Reasoning and Decision Making - Lecture 3

Category

set of things that have something in common

Concepts

mental representation of a category which can differ person to person

allow us to make inferences/predictions - provide heuristics. used in medicine e.g. diagnosing strokes (FAST), important for law e.g. what constitutes murder or rape.

Categorisation

the act of classifying something as being a member of a category

ad hoc category = concept created to achieve a goal (Barsalou, 1983). e.g. behaviours to confuse people.

basic level - first learnt, balances distinctiveness + informativeness (Rosch et al., 1976). superordinate = broad, subordinate = specific.

The Classic Model

• everything in category must have all those attributes

• anything with all those attributes must be in that category

• attributes are necessary and sufficient for category membership

The Classic Model - Collins & Quillian (1969)

hierarchical network model - measured response times to true/false sentences.

the more jumps between category levels = longer verification time.

The Classic Model - Cons

• people struggle to give explicit definitions of a concepts if

  • they don’t know they exist

  • they don’t know the defining features

• many categories not clearly defined - boundaries unclear.

The Prototype Model

prototype = representation of the average/ideal member of a category.

store typical (prototype) examples of each category in brain - only prototype is stored to represented concept.

category judgements of new exemplar made by comparing with prototype.

The Prototype Model - Pros and Cons

Pros:

• captures borderline cases

• captures typicality effects (people agree on prototypes)

• lack of explicit definitions (unnecessary to describe prototypes)

Cons:

• how can we think logically if concepts/boundaries are unclear?

• are concepts flexible if fixed to similarity structures?

Spreading Activation Model

Collins & Loftus (1975) - propose words in lexicon represented as network of relationships in web of interconnected nodes.

connections = categories, typicality, association. retrieval of info occurs via a (limited) spreading activation.

activation of node needs to reach threshold for retrieval.

Spreading Activation Model - Pros

represents and explains semantic priming effects through spreading of activation between concepts. E.g., faster to say ‘nurse’ when preceded by ‘doctor’ than ‘fireman’.

Parallel Distributed Processing Models

McClelland & Rumelhard (1985); Rogers & McClessand (2003).

spreading activation plus distributed parallel features.

new/repeated experiences alter ‘weights’ between objects and attributes = allows correct association of attribute and object.

Impairments of naming specific objects’ categories

Patient RS (Samson & Pillon, 2003) - <20% correct naming fruit and veg, 60% correct naming non-fruit and veg.

Patient MD (Hard et al., 1985) - 60% correct naming fruit and veg, >80% correct for non-fruit and veg.

Double Dissociations between different classes of objects

Patient KC (Blundo et al., 2006) - 90% correct naming non-living (+ fruit and veg) objects, ~40% correct naming living objects.

Patient CW (Sacchett & Humphreys, 1992) - >90% correct naming living objects, ~40% correct for non-living objects.

Category-Specificity

Gen 1 Models:

• domain-specificity

• confounding factors

• visual accounts

• sensory-functional accounts

• distributed feature accounts

Gen 2 Models:

• distributed PLUS hub model

Domain-Specificity

mind has many specialised learning systems designed for processing different inputs e.g. animate, vegetation, objects.

model assumes specialisations have innate localisations in the brain.

Domain-Specificity - Studies

Mahon, Schwarzbach & Caramazza (2011) - blind people all show response in parietal cortex to tools. argued motor and somatosensory (not visual) input drives neural specificity for tools.

Martin et al. (1996) - PET study - silently naming pictures of animals, tools and non-objects matched for frequency and typicality.

• anterior cingulate - linked to action words - exhibit larger responses to tools than animals

• occipital cortex - linked to visual processing - exhibit larger response to animals than tools.

Confounding Factors

differences due to confounding differences across categories e.g. familiarity, name frequency, or complexity.

doesn’t explain double dissociations:-

• patient 1 - can name non-living, cannot name living

• patient 2 - can name living, cannot name non-living

• if living items more complex, explain patient 1 but not patient 2.

Visual Accounts

category-specificity due to visual similarities/differences between items within category.

Humphreys & Forde (2001) - animal and human faces more similar than tools and objects.

Gerlach et al. (2009) - strong activation in occipital cortex during object decision task.

CON = unclear what similarity actually is. unlikely to explain reason behind all deficits.

Sensory-Functional Accounts

deficits the result of patient losing either sensory OR functional semantic features associated with an object category.

natural objects more visual/sensory, man-made objects more functional.

Sensory-Functional Accounts - Farah & McClelland (1991)

lists of living v non-living words taken from dictionary. Ps underline visual or functional descriptors.

• 7.7:1 visual to functional features for living things

• 1.4:1 for non-living things

visual features more important for categorising living things.

BUT - some patients impaired for living things show knowledge of difference for sensory versus function - suggests associative knowledge.

Distributed Feature Accounts

deficits for individual category occur because items within category have more features in common. deficits for some items likely to affect other items that share features.

Distributed-plus-hub Model

semantic network AND amodal hub in the anterior temporal lobe (ATL).

hub:-

• info converges from mult. distributed networks

• contains a prototype concept

Distributed-plus-hub model - Face/Voice Impairments

modality specific impairment where the hub is intact, but modality-specific region is not. leads to loss of ability to recognise faces/voices, depending on impaired region.

cross-modal impairments - no hub but intact modality-specific regions. could lose concept of a person - cannot match faces.voices to the concept.

Distributed-plus-hub model - Semantic Dementia

gradual deterioration of semantic memory (Patterson et al., 2007).

specific/category specific information lost before basic and general information:

• word-picture matching task

• greatly decreased accuracy in severe SD in basic (dog) and specific (labrador) info compared to controls.

Semantic Dementia - Pobric et al. (2010)

transcranial magnetic stimulation - ‘knocking out’ ATL impairs general and specific semantic knowledge.

• inferior parietal - specific impairment for naming non-living things only

• anterior temporal - general impairment for naming living and non-living things