Transcript Slide 1

Today
1. Misanalysis of Foreign Language Prosody
(word size, word segmentation, tone)
2. Order of Acquisition based on perceptual
salience
Selayarese
• Native Language: add copy vowel after /r,l,s/
• Loan adaptation: add copy vowel after /r,l,s/
BUT add high vowel (copy [round]) after /s/
in monosyllable
BI kípas > kípasa
BI pas > pási
Conservative Learning?
Alderete and Tesar (2002): two possible
learning paths for child who hears only
occasional antepenultimate stress:
1.Reckless: Assume all stress is lexically
marked. But then child can never retreat.
2.Conservative: Assume antepenultimate stress forms
are epenthetic.
Selayarese learners:
superconservative?
Selayarese speakers are accustomed to
alternations like
lámbere~lambér-aŋ
i.e., the combination of
antepenultimate stress plus copy V
following /r,l,s/ indicates inserted V.
But
Selayarese speakers will NEVER hear
alternations in their NL like
pása ~ pás-aŋ
(There are no monosyllabic roots in
Selayarese)
BUT
• Alderete & Tesar’s procedure could not
produce such a restrictive grammar.
• Would we even want it to?
Difference related to stress?
Perhaps the generalization is:
• Copy vowel after unstressed syllable
• High vowel after stressed syllable and
following /s/.
NL Stress Grammar
FtBin: Feet must contain two syllables.
Head-Dep (Alderete 1995): Main stress foot
may not contain an epenthetic V.
Align-Right (Word, Foot): The right edge of the
word must coincide with the right edge of a
foot.
/sahala/ ‘sea cucumber’
Head Dep
Align-Right
> a. sa(hala)
b. (saha)la
*!
/sahal/ ‘profit’
a. sa(hala)
b. (saha)la
*!
*
But with two epenthetic vowels, violating
HeadDep is unavoidable (Broselow 1999).
/kornel/ ‘corner kick’
Head Dep
> a. koro(nele)
*
b. ko(rone)le
*
Align-Right
*!
/orkes/ ‘orchestra’
> a. oro(kese)
*
b. or(oke)se
*
OR c. oro(kesi)???
*
*!
Hasan Basri (and wife)
“if I were asked to say these words in
Selayarese, without a moment hesitation
I'd definitely say: orokese, marakasa,
borokasa, and porokasa with penultimate
stress respectively.”
Word Size Restrictions?
Hypothesis:
Minimal word size constraint in Selayarese
leads listeners to hear CVC as CVCV.
Why high V after [s]?
• [s] release interpreted as high.
• More cues to quality of preceding V in
liquid than in [s].
• Needs to be tested experimentally: Can
word size restrictions trigger
misperception?
More on Words
Problem One
English speakers misanalyze Cairene Arabic
VC#V
miʃ ana ‘not I’ (VC#V)
heard as
mi ʃana (V#CV) (=nonexistent words)
English Word-Syllable Relationships
‘a name’
‘an aim’
Possible phonetic differences:
glottal stop before ‘aim’
longer /n/ in ‘name’ (Lehiste 1960)
Kahn 1975: ambisyllabicity
a. ‘a Neil’ = c. ‘anneal’
b. ‘an eel’
/n/ is solely in onset: ‘a Neil’, ‘anneal’
/n/ is ambisyllabic: ‘an eel’
(effect of constraints Onset and Align-R
(Morpheme, Syllable)
Syll
|
a
Syll
|\\
nil
Syll Syll
|\ / | \
a n i l
‘a Neil’, ‘anneal’
‘an eel’
Cross-linguistic differences in
word-syllable alignment
English
Syll Syll
|\ / | \
a n i l
Cairene Arabic (French, Spanish)
Syll Syll
| /|\
a n i l
‘an eel’
‘an eel’
Evidence of Misanalysis
Cajun music ‘zydeco’ [zajdəko]
Origin: ‘les haricots’ [lez aRiko] (Cajun French)
Cairene Arabic: VC#V syllabified as (V)(C#V)
(full resyllabification)
English: VC#V syllabified as (V(C#)V)
(partial resyllabification)
Source of Misanalysis
English listener:
hears acoustic cues signalling syllable
structure (V)(CV).
this structure is not compatible with
word structure VC#V in English.
