2024 Categorisation Concepts PDF

Concepts and Categorisation PSYC20007 Cognitive Psychology Semester 2 2024 A/Prof Daniel Little Learning Objectives Identify in everyday life common categories that are experienced Understand the flexible structure of categories Explain the benefits of forming categories for cognition Give...

Concepts and Categorisation PSYC20007 Cognitive Psychology Semester 2 2024 A/Prof Daniel Little Learning Objectives Identify in everyday life common categories that are experienced Understand the flexible structure of categories Explain the benefits of forming categories for cognition Give examples of hierarchical nature of categorisation as the basic, superordinate, and subordinate levels Describe evidence against the classical view of cognition Outline reasons why basic-level categories are more fundamental than superordinate or subordinate-level categories Understand how features and attributes of categories can be used to determine category membership Lecture Overview Case Study: The Mysterious Disappearance of the HMAS Sydney Can we use Cognitive Psychology to solve the mystery? Social Categorisation and Stereotyping Selective Attention and Categorisation Evidence for Selective Attention in Categorisation If categorisation is harmful, why do we do it? What are the benefits of forming categories? What is the Mental Structure of Concepts? How do Categories allow for Generalisation? Conclusion: How can we use concepts from Cognitive Psychology to solve the mystery of the HMAS Sydney? WWII: 19 November 1941 Somewhere off the coast of Western Australia HMAS Sydney HSK Kormoran The Mysterious Disappearance of the HMAS Sydney  Timeline: Nov 1941 11 Nov 1941 HMAS Sydney battled23 Nov 1941 HSK Sydney departs Kormoran Victoria Quay somewhereNaval board signals HMAS Sydney off the Fremantle, WA coast of WA …No Reply 11 Nov 1941 19 Nov 1941 23 Nov 1941 25 Dec 1941 25 Dec 1941 German survivors interrogated 17 Nov 1941 20 Nov 1941 24 Nov 1941 Sydney relieved by HMAS 20 Nov 1941 24 Nov 1941 Durban at Sunda Strait Sydney is reported overdue 25 German in Fremantle. Seaman HMS rescued between the islands of Durban was delayed duefrom to industrial a raft WNWaction. of Carnarvon, Sumatra and Java Sydney also presumed delayed WA Survivors of the HSK Kormoran Survivors asked about: what happened? location of the HMAS Sydney? Interrogated to 7-21 days afterwards The accounts of the Kormoran survivors didn’t match up HMA Sydney couldn’t be found  Were the Germans lying? For several decades, most Australians concluded that the Germans must be lying Conflicting accounts were part of a ploy to mislead the Australians Ship hunters typically ignored the German accounts Until recently, the location of the HMAS Sydney was a mystery Many elements that make the story of the HMAS Sydney compelling to a Cognitive Psychologist Folk Devil Theory Folk Devil Theory introduced by sociologist Stanley Cohen in 1972, in his study Folk Devils and Moral Panics Three stages in the media's reporting on folk devils: 1. Symbolisation: the folk devil is portrayed in an oversimplified, easily recognizable, stereotyped fashion [you’re explicitly told what important attributes you should pay attention to] 2. Exaggeration: the facts of the controversy surrounding the folk devil are distorted or simply made up [those attributes are linked to something negative] 3. Prediction: [changefurther immoral actions in representation on the affects part ofmaking: decision the folk devil are anticipated. prediction of future negative actions based on selective attention to negative attributes]  Folk Devil Examples: Witch hunts West Memphis 3 Categorisation determines the answer to the question “why don’t their reports match?” Sailors/Experts Enemy Skilled in Untrustworthy navigation Lying Information about their location was critical to their survival Can you think of other examples where this happens? Selective attention and categorisation When we categorise, our attention is drawn is salient (attention-grabbing) or relevant attributes In turn, this influences how we generalise to new examples How do we know attention is important in categorisation? Shepard, Hovland & Jenkins (1961) Garner (1974), Lin & Little (2024) Nosofsky (1986) Shepard, Hovland & Jenkins (1961) Purpose of this study was to examine how people learn categories that differ in the number of features that define the category A