sign in sign up
<<
Home , Open vowel, Schwa, Tosk Albanian

1 Northern Tosk Albanian

1 1 2 Stefano Coretta , Josiane Riverin-Coutlée , Enkeleida 1,2 3 3 Kapia , and Stephen Nichols

1 4 Institute of and Speech Processing,

5 Ludwig-Maximilians-Universität München 2 6 Academy of Albanological Sciences 3 7 and English , University of Manchester

8 29 July 2021

9 1 Introduction

10 Albanian (endonym: Shqip; Glotto: alba1268) is an Indo-European lan- 11 guage which has been suggested to form an independent branch of 12 Indo-European family since the middle of the nineteenth century (Bopp 13 1855; Pedersen 1897; Çabej 1976). Though the origin of the language has 14 been debated, the prevailing opinion in the literature is that it is a descend- 15 ant of Illyrian (Hetzer 1995). Albanian is currently spoken by around 6–7 16 million people (Rusakov 2017; Klein et al. 2018), the majority of whom 17 live in and , with others in Italy, , 18 and Montenegro. Figure 1 shows a map of the main Albanian-speaking 19 areas of Europe, with major linguistic subdivisions according to Gjinari 20 (1988) and Elsie & Gross (2009) marked by different colours and shades. 21 At the macro-level, Albanian includes two main varieties: Gheg, 22 spoken in Northern Albania, Kosovo and parts of Montenegro and North

1 Figure 1: Map of the Albanian-speaking areas of Europe. Subdivisions are based on Gjinari (1988) and Elsie & Gross (2009). CC-BY-SA 4.0 Stefano Coretta, Júlio Reis.

2 23 Macedonia; and Tosk, spoken in Southern Albania and in parts of Greece 24 and Southern Italy (von Hahn 1853; Desnickaja 1976; Demiraj 1986; Gjin- 25 ari 1985; Beci 2002; Shkurtaj 2012; Gjinari et al. 2007). At the meso-level, 26 Gheg and Tosk are usually further divided into the following sub-varieties 27 (Gjinari 1988): Gheg includes Northwestern Gheg, Northeastern Gheg, 28 Central Gheg, and Southern Gheg; while Tosk includes Northern Tosk, 29 Southern Tosk, Cham, , and Arbëresh. The speakers recorded 30 for this illustration were all born and raised in areas where Northern Tosk 31 is spoken, namely the counties of Përmet and Fier. Thus, we consider 32 the doculect (Cysouw & Good 2013) discussed here to belong to Northern 33 Tosk. 34 The five consultant speakers (2 females and 3 males) wereaged 35 between 25 and 33 years old. They were recorded in Munich while read- 36 ing a word list, a list of sentences, and the North Wind and the Sun story. 37 Following the suggestion by Whalen et al. (2020) that more acoustic and 38 articulatory details be included as part of IPA Illustrations, here we dis- 39 cuss a set of widely-used measures pertaining to and as 40 complementary evidence to our impressionistic observations. Though we 41 recognise that this set is not exhaustive, we hope it may serve as a base 42 on which more thorough future work on Albanian may build. For a full 43 report of the methods employed in this study, the reader is referred to the 44 Appendix. 45 Following recommendations for Open Science in Crüwell et al. (2019) 46 and Berez-Kroeker et al. (2018), data and code used to produce the ana- 47 lyses discussed in this paper are available on the Open Science Framework 48 at https://osf.io/vry3h/ (Coretta et al. 2021).

3 Table 1: Phonemic consonants of Northern Tosk Albanian.

Bilabial Labiodental Dental Alveolar Postalveolar Retroflex Palatal Velar Glottal Plain Velarised Apical Laminal p b t d k ɡ ts dz tʃ dʒ t̻ʃ̻ d̻ʒ̻ Nasal m n ɲ Trill r Flap ɽ f v θ ð s z ʃ ʒ h j Lateral approximant lˠ l

49 2 Consonants

50 p /pata/ pata ‘duck’ r /rapi/ rrapi ‘oak tree’ b /baɽi/ bari ‘grass’ ɽ /ɽaʃɜ/ rashë ‘I fell’ t /tapa/ tapa ‘cork’ f /fati/ fati ‘luck’ d /data/ data ‘date’ v /vata/ vata ‘epaulette’ k /kati/ kati ‘floor’ θ /θatɜ/ thatë ‘dry’ ɡ /ɡati/ gati ‘ready’ ð /ðatɜ/ dhatë ‘you gave’ ts /tsitsa/ cica ‘boobs’ s /salˠa/ salla ‘big room, hall’ dz /dzidza/ xixa ‘sparkles’ z /zalˠi/ zalli ‘pebbles’ tʃ /tʃaji/ çaji ‘tea’ ʃ /ʃava/ shava ‘I scolded’ dʒ /dʒaja/ xhaja ‘uncle’ ʒ /ʒaba/ zhaba ‘lizard’ t̻ʃ̻ /t̻ʃava/̻ qava ‘I cried’ h /hanɜ/ hanë ‘they eat’ d̻ʒ̻ /d̻ʒaku/̻ gjaku ‘blood’ j /janɜ/ janë ‘they are’ m /mana/ mana ‘mulberries’ l /lava/ lava ‘I washed’ n /nana/ nana ‘grandma’ lˠ /lˠava/ llava ‘lava’

51 ɲ /ɲeɽka/ njerka ‘stepmother’

52 The 29 phonemic consonants of Northern Tosk described here are 53 shown in Table 1. For each , our description is based on the 54 consonant that occurs as the first of near-minimal pairs consist- 55 ing of trochaic words. The illustrative words are taken from Beci (2004), 56 with these exceptions: bari, xhaja, mana, nana, njerka, fati, vata, lava, 57 llava. Northern Tosk contrasts eight manners of articulation (as defined 58 in International Phonetic Association 1999): plosive, affricate, nasal, trill, 59 flap, fricative, approximant and lateral approximant. In total, there are

4 Figure 2: Segmentation of release and onset in /pata/ ‘duck’ and /kati/ ‘floor’, uttered by S04. Tiers from top to bottom: word, segments, C1 release, voice onset.

60 nine contrastive places of articulation: labial, labiodental, dental, alve- 61 olar, postalveolar, retroflex, palatal, velar and glottal. The postalveolar 62 further contrasts apical and laminal articulators in 63 the . , affricates and (with the exception of 64 /h/) contrast voiceless and voiced phonemes.

