Phonological (un)certainty weights lexical activation Laura - - PowerPoint PPT Presentation

▶
phonological un certainty weights lexical activation
SMART_READER_LITE
LIVE PREVIEW

Phonological (un)certainty weights lexical activation Laura - - PowerPoint PPT Presentation

Oct 21, 2023 273 likes •573 views

Phonological (un)certainty weights lexical activation Laura Gwilliams , David Poeppel, Alec Marantz & Tal Linzen 7th January 2018 1 big ballet blind based on the Cohort model of spoken word recognition: Marslen-Wilson, 1978 bath baptist

slide-1
SLIDE 1

Phonological (un)certainty weights lexical activation

Laura Gwilliams, David Poeppel, Alec Marantz & Tal Linzen 7th January 2018

1

slide-2
SLIDE 2

b

bond book break band bind bath band baptist ballot black boast blind ballet back balance big

based on the Cohort model of spoken word recognition: Marslen-Wilson, 1978

slide-3
SLIDE 3

b a

band bath band baptist ballet back balance ballot

based on the Cohort model of spoken word recognition: Marslen-Wilson, 1978

slide-4
SLIDE 4

b a l

ballet ballot balance

based on the Cohort model of spoken word recognition: Marslen-Wilson, 1978

slide-5
SLIDE 5

b a l ə

ballot balance

based on the Cohort model of spoken word recognition: Marslen-Wilson, 1978

slide-6
SLIDE 6

b a l ə n

balance

based on the Cohort model of spoken word recognition: Marslen-Wilson, 1978

slide-7
SLIDE 7

Laura Gwilliams | CMCL | January 2018

But what about ambiguity?

  • Real world speech is noisy and ambiguous; there is not a

direct mapping between speech and phonemes

7

p b b p p b b p

slide-8
SLIDE 8

b

bond book pants balance bind bath palate pacify beef paddle boast prove ballet panda poke pin

p

slide-9
SLIDE 9

bond book pants balance bind bath palate pacify beef paddle boast prove ballet panda poke pin

b

bond book pants balance bind bath palate pacify beef paddle boast prove ballet panda poke pin

p b

p

slide-10
SLIDE 10

b

bond book pants balance bind bath palate pacify beef paddle boast prove ballet panda poke pin

p

bond book pants balance bind bath palate pacify beef paddle boast prove ballet panda poke pin

b p

slide-11
SLIDE 11

b

bond book pants balance bind bath palate pacify beef paddle boast prove ballet panda poke pin

p b

p

slide-12
SLIDE 12

b a

pants balance bath palate pacify paddle ballet panda

p

slide-13
SLIDE 13

b a l

balance ballet

p

palate

slide-14
SLIDE 14

b a l ə p

balance palate

slide-15
SLIDE 15

b a l ə n p

balance

slide-16
SLIDE 16

Laura Gwilliams | CMCL | January 2018

Two Computational Models

16

ACOUSTIC WEIGHTED SWITCH-BASED = phonemea = acoustic input

  • 1 cohort of words
  • binary acoustic term
  • 1+ cohort of words
  • continuous acoustic term
slide-17
SLIDE 17

Laura Gwilliams | CMCL | January 2018

Research Question

Does acoustic-phonetic uncertainty weight activation at the lexical level?

17

slide-18
SLIDE 18

Laura Gwilliams | CMCL | January 2018

Prediction aids speech comprehension

  • The brain predicts future linguistic content in terms of

phonemes, morphemes, words and syntactic structures

  • When input is predictable, it is easier to process; reflected

as a relative reduction in neural amplitude

18

x x x x x x x x x x

predictability brain response

slide-19
SLIDE 19

Laura Gwilliams | CMCL | January 2018

Quantifying predictability

19

  • Surprisal:

Probability of an outcome

  • Entropy:

Uncertainty over future input

slide-20
SLIDE 20

Laura Gwilliams | CMCL | January 2018

Critical Variables

20

  • Surprisal:

Switch-based Acoustic-weighted

  • Entropy:

Switch-based Acoustic-weighted

slide-21
SLIDE 21

Laura Gwilliams | CMCL | January 2018

barricade parricade

p b b p p b b p

Stimuli

21 2 4 6 8 10

Frequency (kHz)

barricade parricade

.95

.75

  • 140
  • 120
  • 100
  • 80
  • 60
  • 40

Power/frequency (dB/Hz)

= 1 = 0

Switch-based:

= .75 = .25

Acoustic weighted:

slide-22
SLIDE 22

Laura Gwilliams | CMCL | January 2018

Protocol

22

+

∞ ms

palate

< 2000 ms

+ +

500 ms

slide-23
SLIDE 23

Laura Gwilliams | CMCL | January 2018

Procedure & Analysis

23

x 25 time (ms) 208 sensors (1) (2) (3) averaged 200:250 ms

slide-24
SLIDE 24

Laura Gwilliams | CMCL | January 2018

Procedure & Analysis

24

slide-25
SLIDE 25

Laura Gwilliams | CMCL | January 2018

Model Setup

25

  • Critical variables:

acoustic-weighted entropy acoustic-weighted surprisal switch-based entropy switch-based surprisal

  • Control variables:

phoneme latency (ms) phoneme latency (number of phonemes) trial number block number stimulus amplitude phoneme pair ambiguity

slide-26
SLIDE 26

Laura Gwilliams | CMCL | January 2018

Results

26

2 4 6 8 2 3 4 5 6

Phoneme Position Chi−Squared

Acoustic−Weighted Switch−Based

*

n.s n.s

‘

n.s n.s n.s n.s

* ‘

slide-27
SLIDE 27

Laura Gwilliams | CMCL | January 2018

Discussion

  • Fine-grained acoustic information does weight lexical candidates
  • There is a dynamic interaction between different levels of linguistic

description: phonological <-> lexical

  • Not a single heuristic applied in all situations: perhaps reflects that

the brain commits to an interpretation of the phonological category after a certain period of time

27

2 4 6 8 2 3 4 5 6

Phoneme Position Chi−Squared

Acoustic−Weighted Switch−Based

*

n.s n.s

‘

n.s n.s n.s n.s

* ‘

slide-28
SLIDE 28

Laura Gwilliams | CMCL | January 2018

Research Answer

Acoustic-phonetic uncertainty can weight activation at the lexical level

28

slide-29
SLIDE 29

With big thanks to:

Funding: G1001 Abu Dhabi Institute

  • My supervisors, Alec Marantz and

David Poeppel, as well as everyone in the Neuroscience of Language Lab and Poeppel Lab!

laura.gwilliams@nyu.edu @GwilliamsL