This study examined the neural correlates of gated word recognition in bilinguals and monolinguals using EEG. Two experiments were conducted using a gating paradigm to present English words to monolingual, early bilingual, and late bilingual participants. The results showed that prosody contributed to faster word recognition times, and adding a Spanish accent to words delayed recognition for monolingual English speakers. A pilot EEG study was also conducted to investigate how prosody is processed neurologically in monolinguals versus late bilinguals during word recognition.
1. Neural Correlates of Gated
Word Recognition in
Bilinguals and Monolinguals
Derick D. Deweber, MS, CCC-SLP
Frank R. Boutsen, PhD, CCC-SLP
2009 ASHA Convention New Orleans, LA
2. Presentation Overview
• Review of models of word recognition
• Discussion of first bilingual word recognition
study
• Discussion of pilot EEG study
• Future directions
3. Word Recognition in Monolinguals
• Methods used to examine spoken word
recognition
• Findings from studies on word recognition
• Models of spoken word recognition
4. Experimental Methods
• Lexical decision
• Subject required to make decision based on stimulus; could
be affected by word status, word frequency, and context
• Priming
• Subject makes decision following two words that may be
related or unrelated (i.e., semantic priming) or may inform
subject which language to use
• Shadowing
• Subject required to repeat back word as they hear it; words
may contain semantically unpredictable errors
• Gating
• Subject makes decision about word after hearing increasing
increments of a target word
5. Experimental findings
• Word frequency effects
• More frequent words recognized faster than non-
frequent words
• Word supremacy effects
• Target recognized more easily when it is a word;
non-words less easily recognized
• Context effects
• Words are more easily recognized in context than
when they occur alone
• Distortion effects
• Words are more easily recognized when distorted at
the end than if distorted at the front
6. Models of Word Recognition
• Initial Cohort Model
• Neighborhood Activation Model (NAM)
7. Initial Cohort Model
• Marslen-Wilson, 1990
• As initial phonemes of a word are heard, all words in the
lexicon that share these phonemes are activated
• Words that share phonemes and that are
simultaneously activated are called a “cohort”
• As more of the spoken word unfolds over time the
relative size of the word’s cohort diminishes
• This process unfolds until the target word is recognized
8. Neighborhood Activation Model (NAM)
• Luce, Pisoni, & Goldinger, 1990
• “Goodness of fit” model where word is identified
based on fit of stimulus and features of
alternatives (Connine, Blasko, & Titone, 1993)
• Density of word’s “neighborhood” is based on
shared phonemes of word and other words as
well as the frequency of the words neighbors
Frequency -
• Does not account for prosodic sensitivity
9. Gating Paradigm in WR Research
• Wingfield et al., 1997
• Found that when a word is constrained by stress,
cohort size is a better predictor of when the word will
be recognized than when stress pattern is ignored
• Lindfield, Wingfield, and Goodglass, 1999
• Found that words were correctly recognized with
much less segmental (initial cohort) onset
information when word prosody was available to
subjects
10. Experiment 1
• Application of gating paradigm established by Lindfield,
Wingfield, and Goodglass, 1999
• the effect of prosody on word recognition in adult
speakers of English,
early bilingual (Spanish/English)
late bilingual (Spanish/English)
• How does English “accent” affect word recognition
scores?
11. Participants
Group 1 Group 2 Group 3
English Early Bilinguals Late Bilinguals
Monolinguals
Age 40.2 years 32.5 years 37.8 years (8.80
(12.57 SD) (15.39 SD) SD)
Gender 4 M - 6F 1M - 7F 1M - 9F
Age Began 6.5 years (1.60 29 years (5.68
Speaking From Birth SD) SD)
English
Time Speaking From Birth 26 years (14.83 11 years (6.83
English SD) SD)
12. Materials
• 30 English words
• 12 two-syllable
•
6 with stress on 1st syllable
•
6 with stress on 2nd syllable
•
Ballet, diamond, dolphin, penguin, pumpkin, sparrow,
window, cartoon, exhaust, guitar, quartet, hotel
• 18 three-syllable
•
6 with stress on 1st syllable
•
6 with stress on 2nd syllable
•
6 with stress on 3rd syllable
•
Celery, photograph, pineapple, radio, stadium, telescope,
apartment, cathedral, foundation, horizon, umbrella,
suspender, chandelier, engineer, gasoline, kangaroo,
referee, violin
13. • All words are common nouns with a mean
frequency of 37 words per million in print
(Francis & Kucera, 1982)
• Spoken with carrier phrase “The word is” with
normal intonation by a female speaker of
American English
• Words were digitized and later manipulated in
Audacity and Praat sound software applications
• Target words began at 50 ms gate size and
increased incrementally by 50 ms
14. • Gated stimuli were presented in 3 conditions
1. Onset only
2. Onset plus duration
3. Onset plus prosody
15. Onset Only
50 ms gate
100 ms gate
150 ms gate
200 ms gate
250 ms gate
300 ms gate
350 ms gate
400 ms gate
450 ms gate
500 ms gate
16. Onset plus Duration
50 ms gate
100 ms gate
150 ms gate
200 ms gate
250 ms gate
300 ms gate
350 ms gate
400 ms gate
450 ms gate
500 ms gate
17. Onset Plus Prosody
50 ms gate
100 ms gate
150 ms gate
200 ms gate
250 ms gate
300 ms gate
350 ms gate
400 ms gate
450 ms gate
500 ms gate
18. Procedure
• Each subject heard all 30 words (10 from each
condition)
• Words presented in increasing increments of 50
ms beginning with the first presentation at 50 ms
• Procedure continued until subject correctly
identified word
• Words were blocked by condition, with
conditions being counter-balanced across
subjects
19. Results
Group Onset Only Onset + Onset +
Duration Prosody
English 323.5 359 (163.36) 283 (158.94)
Monolinguals (112.02)
Early 349.38 370.63 360.63
Bilinguals (113.21) (164.72) (158.85)
Late 415.5 402.5 412.9
Bilinguals (121.98) (139.33) (133.39)
23. Experiment 2
• Same paradigm as experiment 1
• Speaker with Spanish accent
• How does Spanish “accented” English affect
word recognition scores?