65 PLOSIVES Northern Tosk possesses six plosives: bilabial /p b/, 66 (apico-)alveolar /t d/ and velar /k ɡ/. Figure 2 shows an example of 67 the segmentation of stop release and voice onset in pata and kati. Fig- 68 ure 3 shows raw VOT values obtained for the initial plosive of the words 69 pata, bari, tapa, data, kati and gati as produced by the five speakers. Each 70 speaker repeated each word three times, so that there are three points per 71 speaker/consonant in the figure. The voiceless plosives /p t k/ have apos- 72 itive VOT, with /k/ being more post-aspirated (and more so for S05 than

5 Table 2: Mean VOT and standard deviation of word-initial plosive (15 tokens per word).

word mean (ms) sd word mean (ms) sd pata 14 6 bari -108 27 tapa 15 10 data -117 19 kati 55 26 gati -106 23

73 other speakers). The mean VOT of /p/ is 14 ms (SD = 6), /t/ 15 ms (10), 74 /k/ 55 ms (26). The voiced plosives /b d ɡ/ have robust vocal fold vi- 75 bration during their closure (i.e. pre-voicing or negative VOT). Their VOT 76 values tend to show more intra- and inter-speaker variability than those 77 of the voiceless plosives. The mean VOT values are similar across these 78 consonants and they show greater standard deviations than the voiceless 79 plosives: /b/ -108 ms (SD = 26), /d/ -117 ms (19), /ɡ/ -106 ms (26).

80 FRICATIVES Northern Tosk has nine fricatives at the labiodental, dental, 81 alveolar, postalveolar and glottal places of articulation. In all places but 82 glottal, a voiceless fricative contrasts with its voiced counterpart. The 83 fricatives /f v/ are labiodental, /θ ð/ interdental, /s z/ (apico-)alveolar, 84 /ʃ ʒ/ (lamino-)postalveolar and /h/ (voiceless) glottal. The left panels 85 of Figure 4 show the spectral centre of gravity (CoG) of voiceless (top) 86 and voiced fricatives (bottom). As is seen cross-linguistically, /s z/ are 87 well separated from /ʃ ʒ/ in having a markedly higher CoG than the lat- 88 ter (see for example Maniwa et al. 2009 and references therein). Notably, 89 the labiodental and dental fricatives are not distinguished by CoG. This 90 is not surprising in light of the common sound change from dental to la- 91 biodental fricatives, as seen in varieties of English like Northern British 92 English (Baranowski & Turton 2015). 93 In the isolated words from the word list, the voicing contrast is con- 94 sistent and robust for all speakers: the voiceless fricatives show no passive 95 voicing from neighbouring segments and the voiced fricatives are usually 96 produced with vocal fold vibration throughout the entire consonant. How-

6 Figure 3: Voice-onset time (VOT) of voiceless and voiced plosives as meas- ured in the first consonant of pata, bari, tapa, data, kati and gati. VOT = 0 corresponds to the time of consonant release. Each speaker repeated each word three times (each repetition is represented by a dot in the figure).

7 Figure 4: Spectral centre of gravity of fricatives and affricates.

97 ever, the glottal fricative /h/ is at times realised as voiced [ɦ] in connected 98 speech, as can been seen in the of The North Wind 99 and the Sun in Section 5 (e.g. /ðe meɲɜheɽɜ uðɜtaɽi/ [ðɛ miɲɜˈɦɛɹoθˈtaɽi]̞ 100 ‘and immediately the traveller’).

101 AFFRICATES The consonants represented by ⟨q⟩ and ⟨gj⟩ in writing are 102 traditionally described as plosives (Bevington 1971; Newmark et al. 1982; 103 Dodi 1996; Memushaj 2005, 2011; Jubani-Bengu 2011, 2012), though 104 Lowman (1932) classifies them as affricates and Belluscio (2014) reports 105 that they are realised with strong frication. In our data, in fact, they are 106 never realised as plosives and are instead clearly affricated for all speak- 107 ers. The waveforms and spectrograms of qava ‘I cried’ and gjaku ‘blood’ 108 as produced by speaker S04 are provided in Figure 5. From these it can 109 be seen that the plosive closure in word-initial position is followed by a 110 period of frication of substantial duration. 111 Another contested aspect of these consonants concerns their place of ar-

8 Figure 5: Waveforms and spectrograms of qava ‘cried’ and gjaku ‘blood’ as produced by speaker S04.

112 ticulation, which is generally reported to be palatal or pre-palatal/alveolo- 113 palatal (Lowman 1932; Bevington 1971; Dodi 1996; Kolgjini 2004; 114 Memushaj 2005, 2011; Jubani-Bengu 2011, 2012; Belluscio 2014). On 115 the other hand, Newmark et al. (1982) group ⟨q⟩ and ⟨gj⟩ together with 116 the postalveolar fricatives and affricates /ʃ ʒ tʃ dʒ/ under the label “lam- 117 inal” and observe that the tongue tip is raised during the production of 118 the former. To ascertain the place of articulation of ⟨q⟩ and ⟨gj⟩, we car- 119 ried out a palatographic investigation of various consonants, displayed in 120 Figure 6. Palatograms were produced for the lingual plosive and affricate 121 consonants by applying a mixture of vegetable oil and charcoal powder to 122 the tongue of one of the authors (EK) who then uttered one of the conson- 123 ants flanked by /a/ (e.g. /ata/). A picture of the impression of thetongue 124 on the hard palate was taken with the help of a mirror (Ladefoged 2003). 125 The black markings on the palate indicate the place of contact with the 126 tongue surface. 127 The affricates ⟨q⟩ and ⟨gj⟩ (second row from the bottom) arepro- 128 duced with greater tongue contact than the postalveolar affricates /tʃ dʒ/ 129 (third row), though the location of the anterior contact is approximately

9 130 the same. We interpret this as an indication that ⟨q⟩ and ⟨gj⟩ are ar- 131 ticulated with a lamino-postalveolar closure (cf. Ladefoged & Maddieson 132 1996: 29) while /tʃ dʒ/ have an apico-postalveolar closure. This is re- 133 flected in our choice to use the plain postalveolar symbols /tʃ dʒ/forthe 134 apico-postalveolar affricates and the same symbols with the addition of 135 the laminal /t̻ʃ̻ d̻ʒ/̻ for the lamino-postalveolar affricates ⟨q⟩ and 136 ⟨gj⟩. 137 The right panels of Figure 4 show the spectral centre of gravity (CoG) of 138 the fricative portion of the voiceless (top) and voiced affricates (bottom). 139 The affricates /ts dz/ have the highest CoG, consistent with /sz/.Fig- 140 ure 6 further shows that the tongue contact area of the alveolar affricates 141 and that of the alveolar plosives /t d/ is virtually identical. The apico- 142 postalveolar affricates /tʃ dʒ/ have CoG values that match those of/ʃʒ/. 143 Finally, the lamino-postalveolar /t̻ʃ̻d̻ʒ/̻ have lower CoGs than the other af- 144 fricates, consistent with the spectral properties of lamino-postalveolar con- 145 sonants discussed in Ladefoged & Maddieson (1996: 30). As for voicing, 146 the voiceless affricates are produced with no vocal fold vibration, while 147 voiced affricates are produced with fold vibration during both closure and 148 post-release frication. Thus, in total, we identified three voiceless–voiced 149 pairs of affricates: alveolar /ts dz/, apico-postalveolar /tʃ dʒ/ and lamino- 150 postalveolar /t̻ʃ̻ d̻ʒ/.̻

151 RHOTICS Northern Tosk contrasts two rhotics, written as ⟨r⟩ and ⟨rr⟩. 152 While our speakers always realised ⟨rr⟩ as an alveolar trill [r], ⟨r⟩ was 153 characterised by a great deal of allophonic variation, as also previously 154 noted by Belluscio (2014). Among the range of realisations for ⟨r⟩, we 155 found retroflex flaps [ɽ] and retroflex [ɻ]. These seemto 156 be in complementary distribution, with [ɻ] tending to appear in coda po- 157 sition, although for some speakers and in some words, word-initial /r/ 158 was also realised as a retroflex approximant [ɻ]. Impressionistic studies 159 and some case-study analyses suggest that /r/ may increasingly be real- 160 ised as [ɻ] or variants thereof in different varieties of Albanian (Hysenaj 161 2009; Jubani-Bengu 2012; Belluscio 2016). Figure 7 shows illustrative ex-

10 Figure 6: Palatograms of lingual plosives and affricates. The areas in black indicate the area of contact between the tongue and the palate.

