Phonetic evidence on phonology-morphosyntaxinteractions: sibilant voicing in Quito Spanish

Patrycja Strycharczuka, Marijn van ’t Veerb, Martine Bruilb

and Kathrin Linkeb

aUniversity of Manchester, bLeiden University Centre for Linguistics

19th May, 2011

The data

I /s/ + pause /gas/ [gas] ‘gas’

I /s/ + consonant/plasma/ [plaz.ma] ‘plasma’/gas#noble/ [gaz.no.Ble] ‘noble gas’/rasgo/ [raz.îo] ‘feature’/gas#blaNko/ [gaz.BlaN.ko] ‘white gas’/gas#kaRo/ [gas.ka.Ro] ‘expensive gas’

I /s/ + vowel in Quito Spanish:/gasita/ [ga.si.ta] ‘gauze, dim.’/gas#akRe/ [ga.za.kRe] ‘acrid gas’

I No lexical voicing contrast in sibilants.

The problem

I The Quito dialect of Spanish displays an opaque interactionbetween sibilant voicing in codas and resyllabification.

I Thus, Prevocalic /s/ in onsets behaves opaquely.I /s/ surfaces as voiceless in word-medial onsets, e.g. in

[ga.sa]I /s/ surfaces as voiced in derived onsets, i.e. word-final

sibilants resyllabified into an onset, e.g. [ga.za.kRe]

Opacity

I No-surface true generalisation can be made for the/s/-voicing that can be expressed with a singlerule/constraint

I Syllabic generalisations;I Segmental generalisations;I Word-level generalisations.

I The problem can be solved with a stepwise derivation.

Evidence for resyllabification (1)

Native speakers’ intuitions.

Evidence for resyllabification (2)

Emphatic trilling (Harris, 1983)

I [R] undergoes optional emphatic trilling in a coda:[maR]∼[mar] ‘sea’[maR.ne.îRo]∼ [mar.ne.îRo] ‘Black Sea’

I but trilling is not possible in a word-final rhotic followed bya vowel:[ma.Re.xe.o], but *[ma.re.xe.o] ‘Aegean Sea’

Outline

Introduction

Previous analyses

Experiment

Results

Discussion

Rule ordering

I Lipski (1989): rule ordering works . . .

Lexical /gas/ /akRe/Syllabification [gas] [a.kRe]

postlexical/s/-voicing [gaz.a.kRe]resyllabification [ga.za.kRe]

output [ga.za.kRe]

I . . . but lacks explanatory adequacy

Output-Output Correspondence

I OO-constraints require identity between surface formsI In our case, [gaz akRe] must be related to some baseI Ident-OO(Voice): The output voice specification is

identical to that of its correspondent in isolation

I Preferred output: citation form (containment principle)I BUT: citation form is [gas]

Output-Output Correspondence

I Solution (Colina, 2009): surface voicing underspecificationI License[lar]: codas cannot license laryngeal features

input output/gas/ [gaS]/gas/ /akRe/ [ga.Sa.kRe]

I Hence, all sibilant voicing is phonetic, coarticulatory

Passive voicing

I Not conditioned by a voicing target.

I No specific gesture made to initiate voicing.

I Vocal folds slightly adducted.

I Coarticulatory voicing from the neighbouring segments canbe present.

I Voicing less likely before a pause than when /s/ is flankedby voiced segments.

Cyclic approach

I “A phonological process p will apply transparently in somecyclic domain”

I Cyclic domains are defined by morphosyntacticconstituents, in casu:

I StemI WordI Phrase

I Stratal OT (Bermudez-Otero, 2011)

Cyclic approach

I *z rules out [z] at the stem level

I License[lar] delaryngealizes at the word level

I Resyllabification and laryngeal respecification at the phraselevel

I no opacity between resyllabification and respecification!

I Hence voicing is phonological, assimilatory

Cyclic approach

input gas akRe, gaz akRe gasita, gazita

Stem level: *z [gas] [a.kRe] [ga.si.ta]Word level:

License(Lar) [gaS] [a.kRe] [ga.si.ta]Phrase level:

Resyllabification &Respecification [ga.z a.kRe] [ga.si.ta]

Different predictions

I Phonetic underspecification predicts that the /s/-voicing iscoarticulatory and not assimilatory.

Coarticulation Assimilation

always variable can be variablegradient categorical

Outline

Introduction

Previous analyses

Experiment

Results

Discussion

Research question

Is the /s/-voicing in Quito Spanish a categorical (phonological),or a gradient (phonetic) process?

Stimuli

I Sequences of /s/ + vowel/sonorant consonant, both withina word, and straddling a word boundary

Prevocalic Pre-sonorant

Wd-int gasita ‘gauze (dim)’ entusiasmo ‘enthusiasm’Wd-fin gas acre ‘acrid gas’ gas noble ‘noble gas’

I Six items per condition

I Embedded in a carrier phrase:Diga ‘gas acre’ otra vez.‘Say ‘acrid gas’ one more time.’

