PSY2002 SEMESTER 1 - WEEK 2 Flashcards
define bottom up processing
based on sensory input and built up to semantic understanding
define top down processing
semantic understanding paired to sensory inputs
is speech comprehension concious or not
not conscious, automatic
define prosodic cues
hints to sentence structure and intended meaning by pitch, intonation, stress, timing
outline processes in mental lexicon
- hear word “text”, access it in mental lexicon to gain understanding
- phonology breaks it down (tekst)
- activate relevant syntax (noun/verb)
- activate relevant semantic meanings
- represent an orthographic representation (how it looks)
name challenges to lexical access
- continuous speech stream (no gaps inbetween words)
- homonyms (money bank vs river banks) and homophones (aisle vs isle)
- co-articulation (need tongue dexterity, influenced by prior sentance phoneme0
- accent
- invariance problem
explain segmentation in relation to lexicon access
having to separate out phonemes and words from pattern of speech sounds
give example of a homonym
money bank vs river bank
give a example of homophone
aisle vs isle
why is coarticulation difficult
require tongue dexterity, also dependent on speakers pronunciation of prior phonemes, and next one
increases variability of speech signal but is a useful cue for predicting next phonemes
what is invariance problem
problems of definition of acoustic properties that can occur at phoneme, syllable or word level
name ways of solving challenges of lexical access “disambiguating speech streams”
- categorical perception
- segmentation
- perceptual learning
- top-down processing
explain categorial perception as method of solving challenges of lexical access “disambiguating speech streams”
ability to distinguish between sounds on continuum based on Voice Onset TIme
explain segmentation processes as method of solving challenges of lexical access “disambiguating speech streams”
breaking speech signals into consituent words, using cues and where speaker places stress, with 3 categories of cue:
lexical - syntax + word knowledge
segmental - coarticulation
metrical prosody - word stress
explain perceptual learning as method of solving challenges of lexical access “disambiguating speech streams”
adjust categorical perception based on sounds we hear. become accustomed to different VOT’s of our own language, pick up on sounds which are important to us
explain top-down processing as method of solving challenges of lexical access “disambiguating speech streams”
phonemic restoration effect to process missing phoneme as if it were present
what 3 lexical characteristics affect speed of lexical access
- word length: long words slow to process
- frequency: more frequently word are accessed in lexicon, quicker it can be accessed
- neighbourhood density: lots of neighbours, processed more slowly
outline the McGurk effect
mismatch between spoken and visual info leads listeners to perceive sound/word involving blending of auditory and visual info
when is McGurk effect strongest
when auditory input lags 100ms behind visual input, as lip movement used to predictively anticipate next sound
why is top down processing important for McGurk effect
effects stronger when crucial word formed by blending auditory and visual input presented in semantically congruent sentence
in summary what are lexical access based on
- bottom up= acoustic input
- top down= disambuguate speech streams
- lexical characteristics
- context
- spreading activation that facilitates predictions
outline Elman & McClelland TRACE model
predicts that features activate phonemes that activate words with gradual increase in activation of words that match all features so that word with most activations win
explain evidence for TRACE model
gradually activate matching word more than others eg; ei____ activate say and pay, then hear ei____pr and activate more like april, apricot
gradual activations of items which match input, also works with rhyming competitor = lexical competitive inhibition
explain why TRACE is a implemented computational model
based on connectionist principles
- processing units/nodes at 3 levels correspond to mental representations of features, phonemes, words
- still rely on bottom-up from acoustic input
- nodes influence each other according to their activation levels and strengths of connections
- activation develops as pattern of excitation from faciliations/inhibition
- candidate words activated based on activation patterns
- bottom up/top-down process
all activated word involved in competitive process to inhibit each other words
explain how TRACE involves nodes
nodes influences each other depending on activation level and connections strength
feature nodes are connected to phoneme nodes which are connected to word nodes
explain how TRACE activation develops
as pattern of excitation from facilitation and inhibition
nodes influence one another in proportion to activation level, strength of interconnection