11 Figure 7: Waveforms and spectrograms of rrapi ‘oak tree’, peri ‘thread’ and perde ‘curtains’ as produced by speaker S04.

162 amples of the three main rhotic allophones. The limited set of contexts in 163 the words collected does not allow us to make definitive generalisations 164 about the distribution of these allophones and future work is warranted 165 on this matter. We have chosen to use /r/ for ⟨rr⟩ and /ɽ/ for ⟨r⟩, though 166 as noted above, /ɽ/ is frequently realised as [ɻ].

167 LATERALS Northern Tosk contrasts two lateral consonants, namely a 168 plain alveolar lateral approximant /l/ and a velarised dental lateral ap- 169 proximant /lˠ/. As evidenced by the palatograms in Figure 8, the tongue- 170 tip closure is produced just behind the alveolar process for /l/ while it is 171 clearly dental for /lˠ/. Figure 9 shows the smoothed midsagittal tongue 172 contours of /l/ and /lˠ/ as gathered from ultrasound tongue imaging, ex- 173 tracted from a single utterance of the nonce words ala and alla respectively. 174 In this figure, the tongue tip is on the right-hand side. The dashed linein- 175 dicates the surface of the hard palate. The posterior part of the tongue 176 dorsum is raised in /lˠ/, making it velarised. We take these results from 177 palatography and ultrasound imaging to indicate that /lˠ/ is a velarised 178 (lamino-)dental lateral approximant (Dodi 2004; Belluscio 2014), while

12 Figure 8: Palatograms of the lateral approximants /l/ and /lˠ/. Note the more front (dental) closure in /lˠ/.

179 /l/ is an (apico-)alveolar lateral approximant.

180 NASALS AND APPROXIMANT Finally, Northern Tosk has three nasal con- 181 sonants at the bilabial /m/, alveolar /n/ and palatal /ɲ/ places of articula- 182 tion and the palatal approximant /j/. We did not identify any peculiarities 183 in these consonants that would set them apart from their canonical real- 184 isation as seen cross-linguistically.

185 3 Vowels

i /pikɜ/ pikë ‘drop, dot’ /pykɜ/ pykë ‘head (of a hammer)’ u /puɽɔ/ puro ‘cigar’ 186 e /peɽi/ peri ‘thread’ ɜ /pɜɽ/ për ‘for’ ɔ /pɔɽ/ por ‘but’ a /paɽa/ para ‘before’

187 Northern Tosk has seven vowels: /i y u e ɜ ɔ a/. Figure 10 shows 188 a trapezoid with the vocalic symbols placed approximately according to 189 their mean position in the F1/F2 normalised-Hertz space displayed in Fig- 190 ure 11. Hertz values were normalised across speakers to reduce the effects

1We chose to use ⟨a⟩ and ⟨y⟩ over ⟨ɐ̞⟩ and ⟨y̽⟩ for typographical simplicity.

13 Figure 9: Smoothed midsagittal tongue contours of /l/ and /lˠ/ from ul- trasound tongue imaging of one utterance of /ala/ and /alˠa/. The dashed line marks the surface of the hard palate.

i u y e

ɜ ɔ a

Figure 10: chart of Northern Tosk. The position of the vowels in the chart reflects approximately their position in the F1/F2 space (see Figure 11).1

14 Figure 11: F1 and F2 space in normalised Hertz of the Albanian vowels, with 95% confidence ellipses and vowel labels at the centroids. Seetext for a description of the normalisation procedure.

191 of anatomical and gender differences on formants (see Appendix for afull 192 explanation of the normalisation procedure). 193 The labiograms of the seven vowels uttered in isolation are shown in 194 Figure 12. Furthermore, Figure 13 shows midsagittal tongue contours 195 from ultrasound imaging extracted from around the mid-point of the sus- 196 tained tokens of each vowel (the front of the oral cavity is on the right 197 side). In this figure, the tongue contours have been smoothed using agen- 198 eralised additive model (Hastie & Tibshirani 1986; Wood 2006) with polar 199 coordinates (Coretta 2019). 200 Note that the description to follow is restricted to data from lexically- 201 stressed vowels. The characteristics of unstressed vowels may differ (cf. 202 Belluscio 2014), although we did not investigate this thoroughly and leave 203 it to future work.

15 Figure 12: Labiograms of a sustained token of each vowel produced in isolation. A picture of the lips was taken first from the front then from the side.

Figure 13: Smoothed midsagittal tongue contours from ultrasound tongue imaging, taken from the mid-point of sustained utterances of each vowels. The tongue tip is on the right and the tongue back on the left, the solid grey line indicates the surface of the hard palate.

16 204 CLOSE VOWELS Among the close vowels, /i/ is articulated as an unroun- 205 ded, close and front [i]. It is tense; this is reflected in the peripheral po- 206 sition of /i/ in the acoustic F1/F2 space in Figure 11 and the position of 207 the tongue, which is markedly high and front, as evidenced in Figure 13. 208 The vowel /y/ is a mid-centralised (i.e. lax) close 209 [ʏ]. The mid-centralised articulation of this vowel is clearly visible in Fig- 210 ure 13, which shows that the highest point of the midsagittal contour of 211 the tongue in /y/ is lower than that of /i/ and /e/. In Figure 11, /y/ also 212 occupies a central position, although this could in part be attributable to 213 both F3 and F2 having lower frequency in rounded vowels (Schwartz et al. 214 1993). The mid-centralised quality of /y/ has also been observed by New- 215 mark et al. (1982). /u/ is a rounded close . Its quality ranges 216 from close [u] to a more open [u̞], as illustrated by its spread distribution 217 in Figure 11. Figure 12 illustrates the very similar, if not identical, labial 218 profiles of /y/ and /u/.

