Aphonologicallyweakcontrastcaninducephone5coverlap

MargaretE.L.Renwick1,IoanaVasilescu2,CamilleDutrey2,LoriLamel2,BiancaVieru2,3

1TheUniversityofGeorgia,USA,mrenwick@uga.edu2LIMSI/CNRS,France3VocapiaResearch,France

TheRomanianvowelsystemisuniqueamongtheRomancelanguages,inpar5cularforitsphonemiccentralvowelsandrarediphthongs(Chitoran2002)

MarginalcontrastinRomanian(Renwick2014)²  Central/ɨ,ʌ/arehistoricalallophonesinnear-complementarydistribu5on²  /ɨ/:Typically(90%)foundinpre-nasal,stressedcontexts;neverpost-tonic

[ˈkɨmp] ‘field’ [ˈlɨna] ‘wool (def.)’ [ˈvɨna] ‘vein (def.)’²  /ʌ/:Usuallyunstressed,word-final(46%);hasalargemorphologicalrole

[ˈkasʌ] ‘house’ [ˈsutʌ] ‘hundred’ [pʌˈtuts] ‘bed (dim.)’ ²  Fewminimalpairsexist

[rɨw] ‘river’ [rʌw] ‘bad’ [vɨr] ‘I thrust’ [vʌr] ‘cousin’ [tsɨrj] ‘sea mackerels’ [tsʌrj] ‘lands (n.)’

²  Both/ɨ,ʌ/havelowtypefrequency,sugges5nglowfunc5onalload²  Itwashypothesizedthatthephonologicallyweakcontrastbetween/ɨ,ʌ/

wassubjecttophone5cmerger;however,studiesofproduc5onandpercep5oninlaboratoryspeechfoundli`leevidenceforthis.

ReferencesChitoran,I.(2002).ThePhonologyofRomanian:AConstraint-BasedApproach.Berlin;NewYork:MoutondeGruyter.Fougeron,C.,&Audibert,N.(2011).Tes5ngvariousmetricsforthedescrip5onofvoweldistor5onindysarthria.InProceedingsofthe17th

Interna>onalCongressofPhone>cSciences(pp.1–4).HongKong.Gendrot,C.,&Adda-Decker,M.(2005).ImpactofDura5ononF1/F2formantvaluesoforalvowels:anautoma5canalysisoflargebroadcastnews

corporainFrenchandGerman.InProceedingsofEurospeech(pp.2453–2456).Lisbon.Hall,K.C.(2013).Atypologyofintermediatephonologicalrela5onships.TheLinguis>cReview,30(2),215–275.Hall,K.C.,Allen,B.,Fry,M.,Mackie,S.,&McAuliffe,M.(2015).PhonologicalCorpusTools,Version1.1.0.[Computerprogram].

Retrievedfromh`ps://github.com/PhonologicalCorpusTools/CorpusTools/releasesHocke`,C.F.(1966).Thequan>fica>onoffunc>onalload:Alinguis>cproblem(MemorandumNo.RM-5168-PR).Lobanov,B.M.(1971).Classifica5onofRussianvowelsspokenbydifferentspeakers.TheJournaloftheAcous>calSocietyofAmerica,49(2B),606–

608.

Mahalanobis,P.C.(1936).Onthegeneralizeddistanceinsta5s5cs.ProceedingsoftheNa>onalIns>tuteofSciences(CalcuNa),2,49–55.Nadeu,M.,&Renwick,M.E.L.(2016).Varia5oninthelexicaldistribu5onandimplementa5onofphone5callysimilarphonemesinCatalan.Journal

ofPhone>cs,58,22–47.h`p://doi.org/10.1016/j.wocn.2016.05.003Renwick,M.E.L.(2014).ThePhone>csandPhonologyofContrast:TheCaseoftheRomanianVowelSystem.Berlin,Boston:DeGruyterMouton.Vasilescu,I.,Vieru,B.,&Lamel,L.(2014).ExploringPronuncia5onVariantsforRomanianSpeech-to-TextTranscrip5ons.InProceedingsofSLTU.St.

Petersburg,Russia.Wedel,A.,Jackson,S.,&Kaplan,A.(2013).Func5onalloadandthelexicon:Evidencethatsyntac5ccategoryandfrequencyrela5onshipsin

minimallemmapairspredictthelossofphonemecontrastsinlanguagechange.LanguageandSpeech,56(3),395–417.

ResearchQues6on:Howisthephonologicallyweakcontrastbetween/ɨ,ʌ/realizedincon5nuousspeech?

DataandMethodsAcous>csof7monophthongscomparedacrosslaboratoryvs.broadcastspeech.Laboratoryspeech(Renwick2014)²  Stressedandunstressedvowels;targetwordsinaframesentencewith3repe55ons²  18na5vespeakers(3male)²  Formantvalues(F1,F2)extractedatmidpoint;hand-checked²  5,261tokens(2,396centralvowels)

Broadcastspeech(Vasilescu,Vieru&Lamel2014)²  Mixtureofpreparedspeechfromnewsshows,andsemi-spontaneousdebatesfrom

TVchannelAntena,inthestandardSoutherndialect²  7hours,86speakers(male♀adult)²  Segmentboundariesautoma5callyaligned²  Formantvaluesextractedatmidpoint(Gendrot&Adda-Decker2005)²  Acous5cfilteringofdata

²  Tokenswithvoicingin<40%ofvowelwereexcluded²  Eachtoken’sMahalanobisdistance(Mahalanobis1936)wascalculatedrela5ve

toaspeaker-andvowel-specificcentroid;tokenswithhighdistance(basedonaχ2distribu5on)wereexcludedasoutliers

