Exotic vowels in Swedish: a project description andan articulographic and acoustic pilot study of /i:/

Susanne Schötz1 , Johan Frid1 , Anders Löfqvist2

1Humanities Lab, Centre for Languages and Literature, Lund University2Department of Logopedics, Phoniatrics and Audiology, Lund University

Abstract

This paper introduces the research project Exotic vowels in Swedish – an articulo-graphic study of palatal vowels [VOKART], which aims at increasing the empiricalknowledge of vowel production in general, and at extending our knowledge of thearticulatory dynamics of palatal vowels in Swedish in particular. In a pilot studyof the realisation of the vowel /i:/, we analysed articulatory and acoustic recordingsof two male speakers of different regional varieties of Swedish– one South Swedishwith regular [i:] and one East Central (Standard) Swedish with “damped” so called“Viby-coloured” [1:]. Results showed that [1:] was pronounced further back with anoverall lower tongue position, but with a higher tongue tip position than [i:]. Acous-tic analysis showed a lower F2 for [1:] than [i:], indicating a more centralised vowelquality. These tentative results will be followed up with larger studies. In addition,one of the analysis tools being developed within the project is described briefly.

General introductionIn a cross-language comparison, Swedish has afairly rich vowel system with some particularlyexotic and distinctive features. One such fea-ture is that among the front, close vowels thereare three contrastive long vowel sounds /i:, y:,0ff:/, characterised by a relatively small acous-tic and perceptual distance, exemplified by min-imal triplets such as /ni:, ny:, n0ff:/ (‘you’, ‘new’,‘now’). Specifically the contrast between /y:/ and/0ff:/, two similar but still phonemically distinctrounded vowels, is considered highly unusual andexotic among the world’s languages. The acquisi-tion of these two vowel sounds by Swedish chil-dren as well as adults learning Swedish typicallypresents a major difficulty (e.g., Johansson, 1973;Linell and Jennische, 1980). For these three vow-els, the tongue articulation is assumed to be ba-sically identical, but the documentation is incom-plete.

Yet another exotic feature is the today fairlywide-spread realisation of /i:/ and /y:/ in Swedishwith a “damped”, “buzzing” so called “Viby-coloured” (named after the small town of Vibyin Central Sweden) quality, generally phoneti-cally transcribed as [1:]. This vowel quality hasbeen recognised as a dialectal feature in severalSwedish regions (Bruce, 2010), and is consid-ered to be very rare among the world’s languages(Ladefoged and Maddieson, 1996).

Phonetic investigations of vowels in differentlanguages have been mainly acoustic (Ladefoged,2003). Acoustic studies of Swedish vowels usingformant frequencies include Fant (1959), Eklundand Traunmüller (1997), and Kuronen (2000).However, it does not seem possible to uniquelydetermine the underlying articulation of a vowelfrom its formant frequencies.

Object of study and goalsThe general object of study of the project is theproduction of vowels, specifically the articulationof the Swedish long palatal vowels /i:/, /y:/, /0ff:/,/e:/ and /ø:/. We will focus on three specific is-sues: (1) the crowding of vowels among the front,close vowels, particularly /y:/ and /0ff:/, (2) thediphthongisation of all five, long vowels, and (3)the special realisation of /i:/ and /y:/ vowels witha “damped” quality in contrast to the regular real-isation of these vowel sounds. The major goal ofour project is to describe and understand the artic-ulation of Swedish palatal vowels, including theirarticulatory dynamics (diphthongisation). Fur-thermore, we will also elucidate the dialectal vari-ation among Swedish vowels. For this purpose,we will restrict ourselves to studying vowels pro-duced by speakers from the three metropolitan ar-eas of Stockholm, Göteborg and Malmö, whichrepresent East Central (Standard), West Centraland South Swedish.

TMH - QPSR Vol. 51

25

Material and methodSpeech material containing vowels from at least15 subjects from each of the three dialectal ar-eas Stockholm, Göteborg and Malmö will berecorded. For the sake of completeness, wewill record realisations of all Swedish vowelphonemes, and then focus our study on the palatalvowels. Articulographic and acoustic (16 kHz/16bit) recordings of the vowels will be made inseveral consonantal contexts and different typesof speech material. The recorded movements ofeach vowel will then be analysed using automaticmethods. As indicated above, our focus will be onarticulation, but also the formant frequencies (F1,F2, F3) and their dynamics (possible diphthongi-sation) will be analysed and related to the articu-latory trajectories. We will use the Carstens Artic-ulograph AG500 to record the articulatory move-ments (at 200 Hz). This method is based on theprinciple that when a coil (sensor) moves insidea magnetic field, a voltage is induced in the coil,which is proportional to the distance between thecoil and the transmitter coil generating the mag-netic field. Articulography tracks movements in3D of the discrete points where the sensors are at-tached to the tongue or to other articulators. Wewill record twelve sensors simultaneously. Figure1 shows the positions of these.

1

2

3 54

8

7

6

12

119 10

Figure 1: Positions of the 12 sensors to be used.