219 MID VOWELS Regarding the mid vowels, /e/ is front, close-mid and un- 220 rounded [e]. Figure 12 illustrates that /e/ is produced with lips as retrac- 221 ted as /i/ but with a greater degree of aperture. /ɔ/ is a back, open-mid 222 and rounded vowel [ɔ]. Both the position of /ɔ/ in the acoustic space and 223 its tongue configuration show the open-mid nature of this vowel. /ɔ/is 224 produced with rounded lips, though they are not as protruted as for /y/ 225 and /u/, unsurprisingly, likely due to the greater lip aperture. 226 The vowel written ⟨ë⟩ in the Albanian writing system has traditionally 227 been described as a mid- (i.e. schwa). However, work on 228 different varieties of Albanian by Granser & Moosmüller (2002) shows that 229 this vowel is characterised by a great deal of variation across speakers and 230 contexts and that it deviates from the canonical mid-central vowel [ə]. We 231 identify this vowel as a central open-mid unrounded /ɜ/. In our data, /ɜ/ is 232 more open than canonical [ə] and partially overlaps with /a/ in the F1/F2 233 space. Remarkably, a similar kind of variation for this type of vowel has 234 been reported for Nalögo, an Oceanic language of the Temotu (Alfarano 235 2021). Furthermore, the tongue contour in Figure 13 highlights the low

17 236 position of the tongue in the articulation of /ɜ/, which is as low as that of 237 /a/. What distinguishes these two vowels in terms of tongue configuration 238 is in fact the tongue root, which is more front for /ɜ/ than /a/. Moreover, 239 /ɜ/ shows some degree of lip protrusion, similar to that of /ɔ/, while it 240 is totally absent for /a/. In sum, /ɜ/ is realised as a partially-labialised 241 open- [ɜ̹ ~ ɞ̜].

242 OPEN VOWEL The only open vowel /a/ is articulated between an open 243 central vowel [ä] and an open back vowel [ɑ].

244 VOWEL SEQUENCES Albanian also has various vowel sequences, al- 245 though their phonological status is still a topic of debate and different 246 authors propose different sets of diphthongal versus hiatus sequences (see 247 for example Beci 2002).

248 4 Prosody

249 As we are not aware of any study that has specifically addressed the 250 and intonation system of Northern Tosk in particular, we will review here 251 those that exist on Standard Albanian, given that Standard Albanian is 252 based primarily on Northern Tosk (Kostallari 1984).

253 4.1 Lexical stress

254 Albanian is a stress-accent language (Hyman 2006), although traditional 255 grammars refer to it as having “dynamic stress” (rather than “melodic 256 stress”, i.e. a pitch-accent system or lexical tones). More specifically, 257 Çabej (1976); Demiraj (1984); Topalli (1995), among others, argue that 258 the primary acoustic correlates of stress are duration and intensity (al- 259 though cf. Jubani-Bengu & Conforti 2008 for a less robust role of intens- 260 ity). The location of the lexical stress has been argued to be either pre- 261 dicted from phonological structure or morphology (e.g. Memushaj 2017;

18 262 Demiraj 1984; Trommer & Grimm 2004; Bermúdez-Otero 2011). Gener- 263 ally speaking, primary lexical stress can be found on the final, penultimate, 264 and antepenultimate of the word, as illustrated by the following 265 words.

final /liˈɽi/ liri ‘freedom’ 266 penultimate /muˈlˠiɽi/ mulliri ‘mill’ antepenultimate /ˈflutuɽa/ flutura ‘butterfly’

267 To assess which acoustic features might cue lexical stress in Northern 268 Tosk, we have extracted three measures from each vowel in the words liri 269 /li.ˈɽi/, mulliri /mu.ˈlˠi.ɽi/ and flutura /ˈflu.tu.ɽa/, as uttered by the speakers 270 in our corpus. From the total number of word tokens (N = 45, 3 words 271 times 3 repetitions times 5 speakers), we removed tokens that were uttered 272 with a rising contour, which we attribute to the typical list effect when 273 reading from a word list. Thirteen such word tokens were excluded from 274 further analysis, leaving 32 tokens. From each vowel of these tokens we 275 extracted vowel duration, maximum intensity, and maximum f0, which 276 we then z-scored to allow for across-subject comparison. These are shown 277 in Figure 14. Each dot represents a vowel and the grey lines link vowels 278 belonging to the same word token. The stressed vowel in each token is 279 marked by a red dot. 280 The top panel of Figure 14 shows measurements of vowel duration. 281 The slope of each connecting grey line indicates whether duration in- 282 creases (positive slope), decreases (negative slope), or does not change 283 (horizontal line) when comparing one syllable to the next. For vowel dur- 284 ation, a general pattern emerges which indicates that stressed vowels tend 285 to be longer than unstressed vowels. This is especially evident when com- 286 paring the antepenultimate and penultimate in mulliri and flutura. 287 Note that the increase of vowel duration in word-final position, albeit not 288 universal, is a well-known phenomenon (White et al. 2020), so that the 289 increase observed in final syllables is not surprising.

19 Figure 14: Three acoustic correlates of lexical stress: duration (top), in- tensity (middle) and f0 (bottom). Each dot represents a vowel and the grey lines link vowels belonging to the same token (N=32). Red dots represent stressed vowels.

20 290 The mid panel reports maximum intensity (i.e. peak intensity within 291 each vowel). While there is a general tendency for stressed vowels to show 292 higher intensity than unstressed vowels, the opposite pattern can also be 293 observed in the data. Moreover, especially in flutura, some tokens are 294 characterised by a stable maximum intensity value across two consecutive 295 syllables. Finally, the maximum intensity values of the stressed vowels 296 in mulliri and flutura show substantially less variation than those in the 297 unstressed vowels. In sum, intensity might be playing a role as a cue to 298 stress, but less systematically so than vowel duration, which is also in line 299 with Jubani-Bengu & Conforti (2008). 300 Maximum f0 is shown in the bottom panel of Figure 14. Fundamental 301 frequency is systematically higher in stressed compared to unstressed vow- 302 els. This result is surprising in light of the traditional view that Albanian 303 lexical stress is cued by duration and possibly intensity (Çabej 1976; De- 304 miraj 1984; Topalli 1995; Jubani-Bengu & Conforti 2008), but not by f0 305 (Jubani-Bengu & Conforti 2008). However, f0 modulations are not un- 306 usual in stress-accent , and the presence of durational, intens- 307 ity, and f0 differences in stressed vs unstressed vowels in production has 308 been argued, for example, for Italian (Albano Leoni & Maturi 1998; Alfano 309 2006; Alfano et al. 2009; Sulpizio & McQueen 2012). 310 To sum up, the data in our corpus suggest that vowel duration and f0 311 can differentiate lexically stressed from unstressed vowels (with the former 312 being longer and having higher f0). While intensity might also play a role, 313 the patterns we found for intensity are less robust compared to the other 314 two measures.

315 4.2 Intonation (a) Lena lau murin. ‘Lena washed the wall.’ (b) Lena lau murin, jo lulen. ‘Lena washed the wall, not the flowers.’ 316 (c) A e lau Lena murin? ‘Did Lena wash the wall?’ (d) Çfarë lau Lena? ‘What did Lena wash?’

21 Figure 15: Smoothed intonation contours (f0) of (a) declarative sentence, (b) contrastive focus, (c) polar question, (d) content question. Normalised times of syllabic boundaries are marked by vertical lines within each panel.

