Kiekybinis diferencinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis

AGNĖ ČEPAITIENĖ Institute of the Lithuanian Language Research fields: geolinguistics, dialectology, sociolinguistics, features of spoken Lithuanian. QUANTITATIVE STUDY OF DIFFERENTIAL DIALECTAL FEATURES: THE CASE OF EASTERN AUKŠTAITIANS OF PANEVEŽYS AND ŠIRVINTOS Quantitative Study of Differential Dialectal Features: The Case of the Eastern Aukštaitians of Panevėžys and Širvintos A N N O T A T I O N This article uses quantitative methods of dialectometry to examine Lithuanian dialects for the first time. By quantitatively assessing the distinguishing features of the Eastern Aukštaitian subdialects of Panevėžys and Širvintos, the study seeks to determine the dialectal affiliation of the Atlas of the Lithuanian Language (LKA) survey points: Stum̃briškis (171), Šilai (338), Kavársas (340), Taujenai (374), Vidiškės (375 a), Dėltuva (409), Jakūtiškės (446), Pabaiskas (475), Mančiūšėnai (500), and Čiobiškis (520). The research material consists of 12 hours of audio recordings from the Panevėžys and Širvintos subdialects. The distinguishing dialectal features selected from them are analyzed using Gabmap, a computer program based on dialectometric methods and designed to investigate linguistic (and geographic) variation. On the basis of phonetic word transcriptions, it calculates the Levenshtein distance between LKA survey points, performs statistical calculations, and identifies differential dialectal features of the Panevėžys and Širvintos subdialects. The results indicate that the traditional dialectal features of the Panevėžys and Širvintos subdialects coincide from both linguistic and linguistic-geographic perspectives. Statistically identical dialectal features are more often used not at LKA survey points belonging to the same subdialect, but in the residential areas of the individual subdialects under study. The possible distinguishing features revealed among the clustered LKA survey points are shared by both subdialects and statistically substantiate the similarity between the Panevėžys and Širvintos subdialects. Straipsniai / Articles 143 AGNĖ ČEPAITIENĖ KEYWORDS: Eastern Aukštaitians of Panevėžys, Eastern Aukštaitians of Širvintos, differential dialectal features, dialectometry. ANNOTATION The article presents the first-time use of quantitative methods of dialectometry to analyse Lithuanian dialects. After quantifying the differential features of the Eastern Aukštaitian subdialects of Panevėžys and Širvintos, it aims to identify the dialectal affiliation of the Lithuanian Language Atlas (LLA) points under analysis – Stum̃briškis (171), Šilai (338), Kavárskas (340), Taujénai (374), Vìdiškiai (375 a), Déltuva (409), Jakūtiškiai (446), Pabáiskas (475), Mančiūšénai (500), Čiõbiškis (520). The research material consists of 12 hours of audio recordings of the Eastern Aukštaitians of Panevėžys and Širvintos. The selected differential dialectal features are analysed by the web application Gabmap based on dialectometry methods. It is designed for the research of linguistic (and geographical) variability. On the basis of phonetic transcriptions of words, the Levenshtein distance is calculated between LLA points, the statistical computations are performed and the differential dialectal features of the Eastern Aukštaitians of Panevėžys and Širvintos are determined. The results suggest that dialectal features of the traditional Eastern Aukštaitians of Panevėžys and Širvintos coincide from the linguistic and linguistic-geographical points of view. Statistically equal dialectal features are frequently used not in the same LLA points of subdialects but in the localities inhabited by the speakers of separate subdialects under analysis. The potential differential features of clustered LLA points under analysis are common to both subdialects and statistically justify the similarity of the Eastern Aukštaitians of Panevėžys and Širvintos. KEYWORDS: Eastern Aukštaitians of Panevėžys, Eastern Aukštaitians of Širvintos, differential dialectal features, dialectometry. 1. INTRODUCTION The relationship between the Eastern Aukštaitians of Panevėžys and Širvintos is a subject of ongoing debate: there is no agreement on the basis for distinguishing them, the boundaries of the subdialects, or the affiliation of the Lithuanian Language Atlas (LKA) points of Kavársko and Vìdiškiai to the Panevėžys subdialect1, 1 LKA 1977; 1982; 1991. 144 Acta Linguistica Lithuanica LXXIV Kiekybinis diferenciacinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis the Širvintos subdialect or the Anykščiai subdialect (cf. the boundaries between the Panevėžys and Širvintos subdialects according to studies conducted in different periods in Figs. 1 and 2)2. 1 PAV. Širvintiškiams priskiriami tiriamieji LKA punktai pagal Aliūkaitė ir kt. 2014: 1833 2 Žr. Urbanavičiūtė-Markevičienė 1994: 85–87; Markevičienė, Markevičius 1995: 69–72; Zinkevičius 1997: 237–241; Markevičienė 1999: 18–20; Markevičius 2002: 101–105; Markevičienė ir kt. 2009: 11–56. Markevičienė ir kt. 2009 žr. ir apie iki šiol atliktus širvintiškių tyrimus, o panevėžiškių tyrimų literatūros sąrašą žr. Pukevičiūtė 2014: 175–195. 3 Remiamasi LKA 1977: žemėlapis Nr. 1. Straipsniai / Articles 145 AGNĖ ČEPAITIENĖ 2 PAV. Širvintiškiams priskiriami tiriamieji LKA punktai pagal Markevičienę ir kt. 2009: 24 146 Acta Linguistica Lithuanica LXXIV Kiekybinis diferenciacinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis In this article, methods of dialectometry4 are used for the first time to study Lithuanian dialects5. The study examines 12 hours of audio recordings6 collected in the late 20th and early 21st centuries from 10 LKA sites: 6 Panevėžys-area sites (Stumbriškis (171), Šilai (338), Kavarskas (340), Taujėnai (374), Kunigiškiai (409)7, Jakūtiškiai (446)), 1 Anykščiai-area site (Vidiškiai (375 a)), and 3 Širvintos-area sites (Pabaiskas (475), Mančiūšėnai (500), Čiobiškis (520)) (see Fig. 3). The aim of the article is to determine the dialect affiliation of the LKA sites under study by quantitatively assessing the distinguishing features of the Panevėžys-area and Širvintos-area subdialects. This will also test the similarity of the LKA sites of Kavarskas and Vidiškiai to localities of either the Panevėžys-area or Širvintos-area subdialects. The object of the study is the distinguishing dialectal features of the Panevėžys-area and Širvintos-area subdialects. The Panevėžys-area subdialect is considered the largest and strongest East Aukštaitian subdialect, so it may be expected to influence the Širvintos-area subdialect as well—the smallest and a declining East Aukštaitian subdialect. This hypothesis is supported by the relatively small difference between the subdialects: the Širvintos-area speakers are distinguished from the (southern) Panevėžys-area speakers by their different pronunciation of short a, e, i, u in word-final position: the Širvintos-area speakers preserve their quality, whereas the Panevėžys-area speakers merge these sounds into one (v after a hard consonant, ʋ after a soft consonant) (Girdėnis, Zinkevičius 1966: 139–147; Zinkevičius 1966: 15; Zinkevičius 1994: 56)8. In addition to this feature, a further 7 dialectal features of the Panevėžys-area and Širvintos-area subdialects are examined to achieve the aim (see Table 1). 