as excit/inhib spreads among nodes, pattern of activations (trace) develop
explain bottom-up and top-down process for TRACE
bottom up in activation from feature to word level
top down, in activation from word to feature level
outline Radical Activation model (stemming from TRACE)
any consistency between input and representation may result in some degree of activation
eg; show beaker but look at speaker
explain speaker/beaker study for radical activation TRACE model
ppts hear beaker, see number of objects (beaker, beetle, speaker, carriage)
eye tracking showed highest activation for object corresponding to spoken word (beaker)
and then phonological competitor= speaker
phonological competitor sharing first phoneme (beetle)
explain how eye-tracking (Allopenna, 1998) shows evidence for TRACE
show pictures, some in same cohort as semantic/phonological relate
ppts observe, do different tasks and track eyes (see which item activated in lexicon)
1. ppt activation trace showed words with overlapping phonology that don’t start with same onset as speech input (rhyme competitors) are activated in speech perception eg; beaker, speaker
2. initial cohort of words activated in response to speech streams not limited to word with same onsets
what should top-down processing with facilitatory links result in (TRACE)
facilitatory links between words and phonemes should result in more accurate detection of phonemes in words compared to non-words
ppts asked to detect a /t/ or /k/ in words (eg; heighten) and non-word (eg, veighten)
should be easier to identiy /t/ in word
give evidence for top-down processing being involved in TRACE
Mirman (2008)= ppts detect target phoneme (t or k) in words and non-words
faster identification in /t/ and /k/ words demonstrates effect of top-down processing
give limitations of TRACE
- assumes top-down influences originates at word level, and if were true then top-down effect should benefit target identificaiton more when target is word than just sound (sheep vs a baaaa), but context effects are just as much with environmental sound
- Fraudenfelder (199), ppts detect phoneme when was non-word resembling actual word (vocabularly, vocabutary). model predicts top-down effect correspond to vocatuary should impair identifying t, but doesn’t
- ppts fail to complete ambiguous phonemes with phoneme that would create word unless stimuli were degraded eg; identify /sh/ as final phoneme for “fiss”. predicts ppt showing preference for perceving phonemes as completing words but instead was when stimulus degraaded, not when not being degraded
- ignore role of context provided by verbs by influencing word recogs
outline Cohort model
predicts that we access words in lexicon via activation of all words sharing initial features and gradually deactivate words that stop matching features
give examples for cohort model
when receiving sounds, activate all words matching and then eliminate sounds that gradually stop matching what we hear
eg; ei___ activate April, Ape, Apricot then hear ei__pr___ and narrow it down to April
does Cohort model focus more on top-down or bottom-up
focus on process involved in spoken word recognition, more on bottom-up
name main assumptions of cohort model
- early in auditory presentation of word, all words conforming to sound sequence heard so far are activated (word-initial cohort, competes for selection)
- words within cohort eliminated if dont match further info like semantics, context
- processing continues until info from word itself and contextual info permits elimination of all but one cohort word (this is uniqueness point)
explain lexical activation in cohort model
lexical activation for all possible word when first hear a sound, then gradually deactivate words no longer matching acoustic input
reaches uniqueness point
what is uniqueness point in cohort model
reached candidate word that’s only remaining word to match acoustic input, then can access semantic meaning of word
explain neighbourhood effect for cohort model
words that match acoustic input compete for activation eg; apricot, aprikol
if we learn word aprikol, slows down recognition for apricot
explain frequency effect for cohort model
increased frequency of word activation means activated easier (eg; use apricot more than aprikol, activate apricot easier)
word with high freuqencies have high resting state so less activation needed to recognise high frequency words
explain gating experiments as evidence for cohort
ppts presented with word fragment, gradually reveal whole word, asked to guess what word is
eg; el….elephant, ellie, elevate
ele… elephant, elevate
eleph…. elephant
constrains cohort until reached uniqueness point
in cohort model, where are facilitatory signals, and inhibitory signals sent
facilitatory signals sent to word matching speech inputs
inhibitory signals sent to word not matching speech inputs
name 3 stages of cohort model word recognition