317 Figure 15 shows the f0 contours of a declarative sentence (a), contrast- 318 ive focus (b), a polar question (c), and a content question (d), from three of 319 our informants. Early descriptive studies have suggested that declarative 320 sentences have a falling intonational contour, as seen in our corpus in sen- 321 tences (a) and (b), while questions have a rising one, as in sentences (c) and 322 (d) (Boriçi 1987; Memushaj 2015). More recent work has proposed that 323 different informational structure constructs, such as topic, informational 324 focus and contrastive focus associate with different tunes in Standard Al- 325 banian (Kapia & Brugos 2016, 2019; Kapia et al. 2020, forthcoming). The 326 narrow focus pattern can be seen in the word murin in the polar question 327 (c).

22 328 5 The North Wind and the Sun

329 5.1 Phonemic transcription

330 /eɽa e veɽiut | ðe dielˠi | pɔ ziheʃin se kuʃ iʃte mɜ i fɔɽtɜ ǁ kuɽ aty kalɔi ɲɜ 331 uðɜtaɽ t̻ʃɜ̻ kiʃte veʃuɽ ɲɜ palˠtɔ tɜ nɡɽɔhtɛ ǁ ata ɽanɜ dakɔɽd | t̻ʃɜ̻ kuʃ do ta 332 bɜnte uðɜtaɽin tɜ hit̻ʃte̻ palˠtɔn mɜ pɜɽpaɽa do tɜ t̻ʃuhej̻ mɜ i fɔɽti ǁ filˠ pas 333 kɜsaj ǀ eɽa e veɽiut filˠɔi tɜ fɽynte me d̻ʒiθɜ̻ fut̻ʃinɜ̻ e saj ǁ pɔɽ sa mɜ ʃumɜ t̻ʃɜ̻ 334 fɽynte | at̻ʃ̻ mɜ ʃumɜ kapej uðɜtaɽi pas palˠtɔs sɜ tij ǀ deɽisa mɜ nɜ fund | eɽa 335 | e veɽiut | u doɽɜzua ǁ pastaj dielˠi lɜʃɔi rezet e tij tɜ nɡɽɔhta ǀ ðe meɲɜheɽɜ 336 uðɜtaɽi e hɔt̻ʃi̻ palˠtɔn ǁ e kɜʃtu | eɽa e veɽiut | u detyɽua tɜ pɽanɔnte t̻ʃɜ̻ 337 dielˠi | iʃte mɜ i fɔɽtɜ se ajɔ ǁ/

338 5.2 Narrow phonetic transcription

339 [ˈeɹä e veˈɽiut | θe diˈelˠi | po ˈzieʃin se kuʃ ˈiste mɜj ˈfɔɹt ‖ kuɹ ätʏ̈ käˈloi 340 ɲuðɜˈtäɹ t̻ʃɜ̻ ˈkiʃte ˈveʃu ɲɜ ˈpäl̞ˠto tɜ ˈŋɡɹäu̯t ‖ äˌtä ɹän täˈkoɹt | t̻ʃɜ̻ ˈkuʒ dɔ 341 täˈ bɜntoːθˈtäɽin tɜˈhiʃte̻ ˈpäl̞ˠton mɜ pɜɹˈpäɽä̞ do tɜ ˈt̻ʃuɦej̻ ˈmɜ i ˌfɔɹti ‖ ˈfilˠ 342 päs kɜˈsäj | ˈeɹä e veˈɽiut fiˈl̞ˠoi tɜ ˈfɹʏ̈nte me ˈd̻ʒiθ̻ fuʃin̻ e ˈsäj ‖ poɹ ˈsä mɜ 343 ˈʃum t̻ʃɜ̻ ˈfɹʏ̈ənte | äʃ̻ mu ˈʃuŋkäpeuθˈtäɽi päs ˈpäl̞ˠtosɜˈti | deɹiˈsä mɜ nɜ ˈfunt 344 | ˈeɹä | e ve˞ˈɽiut | u d̥oɹˈz̥uä ‖ ˈpästɛj ˈdielˠi lɜˌʃojɜ ˈɽɛzet e ˈti tɜ ˈŋɡɹäu̯tä | 345 ðɛ miɲɜˈɦɛɹoθˈtäɽi̞ ɜ ˌhɔt̻ʃi̻ ˈpäu̯tɔn ‖ e kɜʃˈtʊ | ˈeɻä e veˈɽiut̞ | u dɜtʏ̈ˈɹuä tɜ 346 pɻäˈnɔ̃nte t̻ʃɜ ˈdielˠi | ˈʔiʃtɛ mʷɨj ˈfɔltsɛ̆ ˈäjɔ̃ ‖]

347 5.3 Albanian

348 Era e Veriut dhe Dielli po ziheshin se kush ishte më i fortë, kur aty kaloi 349 një udhëtar që kishte veshur një pallto të ngrohtë. Ata ranë dakord që 350 kush do ta bënte udhëtarin të hiqte pallton më përpara do të quhej më 351 i forti. Fill pas kësaj, Era e Veriut filloi të frynte me gjithë fuqinë esaj, 352 por sa më shumë që frynte aq më shumë kapej udhëtari pas palltos së tij, 353 derisa më në fund Era e Veriut u dorëzua. Pastaj Dielli lëshoi rrezet e tij 354 të ngrohta dhe menjëherë udhëtari e hoqi pallton. E kështu Era e Veriut u 355 detyrua të pranonte që dielli ishte më i fortë se ajo.

23 356 5.4 English

357 The North Wind and the Sun were disputing which was the stronger, when 358 a traveller came along wrapped in a warm cloak. They agreed that the one 359 who first succeeded in making the traveller take his cloak off should be 360 considered stronger than the other. Then the North Wind blew as hard 361 as he could, but the more he blew the more closely did the traveller fold 362 his cloak around him and at last the North Wind gave up the attempt. 363 Then the Sun shone out warmly and immediately the traveller took off his 364 cloak. And so the North Wind was obliged to confess that the Sun was the 365 stronger of the two.

366 6 Open Science statement

367 The research compendium containing data and code used in this illustra- 368 tion can be found at https://osf.io/vry3h/ (Coretta et al. 2021).

369 7 Authors’ roles

370 We follow the CRediT (Contributor Roles Taxonomy) standard in provid- 371 ing information about the roles and contribution of each author. A visual 372 representation of the authors’ roles is available here: https://github. 373 com/stefanocoretta/alb-ipa/blob/main/img/credit-taxonomy.png.A 374 description of the roles can be found here: http://credit.niso.org/ 375 contributor-roles-defined/.

376 8 Acknowledgements

377 We would like to thank our consultant speakers who enthusiastically par- 378 ticipated in this study. This work was funded by the European Research 379 Council Grant No. 742289 Human interaction and the evolution of spoken 380 accent (2017–2022).

24 381 TBA.

