Please enable JavaScript.
Coggle requires JavaScript to display documents.
SEMANTIC MEMORY AND THE MEANING OF WORDS :PENCIL2: D+L2 LECTURE 7 -…
SEMANTIC MEMORY AND THE MEANING OF WORDS
:PENCIL2:
D+L2 LECTURE 7
Mapping thoughts and concepts to words
Word meanings are linked to concepts and world knowledge = semantic
Semantic memory
Concept = collection of features
How are these features organised?
Encyclopaedic knowledge.
Hierarchically structured models
Collins & Quinlan (1969)
Hierarchical models: organisation based on basic level categories and superordinate categories.
Evidence for this came from category verification tasks:
“A canary is an animal” Vs. “A canary is a bird” (yes / no?)
Reaction times increased as a function of nodes going up in the hierarchy
But this model did not explain why “A penguin is a bird” took longer than other cases.
Prototype models
Concepts made of frequent / typical features of typical category members
Graded internal structure as a function of similarity to prototype (whales, penguins)
Prototype models explained that because they noted that an important aspect in categorising (i.e., determining category membership) was the similarity to the prototype.
So whales and penguins are not very similar to the prototype of mammals & birds respectively and share features with fish.
This makes it more difficult to decide which category they belong to.
Embodied concepts
Grounding concepts in body actions and perception (Sensory-motor features)
JUG
→ perceptual simulations (shape, size) + actions it affords (drinking, grabbing, pouring); activates planning execution structures
Action verbs: execution-related features (kick, pick, lick)
Motor cortex is organised
somatotopically
according to the body part controlled.
Imaging studies indicate that when people observe actions or perform actions, similar brain regions are active as a function of body part involved (e.g., mouth, hand, foot).
Hauk et al. (2004):
Language can elicit similar activity, suggesting that motor aspects of actions are activated when processing language and action verbs, in particular.
Embodied cognition theory of semantic knowledge in the brain.
Predicts that whenever you hear a word, you are activating all of those sensory motor features, associated with that object or word.
Distributed models of semantic memory in the brain
Concepts: distributed network organised around sensory-motor functions
Different bits of the brain respond to different categories/concepts.
Networks of features make up categories, e.g. tools vs. animals
No evidence of a single meaning centre (Hauk, Johnsrude and Pulvermuller)
Spatial
features/spatial attention →
parietal lobe
Summary of results from imaging experiments in which activity elicited by pictures, videos or words is observed as a function of the features encoded in the pictures or words.
One problem for embodied accounts is how to explain abstract concepts such as TRUTH and BELIEVE, as it is not obvious what one would simulate or what sensory-motor features would be activated in those cases.
Distributed account, hierarchical account, prototypical account
What do impairments tell us about these models?
Issue =
correlational
→ need to look at brain damage to see if there is any causal link.
The case of category specific impairments...
Patient JRB:
difficulty understanding and naming some categories (Picture naming and object definition for living things)
Only a specific category elicits impaired responses.
Warrington & Shallice (1984)
Other patients showed the opposite pattern
Picture naming: Living = 6%, Non-living = 90%
Object definition: Living = 8%, Non-living = 79%
Living things tend to be known for their
sensory-perceptual properties
(e.g. shape)
Non-living things tend to have more
functional
properties (actions and uses)
Sensory vs. Functional features
JBR’s type of impairment can be explained as:
Damaged to category knowledge (as in hierarchical models) top node: living vs. non-living
Selective damage of sensory-features
→ knowledge representations are clustered (as in distributed models)
Okay with objects that are defined in terms of their functionality (e.g. Hammer)
Semantic dementia
Semantic dementia and sometimes
herpes simplex encephalitis
affects the anterior temporal lobe and leads to non-category specific semantic impairments
Inconsistent with a distributed view of semantic features.
Suggests localisation of function.
Mummery et al. (2000)
A progressive, degenerative brain disease (dementia) which particularly affects anterior temporal regions.
Loses grey matter.
