Lexical Processing Flashcards
What is the mental lexicon?
[definition] a theory of how words are learned, represented, produced, and understood (theory of lexical activity)
What is the classical view of the mental lexicon?
A “lexicon-in-the-head” view.
- frequency is the most robust predictor of lexical processing (and human performance)
- therefore conceptualized as repeated exposure to a given lexical unit (repeat exposure results in better entrenchment in memory)
- “counters-in-the-head”
What is the (semantic) network theory?
That there is word form-to-meaning (a pattern of activity in a network)
State some evidence for semantic network representations.
[priming studies] a subject responds faster to a target stimulus because a related stimulus appeared in the context
- duck primes goose; horse does not (latency tests)
- neural activity related to goose differs after duck than horse
- easier to access if lions have stripes than something else, because lion is close to feline which is close to tiger which is really close to stripes
What are the stages of word recognition?
- initial contact
- - perceptual input acts as an interface with and activates lexical representations - lexical selection
- - the activated lexical candidates are evaluated against the perceptual-sensory input - word recognition
- - candidate is selected - lexical access
- integration + comprehension proper
What is lexical access?
The process of identifying words and recovering word-related information from LTM
- most models of lexical access deal with activating word form information
- it very fast (speech ~ 5 syllables/second; print ~ 200 - 250 ms)
How does modality factor into processing models?
Most models assume modality specific input representations
– separate for auditory (Wernicke’s) and visual word processing (visual word form area near visual cortex)
What is the cohort model for lexical access?
That multiple candidates (word initial cohorts) are activated by onset.
- there is a uniqueness point at which the last compatible candidate wins
- all or nothing (not sensitive to frequency)
What role does competition have in these models?
ALL MODELS ASSUME COMPETITION!
– words can compete even if they do not share any segments (ie. shipping vs. ship inquiry; shipping and inquiry compete for the same portion of the utterance)
What is the TRACE model?
There are complex input representations (“pseudo-spectral” acoustic-phonetic features).
- these encode the degree to which the feature is found in the input
- units hypothesized at the featural, phonemic, and word levels are replicated at each level
- at any given time slice may activate several different phonemes
[lateral inhibition] – units that span the same portion of the input inhibit each other
[OVERALL]
There is immediate cascaded activation of input representations – competition (the strongest hypothesis wins!)
What is spreading activation?
Activation of similar words, essentially.
What is the neighbourhood activation model/how does it account for spreading activation?
Looks at global similarity rather than onset.
– so /cat/ is similar to /bat/ and /at/ but not /castle/, since the difference in phonemes is too large
How do cohort models incorporate spreading activation?
With onset similarity.
– so /cat/ will activate /cab/, /cap/, /castle/ and so on, but not /bat/ (because initial sound segment doesn’t match)
How does the TRACE model account for spreading activation?
With onset and rhyme overlap (activated even with onset mismatch).
– time is important! early activated words are advantageous over later activated ones (the more information that comes in the better); the selection in favouring the best candidate
What are some problems with processing speech?
- non-linearity!
- - co-articulation
- - assimilation
- - stops are actually silent - segmentation problem
- - speech does not have sequences of discrete sounds
- - the listener must have a way to assign a certain acoustic features in the input to a given segment - embedding problem
- - most long word have multiple short words embedded (so /unitary/ contains /unit/, /you/, /knit/, /it/, /tarry/, /air/, and so one) - lack of invariance
- - acoustic properties vary by context (speaker characteristics, phonetic context, speech temp, environmental noise, foreign accents, etc.)
What is categorical perception?
Acoustic continuum becomes discrete representations.
- many signals but few phonemes
- gradual physical change but abrupt perceptual change
Is human speech special then?
Yes and no.
[NO]
- there is categorical perception in animals (chinchillas and quails, for example)
- there is categorical perception of non-speech sounds as well
- - these point to a non-special sound-processing development (no argument for genetically determined speech-processing module)
[YES]
- innately guided learning processes
- - we instinctively pay more attention to certain aspects of the environment (ie. speech)
- - especially refined/detailed processing of speech sounds - infants come ready equipped with all possible sound inventories (??)
- - babies recognize contrasts they haven’t learned
- - up to about 6 - 7 months
- - at 8 - 10 they start making native language distinctions only
What is the McGurk effect?
A perceptual phenomenon that demonstrates an interaction between hearing and vision in speech perception!
- see mouth moving in a certain way can change how we hear phonemes
- shows that other domains can affect speech perception (no strict modularity)
What is the relationship between conceptual knowledge and knowledge about word forms?
NONE~~
- people can have seemingly intact knowledge of concepts/meanings but no ability to recover information about the word forms
- the brain’s response to words depends on properties of words and properties of task
Where is activity in the brain that is specific to words?
LEFT inferior parietal region
Where is activity in the brain that is specific to pictures?
RIGHT middle occipital region
How do concepts and knowledge about words related to semantics in the brain?