382 Appendix: Methods

383 Our five informants were digitally recorded (44,100 Hz, 16 bits) inanan- 384 echoic chamber at the Institute of Phonetics and Speech Processing at the 385 Ludwig-Maximilians-Universität München (Germany), using a Neumann 386 TLM 103 microphone connected to a Steinberg UR824 console, while 387 reading a word list, a list of sentences and The North Wind and the Sun 388 story. Two of the consultants were visiting family when recorded and still 389 lived in Albania whereas the remaining three lived in Munich but repor- 390 ted speaking Northern Tosk with a member of their household on a daily 391 basis. Immediately prior to the recordings, each speaker engaged in a 10- 392 minute conversation with one of the authors (EK), who is a native speaker 393 of Northern Tosk, to reduce possible effects from German or English (Ant- 2 394 oniou et al. 2010). 395 The speech signal recorded with our five consultants was first forced- 396 aligned with the WebMAUS General BAS service (Schiel 1999; Kisler et al. 397 2017) using a custom grapheme-to-phoneme mapping and the language- 398 independent acoustic model. The automatic segmentation was then cor- 399 rected by hand by the authors in the EMU-SDMS suite (Winkelmann et al. 400 2017). Manual correction was performed for all segments in the words of 401 the word list and for syllable boundaries in the sentences. With respect 402 to the acoustic measurements, we extracted the frequency of the first two 403 formants in vowels, the voice-onset time (VOT) in plosives and the spectral 404 centre of gravity (CoG) of fricatives and the fricated portion of affricates. 405 The frequency of the first five formants in the 0–5500 Hz range was 406 estimated with linear predictive coding (LPC) using the Burg algorithm 407 via the Praat procedure Sound: To Formant (Burg), with Praat’s default

2Participants completed an online informed consent form and a basic sociolinguistic questionnaire.PDF versions of these Google Forms can be found in the forms/ folder at https://osf.io/u43sg/. The recording sessions took place between February and March 2020.

25 408 settings (Boersma & Weenink 2021). Where needed, the automatically de- 409 tected formant trajectories were manually adjusted by one of the authors 410 in EMU-SDMS to fit those visible on the spectrogram. This manual correc- 411 tion was performed only in the words selected to illustrate the vowels. The 412 frequency of F1 and F2 were then extracted at the midpoint of the vowel 413 and normalised for visualisation purposes through the following proced- 414 ure. First, for both F1 and F2, the values were transformed into z-scores 415 within each speaker by subtracting the overall within-speaker mean from 416 the formant value and dividing by the overall within-speaker standard 417 deviation (for F1 and F2 separately); second, the z-scored formant val- 418 ues were transformed back into Hertz by multiplying the z-scored values 419 by the overall cross-speaker standard deviation and by adding the overall 420 cross-speaker mean. This procedure produces a normalised vowel space 421 in Hertz which can be interpreted as the vowel space of an ideal “average” 422 speaker. 423 The voice-onset time (VOT) of plosives was measured as the interval (in 424 milliseconds) between the release of the stop and the onset of vocal fold 425 vibration (voicing) as signalled by the onset of periodicity in the wave- 426 form. We refer to the time lag between the plosive release and the onset of 427 voicing following the release as positive VOT, and to the time lag between 428 the onset of voicing preceding the plosive release and the release as negat- 429 ive VOT. Figure 2 shows how boundaries were placed for the measurement 430 of VOT. The spectral centre of gravity (CoG) of fricatives and the fricated 431 portion of affricates was obtained with the procedure described in Winkel- 432 mann (2021: Ch. 21): the first spectral moment (mean) of the output of 433 a Fast Fourier Transform (of the audio signal spanning the entire frica- 434 tion period) was taken as the CoG (see also Forrest et al. 1988; Harrington 435 2010). 436 The articulatory measurements (palatograms, labiograms, 2D mid- 437 sagittal tongue contours) were taken from one of the authors (EK). Palato- 438 grams were produced for the lingual plosive and affricate consonants (i.e. 439 /t d k ɡ ts dz tʃ dʒ t̻ʃ̻ d̻ʒ/)̻ following the procedure in Ladefoged (2003). 440 We applied a mixture of vegetable oil and charcoal powder to the tongue

26 441 of the subject who then uttered one of the consonants flanked by /a/ (e.g. 442 /ata/). We then took a picture of the palate with the aid of a small mir- 443 ror placed in the subject’s mouth facing up. After rinsing the tongue, the 444 procedure was repeated for the next consonant and so on. 445 Additionally, midsagittal tongue contours were recorded using ultra- 446 sound tongue imaging to illustrate the articulatory differences between 447 the lateral consonants (/l lˠ/) and the articulation of the vowels. We em- 448 ployed the set-up provided by Articulate Instruments Ltd™ (2011), with 449 a TELEMED Echo Blaster 128 unit and a TELEMED C3.5/20/128Z-3 ul- 450 trasonic transducer (20mm radius, 2-4 MHz). An Articulate Instruments 451 P-Stretch unit was used in order to synchronise the audio and video sig- 452 nals. The ultrasound probe was positioned midsagittally under the chin of 453 the subject and a headset manufactured by Derrick et al. (2018) was fitted 454 to the subject’s head in order to stabilise the probe. Articulate Assistant 455 Advanced (Articulate Instruments Ltd™ 2011) was used for both recording 456 and data processing with the following settings: 38 scan lines, 946 pixels 457 per scan line, field of view 88.1, pixel offset 105, depth 90 mm, frame rate 458 137.3. The palate was traced by asking the participant to hold water in her 459 mouth and swallow (Epstein & Stone 2005). The speaker’s occlusal plane 460 was obtained with a bite plate, following the method in Scobbie et al. 461 (2011). The occlusal plane was used to rotate the images of the tongue 462 contours such that the y-axis becomes perpendicular to the occlusal plane. 463 The contours of the midsagittal section of the tongue surface were manu- 464 ally drawn by one of the authors based on the ultrasound images. Tongue 465 contours were extracted from a single ultrasound video frame from around 466 the mid-point of each and each vowel.

467 References

468 Albano Leoni, F. & P. Maturi. 1998. Manuale di fonetica. Roma: Carocci.

469 Alfano, I. 2006. La percezione dell’accento lessicale: Un test sull’italiano 470 a confronto con lo spagnolo. In C. Savy, R. & Crocco (ed.), Atti del ii

27 471 convegno nazionale dell’associazione italiana di scienze della voce, 632–– 472 656.

473 Alfano, I., Savy, R. & J. Llisterri. 2009. Sulla realtà acustica dell’accento 474 lessicale in italiano ed in spagnolo: La durata vocalica in produzione e 475 percezione. In L. Romito, V. Galatà & R. Lio (eds.), Atti del iv convegno 476 nazionale dell’associazione italiana di scienze della voce, 22–39.

477 Alfarano, Valentina. 2021. A grammar of Nalögo, an Oceanic language of 478 Santa Cruz Island: Institut National des Langues et Civilisations Ori- 479 entales (LACITO), Paris dissertation.

480 Antoniou, Mark, Catherine T. Best, Michael D. Tyler & Christian Kroos. 481 2010. Language context elicits native-like stop voicing in early bilin- 482 guals’ productions in both L1 and L2. Journal of Phonetics 38(4). 640– 483 653. doi:10.1016/j.wocn.2010.09.005.

484 Articulate Instruments Ltd™. 2011. Articulate Assistant Advanced user 485 guide. Version 2.16.

486 Baranowski, Maciej & Danielle Turton. 2015. Manchester english. In 487 Raymond Hickey (ed.), Researching northern english, 293–316. Amster- 488 dam/Philadelphia: John Benjamins.

