REGIONAL VARIATION OF SATERLAND FRISIAN VOWELS Heike Schoormann, Wilbert Heeringa, Jörg Peters Institute of German Studies, University of Oldenburg, Germany [email protected], [email protected], joerg.peters@uni- oldenburg.de ABSTRACT thus includes closed short tense /i, y, u/ as well as open-mid long lax vowels /ɛː, œː, ɔː/ [9, 14, 18]. This paper reports on the acoustic investigation of As in other Frisian languages, the short tense vow- Saterland Frisian vowels, including their regional els are especially likely to undergo a merger and variation. The study aims at identifying merged become phonetic variants due to their low func- vowel categories as well as supplementary acoustic tional load and overall markedness [9, 21, 22]. dimensions, which enhance the discrimination of Moreover, language contact with High German spectrally adjacent categories. All vowels were eli- and Low German may further add to this develop- cited in a /hVt/ frame. Acoustic measurements in- ment. According to Fort [8] the merger is least ob- cluded vowel duration, mid-vowel F1 and F2, servable in Ramsloh, being the most conservative Vowel Inherent Spectral Change (VISC) [16], and of the three varieties. In addition, the number of the spectral rate of change [10]. Results confirm distinct diphthongal categories is disputed, ranging large inventories for the varieties of Saterland Fri- from 6 to 16 [3, 9, 14, 18]. sian, although some vowel categories have under- Figure 1: Monophthongs and diphthongs of Sa- gone a merger. The comparison of spectral features terland Frisian according to Fort [9]. /ə/ is re- of single vowel categories in the three varieties re- stricted to unstressed syllables. vealed an effect for Scharrel, in which most mon- ophthongs are more centralized in the F1 dimen- front central back sion than in Strücklingen. These findings are dis- close i y u cussed in light of the natives’ perception of region- iː yː uː yːi̯ uːi̯ iːu̯ (w) ɛːu̯ (w) close-mid ɪ ʏ ʊ al differences in speech rate. œːi̯ oːi̯ iu̯ (w) ɛu̯ (w) eː øː oː Keywords: Saterland Frisian, regional variation, open-mid ɛ œ ɔ ɛːi̯ ɔːi̯ ɪu̯ (w) oːu̯ ɛː œː (ə) ɔː phonetic feature enhancement, VISC aːi̯ ɔy eːu̯ (w) aːu̯ open a aː ̯ 1. INTRODUCTION Because of the high number of distinct catego- ries, Saterland Frisian is most likely to employ Saterland Frisian is spoken in Ramsloh, Scharrel, supplementary acoustic dimensions to support and Strücklingen, which constitute the municipali- vowel distinction within the crowded vowel space ty of Saterland, located in northwest Germany. De- (cf. phonetic feature enhancement [4]). [11] shows spite small geographic distances natives perceive that f0-dynamics and additional centralization in the Saterland Frisian spoken in these villages as F1 and F2 may be used to enhance distinctiveness three distinct regional varieties. They seem to be among closed vowels in Saterland Frisian. As highly aware of the small regional differences, within other large inventories, vowel dynamics which are described in the literature as being most (i.e. Vowel Inherent Spectral Change (VISC) [16]) evident in the use of different vowel qualities, vo- and vowel duration may constitute two additional calic durations, and overall speech rate [8, 14, 17]. possible enhancing cues as they have been attested Among the three villages, Scharrel is regarded as in other languages (cf. [2, 10, 12, 20]). the most divergent variety [18]. This study is an acoustic phonetic investigation With up to 20 distinct monophthongs and 16 of the inventory of Saterland Frisian vowels and its diphthongs in stressed position, Saterland Frisian is regional variation. In particular, we examine (1) reported to have an exceptionally large vowel in- the depicted inventory and possible mergers, (2) ventory [7, 9, 14, 18] (see Figure 1). The high supplementary acoustic dimensions that support number of distinct vowel categories is partly due to vowel distinction, and (3) regional variation regar- the fact that vowel length is not linked to tenseness ding durational differences as well as static and dy- in Saterland Frisian. Both phonological features namic spectral features of corresponding catego- may occur independently. The reported inventory ries. 2. METHOD formant dynamics. Window length was set to 0.025 seconds. Formant settings for the LPC Recordings were made of 35 speakers (13 from analysis were adapted for each speaker individual- Ramsloh, 11 from Scharrel, and 11 from Strück- ly in the script by de- or increasing the LPC order lingen). The informants were all born in Saterland in steps of 1 (default order of 10) and/or the maxi- and lived in the respective village all their lives. mum frequency in steps of 500 Hz (default 5000 All speakers were trilingual with Saterland Frisian, Hz). Outliers due to measurement errors were man- High German, and Low German. They were male ually corrected. Formant frequencies were nor- and aged between 50 and 75. malized, using a version1 of the Watt and Fabricius method [5] modified by Flynn [6]. 