6-Auditory word recognition and auditory verbal agnosia Flashcards
Key problems in speech processing: Continuity
Listeners perceive whole words as discrete entities. In fact, gaps in the speech string are very rare.
which is why we think foreign speakers “talk so quickly”
Key problems in speech processing: Variability
Speech sounds can be highly variable.
Assimilation: speech sound becomes more similar to the neighbouring sound e.g. “handbag” -> /hæmbæg/ and “input,” the /n/ might become [m] ([ˈɪmpʊt]) due to the influence of the bilabial /p/.
Coarticulation: /n/ in “tense” vs “tenth”. How one sound is realised is influenced by next sound along, and next word along!
Between-speaker variation, also within-speaker variation.
Also: speech is very often perceived against competing sounds.
Despite variability, fairly consistent mapping
Key problems in speech processing: Sequentiality
We can understand speech despite fleeting exposure to 20 or more individual sounds per second.
Key debates in speech processing:
Modularity- is speech processing special? or do we process it like every other sound?
Top down vs Bottom up processing- Influence of context on perception of individual speech sounds.
Acoustics to meaning- “…the listener can be thought of as a device for the conversion of acoustic input into meaning…”
(Cutler & Clifton, 2000)
Informed by deficits- including auditory verbal agnosia- speech is heard and to some extent processed but meaning is not accessed.
Proposed solutions to speech processing problems- continuity
Cues to word boundaries
- lexical knowledge (finding word we know), sentential context
- segmental-acoustic (phonotactics (-ch followed by -st very rare so likely to be a boundary), allophony, timing)
Metrical segmentation (IN ENGLISH)
Most words begin with a strong stress syllable.
between languages e.g. garage vs garage
85-90% of English words begin with syllable stress- Cutler & Carter (1987)
Hierarchical model of word segmentation.
Starts with lexical–> segmentational-acoustic –> metrical
Metrical Segmentation Strategy hypothesis:
Listeners use placement of stressed syllables to determine word boundaries in speech.
Evidenced by, e.g., slips of the ear:
In speech in noise or reduced intensity:
achieve -> a cheap; how big is it? -> how bigoted?
Variability models of speech perception.. What is Motor Theory of Speech Perception?
Critical problem in perception of individual speech sounds is their variability:
MOTOR THEORY OF SPEECH PERCEPTION
When we listen to someone speak, we are trying to produce what they are saying, running motor programming sub vocally. Use motor system to generate a copy of what they are saying to understand it.
What we perceive is underlying gestural intention which is more stable than the acoustics.
Listeners recreates articulatory gestures of speakers (“analysis-by-synthesis”).
Utilisation of speech production mechanism for perception.
But… we can perceive accents we cannot imitate.
Later revival of Motor Theory informed by discovery of mirror neurons.
Variability models of speech perception.. What is Fuzzy Logic Model?
Evaluate input on multiple features for closeness-of-fit to templates.
Step 1: Feature evaluation:
e.g. Does sound include initial lip closure?
Answer can be: 1 - “yes”; 0 - “no”; or some degree of possibility, “fuzzy” truth value between 0 and 1.
Step 2: Prototype matching:
Feature evaluations combined and compared to templates stored in memory.
Step 3: Pattern classification:
Best match maximises “goodness of fit” between input and candidate templates.
Evaluate input on multiple features for closeness-of-fit to templates.
Advantages:
Deals well with lack of invariance in speech signal.
Gradient distance from ideal taken into account.
Not “all or nothing”.
Problems:
Falsifiability:
Hard to make concrete predictions given so much flexibility.
Variability models of speech perception.. What is Acoustic Landmarks and Distinctive Features model?
Claim that reliable correlates of all phonological features are actually availlable in the speech signal.
Need to look in right place, hence “landmarks”
e.g. [-voice] for stop consonants signalled by
voice-onset time greater than ~25ms.
Listener identifies individual segments via cues to abstract phonological features in the acoustic signal.
Differences due to coarticulation are limited and systematic
But Not all clear distinctive features have clear acoustic landmarks
Not all phonetic variation is predicable from phonological specification… speaker differences, rate differences.
Proposed solution to the speech processing problem of sequentiality (fleeting nature of speech)
Models of word recognition
Continuous testing of hypothesis about what words are present.
What is the mental lexicon?
Cognitive representation of word forms linked to meanings
What is the uniqueness point?
The point in the sequence of letters/sounds where we can say “that’s definitely this word” because of uniqueness to other letter/sound sequences
What is the recognition point?