4 Dialectometry is a branch of geolinguistics whose aim is to calculate, represent, and analyze overall dialectal similarities or differences as a function of the characteristics of geographic space (Szmrecsanyi 2011: 45). 5 Dialectometric methods are used in contemporary studies by foreign scholars: English (Shackleton 2007: 30–102), Dutch (Spruit et al. 2009: 1624–1642), Norwegian (Heeringa et al. 2009: 167–183), Swedish (Leinonen, 2010, 2011: 75–95), Swiss (Scherrer 2012: 63–71), Chinese (Stanford 2012: 247–278), Bulgarian (Prokić, Cysouw 2013: 147–168), Dutch and Belgian (Heeringa, Hinskens 2014: 26–52; 452–453), and other language varieties have been examined quantitatively. 6 Registration Nos. CD 1204 RAS 31, CD 484 RAS 26, CD 456 RAS 10, CD 958 RAP 69; the others have not yet been assigned registration numbers. The audio recordings are held in the Dialect Archive of the Institute of the Lithuanian Language. 7 Studies of dialects in the early 21st century found that the LKA site of Kunigiškiai had become extinct, so it was relocated to Deltuva. Hereafter, LKA site 409 is referred to as Deltuva. 8 It has been proposed that other dialectal features be used as the basis for distinguishing the Širvintos-area and Panevėžys-area subdialects, for example, the change of unstressed stem o to a (Markevičius 2002: 103). Straipsniai / Articles 147 AGNĖ ČEPAITIENĖ 3 PAV. Tiriamieji LKA punktai (LKA 1977: žemėlapis Nr. 1) TABLE 1. Dialectal features of the Panevėžys and Širvintos subdialects under study9 Dialectal features Words recorded Lengthening of stressed short vowels Was, big, my, own, has, everything Shortening of unstressed long vowels Lives Merger/non-merger of short stressed word-final a, e, i, u Is, is not, difficult, person Realization of unstressed word-final o as a or o; realization of e as short e My, nothing, says, own, had Unclear pronunciation of the first component of stressed falling diphthongs ai, au I was, well, how, very, I don’t know, yes Stress retraction I was, well, is, I don’t know, is not, difficult, person Ending reduction Big, lives, is, years, is not, one, everything, person, people Rendering of unstressed stem o as a Person, people 9 LKTCH 2004: 160–164, 173–183. 148 Acta Linguistica Lithuanica LXXIV Kiekybinis diferencinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis The research material was processed in several stages: 1) individual words (26 in total) reflecting the characteristics of the Panevėžys and Širvintos subdialects under investigation were noted while listening to dialect audio recordings; 2) the recorded words were transcribed using the transcription of the International Phonetic Alphabet (IPA) (see Figs. 4 and 5); 3) the data were entered into the MS Excel program and saved in Unicode (Unicode) (a text document was created); 4) the material was processed using the computer program Gabmap10; 5) a map of the research locations was created using the computer programs RuG/L0411 and Google Earth12, and was also processed using the computer program Gabmap (see Map 1 and Fig. 6). When interpreting the results, informants’ sociolinguistic information is not taken into account. 4 PAV. IPA balsių klasifikacija (Kasparaitis 2013: 4) In the computer program Gabmap, the basis for dialectometric analyses performed using different computational tools is the linguistic differences (distances) between pairs of locations under investigation. These are assessed by calculating the mean difference across the variables (i.e., words reflecting dialectal features) of the locations. Phonetic transcriptions are most often analyzed, with the Levenshtein distance (Levenshtein distance13) calculated (Leinonen et al. 2015: 6)14. The dialect-feature measurement tools available in the computer program Gabmap and the principles for calculating and interpreting differences used in this study are briefly described separately in each subsection of the following chapter. 10 Main developer: Çağrı Çöltekin. The computer program Gabmap. Available online: https://www.gabmap.nl/; Nerbonne et al. 2011: 65–89; Leinonen et al. 2015: 1–21. 11 Kompiuterinė programa RuG/L04. Prieiga internetu: http://www.let.rug.nl/~kleiweg/L04/. 12 Kompiuterinė programa Google Earth. Prieiga internetu: https://www.google.com/earth/. 13 Apie Levenšteino atstumo algoritmą žr. Kessler B. 1995: 60–66; Nerbonne ir kt. 1996: 185–202; Nerbonne ir kt. 1999: 5–15; Heeringa 2004: 121–124 ir kt. 14 Numerical data (vowel formant frequency vectors) and categorical data (lexical or syntactic) may also be examined (Leinonen et al. 2015: 6). Straipsniai / Articles 149 AGNĖ ČEPAITIENĖ 5 PAV. IPA priebalsių klasifikacija (Kasparaitis 2013: 5) MAP 1. Map of the LKA survey sites, created using the computer program Gabmap.15 FIG. 6. Excerpt from the map of the LKA survey sites 15 Here and in the other maps, RAA denotes Eastern Aukštaitians of Anykščiai, RAP – Eastern Aukštaitians of Panevėžys, and RAŠ – Eastern Aukštaitians of Širvintos. 150 Acta Linguistica Lithuanica LXXIV Kiekybinis diferencinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis 2. ANALYSIS OF DIFFERENTIAL DIALECTAL FEATURES 2.1. Statistical reliability of the calculations performed The reliability of the calculations performed was statistically assessed using the computer program Gabmap to calculate Cronbach’s alpha (Eng. Cronbach‘s alpha)16. This assesses not the reliability coefficient of the data, as is customary, but the reliability coefficient of the different measurements of the data as a whole. It depends on the quality of the data, the quality of the method used for the calculations, and the number of items studied. The coefficient can range from 0 to 1: the closer the value is to one, the more closely related the items under study are to one another17. It is desirable for Cronbach’s alpha coefficient to be at least 0.5. This value may vary slightly depending on the research objectives and the material being studied: the larger the number of items studied, the higher the coefficient (Vaitkevičius, Saudargienė 2010: 98). The