ghosts in the bilingual machine : consonant cluster production in spanish and english bilinguals
DESCRIPTION
Ghosts in the bilingual machine : Consonant cluster production in Spanish and English bilinguals. Grant M. Berry The Pennsylvania State University CASPSLaP 2014. Download this Presentation: http :// goo.gl / CnxXCJ. Difficult Names…. Christopher Nwankwo Chukwuma Senator, Nigeria - PowerPoint PPT PresentationTRANSCRIPT
Ghosts in the bilingual machine: Consonant cluster production in Spanish and English bilinguals
Grant M. BerryThe Pennsylvania State University
CASPSLaP 2014
Download this Presentation:
http://goo.gl/CnxXCJ
2
Difficult Names…
• Christopher Nwankwo Chukwuma• Senator, Nigeria
• hridôe (হৃদয)• Bengali, heart
• Quvenzhané Wallis• Actress, Beasts of the Southern Wild
Introduction Participants Setup Vowels Strategy Discussion Conclusion
3
Global Differences in CC Production• When speakers encounter new words with unfamiliar consonant
clusters, they are likely to:• Epenthesize or use Anaptyxis (considered equivalent here)
(Altenberg 2005, Eddington 2001, Fleischhacker 2000, Rose and Demuth 2006, Ramírez 2006)
• Delete a member of the cluster (Jabbari and Samarvachi 2011, Chang 2008, Bonet 2006)
• Resyllabify the cluster with a nuclear sonorant (Zec 1995: 88)
Introduction Participants Setup Vowels Strategy Discussion Conclusion
mbatə
4
Some Strategies in Action:
Epenthesis
æ b ɪ n
Introduction Participants Setup Vowels Strategy Discussion Conclusion
5
Some Strategies in Action:
Resyllabification
b
m
æ t
Introduction Participants Setup Vowels Strategy Discussion Conclusion
6
Some Strategies in Action:
Deletion
b ʌ k
Introduction Participants Setup Vowels Strategy Discussion Conclusion
7
Global Differences in CC Production
• Strategies like these frequently occur with: • heterorganic clusters (Byrd and Tan 1996)
• violations in sonority sequencing (Zec 1995, Sherwin 1999, Sing 1985)
Introduction Participants Setup Vowels Strategy Discussion Conclusion
Sherwin 1999:56
Onset Nucleus Coda
8
Phonotactics in Spanish and English• The strategies used are constrained by the phonotactic systems of the
speakers in question:
• Spanish• Simple syllable structure• Few clusters permitted; many contain a liquid (pl, bl, br, pr, kr, etc.)• Coda clusters are rare
• English• Allows a wide variety of onset and coda clusters
• (sixths=[sɪksθs])• Permits liquids and nasals as syllabic nuclei (as well as other sonorants)
Introduction Participants Setup Vowels Strategy Discussion Conclusion
9
Language-specific differences in Strategy• Consequently, the strategies frequently used by speakers of these
languages are:
• Spanish• Preference for deletion of one member of the cluster (Bonet 2006)
• English• Preference for resyllabification with a liquid or nasal
• Both• Anaptyxis (the insertion of a small ghost vowel) is a common cross-linguistic
strategy which may be produced by gestural mistiming or misalignment (cf. Byrd and Tan 1996, Hall 2006) • This is reflected through lower vowel durations for ghost vowels compared to full vowels
Introduction Participants Setup Vowels Strategy Discussion Conclusion
10
Previous Research
• Most research has focused on monolingual speakers, and assumed that a given language’s phonotactics govern strategy use
• However, comparatively little work has looked at bilingual production of consonant clusters
Introduction Participants Setup Vowels Strategy Discussion Conclusion
11
Research Questions
• When one has experience with two phonotactic systems, which system’s preferred pattern predominates in consonant cluster production?• How does experience modulate the choice of strategy for consonant
cluster resolution?• Do late bilinguals and early bilinguals show similar or distinct strategy
patterns?
• Do anaptyctic vowels, which are cross-linguistically common, demonstrate evidence of gestural preparation?• Do they share characteristics with nuclear vowels in their respective words?
