Idea Transcript
ERP-evidence for laryngeal underspecification in English A mismatch negativity study Evan Bradley, Arild Hestvik, Karthik Durvasula , Catherine Bradley Experimental Psycholinguistics Lab, University of Delaware
A version of this talk was presented at the Western Conference on Linguistics 2009, Davis, CA
1
Phonemes • Phonetic categories are extracted from speech based on acoustic properties – e.g., voice onset time (VOT) – Represented by distinctive features – Many phonetic categories make up a phone
• Several (allo)phones can make up a phoneme – Abstract categories, stored in the lexicon – Auditory cortex accesses phonemic categories Phillips, Pellathy, Marantz et al. (2000)
2
Phonemes and ERP • Oddball paradigm – A standard stimulus is presented repeatedly – Activation of the representation of the standard creates a memory trace – a less frequently occurring oddball is interspersed • (e.g., t-t-t-t-t-d-t-t-t-t-t-d…)
• Mismatch Negativity (MMN) ERP component – The oddball/deviant clashes with the memory trace of the standard – Elicits a less positive deflection in the P2 auditory component (100-300ms after onset) over fronto-central electrodes Phillips, Pellathy, Marantz et al. (2000), inter alia
3
Phonemes and ERP • Modified oddball paradigm taps underlying phonological rather than acoustic/phonetic representations • Within-category acoustic variance shifts mismatch to phonological changes MMN
No MMN
Phillips, Pellathy, Marantz et al. (2000)
4
Phonological Underspecification • Recent research has shown compelling reasons to believe that in English: – Underlying representations of voiceless stops are specified for [SPREAD GLOTTIS] – Underlying representations voiced stops are not specified for laryngeal features • Ignoring all other features Iverson & Salmons (1995), Avery & Idsardi (2001), inter alia
Voiceless Stops t | [SPREAD GLOTTIS] ↓ [+ VOT]
Voiced Stops d | [ ] ↓ [0 VOT]
5
Phonological Underspecification • With respect to all other features, the two sets of sounds have the same specifications • The features of the voiceless stop are a superset of the features of the voiced stop
/d/ =
/t/ =
Iverson & Salmons (1995), Avery & Idsardi (2001), inter alia
CORONAL OBSTRUENT
… CORONAL OBSTRUENT SPREAD GLOTTIS
…
6
Underspecification and ERP • All surface features are extracted from the speech stream • Repetition of varying (within-category) standard activates minimally specified representation Phoneme A B
Acoustic features [α feature] [β feature]
Mental representation [ ] [β feature]
Eulitz & Lahiri (2004)
7
Underspecification and ERP • Surface features of [A] clash with the UR of /B/ • Surface features of [B] do not clash with the unspecified UR of /A/
Phoneme A B
Acoustic features [α feature] [β feature]
Mental representation [ ] [β feature]
Eulitz & Lahiri (2004)
8
Underspecification and ERP • A less specified deviant produces a mismatch to a more specified standard B | [β]
B | [β]
B | [β]
B | [β]
B | [β]
B | [β]
A | [α]
B | [β]
MISMATCH
Eulitz & Lahiri (2004)
9
Underspecification and ERP • A more specified deviant does not produce a mismatch to a less specified standard A | []
A | []
A | []
A | []
A | []
A | []
B | [β]
A | []
NO MISMATCH
Eulitz & Lahiri (2004)
10
Predictions • The laryngeal underspecification theory of English stops predicts that: – when a deviant [d] follows a set of standard [t]s, there will be a greater mismatch, because [d] is less specified than [t] – when a deviant [t] follows a set of standard [d]s, there will be less mismatch, because [t] is more specified than [d] Phoneme
Acoustic features
Mental representation
d
[0 VOT]
[
t
[+ VOT]
[spread glottis]
Eulitz & Lahiri (2004)
] 11
Experiment • Stimuli: – Low level Klatt speech synthesizer – 290 ms duration – varying only along VOT dimension • 5msec steps from 0 to 100 • Subjects: – 23 subjects (10 males) – age 18 to 28 (mean 23.5, sd 5.6) – normal hearing, monolingual English – 3 left-handed Phillips, Pellathy, Marantz et al. (2000)
12
Experiment • 128-channel EEG in a sound-attenuated booth • Oddball paradigm – two blocks of 700 standards and 100 deviants pseudorandom order – Standard/deviant alternated by block • [d] or [t] as deviant first – Variable ISI (mean: 903ms; SD: 80ms)
• Distractor task – Identifying voice gender of 50 interspersed ‘ba’ trials – Ensured uniformity of attention to stimulus stream
13
Experiment • Stimuli selected based on subject-specific boundary measured before EEG • Four VOT versions of each phoneme in a range of 15ms, on either side of individual discrimination boundary
/d/
/t/
14
ERP Analysis • 800ms epochs with 200ms baseline – Artifacts removed – Bad channels replaced – Baseline corrected – Average voltage reference
• Difference waves computed – Deviant-D > Standard-D – Deviant-T > Standard-T
15
Comparisons devD must be compared to stdD, (likewise for /t/), because: •Waveforms peak at different latencies due to VOT difference •Interest is in perception of a phoneme depending on its environment
16