computer program Gabmap calculated that the reliability coefficient for the different measurements of the research material does not meet the recommended standard—it is 0.04. It is assumed that the low coefficient may have been caused by the small number of units studied (26 words, used in 10 residential localities); therefore, the value obtained does not necessarily indicate low reliability of the data measurements. Another statistical measure, local incoherence (Eng. Local Incoherence), calculates the differences between units (the residential localities studied) in relation to the geographical distance between them. The optimal value is 0. The computer program Gabmap calculated that the local incoherence measure for the research material is slightly greater than 0—equal to 0.42. This, too, may have been caused by the small number of localities studied, and not all of them are located close to one another (the most distant locality is Stumbriškis). Although the amount of research material is small and the LKA sites are not located within a continuous area, the calculations performed using the computer program will be reliable for testing the hypothesis of this article. 16 Apie Kronbacho alfa koeficientą plačiau žr. Pukėnas 2009: 24–28. 17 In this study, the closer the value is to one, the more consistent and closely related the different data measurements are. Straipsniai / Articles 151 AGNĖ ČEPAITIENĖ 2.2. Levenshtein distance This algorithm calculates distances (differences) between the phonetic transcriptions of spoken words and estimates the minimum number of operations (insertions, deletions, substitutions) required to change one transcription unit into another (Kessler 1995: 62; Nerbonne et al. 2011: 65–89). In calculating Levenshtein distance, the geographic locations of the study sites on the map are taken into account—the greater the geographic distance between locations, the greater the Levenshtein distance (Nerbonne et al. 1999: 8). Since geographic distance affects the calculated value, the greatest Levenshtein distance for all the features examined separates Stum̃briškis, the locality farthest from the study sites, from the other LKA points (see Table 2). TABLE 2. Levenshtein distance between the LKA points under study18 Stum̃briškis | Šilai | Taujėnai | Deltuva | Jakūtiškiai | Kavárskas | Vidiškiai | Pabáiskas | Mančiūšėnai | Čiobiškis Stum̃briškis | — | 0,9 | 0,77 | 0,54 | 0,65 | 0,96 | 0,9 | 0,44 | 0,65 | 0,56 Šilai | — | 0,64 | 0,67 | 0,71 | 0,75 | 0,39 | 0,62 | 0,79 | 0,89 Taujėnai | — | 0,62 | 0,46 | 0,15 | 0,48 | 0,52 | 0,31 | 0,52 Deltuva | — | 0,67 | 0,77 | 0,44 | 0,56 | 0,5 | 0,71 Jakūtiškiai | — | 0,39 | 0,79 | 0,23 | 0,31 | 0,75 Kavárskas | — | 0,56 | 0,67 | 0,39 | 0,6 Vidiškiai | — | 0,58 | 0,71 | 0,58 Pabáiskas | — | 0,56 | 0,69 Mančiūšėnai | — | 0,48 Čiobiškis | — It can be seen that the Levenshtein distance is also quite large between LKA points belonging to the same subdialect, where Panevėžys dialect speakers live (Stum̃briškis and Šilai (0,9)19, Taujėnai (0,77), Jakūtiškiai (0,65), Deltuva (0,54); whereas, for example, the Levenshtein distance between Stum̃briškis, which belongs to the Panevėžys dialect, and Pabáiskas, where Širvintiškiai dialect speakers live, is 0,44). Such results are influenced not only by the greater geographic 18 LKA points where Panevėžys dialect speakers live are marked in dark color, those where Širvintiškiai dialect speakers live in light color, and Anykščiai dialect points in white (according to the traditional classification of Lithuanian dialects; see Fig. 3). 19 Here and elsewhere, the Levenshtein distance calculated using the computer program Gabmap is given in parentheses. 152 Acta Linguistica Lithuanica LXXIV Kiekybinis diferencinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis distance, but also the fact that Stum̃briškis belongs to the northern Eastern Aukštaitian Panevėžiškiai, whereas all the others belong to the southern group20. The results presented in Table 2 show that linguistic and geographic similarity is greater not between localities belonging to the same subdialect, but between the Panevėžiškiai and Širvintiškiai subdialects, that is, between Taujėnai and Kavársas (0.15), Jakūtiškiai and Pabáiskas (0.23), Taujėnai and Mančiūšėnai (0.31), and Jakūtiškiai and Mančiūšėnai (0.31). It is also evident that the same dialectal features as those found at the Panevėžiškiai LKA site are used in Kavárskas. The analysis also highlighted Vidiškiai’s similarity to the Panevėžiškiai: the smallest Levenshtein distance separates this LKA site from Šilai (0.39). The Levenshtein distance between Kavárskas and Vidiškiai is relatively moderate (0.56), which means that the speech of the people living there and speaking the dialect should be similar. Thus, the results show that the overall Levenshtein distance between the Panevėžiškiai and Širvintiškiai is small. Statistically, the features of these subdialects (taking into account the geographic distance between LKA sites) are similar and may point to a process of convergence between the subdialects taking place at the LKA sites. To further substantiate this claim, maps and graphs were generated using the computer program Gabmap, showing the Levenshtein distance from individual localities to other LKA sites. 2.3. Spatial and linguistic relationships between LKA sites Spatial and linguistic relationships are shown in reference-point maps and graphs. They indicate how much a selected locality (marked by a star on the map) differs from others in both dialectal and geographic terms. The darker the color on the map, the more similar the localities are. Map 2 and Figures 7 and 8 show the strongest linguistic and spatial relationships of the LKA site Kavárskas, discussed above, with Taujėnai (0.15), indicating that dialectal features of the Panevėžiškiai are used predominantly in Kavárskas. Features characteristic of the Širvintiškiai are also recorded here—this is indicated by the (markers) lower similarity of Kavárskas not only to Jakūtiškiai, which belongs to the Panevėžiškiai, but also to Mančiūšėnai, which belongs to the Širvintiškiai (0.39 each). In terms of the relationship between geographic distance and dialectal character, Kavárskas differs most from Stum̃briškis (0.96), possibly for the reasons discussed above. 