 2 involves only the defining features of each list. Feature Comparison Empirical Tests of the Models • Used the Sentence Verification Task (True or False, an x is a y?) • Key issues: Cognitive Economy Property Statements Typicality Effects Cognitive Economy • The incorrect notion from the original network model that redundant information is NOT stored in semantic memory. • Redundant information IS stored in memory (e.g., “flies” appears under the bird node, but also under the robin node). • Frequency of properties vs. hierarchical levels Property Statements • • • • Big problem for the Feature List Model. “Things that are Brown” list? What features would be included in this list? General Criticism: Why would the robin list contain every important concept about robins except that a robin is a bird? Typicality Effects • The Feature List Model was built to explain them. • More typical members of a category are judged faster than are less typical ones. • “A robin is a bird” is verified faster than is “A chicken is a bird.” • Adding semantic relatedness to the network model allows it to explain typicality effects. Imagery And Semantic Relatedness • Paivio (1990) – concrete words have a dual-coding advantage, verbal and visual knowledge. • Kounios and Holcomb (1994) examined the mean amplitude of the ERP response. • Concrete words: left = right • Abstract words: left > right Concept Formation • Traditional Research: Show subjects a series of arbitrary patterns and have them judge whether each is an example of the concept being tested. • Limitations of this approach? Natural Categories • Rosch (1973) Natural categories occur in the real world of our experience and have a complex internal structure. A collie is a “better” dog than is a dachshund. Natural categories have fuzzy boundaries-category membership is a matter of degree. Perceptual Categories and Prototypes • Perceptual Categories: Research on the Dani Tribe in New Guinea. • Prototypes: The central or core instance of a category. The prototypical bird may not even exist in the real world. Priming in Semantic Memory • Prime: Stimulus presented first in the hopes of influencing some later process. • Target: The stimulus that follows the prime. • Facilitation: Prime decreases processing time needed for the target. • Inhibition: Prime increases processing time needed for the target. Sample Priming Tasks Lag versus Stimulus Onset Asynchrony • Lag (studying priming across trials): • The number of intervening stimuli between the prime and the target. • SOA (studying priming within trials): The time between the onset of the prime and the onset of the target. The Lexical Decision Task • Meyer and Schvaneveldt (1971). • Subjects judge whether a string of letters is a word. • The semantic relatedness of the prime-target pairs was varied. • Reaction time was the primary index of performance. Studying Priming with the Lexical Decision Task The Automaticity of Priming • Neely (1976): By varying the SOAs, Neely concluded that priming is automatic. • Marcel (1980) Priming is an implicit process. Evidence for subliminal perception? Context • Simpson (1981) Ambiguity and Priming: “We had trouble keeping track of the count.” Context effects are priming effects. • Polysemous words are words with more than one meaning. • How do we assign meaning based upon the sound of a word when that word has multiple meanings? • Ambiguity resolution – Infer meaning from context or – Activate all possible meanings Swinney (1979): Ambiguity Resolution The man was not surprised when he found spiders and other bugs in his room. Word or Nonword ANT SPY SEW Ambiguity Resolution • This study showed that we tend to (initially) activate all the meanings associated with a word. • This effect disappears when the target word (e.g., ant, spy, or sew) is presented after the prime (e.g., bug). • One other qualification on this effect is that word frequency (i.e., how often a word appears in printed material) modulates the number of meanings activated (e.g., MacDonald et al., 1994). Connectionism and the Brain • Also known as PDP models (see chapter 2). • structurally similar to the brain’s network of neurons. • individual units in PDP models are also similar to those in the brain. • positive and negative weights mimic excitatory and inhibitory neural synapses. Lexical Memory and Anomia • Lexical Memory: The mental lexicon where word knowledge is stored. • Anomia: A deficit in word finding. PDP models can be “lesioned” to mimic the effects of anomia. Category Specific Deficit • A disruption where a person loses access to one semantic category of words but not others. • Example: Patient J.B.R. Had problems naming living things, but no problems naming non-living things. Disruption in sensory versus functional knowledge?