489 Beci, Bahri. 2002. Dialektet e shqipes dhe historia e formimit të tyre: Autokto- 490 nia e shqiptarëve në dritën e të dhënave të dialektologjiisë historike shqiptare. 491 Tiranë: Dituria.

492 Beci, Bahri. 2004. Fonetika e gjuhës shqipe. Tiranë: EDFA.

493 Belluscio, Giovanni. 2014. Fonematica contrastiva albanese-italiano. In 494 Gesuato S. & Busà M.G (eds.), Festschrift in onore di Alberto Mioni, 277– 495 292. Padova: CLEUP.

496 Belluscio, Giovanni. 2016. Varietete/variante fonetike “tendenciale” në 497 përdorimin e kontrolluar të shqipes standarde në një emision televiziv 498 të sotëm. In Aktet e Seminarit XXXVI ndërkombëtar për gjuhën, letërsinë 499 dhe kulturën shqiptare, 403–426.

28 500 Berez-Kroeker, Andrea L., Lauren Gawne, Susan Smythe Kung, Barbara F. 501 Kelly, Tyler Heston, Gary Holton, Peter Pulsifer, David I. Beaver, 502 Shobhana Chelliah & Stanley Dubinsky. 2018. Reproducible research 503 in linguistics: a position statement on data citation and attribution in 504 our field. Linguistics 56(1). 1–18. doi:10.1515/ling-2017-0032.

505 Bermúdez-Otero, Ricardo. 2011. Cyclicity. In Marc van Oostendorp, 506 Colin J. Ewen, Elizabeth Hume & Keren Rice (eds.), The blackwell com- 507 panion to , 2019–48. Oxford: Wiley-Blackwell.

508 Bevington, Gary Loyd. 1971. Albanian phonology: University of Massachu- 509 setts Amherst dissertation.

510 Boersma, Paul & David Weenink. 2021. Praat: doing phonetics by com- 511 puter [Computer program]. version 6.1.40.

512 Bopp, Franz. 1855. Über das Albanesische in seinen verwandtschaftlichen 513 Beziehungen. Berlin: Düm̈mler.

514 Boriçi, L. 1987. Intonacioni dhe funksionet e tij në gjuhën shqipe. Studime 515 Filologjike 2. 74–87.

516 Çabej, Eqrem. 1976. Studime etimologjike në fushë të shqipes, vol. 2. Tiranë: 517 Akademia e Shkencave e Shqipërisë.

518 Coretta, Stefano. 2019. Assessing mid-saggital tongue contours in polar 519 coordinates using generalised additive (mixed) models. OSF pre-print. 520 doi:10.31219/osf.io/q6vzb.

521 Coretta, Stefano, Josiane Riverin-Coutlée, Enkeleida Kapia & Stephen 522 Nichols. 2021. IPA Illustration of Northern Tosk Albanian [tosk1239], 523 (Research compendium). OSF. doi:10.17605/OSF.IO/VRY3H.

524 Crüwell, Sophia, Johnny van Doorn, Alexander Etz, Matthew C. Makel, 525 Hannah Moshontz, Jesse Niebaum, Amy Orben, Sam Parsons & Michael 526 Schulte-Mecklenbeck. 2019. 7 easy steps to open science: an annotated 527 reading list. Zeitschrift für Psychologie 227(4). 237–248. doi:10.1027/ 528 2151-2604/a000387.

29 529 Cysouw, Michael & Jeff Good. 2013. Languoid, doculect, and glossonym: 530 Formalizing the notion ‘language’. Language Documentation & Conserva- 531 tion 7. 331–359.

532 Demiraj, Shaban. 1984. Theksi në gjuhën shqipe. Studime Filologjike 2.

533 Demiraj, Shaban. 1986. Gramatikëhistorike e gjuhës shqipe. Tiranë: Aka- 534 demia e Shkencave e Shqipërisë.

535 Derrick, Donald, Christopher Carignan, Wei-Rong Chen, Muawiyath 536 Shujau & Catherine T. Best. 2018. Three-dimensional printable ultra- 537 sound transducer stabilization system. The Journal of the Acoustical So- 538 ciety of America 144(5). EL392–EL398.

539 Desnickaja, Agnija Vasiljevna. 1976. Çështje të dialektologjisë historike të 540 gjuhës shqipe. Jehona 4. 305–310.

541 Dodi, Alastas. 2004. Fonetika dhe fonologjia e gjuhës shqipe. Akademia e 542 Shkencave e Shqipërisë.

543 Dodi, Anastas. 1996. Kundërvëniet fonologjike të bashkëtingëlloreve të 544 shqipes letrare mbi bazën e tipareve akustike. Studime Filologjike 1-4.

545 Elsie, Robert & Josef Gross. 2009. . http://dialects. 546 albanianlanguage.net.

547 Epstein, Melissa A. & Maureen Stone. 2005. The tongue stops here: ul- 548 trasound imaging of the palate. The Journal of the Acoustical Society of 549 America 118(4). 2128–2131. doi:10.1121/1.2031977.

550 Forrest, Karen, Gary Weismer, Paul Milenkovic & Ronald N. Dougall. 551 1988. Statistical analysis of word-initial voiceless obstruents: Prelimin- 552 ary data. The Journal of the Acoustical Society of America 84(1). 115–123.

553 Gjinari, Jorgji. 1985. Marrëdhëniet e sotme tëgjuhës letrare me dialektet. 554 Studime Filologjike 22(1). 43–52.

30 555 Gjinari, Jorgji. 1988. Dialektologjia shqiptare. Tiranë: Akademia e Shken- 556 cave të Shqipërisë.

557 Gjinari, Jorgji, Bahri Beci, Gjovalin Shkurtaj, Xheladin Gosturani & Ana- 558 stas Dodi. 2007. Atlasi dialektologjik i gjuhës shqipe. Napoli: Università 559 degli Studi di Napoli.

560 Granser, Theodor & Sylvia Moosmüller. 2002. Phonemic schwa – a chal- 561 lenge for Natural Phonology? the case of Albanian. In Katarzyna 562 Dziubalska-Kołaczyk & Jarosław Weckwerth (eds.), Future challenges for 563 natural linguistics, 165–178. München: Lincom.

564 von Hahn, Johann Georgw. 1853. Albanesiche Studien. Wien: Friedrich 565 Mauke.

566 Harrington, Jonathan. 2010. Phonetic Analysis of Speech Corpora. 567 Chichester: Wiley-Blackwell.

568 Hastie, Trevor & Robert Tibshirani. 1986. Generalized additive models. 569 Statistical Science 1(3). 297–310. doi:10.1201/9780203753781-6.

570 Hetzer, Armin. 1995. Nominalisierung und verbale Einbettung in Varietäten 571 des albanischen: eine untersuchung zur geschichte der albanischen schrifts- 572 prache am beispiel erweiterter Verbalprädikate auf areallinguistischem Hin- 573 tergrund, vol. 28. Wiesbaden: Harrassowitz.