20 The northern Eastern Aukštaitian Panevėžiškiai live north of Panevėžys, and the southern ones live south of Panevėžys. The main distinguishing feature of the northern and southern Panevėžiškiai is the different degree of vowel reduction (LKTCH 2004: 173). Straipsniai / Articles 153 AGNĖ ČEPAITIENĖ MAP 2. Geographic and linguistic links between Kavársko and other LKA points FIG. 7. Excerpt from the map of geographic and linguistic links between Kavársko and other LKA points FIG. 8. Geographic and linguistic links between Kavársko and other LKA points 154 Acta Linguistica Lithuanica LXXIV Kiekybinis diferencinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis MAP 3. Geographic and linguistic links between Vidiškiai and other LKA points FIG. 9. Excerpt from the map of geographic and linguistic links between Vidiškiai and other LKA points FIG. 10. Geographic and linguistic links between Vidiškiai and other LKA points Straipsniai / Articles 155 AGNĖ ČEPAITIENĖ Map 3 and Figs. 9 and 10 show that, in terms of geographic and linguistic similarity, the LKA point of Vidiškiai can be most closely grouped with localities in the Panevėžys dialect area—Šilai (0.39), Deltuva (0.44), Taujėnai (0.48), and Kavársko (0.56). This locality is separated from the LKA points in the Širvintos dialect area by a greater Levenshtein distance, and the greatest distance is from Stum̃briškis (0.9) (perhaps for the same reasons mentioned above). Thus, analysis of the geographic and linguistic links between Kavársko and Vidiškiai and individual localities shows that, from both geographic and dialectal perspectives, both LKA points can be grouped with the Panevėžys dialect area, rather than the Širvintos dialect area. Applying the same algorithm to unproblematic localities also reveals clear geographic and dialectal similarities between LKA points in the Panevėžys and Širvintos dialect areas. For example, Pabáiskas, classified as belonging to the Širvintos dialect area, is more similar in terms of Levenshtein distance not to a locality in its own subdialect area, but to Jakūtiškiai (0.23), located in the Panevėžys dialect area (see Map 4, Figs. 11 and 12). Pabáiskas’s geographic and linguistic links with the LKA points in the Širvintos dialect area are undeniable (it has moderately strong links with Mančiùšėnai (0.56), which belongs to the Širvintos dialect area); however, the LKA point under discussion has more similarities with settlements in the Panevėžys dialect area—Stum̃briškis (0.44), Taujėnai (0.52), and Deltuva (0.56). As in the case of Kavársko, this case also shows that Panevėžys dialect features may predominate at one LKA point, although features of the Širvintos dialect may also be recorded. Geographically and dialectally, Mančiùšėnai in the Širvintos dialect area is also closest to the Panevėžys-area LKA points of Taujėnai (0.31) and Jakūtiškiai (0.31) (see Map 5, Figs. 13 and 14). A somewhat greater Levenshtein distance separates Mančiùšėnai from Kavársko (0.39), Čiõbiškis (0.48), Deltuva (0.5), and Stum̃briškis (0.65), and the greatest distance is from Šilai (0.79). The results suggest that the LKA point of Mančiùšėnai, located in the Širvintos dialect area, has stronger geographic and linguistic links with the Panevėžys-area LKA points than with those in the Širvintos dialect area. Thus, more Panevėžys dialect features than Širvintos dialect features are also used in this locality. The case of Čiõbiškis is the only one in which an LKA point shows the greatest similarity to localities in the same subdialect area. Geographically and dialectally, it is most similar to Mančiùšėnai (0.48), which belongs to the Širvintos dialect area (see Map 6, Figs. 15 and 16). Yet Mančiùšėnai itself has its strongest geographic and linguistic links with LKA points in the Panevėžys dialect area (see Map 5, Figs. 13 and 14). A similar Levenshtein distance separates Mančiùšėnai from Taujėnai (0.52), located in the Panevėžys dialect area. 156 Acta Linguistica Lithuanica LXXIV A Quantitative Study of Differential Dialectal Features: The Case of the Eastern Highlanders of Panevėžys and Širvintos MAP 4. Spatial and linguistic connections between Pabaiskas and other LKA points FIG. 11. Fragment from the map of spatial and linguistic connections between Pabaiskas and other LKA points FIG. 12. Spatial and linguistic connections between Pabaiskas and other LKA points Straipsniai / Articles 157 AGNĖ ČEPAITIENĖ MAP 5. Spatial and linguistic connections between Mančiūšėnai and other LKA points FIG. 13. Fragment from the map of spatial and linguistic connections between Mančiūšėnai and other LKA points FIG. 14. Spatial and linguistic connections between Mančiūšėnai and other LKA points 158 Acta Linguistica Lithuanica LXXIV Kiekybinis diferencinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis MAP 6. Spatial and linguistic connections between Čiobiškis and other LKA points FIG. 15. Fragment from the map of spatial and linguistic connections between Čiobiškis and other LKA points FIG. 16. Spatial and linguistic connections between Čiobiškis and other LKA points Straipsniai / Articles 159 AGNĖ ČEPAITIENĖ Despite being separated by the greatest geographical distance among the localities studied, Čiobiškis is very similar to Stum̃briškis (0.56), located in the (northern) Panevėžiškiai area, and differs most from the somewhat closer Šilai (0.89). Čiobiškis is also quite similar to the LKA sites of Vidiškiai (0.58) and Kavārsko (0.6), which have stronger spatial and linguistic ties to localities in the Panevėžiškiai area. Based on the results, it can provisionally be argued that the speech of people living in Čiobiškis is dominated by more Panevėžiškiai than Širvintiškiai dialect features. Thus, the maps and graphs of reference points do not bring out separate Panevėžiškiai and Širvintiškiai zones; on the contrary, in terms of the relationship between geographical distance and dialectality, almost all the Širvintiškiai settlements studied are most similar to Panevėžiškiai LKA sites. Nevertheless, Levenshtein distance alone cannot establish which difference—geographical or linguistic—is greater between the localities studied. This can be determined using other difference-measurement tools in the computer program Gabmap. 