OR B Correct ABrillig OR OR Mimsy B Wrong time Shepard, Hovland & Jenkins (1961) Candle/Lightbulb  Color Bolts/Nuts  Shape Violin/Trumpet  Size Over time, people learn that only candle/lightbulb is relevant for sorting into categories Shepard, Hovland, & Jenkins (1961) Category Learning Difficulty Shepard, Hovland & Jenkins (1961)  Category Learning Difficulty Paying attention to one dimension accentuates differences on that dimension Makes the objects which differ in colour appear even more different Because attention has limited capacity, Mental representation of the relationships attention decreases for other dimensions between objects paying equal attention to all features Makes objects that differ in unattended dimensions appear more alike on that dimension Mental representation of the relationships between objects after  Category Learning Difficulty Paying more attention to colour is helpful Mental representation of the relationships between objects paying equal attention to all features Paying attention to shape is not helpful Mental representation of the relationships between objects after  Category Learning Difficulty Because only a single dimension is relevant, learning results in selective attention to only that dimension This has the effect of making all of the black objects seem more similar to each other (likewise for the white objects), and all of the black objects very dissimilar to the white objects (and vice versa)Shepard, Hovland & Jenkins (1961) Category Learning Difficulty Selective attention to one dimension Equal attention to all dimensions Garner (1974) Control Correlated Orthogonal  Control Correlated Orthogonal Line Position Saturation sec sec sec  Control Correlated Orthogonal Line Position sec sec sec Saturation sec sec sec Lin & Little (2023): Increasing the number of stimuli in each category Attending to the relevant dimension makes the categories effectively all the same Collapses the stimuli along the irrelevant dimension Nosofsky (1986) Transfer tests provide a clue as to what people represent about the category Each cell is a stimulus (some combination of two dimensions) Some stimuli were trained to belong to either category 1 or category 2 The unmarked cells were shown after training to see what category people would respond Nosofsky (1986) Transfer tests provide a clue as to what people represent about the category Training Categories Transfer Performance 1 0.9 0.8 0.7 P(Category 1) 0.6 0.5 0.4 0.3 0.2 0.1 0 Summary Copy his notes from lecture capture If categorisation is harmful, why do we do it? What are the benefits of categorisation? The pulsating heartbeat of thought We think in categories Categorisation forms equivalence classes that allow us to decide appropriate action Categories allow us to infer ambiguous or missing features Categories reduce the complexity of the environment Structure our knowledge Allow for generalisation to new examples The pulsating heartbeat of thought At every moment, we are faced with an indefinite number of overlapping and intermingling situations Understanding the world involves the automatic and effortless evocation of categories Examples Language and Communication Speech is not just a series of sounds We group sounds into words and words in the parts of speech (nouns, verbs, adjectives), which we combine into sentences, that allow us to convey ideas Visual scene perception Objects in our perceptual field are not just wavelengths of reflected light We use edge detection, contour and texture perception, colour, movement to identify objects by their categories allowing to make sense of visual scenes Examples Medical Diagnosis Patients are not just a collection of symptoms Symptoms are grouped into diseases and conditions that allow doctors to make diagnoses and treatments Education Subjects are not just a hodgepodge of ideas Disciplines are groups into categories like sciences, humanities, and arts that allows for the organisation and prioritisation of learning, which helps understanding of relationships between fields of study Concepts vs Categories A concept refers to a mentally possessed idea or notion, whereas a category refers to a set of entities that are grouped together. A set of sounds /ba/, /pa/, /ta/, /ga/ can be grouped together into different phoneme categories The concept of the phoneme groups might include an understanding that voice onset timing differentiates /ba/ from /pa/ and that place of articulation differentiates /ta/ from /da/ Concepts vs Categories A concept refers to a mentally possessed idea or notion, whereas a category refers to a set of entities that are grouped together. Different medical conditions like diabetes, hypertension or asthma can be thought of as categories The concept of diabetes would include knowledge of important relations between blood sugar, insulin resistance, and other symptoms Concepts vs Categories A concept refers to a mentally possessed idea or notion, whereas a category refers to a set of entities that are grouped together. The category of Sphynx cats would include all individual Sphynx cats The concept of Sphynx cats would include knowledge of specific characteristics: hairless skin, cheekbones, bat ears Grounding by similarity Natural categories Examples: dogs, birds, apples Man-made artifacts Examples: chairs, clocks, bicycles, cars Ad hoc categories (share a common goal) Examples: Things that you take camping, Things that could be stood on to reach a lightbulb Abstract schema or metaphors (metaphorical qualities) Examples: Events in which a kind action is repaid with cruelty, metaphorical prisons, problems that are solved by breaking a large force into parts that converge on a target Stimuli have features, attributes, and dimensions - Small-to-medium size - Usually have fur - Have: Two eyes, whiskers, claws, tail - Have carnivorous teeth - Exhibit similar behaviours - Require Meat as part of their Diet  Which features do you include? The attribute “doesn’t own a raincoat” is true of almost all cats Usual focus on positive attributes (things which category has rather than things which it does not) Features vary in salience and validity Salience What attracts our attention? Cuteness, funny features, playfulness, is baby, etc Not all salient features are valid for defining the category Validity What is important for defining the category? Carnivorous teeth, retractable claws, flexible spine, sense abilities, whiskers, posture, … , body temperature regulation, vocalisation Benefits of Categorisation 1. Reduces the complexity of the environment 8 Basic Colors: Red Brown Orange Yellow Green Blue Purple Black Benefits of Categorisation 1. Reduces the complexity of the environment Inchworm Outer Space Manatee Copper Pink Flamingo Blush Wild Watermelon Benefits of Categorisation Only a few categories * Not in the 1903 Crayola Set Color universals: Black, white*, red, green, yellow and blue (plus orange, pink*, purple & brown) World Color Survey, Kay & Regier (200 Benefits of Categorisation 2. Provides a basis of deciding what constitutes appropriate action Benefits of Categorisation Subjects code letters by a raw, physical code, but this code rapidly (within two seconds) gives way to a more abstract conceptual code (1967) -Posner & Keele Benefits of Forming Categories 3. Provides a means for identify ambiguous or missing attributes Benefits of categorisation 4. Concepts allow for generalisation If you know that Then you are likely to think this bird has some that these birds also share property that property Benefits of categorisation 4. Concepts allow for generalisation If you know that But that these birds do not this bird has some property Benefits of categorisation 4. Concepts allow for generalisation If you know that But…what about these birds? this bird has some property Benefits of categorisation 1 0.9 generalization Probability of 0.8 The probability of generalization drops off quickly as similarity 0.7 decreases Generalization 0.6 0.5 0.4 0.3 0.2 0.1 0 n row jay rot en go e i ck min ltur Bird o b ar e ar i l u R Sp B P Ch Fla Vu  Benefits of categorisation 5. Enables the organization and relation of classes of objects and events 3 levels Superordinate Fruit Apple Basic Subordinate Pink Lady Envy Granny Smith Importance of Basic Level Categories Superordinate: Function Terms (Makes music) Basic: nouns & adjectives (strings, wooden) Subordinate only Subordinate: 14 adjectives Nonbiological (brown) Biological slightly more than 12 basic Number of Atributes 10 8 6 4 Far more listed for basic and subordinate 2 categories Only a few attributes0 were listed for Super Basic Sub Level superordinate level categories Rosch (1978) Listing Attributes 14 Nonbiological Biological 12 Number of Atributes 10 8 6 4 2 0 Super Basic Sub Level Rosch (1978) Note on biological vs non-biological conditions Rosch