Speakers knowledge of phonological universals: Evidence from nasal clusters Iris Berent Florida...

Post on 27-Mar-2015

214 views 0 download

Tags:

Transcript of Speakers knowledge of phonological universals: Evidence from nasal clusters Iris Berent Florida...

Speakers’ knowledge of phonological universals: Evidence from nasal clusters

Iris BerentFlorida Atlantic University

Tracy LennertzFlorida Atlantic University

Paul SmolenskyJohns Hopkins University

• blif

• lbif

•Challenge: What kinds of knowledge and learning mechanisms support linguistic generalizations?

Two answers

Answer 2: a specialized language acquisition system

Domain-general learning

(e.g., statistical learning) Linguistic experience:b l i f

l b i fBlif

*lbif Universal Grammar• Specialized language-

acquisition device• Universal restrictions on

language structure

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Two answers

L4:bl≻ lb

L3: bl≻ lb

L2: bl≻ lb

L1: bl≻ lb

Answer 2: a specialized language acquisition system

Domain-general learning

(e.g., statistical learning) Linguistic experience:b l i f

l b i fBlif

*lbif Prince & Smolensky (1993/2004):

*lb (marked)

Avoid lb(markednessConstraint)

Markedness constraints are universal

• all grammars ban lb

•regardless of whether bl/lb attested

Do speakers possess universal grammatical preferences for unattested structures?

lb

What’s wrong with lb?

Obstruents p,b,k,g,t,d

1

Nasals n,m 2

Liquids l,r 3

Glides y,w 4

Phonological knowledge:Sonority profile

What’s wrong with lbif?

Large rise

Obstruent-

liquid

bl 2

Small rise

Obstruent- nasal

bn 1

plateau Obstruent- Obstruent

bd 0

fall Liquid- Obstruent

lb -2H1: small sonority distances are universally marked in the grammars of all speakers

Greenberg’s typology (1978): (Berent, Steriade, Lennertz & Vaknin, 2007):

•Frequency: large rise>small rise>plateau>fall•Implications:

•Fall-->plateau•Plateau-->small rise•Small rise-->large rise

Obstruents p,b,k,g,t,d

1

Nasals n,m 2

Liquids l,r 3

Glides y,w 4

Phonological knowledge:Sonority profile

What’s wrong with lbif?

H1: small sonority distances are universally marked in the grammars of all speakers

Grammar Audition

Articulation

Statistical knowledge

Non-grammatical sources

What’s wrong with lbif?

• Source: – Are speakers equipped with grammatical

restrictions on sonority

• Scope: – Do speakers extend sonority restrictions to

unattested clusters?

Previous research (Berent, Steriade, Lennertz & Vaknin, 2007)

• Unattested obstruent sonorant sequences

• Infer markedness from perceptual illusions– Ill-formed onsets are misperceived (e.g., Pitt, 1998)– Misperception is inversely related to sonority distance

• Misperception is not due to– Phonetic failure– Statistical knowledge

• Conclusion: grammatical preference

– Misperception is inversely related to sonority profile• Lbif—>lebif• Bnif—>benif

• Misperception is not due to– Phonetic failure

• English speakers can perceive lbif accurately when attention to phonetic information is encouraged• Misperception observed also with printed materials

– Statistical knowledge• Conclusion: grammatical preference

Grammar

lb

leb

Markedness hierarchy

Blif≻ bnif ≻ bdif ≻ lbifLarge rise>small rise>plateau>fall

English

Why does the grammar favor bn?

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

UG: sonority distance

L4: rise>fall

L3: rise>fall

L2: rise>fall

L1: rise>fall

GEnglish: ob-son

ob-son

Typological evidence: Broad preference for rises

• Obstruent-initial (Greenberg, 1978, Universal 17)

• Nasal-initial (Greenberg, 1978, Universal 24)

OL ~OL

LO 12 0

~ LO 65 15

NL ~NL

LN 7 (1)

~ LN 16 66

Do English speakers extend the preference for sonority rises to

nasal-initial onsets?

Rise mlif

Fall mdif

Test: nasal onsets

Is mlif>mdif

Rationale:Infer markedness from repair

• Assume: Ill formed onsets are repaired epenthetically (e..g, Berent et al., 2007)

• If the grammar broadly disfavors falls– Then, compared to rises

• Falls should be more likely to

– undergo epenthetic repair– Be misperceived as

disyllabic

• Hypothesis:– If the grammfavor rises to falls– Falls should be more likely to be

misperceived as disyllabic

medif

*falls Faith

fall

mdif * *

medif *

Grammar

mdif

Is mdif medif?

Experiment 1syllable count

• One syllable or two?

Rises fallsmonosyllabic mlif mdifdisylalbic melif medif

Method• 12 pairs

– Labial-coronal– Coronal-labial

• Generated by splicing– Melif-->mlif

Rise Fall mlif mdif mlef mdef mlak mdak mleb mdeb mlup mdup mlek mdek nwot nbot nwik nbik nwef nbef nwog nbog nwuf nbuf nwod nbod

Rise mlif melif

fall mdif medif

•Fillers: mnif/nmifOCP manner

(Greenberg, 1978)

procedure

• Hear a word

• One syllable or two?