Problem Two
English speakers misanalyze Cairene Arabic
V#CCV
binti ‘my girl’
simiina ‘heavy (fem.)’
binti s_miina ‘my heavy girl’
(/i/ deletes in the context VC_CV)
binti smiina ‘my heavy girl’ (V#CCV)
heard as
bintis miina (VC#CV)(=nonexistent words)
Cairene syllable structure
Each syllable must have an onset.
VC#V > V.C#V
phrase-initial V > [ʔV]
No complex onsets are permitted.
V#CCV > V#C.CV
(leftward resyllabification)
Compare English
night rate
Nye trait
VC#CV
V#CCV
Vt.ret
V.thret
In English, syllabification VC.CV is not
compatible with V#CCV.
Interim Summary
Selayarese interpretation of Bahasa Indonesian
monosyllables: misanalysis of CVC as CVCV,
based on NL word structure requirements
English interpretation of Cairene Arabic syllable
structure: misananalysis of V.C#V as V.#CV
and of V#C.CV as VC#.CV, based on English
requirements on word-syllable alignment.
More on Prosody:Intonation
Mandarin Tones
Tone 1 ma
Tone 2 ma
Tone 3 ma
Tone 4 ma
‘mother’
‘hemp’
‘horse’
‘scold’
Production Errors (Shen 1989)
Error Rate (NL English, 4 months of study)
Tone 4 (falling) 55.6%
Tone 1 (high) 16.7%
Tone 3 (fall-rise) 9.4%
Tone 2 (rising)
8.9%
Perception Errors
Broselow Hurtig & Ringen 1987
Error Rate, Single syllables
Tone 2 (rising) 33%
Tone 3 (fall-rise)22%
Tone 1 (high) 11%
Tone 4 (falling) 6%
BUT
Tone 4 was more accurately perceived than
other tones in single syllables.
But in strings of 2 and 3 syllables, Tone 4 had
LEAST accurate perception.
Only Tone 3 and Tone 4 had a significant
difference in perception accuracy by position.
Tone 3 has allophonic variation in Mandarin:
final position: fall-rise
nonfinal position: abbreviated rise
Similarity of Tone 4 to English
Declarative Fall Intonation
Misanalysis of Lexical Tone
Mandarin lexical Tone 4 (fall = HL) is
misanalyzed by English speakers as English
H*L% contour
H* = associated with accented syllable
L% = boundary tone associated with end
of declarative
(Pierrehumbert)
Hsieh et al. 2000, Klein et al. 2001
• Chinese speakers: tones processed in left
hemisphere
• English speakers: tones processes in right
hemisphere
• (Methodology: PET)
Wang, Jongman, Sereno, & Hirsch
Adult English speakers, beginning learners of
Chinese
Tone identification task before and after 2
weeks training on tone identification
Wang, Jongman, & Sereno
“improvements in performance were associated
with an increase in activation in Wernicke’s area
(left STG, Brodmann’s area 22) and emergence
of additional activity within adjacent regions
(right STG, Brodmann’s area 42)…findings
indicate that the early cortical effects of learning
a second language involve both the expansion of
preexisting language-related area and
recruitment of additional cortical regions,
suggesting the plasticity of the human brain in
the acquisition of Mandarin tone.”
And now for something completely
different…
Problem: ‘Hidden Rankings’ (Davidson
2001)
M>>M (differential difficulty)
Case Study 1: Jamaican Creole (Meade
2001, Ito & Mester 2001)
Basilect:
dat tik
‘that stick’
Mesolect:
dat stik
Acrolect:
ðat stik
--------------------------------------------------Not found: *ðat tik
---------------------------------------------------[st] onset easier than [ð].
Meade (2001, 46)
“A speaker whose language competence allows
her to regularly produce dental fricatives…can
be assumed to also regularly produce /s/-stop
clusters…”
WHY?