2.4. Phonological differences between LKA sites Phonological differences are analysed using difference maps. These establish whether, in terms of overall phonological differences, the connections between any two neighbouring localities are strong. The darker the line connecting the LKA sites, the more similar the localities are. Map 7 and Fig. 17 show that the strongest connections are between Taujėnai and Kavārsko; connections that are almost as strong are between Jakūtiškiai and Pabāiskas, and between Jakūtiškiai and Mančiūšėnai. This indicates that the dialect features used in these neighbouring localities are the most similar. These results are consistent with the Levenshtein distance data discussed above (see Table 2). Thus, it can be stated with confidence that distinguishing dialect features may be equally frequent not only in localities belonging to the same subdialect, but also at individual Panevėžiškiai and Širvintiškiai LKA sites. An assessment of the results shows that nearby LKA sites belonging to the same subdialect are not always the most similar phonologically. This is illustrated by the cases of Šilai and Kavārsko, and Šilai and Jakūtiškiai, which belong to Panevėžiškiai, and Pabāiskas and Čiobiškis, which belong to Širvintiškiai. These localities are also separated by a large Levenshtein distance. The weakest connections are between the LKA sites of Stum̃briškis and Šilai, which have many phonological differences. These results can be explained by their belonging to different Eastern Aukštaitian Panevėžiškiai subdialects: Stum̃briškis belongs to the northern subdialect, while Šilai belongs to the southern one. 160 Acta Linguistica Lithuanica LXXIV Kiekybinis diferenciinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis MAP 7. Connections between neighboring LKA sites according to phonological differences FIG. 17. Excerpt from the map of connections between neighboring LKA sites according to phonological differences Thus, analysis of the map of differences based on phonological differences between neighboring LKA sites confirmed the results of the calculated Levenshtein distance—the dialectal similarity between the Panevėžys and Širvintos LKA sites. These trends also become apparent when the material is examined using other tools in the computer program Gabmap. 2.5. Cluster analysis of LKA sites according to the similarity of dialectal features21 The clustering performed by the computer program Gabmap is an alternative to the isogloss techniques used in traditional dialectology. However, clustering identifies boundaries that correspond not to a particular individual isogloss, but to the overall dialectal differences among the localities that make up separate groups (Nerbonne et al. 2011: 80). Clustering algorithms group the data so that dialectal differences within a cluster of the localities under study are as small as possible, and those between clusters are as large as possible (Leinonen et al. 2015: 11). 21 For discussion of cluster analysis applied in dialect studies and performed using another computer program—SPSS Statistics—see Pukevičiūtė 2015: 175–195. Straipsniai / Articles 161 AGNĖ ČEPAITIENĖ The data examined using the hierarchical cluster analysis methods of weighted average (Eng. Weighted Average) and Ward (Eng. Ward‘s Method) make it possible to test how the localities under study cluster according to the similarity in their use of dialectal features. When analyzing the material using the weighted average method, all clusters are treated as having the same value (weight), regardless of the number of objects in each cluster. When the material is analyzed using the Ward method, separate pairs of clusters are considered at each step of the analysis, and the sum of the squared deviations of the variables from the center of the combined cluster is calculated. This is done for each pair of clusters, and at the end those clusters are merged for which the sum of deviations is smallest. Data calculated using the two methods may not coincide, because differences are measured according to different principles (Prokić 2008: 156, Leinonen et al. 2015: 11)22 The clustering of the identified similar objects is presented in maps and dendrograms. The number of clusters identified is 2 (the material of two subdialects is studied). The cluster analysis map produced using the weighted average method distinguishes two groups of similar objects: Stum briškis and Čiobiškis form one group, while the remaining localities form the other (see Map 8 and Fig. 18). Thus, the differential dialectal features at the Panevėžys and Širvintos LKA sites may be very similar, regardless of their geographic distribution. MAP 8. Clustering of the LKA sites under study according to the similarity of dialectal features. Weighted average method23 FIG. 18. Excerpt from the map of clustering of the LKA sites under study according to the similarity of dialectal features (using the weighted average method) 22 Plačiau apie svorinio vidurkio ir wardo klasterinės analizės metodus žr. Prokić 2008: 153–172. 23 In the cluster analysis maps and dendrograms, the first group of similar objects is shown in a darker color, and the second in a lighter color. 162 Acta Linguistica Lithuanica LXXIV A Quantitative Study of Differential Dialectal Features: The Case of Eastern Aukštaitian Panevėžiškiai and Širvintiškiai The dendrogram graphically shows the order in which clusters were joined (see Fig. 19). It can be seen that in cluster 2, based on the similarity of dialectal features, Taujėnai and Kavárskas were joined in the first step, Mančiūšėnai and Jakūtiškiai in the second, and so on. The earlier the LKA sites are joined, the more similar the localities are from a dialectal perspective. FIG. 19. Dendrogram of cluster joining. Weighted average method The data obtained confirm that the localities under study belonging to separate subdialects may be more similar to one another than LKA sites belonging to the same subdialect. After applying the wardo method to group the localities, the results changed: based on the use of similar dialectal features, one cluster comprises Čiobiškis, Dél tuva, Stumbrìškis, Šilai, and Vidiškiai, while the other comprises the remaining localities (see Map 9 and Fig. 20). However, the same trend remains: the Širvintiškiai and Panevėžiškiai group together according to the similarity of their dialectal features. A comparison of the results of cluster analyses performed using different methods shows that a different principle for grouping localities led to the regrouping of Šilai, Dél tuv os, and Vidiškiai. However, as with the calculations performed using the weighted average method, the dendrogram of the cluster analysis performed using the wardo method also shows that similar or identical dialectal features are used in both the Panevėžiškiai and Širvintiškiai areas: the most similar are the LKA sites of Vidiškiai and Šilai, and Stumbrìškis and Dél tuv os (in cluster 1), as well as the localities of Taujėnai and Kavárskas, and Jakūtiškiai and Mančiūšėnai (in cluster 2); when the study sites were grouped, these were joined in the first steps (see Fig. 21). Thus, hierarchical cluster analysis using two methods shows that the dialectal features at the LKA sites in both the Panevėžiškiai and Širvintiškiai areas are very similar or even identical; slight differences in the pronunciation of individual words are galimai too scant a basis for distinguishing these subdialects. To substantiate