Participants

I Eight native speakers of the Quito dialect of Spanish.

I Four males aged 16-25, and four females aged 21-28.

I Naive as to the purpose of the experiment.

I Data from speaker 6 later discarded.

Speech rate test

I No comparison of derived [z] to lexical [z] possible (novoicing contrast).

I Speech rate test used instead.

I Two repetitions:I Two at a normal rateI Two at a fast rate

I Speech rate expected to affect coarticulation more thanassimilation (Sole, 1995; Cuartero Torres, 2001).

I Speech rate effects and the phonological vs. phoneticcharacter of voicing:

I In gradient voicing faster speech rate can trigger morevoicing (more coarticulation), or leave the voicing durationunaffected. Voicing ratio will change due to shorter fricativeduration.

I In categorical voicing speech rate affects the voicingduration but not the voicing ratio.

Acoustic analysis

I Acoustic analysis carried out in Praat (Boersma &Weenink, 2010)

I Measurements:I Duration of voicing during fricationI Fricative duration

Outline

Introduction

Previous analyses

Experiment

Results

Discussion

Categorical voicing – external sandhi

e s au

Diga ‘tres autores’ otra vez.‘tres autores’

No voicing word-internally

a s i

Diga ‘gasita’ otra vez.‘gasita’

Variation in the external sandhi cases

e s a

Diga ‘tres autores’ otra vez.‘tres autores’

Variation in the external sandhi cases

e s a

Two types of speakers

“All people fall into two campsThat ever twain shall be.”

I Some apply the /s/-voicing categorically (realise thesibilants with full voicing during frication), but optionally(in some tokens there is no voicing at all).

I Some display a continuum of voicing during frication.

I Optional but categorical (henceforth ‘categorical’)speakers: 1, 2, 7 and 8.

I Gradient speakers: 3, 4 and 5.

Two types of speakers

s#N, i.e. ropas negras, normal rate.

Two types of speakers

s#N, i.e. ropas negras, normal rate.

Statistical analysis

I Method:I Linear mixed effects model (Bates & Maechler, 2009) with

p-values obtained via MCMC sampling

I Data:I /s/-voicing contexts (tres autores, gas noble, esmoquin)

I Dependent variable:I Duration of voicing during frication.I Ratio of voicing duration to fricative duration.

I Predictors:I Presence of the word boundary (e.g. ‘gas noble’ vs.

‘esmoquin’);I Context (sonorant vs. vowel);I Speech rate (normal vs. fast);I Sex of the speaker.

Voicing duration in gradient speakers

Varies with a number of factors:

I Increased voicing duration in the word-internal contextthan in the context of external sandhi (p<0.05).

I Increased voicing duration in male than in female speakers(p<0.01).

I Increased voicing duration before a vowel than before asonorant (p<0.05).

I Increased voicing duration at a faster rate (p<0.05).

Voicing duration in gradient speakers. Significant effects

2030

4050

Word boundary

Voi

cing

dur

atio

n (m

s)

no yes

-

-

-

-

2040

60

Sex of the speaker

Voi

cing

dur

atio

n (m

s)

F M

- -

-

-30

4050

60

Context

Voi

cing

dur

atio

n (m

s)

sonorant vowel

--

--

2030

4050

Speech rate

Voi

cing

dur

atio

n (m

s)

Fast Normal

-

-

-

-

Voicing ratio in gradient speakers

Varies with a number of factors:

I Voicing ratio increases at the faster rate (p<0.001).

I Other predictors not significant.

Voicing ratio in gradient speakers

0.0

0.2

0.4

0.6

0.8

1.0

1.2

Speech raate

Voi

cing

ratio

Fast Normal

-

-

-

-

Voicing duration in categorical speakers

I Increased voicing duration in the normal condition(p<0.001).

I Increased voicing duration before a vowel (p=0.03).

Voicing duration in categorical speakers. Significanteffects

2030

4050

60

Context

Voi

cing

dur

atio

n (m

s)

sonorant vowel

-

-

-

-

2030

4050

60

Speech rate

Voi

cing

dur

atio

n (m

s)

Fast Normal

-

-

-

-

Voicing ratio in categorical speakers

I No significant effects on voicing ratio in the categoricalspeakers.

I Voicing ratio does not change with the speech rate(p=0.21).

No effect of speech rate on the voicing ratio in thecategorical speakers

0.0

0.2

0.4

0.6

0.8

1.0

Speech rate

Voi

cing

ratio

Fast Normal

- -

Summary

I For a subset of speakers it does, indeed, seem that the/s/-voicing is gradient:

I There is more voicing in the fast condition (morecoarticulation).

I The voicing ratio increases with the speech rate.

I However, for 4 out of 7 speakers, the voicing is optional butcategorical:

I The voicing duration is greater at the normal as opposed tothe fast rate.

I The speech rate does not affect the voicing ratio.

