Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Evolution of an Idea

No description
by

Kyna Betancourt

on 9 October 2013

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Evolution of an Idea

But how do bilingual children process phonotactic probabilities from two languages?
Evolution of an Idea
Why can't ELL children read on the same level as their monolingual peers?
Does it have to do with vocabulary development?
I wanted to expand on Storkel (2001)
This literature base doesn't exist for bilinguals
I was going to study fast mapping in bilingual children
What do we know about phonotactic processing in bilingual children?
Bilingual infants and young children can identify legal and illegal phonotactic patterns in both languages (Sebastian-Galles & Bosch, 2002)
Bilingual children learn phonotactic probabilities for both languages, but are more proficient in L1 (Messer et al., 2010)
How do they store phonemes and phonotactic probabilities of two languages?
Is there cross-language activation?
Computational Neural Network Models
Simulate bilingual word learning and hypothesized that bilingual kids stored phonemes and phonotactic probabilities from two languages together
Suggested word processing and organization was based on phonotactic probability alone
No language nodes or task effects
These hypotheses have not been supported in kids using behavioral data
SOMBIP (Li & Farkas, 2002)
DevLex-II (Zhao & Li, 2010)
BLINCS (Shook & Marian, 2013)
Incoming Word
Phoneme Processing
Phoneme Pattern Level
Lexical Level
Neighborhood effects influence activation
Winner chosen
Phoneme sequence "memorized"
Lexical status
No lexical status
Word processed
Meaning for new phoneme sequence
Phoneme sequence
(Luce et al., 2000)
(Luce et al., 2000)
(Luce & Pisoni, 1998)
(Storkel & Morrisette, 2002)
(Zhao & LI, 2010)
(Li & Farkas, 2002)
The first step in testing this model is to investigate what is happening at the phoneme processing and pattern levels.
Research Questions:
Do bilingual children benefit from the high probability advantage during nonword processing?
Are the phoneme systems of a bilingual child's two languages stored together as one unit?
Does a bilingual child's language environment influence his/her sorting of ambiguous nonwords?
Have bilingual children sort words into Spanish and English based ONLY on phonotactic probability
Nonwords avoid linguistic influence and allow for complete control of phonotactic probability
Nonwords have been used with bilinguals to study phonotactic processing
I also wanted more information than RT and judgment.
I wanted to see how children were making decision on-line
MouseTracker (Freeman, 2011)
MouseTracker
Way to measure "fine-grained temporal components" of a decision (Freeman & Ambady, 2010, pg. 226)
Continuously records the mouse cursor while participant is making a decision
Primary analysis is spatial attraction, measured by the Area Under the Curve (AUC)
Nonword Processing in Bilingual Children: Do Phonotactics Count?
Stimuli
Stimuli Characteristics
Spondee nonwords consisting of stressed onsets and rimes from Spanish and English
All started with either the onset /k/ or an empty onset (i.e., no beginning consonant)
Hoosier Mental Lexicon used for English phonotactic probability (Nusbaum et al., 1984)
USF Spanish Frequency Lexicon used for Spanish phonotactic probability (Brea-Spahn & Frisch, submitted)
Computation of Phonotactic Probability
log( P (IO) x P (MR) x P (MO) x P (FR))
/kusar/
log(P (k) x P (u) x P (s) x P (ar))
log((.064) x (.035) x (.065) x (.006))
English Phonotactic Probability = -6.078
Ambiguous
High/Low
Language Exclusive
Consist of phonemes not present in the other language
Half the nonwords had high phonotactic probability and half had low
-6.78
-7.88
-5.88
Bell Curve for Spanish Phonotactic Probability
/usor/
-5.02
/kelfin/
-8.05
Low Probability
High Probability
Consist of phonemes probable in both English and Spanish
High English-Low Spanish
High Spanish-Low English
Spanish Phonotactic Probability
English Phonotactic Probability
Scatterplot of SIO and SMO from both Spanish and English
Consist of phonemes with same probability in English and Spanish
High Both
Low Both
Recorded on Micromic C420 Headset connected directly to Dell Computer
Used Praat software (Boersma & Weenik, 2011) to record at sampling rate of 22.05 kHz
All nonwords recorded by native, female Spanish-English bilingual and native, female English-Spanish bilingual
All nonwords embedded in Spanish carrier phrase and English carrier phrase
Each nonword then excised from carrier phrase to form Spanish and English versions
Nonword Type t-Test Comparison df t-Stat P-value