the data obtained, the results of the cluster analysis must also be confirmed using other statistical methods, that is, by examining how the study objects are distributed in a multidimensional scale (Nerbonnė et al. 2011: 81). Straipsniai / Articles 163 AGNĖ ČEPAITIENĖ MAP 9. Grouping of the LKA sites under study according to the similarity of terminological features. Wardo method FIG. 20. Fragment of a map showing the grouping of the LKA sites under study according to the similarity of terminological features (using the wardo method) FIG. 21. Dendrogram of cluster joining. Wardo method 2.6. The location of LKA points on the multidimensional scale The multidimensional scale examines the extent to which dialects form a continuum. From a linguistic perspective, the most similar localities are positioned closer together on the plane, while dissimilar ones are farther apart. The diagram helps identify non-local residents and recognize dialect groups. It also shows which structures are clustered 164 Acta Linguistica Lithuanica LXXIV Kiekybinis diferenciacinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis groups established during the analysis, composed of LKA points that are similar from a dialectal perspective (Leinonen et al. 2015: 11). The results are statistically evaluated using the Rand index (in English, Rand index) (denoted by the letter r), ranging from 0 to 1: 0 indicates that the two clusters formed from the units under study do not match in any dialectal feature, whereas 1 indicates that the clusters under study are entirely identical (Prokić 2008: 161–163)24. The data obtained correspond to the arrangement of similar-object groups established by cluster analysis using the Ward method: the closest localities from a dialectal perspective are Šilai and Vidiškiai (although they are distant from the plane, they are linked by a small Levenshtein distance (0.39)), which belong to cluster 1, and Taujėnai, Jakūtiškiai, and Mančiūšėnai (the Levenshtein distance between these localities is also among the smallest), which belong to cluster 2 (see Fig. 22). FIG. 22. Distribution of LKA points on the multidimensional scale according to similarities in dialectal features 24 See also, for a fuller discussion, the formulas for calculating the Rand index. Straipsniai / Articles 165 AGNĖ ČEPAITIENĖ The LKA points of Stumbriškis, Deltuva, Pabaiskas, Čiobiškis, and Kavarskas, which belong to different subdialects, also fall outside the defined zones and stand out most among the localities under study. This might be explained by the fairly large Levenshtein distance between these and the other localities discussed. The computer program Gabmap calculated that r = 0.82; thus, the dialectal features used in the groups of similar objects formed from the LKA points differ only slightly. Thus, the multidimensional scale statistically substantiates the results of the cluster analysis and further highlights the similarities between the dialectal features of the Panevėžys and Širvintos subdialects. 2.7. Differential dialectal features of LKA points Differential dialectal features representing a dialect area are identified using the tools of the computer program Gabmap. The data are presented in four tables: the distinguishing dialectal features are indicated for two groups formed from LKA points and identified by cluster analysis using the weighted average method, and for two other clusters identified by cluster analysis using Ward’s method. The Between-cluster difference column (Lith. Between score) presents the differences between the selected cluster and the other group. The higher the score, the more pronounced the distinguishing feature (Lith. item) is among the LKA points that make up the selected cluster. The Variation within cluster column (Lith. Within score) indicates how much a particular feature varies within the selected cluster. The lower the variation score, the more representative the distinguishing feature. The Overall score (Lith. Score) indicates the overall difference between these two scores (Lith. Between–Within score) (Leinonen et al. 2015: 14). The results of the cluster analysis performed using the weighted average method show that the most salient distinguishing feature of the LKA points of Stumbriškis and Čiobiškis, which form the first cluster, is the lengthening of short stressed vowels (these locations are differentiated by the use of the word jis, and partly by didelis) (see Table 3). The overall score for this differential feature is high (more than 2), indicating that Stumbriškis and Čiobiškis differ substantially from the other locations under study (by slightly more than 1). Variation within the cluster is negative, which means that the differential dialectal feature is representative. The difference between clusters in the use of the word didelis, which reflects the same feature or ending reduction, is smaller than the difference in the use of the word jis. However, the pronunciation of this word varies less in Stumbriškis and Čiobiškis than that of jis, while the overall score for both measures is high. It can therefore be asserted that the use of didelis substantiates the fact that lengthening a short stressed vowel or reducing an ending are representative dialectal features distinguishing the LKA points of Stumbriškis and Čiobiškis from the others. 166 Acta Linguistica Lithuanica LXXIV Kiekybinis diferenciinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis TABLE 3. Differential dialectal features of cluster 1 (Stuṁbriškis, Čiobiškis). Weighted average method Overall score Variation within cluster Between-cluster difference Word / differential dialectal feature 2.065730 −0.734482 1.331248 He 0.991031 −1.057100 −0.066069 Big … … … … −1.485962 1.699261 0.213300 Is not The cluster is not at all characterized by the uneven pronunciation of short stressed final a, which is regarded as the basis for distinguishing the Panevėžiškiai and Širvintiškiai subdialects (as indicated by the use of the word nėra). Although a high score for variation within the cluster means that the pronunciation of nėra varies greatly in the LKA points of Panevėžiškiai Stum briškis and Širvintiškiai Čiobiškis (which would suggest that short stressed final a is pronounced differently at these LKA points and that the points (and subdialects) differ according to this dialectal feature), the overall score and the difference between clusters in this word’s pronunciation are very low (approximately −1.5 and approximately 0.2, respectively). Thus, this dialectal feature is unrepresentative: short stressed final a is pronounced almost identically in both Panevėžiškiai and Širvintiškiai locations. Lengthening