“The poacher ignored the sign not to tres…”
Potential to reach recognition point before uniqueness point due to context.
However, in noisy speech/little context recognition after uniqueness point.
Studies for word recognition and sequentially- SHADOWING
Participants listen to continuous speech and repeat it back as quickly as possible.
Delay typically 250 ms.
Input speech contains mispronunciations:
similar: “trachedy”
Less similar: “travedy”
Fluent restoration: Which errors are corrected in repetition?
Restoration most likely when:
Distortion was slight (i.e. phonologically similar to original).
Word was highly predictable from context.
When distortion was in final syllable.
Informativeness of word onsets
Cohort Model of Word Recognition
Constantly trying to work out what words are, based on acoustics and prior context,
1. Access stage, listener hears /sl/ and sets up initial cohort of /sl/ words based on both acoustic input and prior context
- Selection stage, listener hears /ei/ and eliminates non-matching candidates. Listener hears /v/ and eliminates final non-matching candidate
- Integration stage
Listener uses knowledge about syntax and semantics to integrate the word into the sentence
The Revised Cohort Model has a “bottom up” priority, explain this.
Context not used to influence selection/ elimination in initial cohort.
Items that receive negative information decay rather than being eliminated
Can be reviewed by subsequent positive information
Allows for backtracking if initial acoustic information is distorted: “The dinner was very bleasant”. Whereas in linear cohort model words ‘thrown out’
- Recover should be possible, if “pleasant” best overall word candidate.
- Beginnings of words still have high, but not overriding importance.
Ongoing problem for Cohort-type models:
* Reliance on word onsets without explicit mechanism for identifying them in speech.
* …but see “Continuity and word segmentation” section.
What is “Interactive Activation Model”
Miltiple simple processing units each with variable activation level e.g. features, phonemes, words
Inhibitory connections within levels, mutual inhibition between phoneme units can account for categorical perception e.g. voiced vs voiceless boundary
Accounts for importance of word initial sounds
Excitatory connections between consistent units on different levels with higher levels affecting activation at lower levels (top down processing)
Evidence for top down processing in word recognition
Ganong effect:
VOT is primary cue to voicing stop consonants
e.g. voiced /d/ (short VOT) vs unvoiced /t/ (long VOT).
However Ganong effect occurs when a phoneme is ambiguous, lying between two possible sounds (e.g., between /d/ and /t/).
Listeners are more likely to interpret the ambiguous phoneme in a way that creates a real word rather than a non-word.
The Ganong effect demonstrates that lexical knowledge (our mental dictionary of words) can bias the perception of individual sounds
Neuropsychology of speech processing. Information is carried via the auditory pathway. Explain this
Vestibulocochlear nerve takes sounds into the brainstem. Crosses in brainstem travels up to midbrain. Passes through inferior colliculus (bilateral auditory information), to thalamus (sensory info relay station) to primary auditory cortex in superior temporal gyrus.
Primary auditory cortex (awareness of auditory stimuli) next to auditory association cortex (processes, analyses, recognises memory of sounds)
What is auditory verbal agnosia known as?
Pure word deafness or Word meaning deafness
Inability to recognise spoken words
Where is the lesion for auditory verbal agnosia?
Superior temporal gyrus, including posterior superior temporal gyrus.
How is this different to auditory agnosia?
Difficulties recognising or distinguishing SOUNDS IN GENERAL, typically associated with damage to auditory association cortex.
What is agnosia?
A modality-specific disorder of recognition.
Inability to recognise and process certain sensory info.
Therefore auditory agnosia is difficulty recognising sounds despite normal hearing ability.
Not attributable to:
Loss of primary sensory function
Inattentiveness
General mental impairment
Loss of semantic knowledge.
Identifying auditory verbal agnosia
Impairments in speech perception.
Despite intact hearing, cognitive functioning, language abilities.
Normal attentional function.
Client may say “I can hear but idk what it is”
“i don’t hear anything”
“everything sounds mumbled”
Diagnosing auditory verbal agnosia. Diagnostic criteria
Evidence of hearing as indicated by repetition
Semantic representations of words must be intact as evidenced by immediate comprehension of written words
Sometimes deficit is partial
Intervention for AVA
Compensatory strategies e.g. sign language, lipreading, written language
What is phonagnosia?
Inability to recognise familiar voices, although content is understood
What is amusia?
Impairment in processing melodies- may be associated with right-hemisphere lesions
How can we tell someone with auditory verbal aphasia is not deaf?
Discrimination of sounds e.g pa vs ba
Self-report, what can you hear?
Can they repeat words or write them to dictation?
