Cog models/ neural basis of speech perception

Question 1

Motor theory of speech perception
What are the 2 components of the theory?

Answer 1

1. Speech perception is the result of a specialised speech module that operates separately from the mechanisms involved in perceiving non-speech sounds and is uniquely human (“speech is special”)- speech is somehow special

Evidence: Speech (but not other sounds) are perceived categorically

2. The objects of speech perception are intended articulatory events rather acoustic events

Evidence: Speech sounds are highly variable (articulations less so). Lip movements are more stable than acoustics

Question 2

Perceiving intended articulatory events

Answer 2

If you hear a sound- comes through ears, relayed to brain. This is acoustic info

Auditory cortex recognising acoustic cues

According to Lieberman because acoustics is so variable, he’s saying that the sound you’re getting, the brain is trying to latch onto this to perceive the underlying gestures to perceive these sounds. Rather than the acoustic consequences of the speech sounds.

Question 3

What method can we use which might offer support for speech perception?

Answer 3

fMRI

Question 4

Evidence for the motor theory of speech perception
- What does passive listening to meaningless monosyllables activate?
- what other areas are also activated?

Answer 4

Passive listening to meaningless monosyllables activates auditory cortex (region that is normally related to sound processing)

But motor (activate if you make a movement) and premotor (activate if you plan t make a movement) areas are also activated

Question 5

What is TMS and what does it involve?

Answer 5

Transcranial Magnetic Stimulation

Use a coil that makes magnetic pulses over specific brain regions. This will disrupt activity in the underlying brain regions.

Question 6

Evidence for the motor theory of speech perception
Procedure and result

Answer 6

Procedure:
Asking pps to do speech discrimination task (hearing pairs of speech sounds and have to discriminate between sounds- are they different or the same). Also have control task where they have to do a visual task (nothing to do with sound or speech)

Results:
TMS over premotor areas interferes with phoneme discrimination in noise but not colour discrimination

Have a selective disruption of speech when TMS is applied over motor areas but not when it’s applied over control.

Evidence that motor areas are causally involved in speech perception

Question 7

Evidence against the motor theory of speech perception

Answer 7

Categorical perception can also be demonstrated for non-speech sounds
- So not the result of a specialised speech module

With training, chinchillas (rodents) shows the same phoneme boundary for a /da/-/ta/ continuum as humans

Steep change in identification function when you’re near phoneme boundary
- So not unique to humans

Question 8

Motor theory of speech perception
Two components of the theory
Are the theories correct?

Answer 8

X - This aspect of the motor theory isn’t quite right- its not a result of the mechanism

1. Speech perception is the result of a specialised speech module that operates separately from the mechanisms involved in perceiving non-speech sounds and is uniquely human (“speech is special”)
   - Evidence: Speech (but not other sounds) are perceived categorically

? - more evidence of this aspect- it does seem that processing gestures is involved in speech perception. It’s about how extreme you take this on.

2. The objects of speech perception are intended articulatory events rather acoustic events
   - Evidence: Speech sounds are highly variable (articulations less so)

Question 9

Brain basis of speech perception: Classic model

Answer 9

• Superior temporal gyrus for speech perception (Wernicke’s area- posterior section of superior temporal gyrus)
• Inferior frontal gyrus for speech production (Broca’s area)
• Left hemisphere dominant

Question 10

What are the two streams for speech processing that are engaged in a task-dependent manner

Answer 10

Dorsal and Ventral

Question 11

Brain basis of speech perception:
Dorsal stream

Answer 11

Dorsal stream for mapping speech sounds onto articulatory representations
- activated for tasks focussing on perception of speech sounds (e.g. phoneme discrimination)
- Left hemisphere dominant
- Broca’s area part of the dorsal stream but also involved in perception (not just production)

Question 12

Brain basis of speech perception:
Ventral stream

Answer 12

Ventral stream for mapping speech sounds onto lexical representations
- Activated for tasks focussing on comprehension (e.g. word recognition)
- Bilateral (left and right hemispheres)

Question 13

Brain basis of speech perception:
What does it explain?

Answer 13

Explains why some aphasics can’t tell apart phonemes but can recognise words (and vice versa)

This model can explain counterintuitive finding- damage to dorsal stream but ventral stream is intact. For other patients, they will have ventral stream damages.

Question 14

Evidence for ventral stream processing

Answer 14

1. Williams, Nestor and Hodges
   Example of semantic task: Pyramids and Palm Trees test. Present picture of pyramid. Choose between picture of palm tree or foil (other type of tree).