Introduction Participants Setup Vowels Strategy Discussion Conclusion
12
Participants (10)
• English monolinguals (3)• No significant experience with other languages• Functionally monolingual
• Late English-Spanish bilinguals (2)• Age of exposure to Spanish: 13 and 14 years• Spanish Speaking Proficiency: 9 Reading: 8
Introduction Participants Setup Vowels Strategy Discussion Conclusion
13
Participants (10)
• Late Spanish-English bilinguals (2)• Mean age of exposure to English: 13 years• Mean English Speaking Rating: 5.5 Reading: 9
• Early Spanish-English bilinguals (3)• Mean age of exposure to English: 1.6 years• Mean English Speaking Rating: 8 Reading: 8.7
Introduction Participants Setup Vowels Strategy Discussion Conclusion
14
Experimental Stimuli
• 219 Experimental Stimuli• 108 Targets
• 18 sC clusters• 63 onset clusters, 45 coda clusters• Targets were constructed to violate sonority sequencing, or to have heterorganic clusters
• 111 controls (mixture of monosyllabic and bisyllabic words)
• Construction of stimuli was mostly algorithmic• Onset Cluster: Cluster group + {a, i, u}+Consonant• Coda Cluster: Consonant +{a, i, u}+Cluster group
Introduction Participants Setup Vowels Strategy Discussion Conclusion
sC Cluster:
skagOnset Cluster:
fnarCoda Cluster:
eebnControl:
fana
15
Coding• Coding Predictors:• Strategy: Deletion, Epenthesis/Anaptyxis, Syllabic Nasal/Liquid, None, Other• Stimulus Position• Speaker group and participant number• Permissibility in English, Both, or Neither
• Analysis:• Data were analyzed in R (R Core Team, 2013)
Introduction Participants Setup Vowels Strategy Discussion Conclusion
16
Vowel Spaces and Ghost ProductionBy Group
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
Vowels were plotted with the phonR package using F1 and F2 values (McCloy 2013)
18
English Monolinguals English-Spanish Late Bilinguals
Early Spanish-English Bilinguals Late Spanish-English Bilinguals
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
19
English Monolinguals English-Spanish Late Bilinguals
Early Spanish-English Bilinguals Late Spanish-English Bilinguals
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
20Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
Vowel Length Density by Group
The ghost vowel (anaptyxis) is
consistently shortest cross-linguistically, in line with a gestural misalignment claim
21
F1 DevianceGhost F1-Main F1
Negative value implies that the
ghost vowel is higher than the main vowel
In general, we see a trend toward
centralization of the anaptyctic
vowel
This seems to be independent of
the stimulus position
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
22
F2 DevianceGhost F2-Main F2
Negative implies that the ghost vowel is more backed than
the main vowel
In general, we see a trend toward
centralization of the anaptyctic
vowel
However, æ behaves differently for coda and onset
targets. Why?
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
23
F1 and F2 deviances as functions of Euclidean distance
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
24
Factors Contributing to Overall Deviance• To determine how F2 and F1 deviance related to overall Euclidean
distance from the target vowel, an ANOVA was run on the square of the Euclidean Distance with the squares of F1 and F2 deviance as predictors, as well as the target vowel, cluster position, and speaker• Results:• F1 square deviance highly significant (p<.001)• F2 square deviance highly significant (p<.001)• Target Vowel significant (p<.02)
• /e/, /u/ and /i/ were the largest contributors to the effect• No significance for participant or cluster position
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
25
Ghost Vowel Summary• Ghost vowels are consistently much shorter than other vowels,
suggesting they result from gestural misalignment rather than full vowel insertion• Anaptyctic Vowels are usually centralized relative to target vowels,
independent of whether the language has phonological vowel reduction in unstressed syllables• This is true regardless of cluster position
• Both F1 and F2 deviance from the main vowel are strong predictors of total distance (perhaps unsurprisingly), as well as main vowel• /i/, /u/, and /e/ are driving this effect
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
26
CC Resolution StrategiesOnly done on targets, because strategies are not expected to emerge with controls
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
27
Strategy by Group(Targets only)
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
Deletion Epenthesis Other ResyllabifiedLiquid/Nasal
None
ENGLATE
Deletion is highest for late Spanish-English bilinguals