Conceptual knowledge (non-linguistic) and knowledge about words/word-forms (linguistic) combine to enable semantics/meaning when words are either heard or read or signed -- slight complication in considering exactly what a word is (definition)
What are some modality-specific representations in the brain regarding word types?
- more difficult producing nouns than verbs in speech
- more difficulty producing verbs than nouns in writing
– argues that DD results from differential impairment in modality specific output lexicons and makes reference to there being distinct grammatical categories of nouns and verbs
ADDITIONALLY there is differential activation of brain regions for categories (ie. animals [occipital] and tools [inferior frontal regions], or concrete words vs. function words)
What are some task effects on neural responses?
There is differential activation for passive viewing of words compared to staring at a fixation cross, and action generation (name the action the noun; ie. hammer).
- for passive viewing there is activation in the left and right occipital lobes
- for action generation there is activation in the left frontal lobe
So what does brain activity depend on (for lexical access)?
- characteristics of individual words
- task that the person is doing
- - tasks specific to word processing activate a network of left hemisphere regions - modality
- - auditory input more strongly activates areas within the vicinity of Wernicke’s area (visual input may not activate this area at all)
ADDITIONALLY post-lexical (modality independent) activation is wide-spread in the LH and depends heavily on the task
What are localist theories of meaning representation?
- semantic memory represents biologically important (via. natural selection) domains in separate categories
- this conceptual division is reflected in the physical brain as differently localized representations
- category specific deficits (ie. natural vs. man-made objects)
What is the posterior-anterior view of meaning representation?
This idea that more posterior regions are related to simpler concepts, while more anterior regions are related to processing of more complex concepts and relational information
– a functional vs. perceptual distinction (tools, which are defined by their functions, activate posterior regions; animals, defined by what they look like, activate more anterior regions)
- problems with verbs is associated with frontal lobe damage
- problems with concrete nouns is associated with temporal lobe damage
What is the idea of language “networks”?
DORSAL-VENTRAL
[that the superior temporal gyrus contains two streams for early cortical stages of speech perception:
- dorsal stream: towards inferior parietal posterior frontal lobe regions
- - form/articulation - ventral stream: middle and inferior temporal cortices
- - meaning/comprehension
What is the traditional model of lexical access (in the context of looking at dual-stream models)?
lexical access is through a single route with various levels of representation (whether these levels ‘exist’ as part of the neural processing is unclear)
What is the dual-stream model of lexical access?
That there are multiple routes to lexical access in speech (implemented as parallel channels)
VENTRAL STREAM: sound-to-meaning
– further: organization is bilateral; Wada, split brain, unilateral damage etc. speech processing is intact, bilateral damage to superior temporal lobes is associated with severe problems in speech recognition)
DORSAL STREAM:
sound-to-action
– role less clear
– likely it supports an interface between auditory processing and motor system; there needs to be one because learning to speak is a motor task
– at least two levels of auditory-motor interaction (one involving speech segments, another involving sequences of segments(
What is the new theoretical perspective for how cortical systems function for language?
The old one is modular, the new one is distributed representations!
– rather than local encapsulated modules, cortical functions might be served by distributed interactive functional systems
What are some of the reasons to think that there’s clout in the distributed representation model?
CONNECTIONS
- the dorsal and ventral cortex connect with auditory areas in the superior temporal gyrus and sulcus
- there are multiple links between the superior temporal language area and the motor system
- information flows between the systems for language and action
- the cortical basis of learning: cortical systems for language and action develop specific links between each other whenever actions correlated with specific language processes
What is a distributed cell assembly?
Proposed by Donald Hebb as an alternative to both localization and holistic approaches.
- there are cell “assemblies” with well-defined cortical topographies
- [against localization] neurons in different cortical areas may contribute to the same complex process
- [against holistics] representation of a process may involve complete different distributed neuronal substrate than another process
So, in theory, representation of a word is distributed over well-defined areas!
What are the three fundamental assumptions about cortical functioning?
- coactivated neurons become associated
- associations can occur between adjacent or distant neurons (ie. the entire cortex forms an associative memory)
- when neurons become associated, they will develop into a functional unit = cell assembly
What does Pulvermuller have to say on the topic of neural architect?
Action words: abstract semantic links between language elements and motor programs.
- these words are learned in the context of action performance
- synaptic connections between neurons in specific motor and premotor areas and in the language area become stronger
- neuronal assemblies!! with specific cortical distributions!!
How are semantic concepts and actions linked?
Action words elicit specific patterns of activation depending on the motor movement they invoke (reading words silently still shows this).
this means that:
- action word recognition triggers automatically the activation of specific action-related networks
- motor activation is due to late postlexical strategy to plan an action
How does the distributed representation model deal with explaining category-specific deficits?
the correlated features approach!!!
- knowledge about objects is spread throughout the brain
- so hearing “cat” will activate a wide-spread network of neurons
- the loss of knowledge of living things is more common than of artificial entities (localizationists can explain why these two areas of knowledge differ but not why one is more common)
- semantic loss is graded, not absolute (so how well patients do depends on the task; how much detailed knowledge it requires independent of whether it is about animals or tools)
