1
Speech rate Speech rate is lower in CDS (Beta=-70, SE=9, p<0.001). No effect of time. (Beta= 1, SE=0.2, p<0.001) CDS is not what you think - Hypoarticulation in Danish Child Directed Speech Christina Dideriksen & Riccardo Fusaroli Aarhus University Introduction: Research over the past decades has found evidence for specific acoustic features that seem to be consistent in CDS across different languages (Fernald et al. 1989, Kuhl et al. 1997): a high pitch and a high pitch variation a lower speechrate and an increased vowel space. However, Danish is understudied with regards to CDS and with the the vast number of spoken vowels in Danish (appr. 60, compared to ca. 20 in US English, Basbøll 2005) and the delay in language comprehension in Danish children (Bleses 2008) we wanted to investigate this particular aspect of language acquisition. Project details: We wanted to emphasize the importance of working with ecologically valid data. Therefore we collected data from: 23 Danish mothers and their children (between 12-24 months, M= 17.73, SD=4.46) 1 hour of spontaneously occurring CDS 15 minutes of semi-structured mother-adult conversations. All data is collected in the homes of the participating families. Mothers filled out a questionnaire about parenting attitudes after the recordings Procedure: Mothers were instructed to play with their child as they would normally do. Afterwards the mother would have an informal conversation with an adult interviewer. Mothers were not informed about the entire scope of the study before after the recording session. Materials: Orthographic transcription and time coding of 10 mother-child dyads and the corresponding mother-adult dyad This yielded: 3782 utterances from the mother in the CDS-condition 1065 utterances from the mother in the ADS-condition Parenting attitude questionnaires: Negative attitude towards ‘baby talk’ The belief that it is important to talk ‘clearly and lively’ to a child is correlated with an expanded vowel space and a smaller difference between CDS and ADS (Beta: -0.06489, SE=0.02576, t=2.52, p=0.041) Vowel space: Does a smaller vowel space in CDS actually impede language acquisition in Danish or could it help it? Is there a similar pattern in other vowel rich languages? Clustering The model defined 12 different vowel clusters Dispersion of vowels is important. An expanded vowel space can be a signal of a big dispersion of vowels (Martin et al. 2015). Pitch Median pitch was lower in ADS (Beta=-70, SE=9, p<0.001). No effect of time. Interquartile range was lower in ADS (Beta=-26, SE=5.6, p=0.001). No effect of time. Vowel space CDS has a decreased vowel space in CDS (Beta= 0.1, SE=0.04, p=0.046 (Bayes factor 2.2)) References: Basbøll, H. 2005. The phonology of Danish. Oxford: Oxford University Press. Bleses, D. et al. 2008. Early vocabulary development in Danish and other languages: a CDI-based comparison. Journal of Child Language, 35, 619-650 De Jong, N.H. & Wempe, T. 2009. Praat script to detect syllable nuclei and measure speech rate automatically. Behavior research methods, 41, 2, 385 - 390. Drugman, T. & Alwan, A. 2011. Joint Robust Voicing Detection and Pitch Estimation Based on Residual Harmonics. Paper presented at the Interspeech Fernald, A. et al. 1989. A cross-language study of prosodic modifications in mothers’ and fathers’ speech to preverbal infants. Journal of child language, 16, 3, 477-501 Kuhl, P. et al. 1997. Cross-language analysis of phonetic units in language addressed to infants. In Science, 277, 684-686. Martin, A. et al. 2015. Mothers speak less clearly to infants that to adults: A comprehensive test of the hyper articulation hypothesis. Psychological Science, 26, 3. 341-347. Sandoval, S. et al. 2013. Automatic assessment of vowel space area, in Journal of the Acoustical Society of America, 134, 5. 477-483. This project is supported by an Interacting Minds Center seed funding and the Puzzle of Danish, FKK-collective grant Methods and analysis We used linear mixed effects models (lme4 in R, Bates et al. 2014). Individual variation was accounted for by including participant ID as random intercept and register as random slope. All acoustical elements were automatically extracted: Fundamental frequency (summation of the residual harmonics and filtering for outliers, Drugman & Alwan 2011) Speech rate (vowel onset estimation, de Jong & Wempe 2009) Vowel space (Sandoval et al. 2013) We used an automatized cluster analysis to identify Danish vowel clusters. Vowel space was calculated from F1 and F2 in voiced speech segments and defined as the area surrounded by cluster centroids.