Jamaican Creole subgrammars
• Basilect: *[ð] >> *[sT]Onset >> F
• Mesolect: *[ð] >> F >>*[sT]Onset
• Acrolect: F >> *[ð] >> *[sT]Onset
• The ranking *[ð] >> *[sT]Onset (M>>M)
ensures that if /ð/ is faithfully realized,
then /sT/ is also faithfully realized. (Ito &
Mester 1995, 2001, etc.)
Subgrammars
Ito & Mester (1995, 2001): Subgrammars may
differ with respect to ranking of faithfulness
constraints, but not markedness
constraints.
Problems with ranking answer
• The analysis crucially relies on M>>M
rankings.
• Where did these rankings come from?
Rankings
• Constraint rankings are (at least to some
extent) language-specific.
• M>>F represents default.
• Some other rankings may be universally
specified for specific constraints.
Learnability Problem
• This M>>M rankings can’t be learned from
the NL/basilect data (where neither
structure is allowed).
Learnability Problem
• This M>>M rankings can’t be learned from
the NL/basilect data (where neither
structure is allowed).
• This M>>M ranking can’t be learned from
the FL/acrolect/donor language data
(where both structures are allowed).
Learnability Problem
• This M>>M rankings can’t be learned from
the NL/basilect data (where neither
structure is allowed).
• This M>>M ranking can’t be learned from
the FL/acrolect/donor language data
(where both structures are allowed).
• Is this ranking universal?
Context-free vs. Context-sensitive?
• Ito & Mester (2001, 283): “context-free
markedness, as a rule of thumb, tends to rank
higher than context-sensitive sequential
markedness. Extending the distribution of an
existing segment is often less difficult than
acquiring a new segment.”
• But…
German loans
• ‘jongleur’ > [ʒoŋglör]
• *NasalV >> * [ʒ] (Ito & Mester 2001)
context-free M >> context-free M
Japanese Lexical Strata
Most nativized:
ʃiʧibaŋku ‘Citibank’
Less nativized:
ʃitibaŋku
Least nativized:
sitibaŋku
-----------------------------------------------------------Not found:
*siʧibaŋku
------------------------------------------------------------
Subgrammars
• core:
*si >> *ti >> Faith
• middle:
*si >> Faith >> *ti
• periphery: Faith >> *si >> *ti
*[si] >> *[ti] (Ito & Mester 1995)
context-sensitive M >> context-sensitive M
• How could M>>M ranking *si >> *ti be
learned? Neither structure is allowed in core;
both are allowed in foreign vocabulary.
• Furthermore, M>>M ranking *si >> *ti cannot
be universal; Lauan Fijian has [si] but not [ti]
(Schutz 1978, Kenstowicz 2003).
Munson et al. 2007
Japanese range for acceptable sh: wider than for
English
English range for acceptable s: wider than
Japanese
Alternative Analysis
• M>>M ranking represents difference in
perceptibility of FL contrast.
Jamaican Creole Registers
• [s] is salient: [s]-[st] contrast should be easy to
perceive.
[d- ð] contrast is difficult to perceive
• Moroson and Jamieson (1989): after training,
adult Canadian Francophones successfully
distinguished [θ-ð], not [d- ð].
• Polka et.al. 2001: Anglophone babies at 10-12
months showed poor discrimination of [d- ð]
relative to other contrasts.
Subtle Cues
Polka et.al (2001): major cue is F2
•
•
•
•
F2 at onset (Hz)
F2 change (Hz)
noise duration (ms)
noise amplitude (dB)
[d]
1746
604
16.2
53.7
[ð]
1536
456
18.6
56.2
Source of M>>M
• producing /stik/ requires only reranking of
production grammar constraints.
• producing /ð/ requires reranking of
perception constraints (to form new
contrast) followed by reranking of
production grammar constraints (to
produce that contrast).
Summary
• M>>M rankings emerge from differences in
relative perceptibility.
• Choice of different vowel insertion strategies
emerges from combination of misperception
and misproduction.