1

One syllable

2Two syllables

Prediction

Mdif

(falls)

Grammarmedif

Two syllables

Exp. 1: Syllable judgment of nasal clusters

mlif

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

rise fall

Proportion correct

One syllable Two syllables

mdif

melif medif

mlif

mdif—>medif

Experiment 2

• Does sonority profile affect the interpretation of ambiguous CeC sequences?

Incremental splicing

e

Time (s)0 0.66644

Time (s)0 0.66644

0

5000

Full vowel

Incremental splicing

e

Time (s)0 0.66644

Time (s)0 0.66644

0

5000

Cut 1

Incremental splicing

e

Time (s)0 0.66644

Time (s)0 0.66644

0

5000

Cut 6No vowel

task

• Hear an auditory word

• Is there an “e”?

1

yes

2

No

Prediction

• If falls trigger repair, then people should be more likely to perceive epenthesis in falls relative to rises

results

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1 2 3 4 5 6

vowel duration

proportion epenthetic responses

rise (mlif)

fall (mdif)

CCVC CeCVC

Is mdif=medif

Exp. 3: Identity judgment

Markedness of monosyllabic

form

Word 1 Word 2 Identical

* Melif Mlif no

** medif mdif No

* Mlif Mlif Yes

** mdif mdif Yes

medif

Nonidentity trials

0.6

0.65

0.7

0.75

0.8

0.85

0.9

rise fall

Proportion Correctmlif-melif

mdif-medif

Are falls represented less faithfully?

Experiment 4

• Spelling of auditory words

• Question: Is mlif spelled less accurately?

Correct spelling responses

0

0.1

0.2

0.3

0.4

0.5

0.6

rise fall

Proportion correct

Conclusion

mdif medif

•Sonority falls are encoded less faithfully than rises

•Falls undergo epenthesis

Why?

mdif medif?

Grammar*Falls

Nonlinguisticsources

Alternative explanations

• Stimulus artifacts: – Failure to remove the

epenthetic vowel

– Do Russian speakers misperceive falls?

Syllable count

(monosyllabic items)

0.6

0.7

0.8

0.9

1

rise fall

Proportion correct

one-English one-russian

Russian

English

Alternative explanations

• Stimulus artifacts: – Failure to remove the

epenthetic vowel

– Do Russian speakers misperceive falls?

• Phonetic failure

0.6

0.7

0.8

0.9

1

Rise Fall

Proportion correct

English-noniden Rusisan-noniden

Identity judgment

(nonidentity trials)

Russian

English

Alternative explanations

• Stimulus artifacts: – Failure to remove the

epenthetic vowel

– Do Russian speakers misperceive falls?

• Phonetic failure

Identity judgment (nonidentity trials)

0.6

0.7

0.8

0.9

1

Rise Fall

Proportion correct

English-noniden Rusisan-noniden

Russian

English

Phoneticanalysis

Grammarmdif—>medif

Phonetic formmdif

Phonological form (repaired)

lebif

Repair or phonetic failure?

repair

Phoneticanalysis

Grammarmdif—>medif

Phonetic formmdif

Phonetic formmdif

Phonological form (repaired)

lebif

Repair or phonetic failure?

repair

Phoneticfailure

Do markedness effects extend to printed words?

Identity judgment:

* Word1 XXXX word2: Identical?

* mdif XXXX MEDIF

100ms 500ms 2500ms

No

nonidentity trials

0.7

0.75

0.8

0.85

0.9

0.95

Rise Fall

visual

auditorymlifMELIF

mdifMEDIF

Alternative explanations

Phoneticanalysis

Linguistic knowledgeLbif—>lebif

Phonetic formLbif

Phonological form (repaired)

lebif

•UG•Stat. knowledge

Two statistical accounts

• Segment co-occurrence

Two statistical accounts

• Segment co-occurrence

• Familiarity/legality of C2– Mlif

– Mdif

– Mnif

• Prediction: mnif>mdif

Two statistical accounts

• Segment co-occurrence

• Familiarity/legality of C2– Mlif

– Mdif

– Mnif

• Prediction: mnif>mdif

Two statistical accounts

• Segment co-occurrence

• Familiarity/legality of C2– Mlif– Mdif– Mnif

• Statistical prediction: mnif>mdif

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Plateau Fall

Syllable

Spelling

Visual identity

Auditoryidentity

mnif mdif

Phoneticanalysis

GrammarPhonetic formLbif

Phonological form (repaired)

lebifStatisticallearning

Phoneticanalysis

GrammarRises>falls

Phonetic formLbif

Phonological form (repaired)

lebifStatisticallearning

Rises>falls

Some unanswered questions:

• How does the grammar constrain unattested onsets?

• Contribution of experience to grammatical knowledge:

• Experience-independent

• Inferred from experience

– How is inference obtained

– What kind of experience is necessary

– Domain- and species-specificity of learning mechanism

•Take home:–English speakers manifest broad sonority preferences that extend to unattested clusters–Consistent with hypothesis of universal markedness preferences–Source of markedness preferences remains to be seen

UGRises>falls?

Thank you!