Anterior temporal damage associated with semantic impairment (have trouble choosing correct picture)

2. Bates et al.
   Inferior temporal damage associated with comprehension deficits

Question 15

Evidence for dorsal stream processing

Answer 15

1. Wilson et al.
   Listening to syllables activates motor and premotor areas
2. Meister et al.
   TMS ver premotor areas interferes with phoneme discrimination in noise but not colour discrimination

Question 16

Cohort model:
Explain the process of recognising spoken words

Answer 16

You have a set of word representations in your mind

You have a memory trace if what different words should sound like in your head

As you’re hearing speech (not even the whole word) the speech input you’re getting activates words signal in your lexicon that you’ve heard so far.

If it starts with c- it will activate all words starting with this

As time goes on, it will result in a more selective activation

cohert- set of words that respond to speech signal

With cath- many words before can be ruled out

This will play out until moments of time in uniqueness point

Question 17

Cohort model:
Uniqueness point (UP) and key features

Answer 17

Uniqueness point: Time-point in the speech signal when only one word becomes consistent with the speech input

Word is recognised at the UP, even before the whole word has been heard (‘optimal efficiency’)- instead of waiting for the whole word to be heard, the mind will only use what it needs to recognise the word. It will recognise cathedral at this point in time even before hearing the whole word.

Key features:
- Words are activated immediately upon minimal input
- Multiple words are activated
- Words compete for recognition (‘lexical competition’)

Question 18

Cohort model: Evidence from shadowing task

Answer 18

Listeners presented with speech and they have to shadow what they’re hearing as soon as they can

Histogram showing response latencies

• Average response latency is approx. 250 ms
• Average duration of words is 375 ms
• So implies that listeners recognise words even before they have heard the ends of the words
• Consistent with Cohort model

Average duration of words 375 ms. Also point out that it would take around 50-100 ms to make response.

Question 19

What does learning new words slow down?

Answer 19

Recognition of existing words

Question 20

Cohort model: Shortcoming

Answer 20

• The Cohort model is still very influential in the field
• However, it is an example of a ‘verbal model’, which makes it difficult to evaluate
• A better way of being sure what a model or theory predicts, is to implement it as a computer program (‘computational model’)

Average duration of words 375 ms. Also point out that it would take around 50-100 ms to make response.

Question 21

TRACE model of speech perception

Within-layer inhibitory connections vs Bi-directional excitatory connections

Answer 21

Within-layer inhibitory connections- influencing lexical connections- if you have cat activated, this will inhibit other words like bad and bat.

Bi-directional excitatory connections- link between different levels of processing - link between the word stage and phoneme stage

Question 22

TRACE model of speech perception

Answer 22

Different brain areas and stages of processing

Can peak in each of these stages- have neuron like inputs (one responding to b, k ect and a word part of the model)

If you have activation of acoustic features that are consistent with book for example this activation will travel up and excite the neuron thats consistent with codes for b and this will activate words that start with b

In TRACE, also have activations going the other way- if you have activation of a particular word such as bad, for these top down connections, this will enhance activity for phonemes that comprise bad such as ‘b’ and ‘a’

Question 23

TRACE model: Evidence from eyetracking

Answer 23

Camera monitors where pp’s are looking on the screen

They hear a spoken instruction eg. take the beaker and put it on the black circle

This method of eye tracking is a way to see the time course of speech processing

By using eye tracking, you can look at the more intermediate fine grained processes that lead to the final behavioural outcome

Question 24

Eye tracking data and TRACE model results

Answer 24

Initially they’re not looking at any object but as more of the spoken word is revealed, they start to look at the beaker.

Looking at words that start with the same phoneme or that rhythme/ are unrelated

TRACE- Present same stimuli as human pp’s to model and look at activation in word level and compare data to human pp’s data.

The activations and trace closely respond with eye tracking data suggesting that TRACE is doing well at modelling the dynamics of spoken word recognition.

Question 25

TRACE model: Explaining context effects

Answer 25

Lexical context influences how you perceive ambiguous speech sounds.

Question 26

TRACE’s explanation of Ganong effect

Answer 26

• Word-level activity ‘feeds back’ to influence lower-level phoneme activity
• although this explanation has been strongly challenged

For phonemes and because they’re ambiguous they might start off equally activated

• ift through top down connection will bias activity fot that g phoneme unit
• iss will bias activity in the k unit

This is what affects perception of speech contexts, they are influenced by higher level word context