24. Participants
Group 1 Group 2 Group 3
English Early Late
Monolinguals Bilinguals Bilinguals
Age 40.2 years 32.5 years 37.8 years
(12.57 SD) (15.39 SD) (8.80 SD)
Gender 4 M - 6F 1M - 7F 1M - 9F
Age Began 6.5 years 29 years
Speaking From Birth (1.60 SD) (5.68 SD)
English
Time From Birth 26 years 11 years
Speaking (14.83 SD) (6.83 SD)
English
25. Materials
• 30 English words
• 12 two-syllable
•
6 with stress on 1st syllable
•
6 with stress on 2nd syllable
•
Atom, chicken, cotton, football, harbor, magic, award, debate,
expense, remark, routine, affair
• 18 three-syllable
•
6 with stress on 1st syllable
•
6 with stress on 2nd syllable
•
6 with stress on 3rd syllable
•
Accident, artery, camera, funeral institute, graduate, assistant,
composer, dependent, detective, formation, illusion, magazine,
questionnaire, cigarette, personnel, volunteer, refugee
26. • All words are common nouns with a mean
frequency of 36.43 words per million in print
(Francis & Kucera, 1982)
• Spoken with carrier phrase “The word is” with
normal intonation by a female speaker of
American English with a Spanish accent
• Words were digitized and later manipulated in
Audacity and Praat sound software applications
• Target words began at 50 ms gate size and
increased incrementally by 50 ms
27. Procedure
• Each subject heard all 30 words (10 from each
condition)
• Words presented in increasing increments of 50
ms beginning with the first presentation at 50 ms
• Procedure continued until subject correctly
identified word
• Words were blocked by condition, with
conditions counterbalanced across subjects
28. • Gated stimuli in Experiment 2 were also
presented in 3 conditions
1. Onset only
2. Onset plus duration
3. Onset plus prosody
29. Onset Plus Prosody
50 ms gate
100 ms gate
150 ms gate
200 ms gate
250 ms gate
300 ms gate
350 ms gate
400 ms gate
450 ms gate
500 ms gate
30. Results
Group Onset Only Onset + Onset +
Duration Prosody
English 447 (119.73) 465.5 398 (127.51)
Monolinguals (140.47)
Early 363.75 394.38 338.75
Bilinguals (95.80) (121.43) (112.22)
Late 360 (111.24) 387 (130.77) 342 (113.87)
Bilinguals
34. Discussion
• Prosody contributes to the recognition of spoken words
• Experiment 1: Monolingual vs. Late (t=1.9797, df=18, p=.0632)
• Experiment 2: Monolingual vs. Late (t=1.0359, df=18, p=.3140)
• Addition of Spanish-accent to spoken English words delays word
recognition in native speakers of English
• Prosodic mismatch? Not unlike Phonological Mismatch (Imai,
Flege, and Walley, 2003)
• Degree of fluency in second language likely affects how words
are stored and accessed
• Age of SLA also likely contributor to success in word recognition
35. Pilot Study
• Application of gating procedure to EEG
paradigm
• To investigate functional differences that may
exist been monolinguals and late bilinguals in
terms of how and when prosody is used in the
brain to assist with word recognition
36. Participants
• 7 English Monolinguals (from in and around greater
OKC area)
• 7 Spanish-English Late Bilinguals (originally from
Mexico; began learning English after the age of 18)
• All participants reported to have no know neurologic
impairment, speech and/or hearing difficulties, or other
confounding medical condition
37. Materials
• 50 two syllable and 50 three syllable words
• All words are common nouns with a mean frequency of
38 words per million in print (Francis & Kucera, 1982)
• Words were digitized and later manipulated in Audacity
and Praat sound software applications
• Words filtered at 325 hz and gated with prosodic
information only
• Target words began at 100 ms gate size and increased
incrementally by 300 ms
38. Procedure
• Participant is connected to Nicolet version 5.3
EEG analysis unit according to international 10-
20 system with waves collected at 19 electrode
sites (fp1, fp2, f7, f8, t3, t4, t5, t6, 01, 02, f3, f4,
c3, c4, p3, p4, fz, cz, pz)
• Participants instructed to minimize facial
movements, eye blinks, swallows, etc
39. • Each participant hears all five gates (100, 150,
200, 250, 300 ms) of each of the 100 target
words
• Participants are encouraged to only respond to
gate only when they feel that they know what the
word is
• Example:
100 150 200 250 300
40. Early Behavioral Results
Group Mean Gate Size
English 238.86 (55.5)
Monolingual
Spanish-English 226 (49.71)
Late Bilingual
41. 14
12
10
8 Group 1
6 Group 2
4
2
0
100 150 200 250 300
44. Next Steps
• Deal with obstacles to this paradigm
• Further data analysis
• Possibly investigate word frequency effects
and/or accent effects on word recognition
between the two groups (in terms of event
related potentials)
45. For more information
OUHSC Department of
Communication Sciences and
Disorders
1200 N Stonewall
Oklahoma City, OK 73126
(405) 271-4214
Derick-Deweber@ouhsc.edu
Frank-Boutsen@ouhsc.edu