Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm University[iœː] or [iʉː] (for examples see E. Lagman,1979: 53). Considering this variation, it isnearly impossible to predict how /yː/ might berealised in our ES data. Based on E. Lagman’scomment (1979: 5) about ES being most similarto the Nyland variety of FS, we can hypothesisethat ES vowels would be realisedcloser to those of FS than CS. Yet, we wouldnot expect exactly the same distribution ofclose vowels in ES as in FS or CS.Materials and methodSpeech dataAs materials the word list from the SweDia2000 database was used. The data comprisedthree repetitions of four words containing longclose vowels: dis (mist), typ (type), lus (louse),sot (soot). When recording the ES speakers, theword list had to be adapted slightly because notall the words in the list appear in ES vocabulary.Therefore, dis was replaced by ris (rice),typ by nyp (pinch) and sot by mot (against).Four elderly ES speakers (2 women and 2men) were recorded in a quiet setting in Stockholmin March 2009. All the speakers had arrivedin Sweden in the mid 1940s as youngstersand were between 80 and 86 years old (meanage 83) at the time of recording. They representthe largest dialect area of ES, the Rickul variety,having all been born there (also all theirparents came from Rickul). The ES speakerswere recorded using the same equipment as forcollecting the SweDia 2000 database: a Sonyportable DAT recorder TCD-D8 and Sony tiepintype condenser microphones ECM-T140.For the comparison with CS and FS theword list data from the SweDia 2000 databasefrom two locations was used: Borgå in Nylandwas selected to represent FS, while CS was representedby Kårsta near Stockholm. From eachof these locations the recordings from 3 olderwomen and 3 older men were analysed. TheBorgå speakers were between 53 and 82 yearsold (mean age 73), and the Kårsta speakers between64 and 74 years old (mean age 67).AnalysisThe ES data was manually labelled and segmented,and the individual repetitions of thewords containing long close vowels were extractedand saved as separate sound and annotationfiles. Equivalent CS and FS data was extractedfrom the SweDia database using a Praatscript. The segmentation was manually checkedand corrected.A Praat script was used for obtaining thevalues for the first three formant frequencies(F1, F2, F3) of each vowel with the Burgmethod. The measurements were taken at themid-point of each vowel. All formant valueswere subsequently checked and implausible ordeviant frequencies re-measured and correctedby hand. Mean values were calculated for thefemale and male speakers for each variety.One-Bark vowel circles were plotted for thefemale and male target vowels [iː, yː, ʉː, uː] ofeach variety on separate F1/F2 and F2/F3 plotsusing another Praat script.In order to test for statistically significantdifferences between the dialects a two-wayANOVA was carried out with the betweensubjectsfactors dialect (3) and gender (2), anda dependent variable formant (3).Finally, a comparison of the inventory oflong close vowels in the three varieties wasconducted using the Euclidean distance, whichwas calculated for the first three formants basedon values in Bark.ResultsFigure 2 plots the F1 and F2 values separatelyfor female and male speakers for each of thethree dialects. It can be seen that the distributionis roughly similar for both female and malespeakers in all varieties.There is a significant effect of dialect on F2for the vowel /iː/ (F(2, 10) = 8.317, p