Areas showing degeneration in 6 patients with semantic dementia.
Damage at the anterior temporal, greater on the left than on the right.
Deficit appears multi-modal (not specific to a sensory modality)
Impairment in recognition and understanding of words and objects
Independent of whether objects are cued by smell, sound, visual shape or touch
Spoken and written word are affected
Not category-specific, not restricted to a feature type
Relatively preserved...
grammar (syntax)
articulation (pronunciation)
'episodic' memory for events
spatial & geographical knowledge
executive control
Picture naming in SD
Hodges at el (1995)
Single patient
Losing low frequency and highly specific concepts first
Losing specific distinctions first – superordinate and coordinate errors are made (i.e., using the superordinate category or another basic level category to name a picture)
Impairments - Bozeat, Lambon Ralph et al. (2000)
Distinctive features of objects tend to be lost first.
Patients can copy drawings.
But if the picture is then taken away, drawings lose the distinctive features of the objects and become more generic.
Other factors that affect retention or loss of meanings include:
familiarity / frequency
How often objects are seen and their names are heard or read in everyday life.
Retained for longer
age of acquisition
Whether objects and their names are learned in early childhood or later childhood / adulthood.
Learned early = retained for longer periods of time
Mayberry et al. (JoCN, 2011)
Results of categorization experiments with SD patients: Patients are more likely to make errors with
atypical
members of the bird category (e.g., emu) as well as
over-generalization
errors (e.g., say that butterfly is a bird because it has wings).
Theory:
The temporal pole forms a modality-independent hub where features of meaning (appearance, sound, feel, use, etc.) are brought together from different spoke sites.
Hub-and-spoke model
Anterior temporal lobe = 'hub'
Anterior temporal lobe - multimodal integration
This theory can explain most of the features of semantic dementia
Distributed as well as convergent nature of conceptual knowledge
However, it does not so far explain:
Processes involved in semantic decisions
Context-dependent language use
Anterior temporal lobe is ideally located to be some sort of hub where semantic features come together in an a-modal nature.
Feeds in from the ventral and the dorsal route.
Ventral = object recognition pathway
Dorsal = ‘vision for action’ pathway
D
orsal →
D
ynamic
Semantic Processes
We do not only expect distributed semantic representations but also some way of selecting or deciding what is relevant for the context.
“Semantic control” is used to refer to meaning processes that are not automatic and require more effort and attention to detail
Broca's Area:
Left Inferior Frontal Gyrus (LIFG) is involved in retrieving, selecting and maintaining semantic information.
When semantic interpretation is more difficult (non-automatic), LIFG is involved.
LIFG typically acts in concert with PMTG (posterior middle temporal gyrus)
Functional connectivity between the two when making decisions about semantics/concepts.
LIFG is a convergence zone anatomically.
LIFG and PMTG in semantic control
Thompson-Schill et al. (1997)
Competitive word generation
Imaging study comparing activity when generating action words that are more or less strongly associated with the presented noun.
In the high-selection condition, generating the action word is harder because there are several possible alternatives and the process is less automatic.
(Gennari et al, 2007)
Competitive word understanding
Ambiguous words: BOWL vs. TRAY
More activity for ambiguous words than unambiguous ones.
Imaging study comparing activity to comprehension of ambiguous vs. unambiguous words.
LIFG and PMTG in semantic processing
Respond to difficult word selection
Number of alternatives: Higher activity in the LIFG
Weak associations > strong associates
Retrieval of “demanding” word types
Verbs > Nouns (Number of features and complexity)
Abstract > Concrete concepts (justice vs. chair)
Wagner et al. (2001)
Some form of semantic control is involved in selecting weakly associated words because one has to think harder about the alternatives.
Another manipulation was having to choose among 4 alternatives rather than two, which would make the selection process harder.
Language processing and semantic network revisited
Multiple convergence zones and distributed meaning.
Only see the overall activity of a task in fMRI studies.
MEG evidence might give us a lot more clear temporal resolution for what is going on.