short stressed vowels also identifies the area of the other LKA points under study that form another cluster (as indicated by the use of the words jis and, partly, turi) (see Table 4). The between-cluster difference, variation within the cluster, and overall score exactly match the scores for the distinguishing measures of the word jis in the first group. Thus, in both groups of similar objects, despite the LKA points belonging to different subdialects, the lengthening of stressed short vowels is a shared differential dialectal feature. Therefore, the locations do not differ with respect to this dialectal feature. TABLE 4. Differential dialectal features of cluster 2 (Šilai, Tau jėnai, Deltuva, Jakučiškiai, Kavárskas, Vidiškiai, Pabáiskas, Mančiūšėnai). Weighted mean method Overall score Variation within the cluster Difference between clusters Word / differential dialectal feature 2.065730 −0.734482 1.331248 He 0.516433 −0.218049 0.298383 Has … … … … −0.495996 0.202406 −0.293589 Very Straipsniai / Articles 167 AGNĖ ČEPAITIENĖ The unclear pronunciation of the first stressed falling-diphthong ai does not differentiate the cluster under discussion (as shown by the use of the word labai)—it is pronounced similarly in both clusters (the difference in its use between the clusters is small, about -0.3). This dialectal feature also varies little within the cluster composed of LKA survey points. However, the unclear pronunciation of the first stressed falling-diphthong ai is one of the differentiating dialectal features distinguishing the LKA survey points of Stumbrškis, Šilai, Deltuva, Vidiškiai, and Čiobiškis from the other cluster, according to the results of the cluster analysis conducted using wardo methods (as shown by the use of the word kaip; see Table 5). Slightly more representative is word-final reduction (as shown by the use of the word metų). The overall score for the distinguishing power of both dialectal features is high, and they vary little within the cluster; however, word endings are reduced similarly, and the first component of the stressed falling-diphthong ai is pronounced unclearly in both groups formed from LKA survey points (the difference between the clusters is about 0.3 and 0.5, respectively). TABLE 5. Differentiating dialectal features of cluster 1 (Stumbrškis, Šilai, Deltuva, Vidiškiai, Čiobiškis). Wardo method Overall score | Variation within cluster | Difference between clusters | Word / differentiating dialectal feature 1.189237 | −0.924962 | 0.264275 | Metų 1.086116 | −0.641110 | 0.445006 | Kaip... |... |... |... −0.494413 | 0.494413 | −0.000000 | Žmogus −0.494413 | 0.494413 | −0.000000 | Sako The LKA survey points forming the first group of similar objects are not distinguished by the backing of unstressed stem o a²⁵, word-final reduction, merger of short stressed word-final u, backing of unstressed word-final o to a, or retraction of stress. This is indicated by the zero score for the use of the words žmogus, sako in both clusters, meaning that the two clusters under study do not differ at all with respect to these features, which occur equally throughout all the localities studied. However, these features vary on average among the LKA survey points making up the first cluster under discussion. The unclear pronunciation of the unstressed component ai of the falling diphthong is also a distinguishing dialectal feature of the group of similar objects identified by the second cluster analysis conducted using wardo method (Taujėnai, Kavárskas, Jakūtiškiai, Pabáiskas, and Mančiūšėnai are differentiated by the use of the words gerai, kaip; see Table 6). It is also differentiated by stress 25 This dialectal feature has been proposed as a basis for distinguishing the Širvintiškiai and Panevėžiškiai subdialects (Markevičius 2002: 103). 168 Acta Linguistica Lithuanica LXXIV Kiekybinis diferencinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis retraction. In both clusters composed of LKA survey points, these dialectal features are used in similarly uniform ways (about 0.3 and 0.5), while their use varies to some extent in the localities under study that make up the second cluster (about -0.9 and -0.5), although the overall score is high. TABLE 6. Differentiating dialectal features of cluster 2 (Taujenai, Kavárskas, Jakūtiškiai, Pabáiskas, Mančiūšėnai). Wardo method Overall score Variation within cluster Difference between clusters Word / differentiating dialectal feature 1.189237 -0.924962 0.264275 Gerai 0.916410 -0.471405 0.445006 Kaip............ -0.496747 0.531640 0.034893 Žmonių The nondifferentiating dialectal features of this group partly coincide with those of cluster 1: backing unstressed stem o to a and word-final reduction (as shown by the use of the word žmonių). The pronunciation of this word differs only slightly between the two clusters and varies on average in cluster 2, while the overall scores for both indicators are low. Thus, neither word-final reduction nor the backing of unstressed stem o to a can be considered distinguishing dialectal features of Panevėžiškiai and Širvintiškiai. Although the results of analyses of differentiating dialectal features identified by methods based on different calculation principles do not coincide, they nevertheless reveal the same trends. The data obtained can be summarized as follows: since the similar-object groups identified by both cluster-analysis methods each include LKA survey points from both Panevėžiškiai and Širvintiškiai and are represented by the same dialectal features, those features are characteristic of both dialect areas and should not be regarded as differentiating features. The analysis also identified no other possible differentiating dialectal features that might distinguish the Panevėžiškiai and Širvintiškiai subdialects; both are spoken in the inhabited localities of Panevėžiškiai and Širvintiškiai. Therefore, it can tentatively be stated that there are no substantial differences between these subdialects. The article does not rule out the possibility that Panevėžiškiai and Širvintiškiai may previously have had more pronounced dialectal differences. However, the results of the study show that, if such differences did exist, differentiating dialectal features are no longer recorded in dialectal audio recordings from the late 20th to early 21st century. This may have been caused by various Straipsniai / Articles 169 AGNĖ ČEPAITIENĖ factors—for example, the gradual convergence of the two subdialects26. However, this is a subject for further and more detailed research, since, according to Peter Auer et al. (2005: 16), in order to describe linguistic changes as dialect convergence, the two subdialects would need to be compared diachronically in at least two respects. 