arbitrarily assigned words to category levels, but she could have incorrectly classified the biological levels We can explain the difference in results if we assume that words used as superordinate labels were actually at the basic level Note on biological vs non-biological conditions A better method would be to see what level resulted into the most features and call that the basic level We can explain the difference in results if we assume that words used as superordinate labels were actually at the basic level Correlated features Basic level categories have correlated features. This allows the predictions of missing features Shape Similarity Basic level shape averages were identified faster and more accurately when the shape was drawn from a basic level category 700 True False 650 RT (msec) 600 550 500 450 Superordinate Basic Subordinate Level Naming When shown a picture, people tend to use the basic level name Rosch (1978) Basic Level Names are Learned First Rosch (1978) Summary: Basic level categories Provide more information about the attributes of members Are identified by shape more accurately and faster Are retrieved more often as the category name Are learned first Why? Basic level categories contain the most information about the concept Superordinate Subordinate Levels of Categorization According to Rosch: Basic level categories are the level at which cue validities and category resemblance are maximized Cue validity How much does property X define the category (probability that an object belongs to category Y given that it has property X) Category resemblance How much do the members of a category go together? (sum of all features shared by members of a category have in common minus the sum of the distinctive features that are idiosyncratic to category members) Rosch, Mervis, Gray, Johnson, & Boyes-Braem (1976); Rosch, (1978); Rosch  Cue Validity Read this as: p Category Y, Property X  The probability of p Category Y |Property X   an object p Property X  belonging to Y given that it has X Example: Category Y = “birds” Property X = “wings” In words: Cue validity equals the co-occurrence of the category and the property divided by the overall occurrence (or frequency) of the property Cue Validity However, cue validity is not maximized at the basic level ¿ 𝐴𝑛𝑖𝑚𝑎𝑙𝑠 𝑤𝑖𝑡h𝑤𝑖𝑛𝑔𝑠 ¿ 𝐵𝑖𝑟𝑑𝑠 𝑤𝑖𝑡h𝑤𝑖𝑛𝑔𝑠 > ¿ 𝐴𝑛𝑖𝑚𝑎𝑙𝑠 𝑤𝑖𝑡h𝑤𝑖𝑛𝑔𝑠 ¿ 𝐴𝑛𝑖𝑚𝑎𝑙𝑠 𝑤𝑖𝑡h𝑤𝑖𝑛𝑔𝑠 Cue validity is maximized at the superordinate level  Category Validity and Category-feature Collocation We need another quantity: Category Validity Category Validity - the probability of having a feature given you’re a member of the category P(wings | bird) Category-feature collocation Product of cue and category validity Cue validity Category validity This is maximised at the basic level Jones (1983)  Category Resemblance Sum of all common features minus the sum of the distinctive features Maximised at the Basic Level Category: Furniture (Superordinate level) Has legs Has legs Has legs Has legs Has legs Used for sitting Used for sitting Used for lying down Used for placing items onUsed for storing items Has backrest Has backrest Accommodates sleeping Flat surface Shelves Can hold multiple people  Category Resemblance Sum of all common features minus the sum of the distinctive features Maximised at the Basic Level Category: Chairs (Basic Level) Has legs Has legs Has legs Has legs Has legs Used for sitting Used for sitting Used for sitting Used for sitting Used for sitting Has backrest Has backrest Has backrest Has backrest Has backrest Cushion Has wheels Comes with footrest For kids Exploring category resemblance further For any category there are some shared features and some idiosyncratic features What makes a category a category? Classical view: there are rules that define categories Fails to account for family resemblance and typicality effects Classical View of Categorisation Categories are definitions of necessary and sufficient conditions Categorisation requires checking for the presence of these definitional features The definition works for every member of the category People test different hypothesis about what the target category is Challenges to the Classical View Although we can think of some examples where there are category-defining rules For example, biological taxonomies: Domestic cats (Felis catus) share specific genetic, physical, and behavioral traits that distinguish them from other species within the