Approximantisation

e s au

‘tres autores’

I The categorically voicedfricatives are stronglyapproximantised.

I Easily explained if weassume that there is avoiced sibilant target.

I Puzzling as an effect ofpassive voicing.

Outline

Introduction

Previous analyses

Experiment

Results

Discussion

Problems with underspecification

I At least for a subset of speakers the /s/-voicing applies inan optional but categorical fashion, which suggests anoptional voicing target.

I Challenge to passive voicing.

I If voicing is not purely phonetic, stepwise phonologicalderivation is required.

Evidence for cyclicity?

I Our data bear out the empirical predictions made by thecyclical model, as proposed by Bermudez-Otero (2011).

I But that is not hard proof for the model.

Evidence for cyclicity?

I Our data bear out the empirical predictions made by thecyclical model, as proposed by Bermudez-Otero (2011).

I But

that is not hard proof for the model.

Evidence for cyclicity?

I Our data bear out the empirical predictions made by thecyclical model, as proposed by Bermudez-Otero (2011).

I But that is not hard proof for the model.

How could a process like /s/-voicing evolve?

1. Lack of laryngeal contrast in codas.

2. Delaryngealisation: [gas]→[gaS]

3. Passive voicing: [gaS– ]

4. Contextual reinterpretation: [gaS– #no.Ble]→[gaz.no.Ble][gaS– ]→[gas]

5. Analogical change: [ga.sa.kRe]→[ga.za.kRe]

Learning the pattern

I Syllable coda as an environment? Opacity problem.

I Prosodic word final position as an environment? Voicingalso applies where there is no PWd boundary, e.g. in losotros.

I Grammatical word final position as an environment?Potential loss of generalisation, cf. cosmologo.

Learning lexical and postlexical patterns

Is the /s/-voicing in cosmologo and in gas noble the sameprocess?

I Same degree of variation.

I Which, however, might not be that relevant to the learner.I Learning the variable /s/-voicing in gas noble: variation in

the input + abstraction at the level of a segment/feature(evidence: novel environments).

I Learning the variable /s/-voicing in cosmologo: variation inthe input makes the rule redundant.

Arbitrary generalisations

I “Voice in pre-sonorant codas and prevocalically at the endof grammatical words”

I “Voice prevocalically and before a sonorant, except inword-internal onsets”

Conclusions

I Quito /s/-voicing is a phonological process, at least forsome speakers.

I The categorical external sandhi cases constitute evidencefor a phonological generalisation.

I The opaque nature of /s/-voicing challenges monostratalmodels of phonology.

I Multistratal models fare empirically better.

I At this juncture, the choice of a model depends on theexpected degree of synchronic explanation.

I But finding out about the lexicon could help us understandthe phonology better.

Thank you!

Thanks to...

I ...the speakers for their participation in the recordings

I ...Ricardo Bermudez-Otero, Yuni Kim, and Koen Sebregtsfor comments and discussion

The first author acknowledges the support of the Arts andHumanities Research Council grant no. AH/H029141/1The other authors acknowledge the support of the NetherlandsOrganisation for Scientific Research

Speech rate

Could the two division between gradient and categoricalspeakers follow from individual speech rates?

I Speech rate calculated for every individual speaker perrepetition.

1. Using a Praat script (de Jong & Wempe, 2009): number ofsyllables (calculated based on intensity peaks) divided bytotal duration excluding pauses.

2. Standardised approach: the total number of syllables in thetext (352) divided by total duration excluding pauses.

3. No apparent trends found.

Script-based speech rate

Script-based speech rate

Speech rate based on duration comparison

Speech rate based on duration comparison

References

Bates, D. & M. Maechler (2009). lme4: Linear mixed-effects models usingS4 classes. URL http://CRAN.R-project.org/package=lme4. R packageversion 0.999375-32.

Bermudez-Otero, R. (2011). Cyclicity. van Oostendorp, M., C. Ewen,E. Hume & K. Rice (eds.), The Blackwell companion to phonology,Malden, MA: Wiley-Blackwell, vol. 4, pp. 2019–2048.

Boersma, P. & D. Weenink (2010). Praat: doing phonetics by computer[Computer programme].Version 5.1.12, retrieved 15 October 2009 fromhttp://www.praat.org/.

Cuartero Torres, N. (2001). Voicing assimilation in Catalan and English.Ph.D. thesis, Universitat Autonoma de Barcelona.

Harris, J. W. (1983). Syllable structure and stress in Spanish: a nonlinearanalysis. Cambridge, MA: MIT Press.

de Jong, N. & T. Wempe (2009). Praat script to detect syllable nuclei andmeasure speech rate automatically. Behavior research methods 41, pp.385–390.

Lipski, J. M. (1989). /s/-voicing in Ecuadoran Spanish: patterns andprinciples of consonantal modification. Lingua 79, pp. 49–71.

Sole, M.-J. (1995). New ways of analyzing sound change. speech rate effects.Belgian Journal of Linguistics 9, pp. 21–44.