Our articulographic method is particularlysuited for examining the assumed similarity intongue position among the Swedish close frontvowels. The magnitude of the lip opening (fromlarger to smaller) is regarded as the major dis-tinctive feature for these vowels: unrounded (withspread lips) /i:/, outrounded /y:/ and inrounded /0ff:/(Fant, 1959; Ladefoged and Maddieson, 1996).

Pilot study of two realisationsof /i:/ in SwedishOur interest in studying the damped realisation of/i:/ and /y:/ from a production point of view is

related specifically to an old dispute representedin the Swedish linguistics and phonetics litera-ture about its precise place of articulation (Bruceand Engstrand, 2006). The disagreement is aboutwhether the point of major constriction for thesevowel sounds, i.e. damped /i:/ and /y:/, is fur-ther front, as compared to their regular counter-parts (front, close vowels), and basically alveo-lar (Noreen, 1903), or instead further back andrather central (Borgström, 1913). Moreover, theposition of the tongue tip relative to the tonguedorsum in damped /i:/ and /y:/ has also been un-der debate (see e.g., Engstrand et al., 2000). Itshould be stressed that these views about the spe-cific point of major constriction of these vowelsare at best intelligent speculations, as adequate ar-ticulatory data seem to be lacking here. A fronted(alveolar) variant would seem to be more odd as aplace of articulation of a vowel, while a retracted(central) variant would appear to be a vowel ar-ticulation which is less unusual and found in afair number of the world’s languages (Björstenet al., 1999; Engstrand et al., 2000). To learn moreabout the articulation and acoustics of the regularand damped realisations of Swedish /i:/, we con-ducted a small pilot study.

Data and methodArticulographic and acoustic recordings of /i:/with two Swedish male speakers of different re-gional varieties – one South Swedish speakerwith regular [i:] and one East Central Swedishspeaker with damped [1:] – were made using theCarstens Articulograph AG500. Three repetitionsof the two words /bi:bel/ and /papi:pa/ with bil-abial contexts and primary stress on the /i:/ vowelwere recorded using a subset of the sensor posi-tions shown in Figure 1. Sensors were placed onthe tongue tip, tongue blade and tongue dorsum.We placed reference sensors on the nose bridgeand behind the ear. Articulographic analysis wasmade with the MATLAB script Mview (Tiede,2010), and Praat (Boersma and Weenink, 2011)was used in the the acoustic analysis.

ResultsArticulographic analysis

The MATLAB script Mview enables examina-tion of the data in three dimensions. Since therecorded sensors were aligned along the tongueon the same axis, we examined the tongue move-ment pattern at a midsagittal plane. We selectedan articulatory measurement sample from onepoint in time from the steady-state portion of eachvowel. Figure 4 shows the positions of the sensorsat this time instant.

Fonetik 2011

26

Nose ridge(reference point)

Tongue tip

Tongue bladeTongue dorsum

Horisontal position (mm)

Verti

cal

posi

tion

(mm

)

Nose ridge(reference point)

Tongue tipTongue blade

Tongue dorsum

Horisontal position (mm)

Verti

cal

posi

tion

(mm

)

Figure 2: Midsagittal plots of four articulatorymeasurement sample points from one point in timeof the steady-state portion of Swedish regular [i:](top) and damped [1:] (bottom).

The tongue is positioned more forward in themouth (relative to the nose bridge) for [i:] thanfor [1:]. Another difference is that for [i:] thetongue blade is lower than the tongue dorsum andthe tongue tip is lower than any of the other two.For [1:] however, the height pattern is reversed;the tongue tip appears to be higher than the oth-ers. The overall position of the tongue also ap-pear somewhat lower (again, relative to the nosebridge) for [1:].

Acoustic analysis

Table 1 displays mean values of the first four for-mant frequencies (F1-F4) of manually segmentedsteady-state portions of three repetitions each ofregular [i:] and damped [1:]. F1 appears to be al-most identical in [i:] and [1:], while F2 is 601 Hzhigher in [i:]. F3 and F4 show slight differences;F3 is 131 Hz higher in [i:], while F4 is 282 Hzhigher in [1:].

Figure 3 shows an F1/F2 plot with one Barkcircles of regular [i:] and damped [1:], as wellas the primary cardinal vowels pronounced byDaniel Jones (see e.g., IPA, 1999). [i:] appearsto be quite similar to the second primary cardinalvowel [e:], i.e. a front close or close-mid vowel,while [1:] is positioned further back, like a centralclose or close-mid vowel.

Mean Fn (Hz) [i:] [1:]F1 332 330F2 2017 1416F3 2685 2554F4 3239 3521

Table 1: Mean formant frequencies (Hz) of thefirst four formants in the steady-state portionof [i:] and [1:] (three repetitions by one malespeaker for each vowel realisation).

F2 (Bark)

F1 (Bark)F 1

(Hz)

F2 (Hz)

510152

4

6

1200

200

3500 50010002000

500

1000

1

2

3

4

56

78

ii [ɨː][iː][iː] [ɨː]

Figure 3: F1 /F2 plot of mean formant valuesmeasured in the steady-states of regular [i:] anddamped [1:] along with the eight primary cardi-nal vowels as pronounced by Daniel Jones. Eachcircle is one Bark in diameter.