Language Exclusive English high vs. English low 14 10.55 p <.0001
Language Exclusive Spanish high vs. Spanish low 14 11.63 p <.0001
Spanish High-English Low Spanish prob vs. English prob 30 8.46 p <.0001
English High-Spanish Low English prob vs. Spanish prob 30 10.48 p <.0001
Ambiguous-High Both English prob vs. Spanish prob 30 -.747 p >.05
Ambiguous-Low Both English vs. Spanish 30 1.74 p >.05
Ambiguous High prob vs. Low prob 62 18.8 p <.0001

Stimuli Synthesis
F0 standardized at 180 Hz for all nonwords
TANDEM-STRAIGHT used to morph Spanish and English versions (from one speaker) of each word to ensure acoustic and phonetic ambiguity (Kawahara et al., 2009)
Final list of stimuli created from quasi-randomly selecting stimuli so that each nonword type had same number of stimuli from each speaker
Adult Study
Procedures
Listeners seated at computer and given head phones
Listened to nonword and judged how word-like it was
Scale of 1-7
1: Could definitely not be a word
7: Sounds very much like a word
Bilingual listeners judged nonwords on their wordlikeness in both Spanish and English
Experiment lasted roughly 15 minutes
Language Exclusive
Child Study
Results
Discussion
The High Probability Processing Advantage
Bilingual kindergartners did not show a high phonotactic probability processing advantage
Contradicts monolingual findings that high phonotactic probability =
Better fast mapping (Storkel, 2001)
More accurate repetition (Beckman & Edwards, 2000)
More fluent repetition (Munson, 2001)
Contradicts bilingual findings that high phonotactic probability =
More accurate repetition (Brea-Spahn, 2009; Lee & Gorman, 2013)
Current study tested children in bilingual mode using nonwords with mixed phonotactic probabilities
Children did not pay attention to whether or not the nonwords had a high phonotactic probability, but focused on overall phonotactic probability of the nonword
Trend shows that low probability words were judged as English
Participants
31 monolingual English speakers
30 bilingual Spanish-English speakers
Age range 19 - 42 (M=24.9 yrs)
All university students
No history of speech, language or hearing difficulties
Bilingual speakers from Columbia, Puerto Rico and Mexico
Use Spanish for social situations
English used most often during the day (75.6%)
Participants
21 bilingual kindergartners (12 females)
Aged 5-6 years (M = 5.52 years)
In kindergarten for the first time
Title 1 school
Spoke mostly Spanish at home
Received less than one year of formal schooling in English
Passed hearing screening
No speech or language concerns
Language Skills
Windsor et al. (2010) administered Spanish and English CELF-4 along with TONI-3 to differentiate between bilingual children with and without language impairment
Current study did the same, but used the TONI-P
Did not use language skills as exclusionary criteria since so few children passed
English Only Pass: 4 children
Spanish Only Pass: 4 children
Both Pass: 1 child
Did Not Pass: 12 children
Procedure
Criterion Testing
Lasted one hour
Conducted in both English and Spanish
Order of tests counterbalanced
Stickers and books used as reinforcement
When possible, no more than three weeks between two testing sessions (11/21 children tested less than three weeks later, M = 7.2 days)
Some sessions separated by more than three weeks (8/21 children tested more than three weeks later, M = 36.8 days)
Two children had sessions separated by two months
Experimental Testing
Children quasi-randomly assigned to two experimental groups based on CELF-4 results
One group spoken to in English with Spanish code-switching
One group spoken to in Spanish with English code-switching
Examiner re-introduced herself and conversed with child about his/her day at school
Child sat at computer and told he/she was going to play a game
Data Reduction
Two dependent variables
Listener Judgment
Dichotomous variable (0 = Spanish, 1 = English)
AUC
Continuous variable
Independent variables
Word Type
Probability
Language Group
Data organized into stacked format for Mixed Level Modeling (MLM)
Mixed Level Modeling
Step 1: Determine the amount of variability explained by subjects and items
Create an empty model only containing the dependent variable and the overall variance in the model
Result: restricted log likelihood, intercept, estimated residual variance
Step 2: Add the random effects for subjects
Random effects account for individual differences between children
After running this model, restricted log likelihoods between this model and the previous model are compared using Chi-Square. If they are significantly different, then the new model fits the data better
By accounting for variability of subjects, the total amount of unexplained variance is reduced
Step 3: Add the random effects for items
Random effects account for individual differences between items
After running this model, restricted log likelihoods between previous model and this model are compared using Chi-Square. If they are significantly different, then new model fits the data better
By accounting for variability of subjects AND items, the total amount of unexplained variance is reduced
Step 4: Add the predictor variables
Predictors are the independent variables
Language Group
Word Type
This model computes significance of main effects and interaction effects
Also gives the nature of the significant effects
Storage Structure of Two Phoneme Systems
The phonotactic probability systems of bilingual children's two languages are stored together
Supports hypotheses and simulations of bilingual neural network models (Li & Farkas, 2002; Shook & Marian, 2013; Zhao & Li, 2010)
English Only Spanish Only
High English-Low Spanish SAME DIFF
High Spanish-Low English DIFF SAME
Bilingual children perceive nonword, process phonotactic probability, and language with highest activation "wins"
This message is sent to the other language which becomes deactivated
If languages were not stored together, children would have to guess on language membership; no significant results
The Assimilation Effect
Hypothesizes that when a bilingual listener perceives an unknown word, he/she likely uses the language environment to determine language membership (Burki-Cohen et al., 1989)
Bilingual kindergartners did not show an Assimilation Effect
Contradicts adult bilingual findings for:
French-English bilinguals (Burki-Cohen et al., 1989)
German-English bilinguals (Lemhofer & Raddach, 2009)
Bilingual kindergartners were guessing on language membership while sorting ambiguous nonwords
They are placing much more emphasis on the overall phonotactic probability of a nonword than other cues, like language environment