genus Felis The structure of natural categories is not characterised by definitions Natural categories have a fuzzy structure in which some members are more typical than others Typicality and Family Resemblance Typical mammals: Dogs Horses Cats Bunnies Cows Typicality and Family Resemblance Less Typical mammals: Camels Squirrels Bears Wolves Donkeys Raccoons Chimpanzees Typicality and Family Resemblance Atypical mammals: Dolphins Anteaters Seals Koalas Kangaroos Armadillos Jackrabbits Otters Bats Typicality and Family Resemblance Natural categories are not homogenous People view members of a category as being on a continuum of goodness of membership Some members are more typical than others Natural categories are organised based on family resemblance Let’s return to our attribute listing study 14 Nonbiological Biological 12 Number of Atributes 10 8 6 4 2 0 Super Basic Sub Level Rosch (1978)  Attribute listing study Most attributes are listed for a single item Shared attributes are very rare, on average, there is only one attribute shared by all 20 category members Correlations between typicality and number of The correlations show that the shared attributes more typical an item was, the Furniture =.88 higher number of attributes Vehicle =.92 that it shared with other Weapon =.94 objects Fruit =.85 Vegetable =.84 Clothing =.91 Similarity maps for birds (left) and animals (right Rips, Shoben & Smith (1973) Typical category members are verified more rapidly An apple is a fruit: T or F? A pomegranate is a fruit: T or F? Priming results in faster RTs for more typical category members Answe r same: fast Answe r same: slower Answer differe nt Typicality and Family Resemblance Learning generalises more readily when the instances that are learned are typical of the category Dunsmoor & Murphy (2015) Typicality and Generalisation Category Induction tasks - All horses have Tricket’s illustrate key principles of disease how typicality affects “Tricket’s disease” is - All cows ahave made upTricket’s property generalisation disease that a category - All mice member can have have Tricket’s Induction: Generalizing disease from the particular to the - All lions have Tricket’s general diseaseThe subjects of each Given a set of examples, example can thought of examples of a single what is the The general subject of the category or as conclusion that one could conclusion sentence can - All mammals have categories Tricket’s in their own draw be a category or another diseaseright member of the same category or a member of a different category Category Induction -Robins have a higher potassium concentration in their blood than humans -All birds have a high X potassium concentration in their blood than humans ---------------------------- -------------------------- -Penguins have a higher X potassium concentration in their blood than humans -All birds have a high potassium concentration in their blood than humans Osherson et al. (1990) Category Induction: Effect of typical conclusion examples - Robins use serotonin as a neurotransmitter - Bluejays use serotonin as a neurotransmitter XX X - Sparrows use serotonin as a neurotransmitter -------------------------- ------------------------- - - Robins use serotonin as a neurotransmitter XX - Bluejays use serotonin as a neurotransmitter X - Geese use serotonin as a Osherson et al. (1990) Category Induction: Effect of conclusion category size - Bluejays require Vitamin K for the liver to function - Falcons require Vitamin K for XX the liver to function - All birds require Vitamin K for the liver to function ------------------------------- ------------------------- - Bluejays require Vitamin K XX for the liver to function - Falcons require Vitamin K for the liver to function - All animals require Vitamin K for the liver to function Osherson et al. (1990) Category Induction: Effect of premise example variability - Hippopotamuses have a higher sodium concentration in their blood than humans - Hamsters have a higher sodium X concentration in their blood than humans -All mammals have a higher sodium concentration in their blood than X humans ----------------------------------- --------------------------------- - Hippopotamuses have a higher sodium concentration in their blood than humans XX - Rhinoceroses have a higher sodium concentration in their blood than humans -All mammals have a higher sodium concentration in their blood than humans  Summary Members of natural categories share differing levels of family resemblance Typical instances are verified more rapidly, learned faster, primed more easily, and generalised more readily Generalization is affected by typicality of Instances, Typicality of Category, Category Size, Category Variability Any theory of categorisation needs to be able to explain these types of effects  Goals: Furnishing a Bringing it all together house versus buying an office chair Price, comfort, availability, use, etc How does selective attention interact with Low-cost office chairs typicality/family resemblance to influence generalisation? 