DiscussionArticulatory as well as acoustic differences be-tween regular and damped /i:/ were observed. Wefound a difference in tongue position that sug-gests that [1:] is pronounced further back, i.e.not as an alveolar, but more like a central palatalvowel. The shape of the tongue is also differentin the two realisations, i.e. an downward slopefrom dorsum to tip for [i:], but an upward slopefor [1:]. This supports the observations of [1:] byNoreen (1903), but not Borgström (1913). How-ever, the tongue position and shape needs to be in-vestigated further using more subjects of severalSwedish dialects.

The acoustic analysis showed that the formantfrequencies differ mainly in F2, indicating thatthe main difference between the two realisationsis that [1:] is pronounced further back than [i:].This is in line with Engstrand et al. (2000). Both[i:] and [1:] seem closer to the second cardinalvowel [e:] than the first one [i:], suggesting thatthey are close-mid vowels. However, the SouthSwedish speaker used a slight diphthongisation[ei:], and although only the steady-state portionof the [i:] was used in the analysis, this may stillhave had some influence on the results. Our ten-tative results are based on a very limited mate-

TMH - QPSR Vol. 51

27

rial and need to be followed up by larger stud-ies with more subjects and vowel repetitions. Fu-ture work includes recording and analysing allSwedish vowels produced by at least 15 subjectseach from three dialectal areas.

Exemplifying work in progressWe are currently developing tools for visualisingthe data. One way of plotting articulatory data isthe common midsaggital view of the tongue fromthe side. Figure 4 shows an example of such aplot of the tongue profiles for one realisation ofeach of the Swedish long vowels /i:, y:, e:, ø:, E:,0ff:, u:, o:, a:/ along with the palate contour for onespeaker and one moment in time per vowel.

0 10 20 30

-70

-60

-50

-40

-30

back - front (mm)

low

- h

igh (

mm

)

ɑː

uːʉ̟ː

oː

eː iːyː

ɛːøː

Figure 4: Sensors placed on a male south Swedishspeaker and the corresponding midsagittal viewof the palate contour and stylised tongue profilefor one moment in time of the Swedish long vow-els /i:, y:, e:, ø:, E:, 0ff:, u:, o:, a:/.

The palate profile was measured as the subjectpressed the tongue tip sensor up against the palateand slowly pulled it backwards. The tongue pro-files were reconstructed from sensors 1, 2 and 3(see Figure 1) by simple linear interpolation. Fu-ture work includes developing visualisation toolsfor vowel dynamics in three dimensions.

AcknowledgementsThis work is supported by a grant from theSwedish Research Council, grant no. 2010-1599.We also would like to thank Mark Tiede for hiskind permission to let us use Mview.

ReferencesBjörsten S, Bruce G, Elert C C, Engstrand O,

Eriksson A, Strangert E and Wretling P (1999).Svensk dialektologi och fonetik – tjänster ochgentjänster. In Svenska landsmål och svensktfolkliv.

Boersma P and Weenink D (2011). Praat:doing phonetics by computer (version5.2.17) [computer program]. Webpagehttp://www.praat.org/, visited 3-March-11.

Borgström M (1913). Askermålets ljudlära. InSvenska landsmål och svenskt folkliv B.

Bruce G (2010). Vår fonetiska geografi. Lund:Studentlitteratur.

Bruce G and Engstrand O (2006). The phoneticprofile of swedish. Sprachtypologie und Uni-versalienforschung, 12–35.

Eklund I and Traunmüller H (1997). Com-parative study of male and female whisperedand phonated versions of the long vowels ofSwedish. Phonetica, 1–21.

Engstrand O, Björsten S, Lindblom B, Bruce Gand Eriksson A (2000). Hur udda är viby-i?experimentella och typologiska observationer.Folkmålsstudier, 83–95.

Fant G (1959). Acoustic description and classi-fication of phonetic units. Ericsson Technics1. (Reprinted in Fant, G. 1983. Speech Soundsand Features, pp. 32–83. Cambridge, MA: MITPress.

IPA (1999). Handbook of the IPA. Cambridge:Cambridge University Press.

Johansson F A (1973). Immigrant SwedishPhonology. Lund: Gleerups.

Kuronen M (2000). Vokaluttalets akustiki sverigesvenska, finlandssvenska och fin-ska. Ph.D. thesis, Jyväskylä: University ofJyväskylä.

Ladefoged P (2003). Phonetic Data Analysis:An Introduction to Fieldwork and InstrumentalTechniques. Malden, MA: Blackwell.

Ladefoged P and Maddieson I (1996). The Soundsof the World’s Languages. Oxford: Blackwells.

Linell P and Jennische M (1980). Barns uttal-sutveckling. Lund: Liber.

Noreen A (1903). Vårt språk I. Lund:Gleerups.Tiede M (2010). Mview: Multi-channel visual-

ization application for displaying dynamic sen-sor movements. In development.

Fonetik 2011

28