²  Voweltokensanalyzed:104,456(11,006centralvowels)²  Vowels’func5onalloadcalculatedfrom9,032uniquewords(Halletal.2015)

Normaliza5onandanalysis²  Formantvaluesnormalizedbyspeaker(Lobanov1971)²  Acous5coverlapinstandarddevia5onsofF1calculatedamongadjacentvowelpairs

(Fougeron&Audibert2011)

Results:vowelfrequencyandfunc6onalloadinbroadcastspeech²  [ɨ,ʌ]appearinmanyfunc5onwords,butarerare,withcomplementarydistribu5ons

Discussion² Thecontrastbetween[ɨ,ʌ]isseverelydiminishedincon5nuousspeech

²  [ɨ]islower(higherF1)inbroadcastthaninlaboratoryspeech² Greatercentralvoweloverlapoccursinbroadcastspeech² Centraliza5onisnotabyproductofreduc5on(viashortening)

² Func5onalloadofthe/ɨ,ʌ/contrastislowestofallvowelpairs

² Speakers’cogni5verepresenta5onsof/ɨ,ʌ/maynotbeseparablefromcontext² Mergerisstrongestpre-nasally,where/ɨ/isstronglycondi5oned² Underlyingvowelqualityhighlycondi5onedbymorphologyandphonology² Futureperceptualstudieswillindicatelistenersensi5vitytovowelquality,

independentofphonologicalcontext

Context [ɨ] [ʌ] CVC 30.5%[romɨˈnia] ‘Romania’ 26.6%[ˈastʌzj] ‘today’ #VC 67.2%[ɨn] ‘in’ 0.4%[ˈʌsta] ‘this one’ CV# 0.0%Ø 65.8%[sʌ] ‘that (conj.)’ Other 2.3%[ˈmɨ̯ine] ‘tomorrow’ 7.2%[sʌw] ‘his’ Total 100%   100%

Phonologicalcontras6venessanditsconsequences²  Soundsseparatedbyevenoneminimalpairareconsideredcontras5vephonemes.

Butwhatiftheirdistribu5onismostlypredictable,andnearlycomplementary?Dospeakersandlistenerstreatweak,marginalcontrastsdifferentlyfromothers?

²  Characteris5csofmarginalcontrastNearorpar5alpredictability,phone5cvariability,frequencyimbalance,scarcityofminimalpairs,phone5coverlap(Hall2013,Nadeu&Renwick2016)

²  Func5onalloadHowmuchneutraliza5onwouldresultfromthelossofacontrast?(Hocke`1966)Moreminimalpairs=higherfunc5onalload

²  Languagechangeover5meContrastswithlowfunc5onalloadaremorelikelytomerge(Wedeletal.2013)

/i/ /ɨ/ /u/

/e/ /ʌ/ /o/

/e̯a/ /a/ /o̯a/

Func6onalload,calculatedviatypefrequency,usingchangeinentropyalgorithm

[e] [i] [o] [u] [ɨ] [ʌ][a] 0.0428 0.0143 0.0071 0.0135 0.0025 0.0456

[e] 0.0257 0.0032 0.0107 0.0011 0.0545

[i] 0.0018 0.0061 0.0007 0.0078

[o] 0.0025 0.0011 0.0018

[u] 0.0032 0.0036

[ɨ] 0.0004

22.6%

28%

18.2%

3.3%

9.8%10.9%

7.2%

23.2%

28.7%

16.8%

3.3%

9.9%10.8%

7.2%

All data (N = 125,501)

Filtered data (N = 104,456)

0

10000

20000

30000

0

10000

20000

30000

a e i � o u �

a e i � o u �Vowel

Toke

n co

unt

Vowel frequency

Results:acous6csoflabspeechvs.broadcastspeech

Lab Broadcast

-100

0

ea ie �� oa uo �a ea ie �� oa uo �aAdjacent vowel pair

Mea

n F1

ove

rlap

(Hz)

paireaie��oauo�a

Lab Broadcast

-1

0

1

-1012 -1012Z-scored F2

Z-sc

ored

F1 vowel

aei�ou�

Female speakersLab Broadcast

-1

0

1

-101 -101Z-scored F2

Z-sc

ored

F1 vowel

aei�ou�

Male speakers

Lab Broadcast

-100

0

ea ie �� oa uo �a ea ie �� oa uo �aAdjacent vowel pair

Mea

n F1

ove

rlap

(Hz)

paireaie��oauo�a

Inbroadcastspeech,[ɨ]isheavilycentralized

Thecentralvowels(and[u,o])overlapintheF1dimension,inbroadcastspeech

Does[ɨ]centralizeonlyinshorttokens?No.Correla5onofZ(F1)withdura5onisnotsignificant(p>0.05).

Fricative Liquid Nasal Stop-2

-1

0

1

2

-101 -101 -101 -101Z-scored F2

Z-sc

ored

F1

vowel � �

Following manner in broadcast speech

Fricative Liquid Nasal Stop-2

-1

0

1

-2-1012 -2-1012 -2-1012 -2-1012Z-scored F2

Z-sc

ored

F1

vowel � �

Following manner in broadcast speech

Inbroadcastspeech,[ɨ]centralizesmostinpre-nasalcontexts

Lab Broadcast-2

-1

0

1

2

0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4Vowel Duration (sec)

Vow

el F

1 (Z

-sco

red)

vowel��

Lab Broadcast-2

-1

0

1

20.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5

Vowel Duration (sec)

Vow

el F

1 (Z

-sco

red)

vowel��