1) Our goals or context indicate what is salient, relevant, or important 2) We attend to those aspects Typical chairs 3) Attention determines the categories that are activated, which determines typicality 4) We fill in missing or ambivalent features based on those categories 5) We generalise to new examples based on those categories Atypical chairs Returning to the HMAS Sydney What are the features that separate TRUE accounts from FALSE accounts? In 1991, John Dunn & Kim Kirsner set out to determine whether they could use the interrogation reports to deduce the location of the HMAS Sydney NPR Recording – Part I What are the typical features of true stories? True stories are collections of words and text Natural language has a particular statistical structure The most frequent word appears twice as often as the second most frequent word, which occurs twice as often as the fourth most frequent word and so on Zipf's law: the frequency of any word is inversely proportional to its rank  700 6.5 700 6.5 Zipf’s Law Distribution 66 600 600 5.5 5.5 500 500 55 Log(Frequency) 4.5 4.5 Log(Frequency) Frequency 400 400 Frequency 44 300 300 3.5 3.5 200 200 33 2.5 2.5 100 100 22 00 1.5 1.5 11 22 33 44 55 66 77 88 99 10 10 0 1 20.5 3 4 15 6 1.5 7 8 92 10 2.5 Rank Rank Log(Rank) Rank Rank One night--it was on the twentieth of March, 1888--I was returning from a journey to a patient (for I had now returned to civil practice), when my way led me through Baker Street. As I passed the well-remembered door, which must always be associated in my mind with my wooing, and with the dark incidents of the Study in Scarlet, I was seized with a keen desire to see one 1 night 1 it 1 was 4 on 1 the 4 twentieth 1 of 2  If Zipf’s Law is true, then when we plot the logarithm of the ranks against the logarithm of the frequency, it should form an approximately straight line 4 10 TOP 10 Words the 5252 and 2732 10 3 log(# of ocurances) to 2631 of 2618 I 2533 10 2 a 2519 in 1650 1 that 1545 10 was 1359 his 1096 10 0 0 5 10 10 log(rank ordering) We can look at the interrogation reports and see if the locations given by the German survivors correspond to Zipf’s law Dunn & Kirsner compared the Kormoran reports to the War of Ghosts data Kormoran Survivors War of Ghosts data 20 Position to 15 know uncertain Position to 10 know 5 0 26° S 111° E 26 32 111 27 111 coast 150 25 111 26 111-40 F 100 SW F 20 sw land 60 The frequency distribution is consistent with the assumption that the Kormoran database consists of random errors around a single position [From Kirsner, Norman & Dunn, 2003, Finding Sydney Foundation, 2005]  Use of Zipf’s Law to assess how well the reports confirm to the characteristics of true natural language One frequent report – Black triangles like ‘26° S 111° E’ are reports from Kormoran survivors Log Frequency Red triangles are reports from The “War of the ghosts’, a memory study by Bartlett (1932) Grey circles are From a Simu- lation based on Kormoran Many infrequent survivors’ reports reports - like ‘120 nm SW of Fremantle’ Log Rank  Kormora n 111° East Sydney 26° South Position of Kormoran established in 2008 26° 30’ South On 12 March 2008, the Kormoran was found 2.7 nm from the location suggested by Dunn & Kirsner  Scan of ocean floor On the 16 March 2008, the wreck of the Australian light cruiser HMAS Sydney II was discovered in deep water off the west coast of Australia, approximately 300 km south-west of Carnarvon. Underwater photo of HMAS Sydney Underwater photo of HMAS Sydney Cognitive Psychology “…other discipline-based John Dunn approaches were unable to offer a systematic account of human variability …cognitive psychology allowed us Kim Kirsner to generate…the view that the entire corpus could contain much previously neglected information.” Dunn & Kirsner (2010)