CDS is not what you think -Hypoarticulationin Danish …...Speech rate Speech rate is lower in CDS (Beta=-70, SE=9, p

  • Upload
    others

  • View
    5

  • Download
    0

Embed Size (px)

Citation preview

Page 1: CDS is not what you think -Hypoarticulationin Danish …...Speech rate Speech rate is lower in CDS (Beta=-70, SE=9, p

SpeechrateSpeechrateislowerinCDS(Beta=-70,SE=9,p<0.001).Noeffectoftime.(Beta=1,SE=0.2,p<0.001)

CDS is not what you think - Hypoarticulation in Danish Child Directed SpeechChristinaDideriksen &RiccardoFusaroliAarhusUniversity

Introduction:ResearchoverthepastdecadeshasfoundevidenceforspecificacousticfeaturesthatseemtobeconsistentinCDSacrossdifferentlanguages(Fernald etal.1989,Kuhl etal.1997):• ahighpitchandahighpitchvariation• alowerspeechrate and• anincreasedvowelspace.However,DanishisunderstudiedwithregardstoCDSandwiththethevastnumberofspokenvowelsinDanish(appr.60,comparedtoca.20inUSEnglish,Basbøll 2005)andthedelayinlanguagecomprehensioninDanishchildren(Bleses 2008)wewantedtoinvestigatethisparticularaspectoflanguageacquisition.

Projectdetails:Wewantedtoemphasizetheimportanceofworkingwithecologicallyvaliddata.Thereforewecollecteddatafrom:

• 23Danishmothersandtheirchildren(between12-24months,M=17.73,SD=4.46)

• 1hourofspontaneouslyoccurringCDS• 15minutesofsemi-structuredmother-adultconversations.• Alldataiscollectedinthehomesoftheparticipatingfamilies.• Mothersfilledoutaquestionnaireaboutparentingattitudesafterthe

recordings

Procedure:• Motherswereinstructedtoplaywiththeirchildastheywouldnormally

do.• Afterwardsthemotherwouldhaveaninformalconversationwithan

adultinterviewer.• Motherswerenotinformedabouttheentirescopeofthestudybefore

aftertherecordingsession.

Materials:Orthographictranscriptionandtimecoding of10mother-childdyadsandthecorrespondingmother-adultdyadThisyielded:

• 3782utterancesfromthemotherintheCDS-condition• 1065utterancesfromthemotherintheADS-condition

Parentingattitudequestionnaires:• Negativeattitudetowards‘babytalk’• Thebeliefthatitisimportanttotalk‘clearlyandlively’toachildis

correlatedwithanexpandedvowelspaceandasmallerdifferencebetweenCDSandADS(Beta:-0.06489,SE=0.02576,t=2.52,p=0.041)

Vowelspace:• DoesasmallervowelspaceinCDSactuallyimpedelanguage

acquisitioninDanishorcouldithelpit?• Isthereasimilarpatterninothervowelrichlanguages?

Clustering• Themodeldefined12differentvowelclusters• Dispersionofvowelsisimportant.Anexpandedvowelspacecanbe

asignalofabigdispersionofvowels(Martinetal.2015).

PitchMedianpitchwaslowerinADS(Beta=-70,SE=9,p<0.001).Noeffectoftime.InterquartilerangewaslowerinADS(Beta=-26,SE=5.6,p=0.001).Noeffectoftime.

Vowelspace• CDShasadecreasedvowelspaceinCDS(Beta=0.1,SE=0.04,p=0.046(Bayesfactor2.2))

References:Basbøll,H.2005. ThephonologyofDanish.Oxford:OxfordUniversityPress.Bleses,D.etal.2008.EarlyvocabularydevelopmentinDanishandotherlanguages:aCDI-basedcomparison.JournalofChildLanguage, 35,619-650DeJong,N.H.&Wempe,T.2009.Praat scripttodetectsyllablenucleiandmeasurespeechrateautomatically.Behaviorresearchmethods,41,2,385- 390.Drugman,T.&Alwan,A.2011.JointRobustVoicingDetectionandPitchEstimationBasedonResidualHarmonics.PaperpresentedattheInterspeechFernald,A.etal.1989.Across-languagestudyofprosodicmodificationsinmothers’andfathers’speechtopreverbalinfants.Journalofchildlanguage,16,3,477-501Kuhl,P.etal.1997.Cross-languageanalysisofphoneticunitsinlanguageaddressedtoinfants.InScience,277, 684-686.Martin,A.etal.2015.Mothersspeaklessclearlytoinfantsthattoadults:Acomprehensivetestofthehyperarticulationhypothesis.PsychologicalScience,26,3.341-347.Sandoval,S.etal.2013.Automaticassessmentofvowelspacearea,inJournaloftheAcousticalSocietyofAmerica,134,5.477-483.

ThisprojectissupportedbyanInteractingMindsCenterseedfundingandthePuzzleofDanish,FKK-collectivegrant

Methodsandanalysis• Weusedlinearmixedeffectsmodels(lme4inR,Bates

etal.2014).• Individualvariationwasaccountedforbyincluding

participantIDasrandominterceptandregisterasrandomslope.

• Allacousticalelementswereautomaticallyextracted:• Fundamentalfrequency(summationof

theresidualharmonicsandfilteringforoutliers,Drugman &Alwan 2011)

• Speechrate(vowelonsetestimation,deJong&Wempe 2009)

• Vowelspace(Sandovaletal.2013)

• WeusedanautomatizedclusteranalysistoidentifyDanishvowelclusters.

• VowelspacewascalculatedfromF1andF2invoicedspeechsegmentsanddefinedastheareasurroundedbyclustercentroids.