Spatiotemporal PCA • Traditional MMN measures typically use single electrodes as dependent measures (e.g. FCz) – High-density 128 channel EEG increases spatial sampling and can better pinpoint MMN locus – Temporal and spatial Principal Component Analysis guided selection of spatial and temporal “regions of interest” in the raw data
Dien (2005)
17
1: Select time region: Temporal PCA MMN-region • A temporal PCA was conducted to identify which electrode regions “behaved the same” at different underlying temporal factors
• Temporal factor peaking at 196ms:
Inversion at mastoids
Dien (2005)
18
2: Select electrode region: Spatial PCA • Spatial PCA on the MMN temporal PCA factor: – Separates spatial subfactors of the temporal factor – Electrodes with highest spatial factor loadings in this ST factor selected for averaging in the raw data – This average then used as a single dependent measure Dien (2005)
19
Q1: Is there a MMN? Yes: 100-300ms, MMN-region
20
MMN at FCz Deviant vs Standard T
Deviant vs Standard D
peak - 0.7mV at 153ms
peak -1.4mV at 236ms
•Greater mismatch for deviant D 21
Is MMN different for /d/ vs. /t/? Not significant with time as factor
22
Is MMN different for /d/ vs. /t/? Yes With our a priori prediction of the direction of the interaction, we use the one-tailed probability p < .05
23
Block order effect • Roughly half the subjects heard /d/ as deviant in first block, /t/ as deviant in 2nd block, and vice versa • Mismatch was greater to the deviant presented in first block than in second block • Hearing 700 /d/ tokens in 1st block probably leads to habituation, attenuating effect when /d/ is deviant in 2nd block (vice versa for /t/) 24
Effect of block order on MMN for /t/ No /t/MMN when devT is in 2nd block
/t/ MMN
(Data is from averaged electrodes)
25
Effect of block order on MMN for /d/ /d/ MMN
/d/MMN (almost) disappears when devD is in 2nd block
(Data is from averaged electrodes)
26
Block order factor added to ANOVA Significant 3-way interaction: MMN is bigger for deviant-D in first block Smaller mismatch advantage for /t/ as deviant in first block
/t/ MMN
/d/ MMN
Bigger mismatch advantage for /d/ as deviant in first block Caveat: assumes /t/ and /d/ share MMN region
27
Late Negativity (LN) • Anterior distribution, 400-600ms after onset of salient sound – Unlike MMN, LN requires attention – Stimuli engage attention and cause further assessment
• Observed for devD only Čeponienė et al. (2004); Čeponienė, Cheour & Näätänen (1998) Korpilahti, Krause, Holopainen & Lang (2001); Kraus et al. (1993)
28
Late Negativity
29
Late Negativity Approaches significance, no block order interaction
Significant over 3 midline (peak) electrodes
30
Summary • SIGNIFICANT: MMN asymmetry for /d/ vs. /t/ – Deviant /d/ creates a larger mismatch – Pre-attentive processing
• TREND: Greater LN effect for /d/ than for /t/ – Salience of /d/ as deviant reaches level of awareness and catches subjects’ attention
31
Conclusions • MMN indicates an asymmetry between /t/ and /d/ as a deviant stimulus • This is in the direction predicted by the underspecification account • Suggests neurophysiological evidence for (laryngeal) underspecification in English
32
Extensions • Spanish – [d] = negative VOT, [t] = zero VOT (like English [d]) – Unlike English, Spanish /d/ is thought to be specified for [voice], while /t/ is unspecified
• Reverse MMN effect (devT > devD) predicted Phoneme
Acoustic features
Mental representation
d
[- VOT]
[voice]
t
[0 VOT]
[
] 33
References •
•
•
• •
Avery, P., & Idsardi, W.J. (2001). Laryngeal dimensions, completion and enhancement. In Hall, T.A., ed. Distinctive Feature Theory (pp. 41-70). Berlin: Walter de Gruyter. Čeponienė, R., Cheour, M., & Näätänen, R. (1998). Interstimulus interval and auditory event-related potentials in children: evidence for multiple generators. Electroencephalography and Clinical Neurophysiology 108(4), 345-354. Čeponienė, R., Lepistö, T., Soininen, M., Aronen, E., Alku, P., & Näätänen, R. (2004). Event-related potentials associated with sound discrimination versus novelty detection in children. Psychophysiology, 41(1), 130-141. Dien, J.L. (2005). PCA Toolbox. (Computer software) Eulitz, C., & Lahiri, A. (2004). Neurobiological Evidence for Abstract Phonological Representations in the Mental Lexicon during Speech Recognition. Journal of Cognitive Neuroscience, 16(4), 577-583.
•
•
•
•
•
Iverson, G. K., & Salmons, J. C. (1995). Aspiration and Laryngeal Representation in Germanic. Phonology, 12, 369-396. Korpilahti, P., Krause, C.M., Holopainen, I., & Lang, A.H. (2001). Early and Late Mismatch Negativity Elicited by Words and Speech-Like Stimuli in Children. Brain and Language 76(3), 332-339. Kraus, N., McGee, T., Micco, A., Carrell, T., Sharma, A., & Nicol, T. (1993). Something. Electroencephalography and Clinical Neurophysiology 88, 123-130. Phillips, C., Pellathy, T., Marantz, A., Yellin, E., Wexler, K., Poeppel, D., McGinnis, M., & Roberts, T. (2000). Auditory Cortex Accesses Phonological Categories: An MEG Mismatch Study. Journal of Cognitive Neuroscience, 12(6), 1038-1055. Shestakova, A., Huotilainen, M., Čeponienė, R., & Cheour, M. (2003). Event-related potentials associated with second language learning in children. Clinical Neurophysiology, 114(8), 1507-1512. 34
35