2.1. Stimuli and recording procedure 2.3. Dynamic spectral features The 36 vowel categories depicted in Figure 1 were recorded in monosyllabic words in a /hVt/ context. To analyze formant dynamics, the amount of vow- The /hVt/ words were cued by instructing the in- el inherent spectral change was assessed as the tra- formants to read aloud a rhyming monosyllabic Sa- jectory length (TL) (1), calculated as the sum of terland Frisian word immediately preceding the the two vowel sections lengths (VSL50-20, VSL80-50) production of the /hVt/ target word (e.g. Poot (2), i.e. the Euclidean distances between the meas- ‘paw’ – Hoot, cf. [2]). High German translations urement points in the F1-F2 plane [10, 13]. were given with every Saterland Frisian word. Speakers did not read the /hVt/ target word directly (1) TL = VSL50−20 + VSL80−50 off the screen: Only the frame H_t was presented as an aid to build the target word. Where no rhym- VSL = (F1 − F1 )2 + (F2 − F2 )2 ing monosyllabic trigger was available, an inter- (2) 50−20 50 20 50 20 mediate step was introduced to approach the target 2 2 VSL80−50 = (F180 − F150 ) + (F280 − F250 ) word in two steps, e.g. Moite ‘effort’ – Moit – H_t. Target words were elicited via rhymes because Sa- To account for dynamic changes in unnormal- terland Frisian orthography is usually unknown ized time, the spectral rate of change (roc) was cal- and there is a possible influence of the written culated for each vowel section length (e.g. form on the production data. VSL roc50-20) separately (3) as well as for the over- For each speaker, the sequences of trigger and all trajectory length (TL roc) (4) [10, 15]. target words were presented in blocks in a con- trolled randomized order on a computer screen. VSL (3) VSL roc = 50−20 Each sequence of a trigger and the rhyming target 50−20 0.30 × V word was presented twice. Practice sequences pre- dur ceded all blocks. Intonation was monitored for all TL (4) TL roc = target words, ensuring that all /hVt/ words were 0.60 × V dur elicited with a falling contour. All recordings were made in a quiet room with a Tascam HD P2 digital 2.4. Statistical processing recorder at a sampling rate of 48 kHz and a head- mounted microphone (DPA 4065 FR). For statistical processing the 36 vowels were com- pared to each other for each of the three villages, 2.2. Acoustic analysis using mixed models with vowel as a fixed factor and random intercepts for speaker. The dependent Acoustic analysis was done in PRAAT [1]. Onset variables were duration, F1 or F2 at the three and offset of the vocalic segment were labeled measurement points, and the amount of VISC. If manually for each /hVt/ syllable. Vowel onset was no significant differences were found for either of measured at the zero-crossing before the first posi- these dependent variables, two vowels were con- tive peak in the periodic waveform. Vowel offset sidered mergers was set at the last negative-to-positive zero- crossing before an (abrupt) decrement in amplitude 3. RESULTS and/or the end of periodicity in the waveform be- fore the stop closure. The frequencies of the first 3.1. Merger and second formant were estimated automatically with the help of a PRAAT script at three equidis- Table 1 shows the results for the analysis of tant points (20%, 50% and 80%) over the course of merged categories. As predicted, the closed tense the vowel’s duration to allow for an analysis of vowels /i y u/ and /iː yː uː/ have merged. The tri- partite vowel contrast of long tense, short tense, pared to corresponding categories in the other two and short lax monophthongs is thus reduced to a villages. This deviation from the overall pattern twofold distinction of short lax versus long tense may enhance the perceptual contrast to the long lax monophthongs (/ɪ/ – /iː/, /ʏ/ – /yː/, /ʊ/ – /uː/). close-mid vowels /ɛː œː ɔː/ and the short lax close- The analysis of diphthongs reveals a merger of mid vowels /ɪ ʏ ʊ/. /ɪu̯ w/ and /iu̯ w/ for all varieties. Whereas /ɪu̯ w/ and /iːu̯ w/ show distinct F2 values in Scharrel and Figure 2: Normalized mean values of the center Ramsloh, they have merged in Strücklingen. In ad- frequencies of the monophthongs of Saterland dition, Ramsloh shows the merger of /iu̯ w/ and Frisian. /iːu̯ w/, which in Strücklingen and Scharrel are still 0.65 iː yː uː distinguished by the frequency of the second for- mant at the temporal midpoint. Although Ramsloh 0.85 eː øː oː is described as the most conservative of the three varieties [8], our analysis reveals two more mer- ʏ 1.05 ɛː ɪ gers for this variety, /ɛu̯ w/-/ɛːu̯ w/ and /oːi̯ /-/ ɔːi̯ /, œː ʊ and thus the overall highest number of merged cat- ɔː 1.25 egories.