Resyllabification is also lowest for this group
English late bilinguals do not adapt deletion as a repair strategy
They do, however, lower usage of resyllabification in favor of the cross-linguistic strategy of epenthesis
Early Spanish-English bilinguals fall between English monolinguals and late English-Spanish bilinguals in terms of strategy usage
They pattern more with monolinguals than English-Spanish late bilinguals
28
Strategy By Participant(Targets only)
Participant
Individual differences are
apparent, but in general follow similar trends within groups
Spanish Early Spanish-English Late
English Monolingual English-Spanish Late
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
29
Strategy Count Summary
• English monolinguals are the highest users of resyllabified sonorants, and Spanish-English late bilinguals are the highest users of deletion
• Early Spanish-English bilinguals consistently fall in between English monolinguals and English-Spanish late bilinguals wrt strategy
• Individual differences currently abound, but they seem to be patterning similarly within groups• Larger n will most likely wash this out at the group level
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
30
Statistical AnalysisMultinomial Logistic Regression
Multinomial logistic regressions were calculated using the nnet package in R (Venables 2002)
Speaker Group Position
English Monolingual
English-Spanish
Late Bilingua
lSpanish-
English Late Bilingual
Early Spanish-English
Bilingual
Deletion
Epenthesis
Resyllabification None
OnsetCoda
Strategy
English
Neither
Permissibility
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
31
Multinomial Logistic Regression
• Run only on target items to predict repair strategy• Intercept: Spanish-English early bilingual, coda cluster permissible in
English and Spanish with no apparent strategy• Run in R (R Core Team, 2013) with the multinom() function in nnet package• Dependent Variables: • Speaker Group• Cluster Position (onset or coda)• Permissibility
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
32
Type 2 ANOVA on Model
• All 3 predictors highly significant (p<.0001)• In order of decreasing Chi-square value:• Permissibility
• Non-permissibility in both languages significantly increases the chance of a strategy emerging
• Speaker Group• Not all speaker groups behaved identically with respect to resolution strategy
• Cluster Position• Coda clusters were more likely to produce a strategy than onset clusters
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
33
Model Summary:
• Onset Clusters are less likely to produce a strategy than coda clusters• Impermissibility in both Spanish and English increases the probability that a
strategy will emerge, but the opposite is true for English-permissible clusters• Early bilinguals behave similarly to English monolinguals wrt most strategies
Introduction Participants Stimuli Vowels Strategy Discussion Conclusion
(Intercept)Late
Spanish-English English
MonolingualLate
English-SpanishEnglish-
permissibleNot permissible
in Eng/Span Onset ClusterDeletion -2.1 2.6 -0.7 -14.6 -14.7 3.5 -3.0
Epenthesis 0.6 0.3 -0.8 0.4 0.0 3.0 -2.2Other -1.3 1.1 -0.9 -0.7 0.5 2.8 -1.7
Resyllabified -1.3 0.2 -0.5 -0.3 -12.8 3.7 -2.8
34
Discussion• Though there are global tendencies for speakers of a given language to
adopt a certain repair strategy for illicit clusters, these tendencies seem to shift with additional experience in a foreign language• New strategies can develop with increased proficiency in a foreign language
• Early bilinguals, with the highest amount of experience in both languages, effectively demonstrate a hybrid phonotactic system with access to strategies common to both languages• They consistently appear between English monolinguals and late English-Spanish
bilinguals in terms of strategy use
• Ghosts vowels, when produced, demonstrate some degree of centralization reflected in changes in both F1 and F2.• This effect is modulated by vowel, with high vowels showing the greatest change in
position between ghost vowel and target (possibly due to potential outliers)
Introduction Participants Setup Vowels Strategy Discussion Conclusion
35
THANK YOU to:
• Professors/colleagues:• Marianna Nadeu• John Lipski• Matthew Carlson• Members of PHON (Phonetics/Phonology Reading Group at Penn State)
• phonpsu.blogspot.com
• Other members of the Department of Spanish, Italian, and Portuguese at Penn State• Members of the Judith Kroll’s Purple Lab and Giuli Dussias’s ISÍ lab
• Research assistants:• Hope Schmid• SeonGu Lee
Introduction Participants Setup Vowels Strategy Discussion Conclusion