574 Hyman, Larry. 2006. Word-prosodic typology. Phonology 23. 225–257. 575 doi:10.1017/S0952675706000893.

576 Hysenaj, Haki. 2009. Kundërvënia fonologjike /r/-/R/, rrafshim dhe re- 577 flekse në gjuhën e folur dhe të shkruar. In Aktet e seminarit xxviii 578 ndërkombëtar për gjuhën, letërsinë dhe kulturën shqiptare, 403–408.

579 International Phonetic Association. 1999. Handbook of the international 580 phonetic association: a guide to the use of the international phonetic alpha- 581 bet. Cambridge: Cambridge University Press.

31 582 Jubani-Bengu, Aljula. 2011. Karakterizimi funksional i sistemit bashkët- 583 ingëllor të shqipes standarde. Buletini i Universitetit Planetar të Tiranës 584 .

585 Jubani-Bengu, Aljula. 2012. Fonetikë akustike. analizë e sistemit fonologjik. 586 Tiranë: SHPSH.

587 Jubani-Bengu, Aljula & Emilia Conforti. 2008. I parametri acustici 588 dell’accento nella parlata di San Benedetto Ullano e nell’albanese stand- 589 ard. In Quaderni del dipartimento di linguistica, vol. 18, Università di 590 Firenze.

591 Kapia, Enkeleida & Alejna Brugos. 2016. Information structure-prosody 592 interface: Towards a model of Albanian intonational phonology. In 593 Proceedings of the 6th international conference on speech prosody 2016, 594 946–950.

595 Kapia, Enkeleida & Alejna Brugos. 2019. Perception of prosodic correlates 596 of topic and focus in Albanian. ICPhS Satellite Workshop: Intonational 597 Phonology of Typologically Rare or Understudied Languages.

598 Kapia, Enkeleida, Jonathan Harrington & Felicitas Kleber. forthcoming. 599 An autosegmental-metrical analysis of Albanian prosody. In S. A. Jun & 600 S. Dowla Khan (eds.), Prosodic typology, vol. III, Oxford University Press.

601 Kapia, Enkeleida, Felicitas Kleber & Jonathan Harrington. 2020. An 602 autosegmental-metrical analysis of rising contours in Standard Al- 603 banian. In Proceedings of the 10th international conference on speech pros- 604 ody 2020, 171–175.

605 Kisler, Thomas, Uwe D. Reichel & Florian Schiel. 2017. Multilingual pro- 606 cessing of speech via web services. Computer Speech & Language 45. 607 326–347.

608 Klein, Jared, Joseph Brian & Matthias Fritz. 2018. Handbook of Comparat- 609 ive and Historical Indo-European Linguistics. Walter de Gruyter.

32 610 Kolgjini, Julie M. 2004. Palatalization in Albanian: An acoustic investigation 611 of stops and affricates: University of Texas at Arlington dissertation.

612 Kostallari, Androkli. 1984. Gjuha e sotme kombëtare shqipe dhe epoka 613 jonë. Studime Filologjike 21(4). 25–59.

614 Ladefoged, Peter. 2003. Phonetic data analysis. Blackwell Publishing.

615 Ladefoged, Peter & Ian Maddieson. 1996. The sounds of the world’s lan- 616 guages. Oxford: Blackwell.

617 Lowman, Guy Sumner. 1932. The phonetics of Albanian. Language 8(4). 618 271–293. doi:10.2307/408833.

619 Maniwa, Kazumi, Allard Jongman & Travis Wade. 2009. Acoustic charac- 620 teristics of clearly spoken English fricatives. The Journal of the Acoustical 621 Society of America 125(6). 3962–3973.

622 Memushaj, Rami. 2005. Tiparet binare të fonemave të shqipes standarde. 623 Studime Filologjike 43(3–4). 41–64.

624 Memushaj, Rami. 2011. Fonetikë e shqipes standarde. Tiranë: Shtëpia 625 Botuese Toena.

626 Memushaj, Rami. 2015. Fonetika e shqipes standarde. Tiranë: Infbotues.

627 Memushaj, Rami. 2017. Studime gjuhësore. Tiranë: Botimpex.

628 Newmark, Leonard, Philip Hubbard & Peter R Prifti. 1982. Standard Al- 629 banian: A reference grammar for students. Stanford, CA: Stanford Univer- 630 sity Press.

631 Pedersen, Holger. 1897. Die albanesichen l-laute. KZ 33. 535–551.

632 Rusakov, Alexander. 2017. Albanian. In Mate Kapović (ed.), The Indo- 633 European languages, Routledge.

634 Schiel, Florian. 1999. Automatic phonetic transcription of non-prompted 635 speech. In Proceedings of ICPhS 14, San Francisco, 607–610.

33 636 Schwartz, Jean-Luc, Denis Beautemps, Christian Abry & Pierre Escudier. 637 1993. Inter-individual and cross-linguistic strategies for the production 638 of the [i] vs. [y] contrast. Journal of Phonetics 21(4). 411–425.

639 Scobbie, James M., Eleanor Lawson, Steve Cowen, Joanne Cleland & 640 Alan A. Wrench. 2011. A common co-ordinate system for mid-sagittal 641 articulatory measurement. In Qmu casl working papers, 1–4.

642 Shkurtaj, Gjovalin. 2012. Dialekte e shqipes: Tiparet e përbashkëta dhe dal- 643 limet dialektore sipas hartave të adgjsh-së. Të folmet shqipe të diasporës. 644 Tiranë: Morava.

645 Sulpizio, Simone & James M. McQueen. 2012. Italians use abstract know- 646 ledge about lexical stress during spoken-word recognition. Journal of 647 Memory and Language 66(1). 177–193. doi:https://doi.org/10.1016/j. 648 jml.2011.08.001.

649 Topalli, Kolec. 1995. Theksi në gjuhën shqipe. Shtëpia Botuese Enciklo- 650 pedike.

651 Trommer, Jochen & Angela Grimm. 2004. Albanian word stress. https: 652 //home.uni-leipzig.de/jtrommer/papers/alb.pdf.

653 Whalen, Douglas H., Christian DiCanio & Rikker Dockum. 2020. Phonetic 654 documentation in three collections: Topics and evolution. Journal of the 655 International Phonetic Association 1–27.

656 White, Laurence, Silvia Benavides-Varela & Katalin Mády. 2020. Are 657 initial-consonant lengthening and final-vowel lengthening both univer- 658 sal word segmentation cues? Journal of Phonetics 81. 100982. doi: 659 https://doi.org/10.1016/j.wocn.2020.100982.

660 Winkelmann, Raphael. 2021. The EMU-SDMS manual. https://ips- 661 lmu.github.io/The-EMU-SDMS-Manual/.

662 Winkelmann, Raphael, Jonathan Harrington & Klaus Jänsch. 2017. EMU- 663 SDMS: Advanced speech database management and analysis in R. Com- 664 puter Speech & Language 45. 392–410.

34 665 Wood, Simon. 2006. Generalized additive models: an introduction with R. 666 Boca Raton: CRC Press.

35