semantic memory

>type of long-term, declarative memory >contains representations of concepts underlying objects, actions, abstract words etc.

SM retrieval

>not random >Bousfield (1953)- recalled in group clusters after retrieval >Mandler (1967)- organising according to own rules and categories before retrieval improves recall

concept

mental representation capturing what members of a category have in common

category

groups of items belonging together because they share meaningful features

categorisation

cognitive process by which item are organised into categories

why categorise?

>cognitive economy- reduces the complexity of the environment >generalisation- recognise unfamiliar/new examples >cost saving- establish hierarchies

subordinate category level

specific objects with many attributes

basic category level

balances informativeness and economy

superordinate category level

general categories with few attributes

why is basic level special?

>almost exclusively used in free-naming tasks >quicker to identify as a member of a category > learnt first >more common in discourse >common across cultures

category membership

>Wittgenstein (1953)- family resemblance measure (continuous) that measures shared attributes between category members >typicality- differences in how well members relate to their category

traditional view to concepts

>Frege (1952)- concepts characterised by a set of defining attributes >share fundamental features that are individually necessary and collectively sufficient >all members represented equally

issues with traditional view

>Eleanor Rosch >not all members are equal >typical category members listed first >better semantic priming for typical category members

prototype view to concepts

>based on typical members of a category > more characteristic features=more likely to be a member >family resemblance structure- one core representation

types of prototypes

>average of all members >set of characteristic attributes >specific instance of the category

evidence for prototypes

>typicality gradient >most typical member usual named first >learn typical members first >prototypical members more affected by priming >Posner and Keele (1968)- unseen prototypes as good as old stimuli >Rosch and Mervis (1975)- set of defining attributes doesn't work for all items of a category when using superordinate features

strengths of prototype view

>better explanation of typicality effects >plausible that prototypes stored in semantic memory

weaknesses of prototype view

>problem with abstract and goal-oriented categories >problems explaining expertise effects

exemplar approach to concepts

>concept is represented by multiple examples so examples are actual category members and new items are compared to stored items >the more similar an exemplar is to a known category member, the faster its categorised

strengths of exemplar view

>categories more coherent and stable >more flexibility >typicality effects >outperforms prototype model when using complex categories

weaknesses of exemplar view

>too unconstrained

variable abstraction model of concepts

>Vanpaemel and Storms (2008) >high abstraction uses prototypes >low abstraction uses exemplars

knowledge-based view of concepts

>categorisation guided by understanding of the world >previous approaches don't capture causal relations between attributes >concepts change with context

Hub and Spoke model of concepts

>spokes represent different types of information (visual, auditory etc) >hub combines information into stable concept

semantic networks view to concepts

>Collins and Quillian (1969) >concepts arranged to represent the way concepts are organised in the mind

cognitive economy in SN

>shared properties only stored at higher level nodes >exceptions stored at lower nodes >lower-level items shared properties of higher-level items

spreading activation in SN

>activation is arousal level of node >activity spread along all connected links >concepts receiving activation are primed and more easily accessed from memory

issues of SN

>can't explain typicality effects >some sentence-verification results are problematic >issues in falsifiability

modifications of SN

>shorter links for more closely related concepts >longer links for less closely related concepts >based on personal experience

latent semantic analysis (LSA)

>computational method for extracting word and passage relations from 'bags of words' >learns word meaning from patterns of word co-occurrence >words represented as vectors in semantic space >words related if appearing in similar context >ignore grammar and word order

connectionist approach to concepts

>'neuron-like units' >input units are activated by stimulation >input units -> hidden units -> output units >parallel distributed processing >learning can be generalised

graceful degradation in connectionist approach

disruption of performance occurs gradually as parts of the system are damaged

error signals in connectionist approach

>occurs when output unit is different from correct response >Back-propagation- transmitted back through the circuit to change weights to match correct signal