- access = acoustic-phonetic info mapped onto lexical item
- selection = candidate word mismatching acoustic input de-selected, candidate word chosen
- integration = semantic and syntactic properties of word are integrated and checked against sentence
why does sentence context not influence process of lexical access (in cohort model)
lexical selections based on activation of phonology, semantics
but integration affected by sentence context
explain priming paradigm for studying how cohort model processes context
doctor (prime), nurse (target) are semantically related so spreading activation allows nurse to be activated when doctor presented
sheep (prime) and nurse (target) not semantically related
explain how Zwitzerlood studied cross-modal priming in cohort model
prime word presented auditory and target word presented visually
related prime-target pair (hear captain, see ship)
unrelated prime-target pair (hear captain, see wicket)
works for word fragments too, hear cap, should also prime for ship as captain should be activated and money as capital’s also activated, via spreading activations so provides evidence for ‘cohort’
outline Zwitserlood study into impact of context, bias in cohort model
hear sentence to make one word more readily accessible, then presented with word options
“men around grave mourned loss of their cap__”
ship and money activated, but context SHOULD bias and cause quick response to ship however not seen!! instead other 2 still showed activation meaning contextual bias doesn’t work, doesn’t constrain cohort
shows context less influential for early stage, only impacts after uniqueness point when words integrate into sentence
summarise revised cohort model
more flexible than original model
assumed words vary in their level of activation and so membership of word cohort is matter of degree, and assume word initial cohort contains words having similar initial phonemes to presented word rather than consisting only of words having same initial phoneme
in revised cohort model, what is speech perception based on
matching acoustic input to stored representations of words in lexicon
in revised cohort model, how are words recognised
via competitive process that activates word “cohort”
in revised cohort model, when are word eliminated?
when cohort candidate doesn’t match acoustic input
compare original/revised cohort models for elimination of word
original assume word not matching context dropped out of word cohort (this is too extreme)
revised assume context-inappropriate word eliminated later on in processing
context influence selection of words, and those with semantic activation that fit into context of sentence are integrated then
compare TRACE to cohort for top-down process
TRACE emphasise top-down processing
cohort minimise impact of top-down processing
is trace constrained by initial phoneme
no- facilitatory connections travel up and down
does TRACE or cohort accomodate activation of rhyming competitor
TRACE
does TRACE or cohort not provide account of how context might affect speech production
TRACE- cohort does
give support for TRACE
- plausible account for phenomic restoration, categorical perception, word superiority in phoneme monitorings
- assume bottom-up and top-down processes both contribute directly to spoken word recognition so is good example of interactionist model
give limitations of TRACE
- narrow focus, and little to say on speech comp
- assume top-down influence specific word activation but research suggest top-down process initially activate higher-level conceptual meaning not specific word (should add another level onto model)
- exaggerate importance of top-down effect on speech percept, predict top-down activation from word level as causing mispronunc and ambiguous sounds to be identified as words more often than actually happens
give limitations of cohort model
- original state words cant be recognsied if initial phoneme unclear but french-speaking listeners activated words even when initial phonemes distorted
- context sometimes influence word processing earlier than integrations stage, especially with strongly predictive context.(revised accommodates this)
give limitations of revised cohort model
mechanism in spoken word recognition differs from that in model = predictive coding and enhanced processing of speech feature inconsistent with predictions
evaluate when spoken words are identified for cohort model
now thought identified when uniqueness point reached
research shows presenting word with early uniqueness point and late, and ERP found access to word meaning occurs sooner for early uniqueness point
what does evidence suggest is associated with uniqueness point (cohort evaluation)
Kocagoncu (2017) did MEG whilst presenting spoken words with varying uniqueness points and as predicted, each word uniqueness point associated with increased semantic processing, as well as reduction in lexical/semantic processing of competitor words