A word on decision complexity
Incoming Word
Phoneme Processing
Phoneme Pattern Level
Lexical Level
Neighborhood effects influence activation
Winner chosen
Phoneme sequence "memorized"
Lexical status
No lexical status
Word processed
Meaning for new phoneme sequence
Phoneme sequence
(Luce et al., 2000)
(Luce et al., 2000)
(Luce & Pisoni, 1998)
(Storkel & Morrisette, 2002)
(Zhao & LI, 2010)
(Li & Farkas, 2002)
Incoming Word
Phoneme Processing
Phonotactic Level
Phonotactic probabilities computed
Activation of each language compared
"Winner" determined
Language membership assigned
Lexical Level
Lexical items in specific language receive majority of activation
If word match found, word processed
If word match not found, new lexical node added
Phoneme sequence
Phoneme sequence with language membership
Strengths and Limitations
Strengths
Unique stimuli
Use of MLM for data analysis
Weaknesses
Small n
Though this was managed via MLM
Language proficiency measurement
Future Directions
Collect cross-sectional data to determine how language dominance shift (Kohnert, Bates & Hernandez, 1999) correlates with phonotactic proessing
Extend findings to fast mapping by asking children to sort words and then learn their "meaning"
What do bilingual children do with newly created lexical node?
Eye tracking as a different way to study decision complexity
Clinical Implications
Since bilingual children seem to focus on overall phonotactic probability of a word to determine to which language it belongs, it is possible that starting vocabulary instruction with words composed of more frequent phoneme sequences in one language or the other would help children categorize a new word by language
Since the two phonotactic systems seem to be stored together, clinicians could use a child's L1 phonotactic knowledge to help teach L2
Special Thanks
Dr. Ruth Bahr, not only for ten years of wonderful mentorship in both research and academic life but for spending countless hours fixing an unending number of drafts.
Dr. Stefan Frisch, for working so tirelessly with me on the stimuli for this project and for lending me the idea of the robots.
Dr. Maria Brea-Spahn and Dr. Cathy McEvoy, for their support and time in making this dissertation a quality piece of work.
Dr. Michael Barker, who spent his first weeks as a new Assistant Professor in the department helping a desperate doctoral student with her statistics.
Results of predictor model for RQ1 with judgment as the dependent variable
Results of predictor model for RQ2 with judgment as the dependent variable
Results
Full transcript