3. GENERAL CONCLUSIONS The dialectal features of the traditional Eastern Aukštaitian Panevėžiškiai and Eastern Aukštaitian Širvintiškiai coincide from both a linguistic and a linguistic-geographical perspective. This is supported by the marked phonological similarities between the varieties under study and the small Levenshtein distance between the LKA points of these subdialects. The same dialectal features are statistically more often used not at LKA points belonging to the same subdialect, but in the residential areas of the individual subdialects under study. The possible distinguishing features of the clustered LKA points that emerged are shared by both subdialects and statistically substantiate their similarity. The analysis did not identify any dialectal features that could potentially distinguish the Panevėžiškiai and Širvintiškiai. Based on Levenshtein distance and statistical data, Kavárskas and Vidiškiai should be considered more similar to the Panevėžiškiai LKA points. To treat the Širvintiškiai as a separate subdialect, different differential dialectal features will have to be sought. LITERATŪRA Aliūkaitė Daiva, Bakšienė Rima, Jaroslavienė Jurgita, Judžentytė Gintarė, Leskauskaitė Asta, Lubienė Jūratė, Meiliūnaitė Violeta, Pakalniškienė Dalia, Ragaišienė Vilija, Rinkauskienė Regina, Švambarytė-Valužienė Janina, Urbanavičienė Jolita, Vaišnienė Daiva 2014: Punktų tinklo tankis geolingvistinės skirties aukštaičiai: žemaičiai požiūriu. – XXI a. pradžios lietuvių tarmės: geolingvistinis ir sociolingvistinis tyrimas (žemėlapiai ir jų komentarai), sud. Danguolė Mikulėnienė, Violeta Meiliūnaitė. Vilnius: Briedis, 129–196. Aliūkaitė Daiva, Mikulėnienė Danguolė 2014: Geolingvistika: ideologija, teorija ir metodai. Pagrindinės sąvokos. – XXI a. pradžios lietuvių tarmės: geolingvistinis ir 26 Kalbinė konvergencija vadinama kalbėtojo, kuris, siekdamas sumažinti skirtumus, prisitaiko prie kitų kalbos, kalbine apraiška (Auer ir kt. 2005: 6–7). Dar žr. ir Aliūkaitė, Mikulėnienė 2014: 44. 170 Acta Linguistica Lithuanica LXXIV Kiekybinis diferenciacinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis sociolingvistinis tyrimas (žemėlapiai ir jų komentarai), sud. Danguolė Mikulėnienė, Violeta Meiliūnaitė. Vilnius: Briedis, 29–47. Auer Peter, Hinskens Frans, Kerswill Peter, eds. 2005: Dialect change: Convergence and divergence in European languages. Cambridge: Cambridge University Press. Girdenis Aleksas, Zinkevičius Zigmas 1966: Dėl lietuvių kalbos tarminių klasifikacijos. – Kalbotyra 14, 139–147. Heeringa Wilbert 2004: Measuring Dialect Pronunciation Differences using Levenshtein Distance. PhD thesis. The Netherlands: University of Groningen. Heeringa Wilbert, Hinskens Frans 2014. Convergence between dialect varieties and dialect groups in the Dutch language area. – Aggregating Dialectology, Typology, and Register Analysis: Linguistic Variation in Text and Speech, ed. by B. Szmrecsanyi, B. Wälchli. New York: Walter de Gruyter, 26–52, 452–53. Heeringa Wilbert, Johnson Keith, Gooskens Charlotte 2009: Measuring Norwegian Dialect Distances using Acoustic Features. – Speech Communication 51, 167–183. Prieiga internetu: http://www.let.rug.nl/~kleiweg/L04/ Kasparaitis Pijus 2013: Kompiuterinės lingvistikos įvadas. Fonologija ir fonetika. Prieiga internetu: http://www.mif.vu.lt/~pijus/CLF/Fonetika.pdf Kessler Brett 1995: Computational dialectology in Irish Gaelic. – Seventh Conference of the European Chapter of the Association for Computational Linguistics, 60–66. Kompiuterinė programa Gabmap. Prieiga internetu: https://www.gabmap.nl Kompiuterinė programa Google Earth. Prieiga internetu: https://www.google.com/earth/ Kompiuterinė programa RuG/L04. Prieiga internetu: http://www.let.rug.nl/~kleiweg/L04/ Leinonen Therese 2010: An Acoustic Analysis of Vowel Pronunciation in Swedish Dialects. Doctoral dissertation. Groningen: University Library Groningen. Leinonen Therese 2011: Aggregate analysis of vowel pronunciation in Swedish dialects. – Oslo Studies in Language 3(2), 75–95. Leinonen Therese, Çöltekin Çağrı, Nerbonne John 2015: Using Gabmap. – Lingua (in press). Prieiga internetu: http://dx.doi.org/10.1016/j.lingua.2015.02.004 LKA 1977: Lietuvių kalbos atlasas 1. Leksika, ats. red. K. Morkūnas. Vilnius: Mokslas. LKA 1982: Lietuvių kalbos atlasas 2. Fonetika, ats. red. K. Morkūnas. Vilnius: Mokslas. LKA 1991: Lietuvių kalbos atlasas 3. Morfologija, ats. red. K. Morkūnas. Vilnius: Mokslas. Straipsniai / Articles 171 AGNĖ ČEPAITIENĖ LK TCH 2004: Lietuvių kalbos tarmių chrestomatija, sud. Rima Bacevičiūtė, Audra Ivanauskienė, Asta Leskauskaitė, Edmundas Trumpa. Vilnius: Lietuvių kalbos instituto leidykla. Markevičienė Žaneta 1999: Aukštaičių tarmių tekstai 1. Vilnius: Vilniaus universiteto leidykla. Markevičienė Žaneta, Markevičius Valdimantas 1995: Kelios pastabos apie Vidiškių šnektos balsių ir dvibalsių trukmę. – Kalbotyra 44(1), 69–72. Markevičienė Žaneta, Markevičius Valdimantas, Markevičius Aurimas 2009: Širvintiškių tekstai. Vilnius: Lietuvių kalbos institutas. Markevičius Valdimantas 2002: Kelios rytų aukštaičių širvintiškių ir pietų panevėžiškių tarmių vokalizmo ypatybės. – Baltistika 37(1), 101–105. Nerbonne John Colen Rinke, Gooskens Charlotte, Kleiweg Peter, Leinonen Therese 2011: Gabmap – A Web Application for Dialectology. – Dialectologia Special Issue 2, 65–89. Nerbonne John, Heeringa Wilbert, Hout Erik van den, Kooi Peter van der, Otten Simone, Vis Willem van de 1996: Phonetic distance between Dutch dialects. – CLIN VI. Papers from the sixth CLIN meeting, ed by. G. Durieux, W. Daelemans, S. Gills. Antwerpen: University of Antwerpen, 185–202. Nerbonne John, Heeringa Wilbert, Kleiweg Peter 1999: Edit distance and dialect proximity. – Time Warps, String Edits and Macromolecules: The Theory and Practice of Sequence Comparison. Second edition, 5–15. Prokić Jelena, Cysouw Michael 2013: Combining regular sound correspondences and geographic spread. – Language Dynamics and Change 3.2, 147–168. Prokić Jelena, Nerbonne John 2008: Recognizing Groups among Dialects. – International Journal of Humanities and Arts Computing, Special Issue on Language Variation, ed. by J. Nerbonne, Ch. Gooskens, S. Kürschner, R. van Bezooijen, 153–172. Pukėnas Kazimieras 2009: Kokybinių duomenų analizė SPSS programa: mokomoji knyga. Kaunas: LKKA. Pukevičiūtė Agnė 2015: Rytų aukštaičiai panevėžiškiai: klasterinė analizė ieškant tarmiškumo. – Acta Linguistica Lithuanica 72, 175–195. Scherrer Yves 2012. Recovering dialect geography from an unaligned comparable corpus. – Proceedings of the 2012 Joint Workshop of LINGVIS and UNCLH. Stroudsburg, PA: Assoc. Comput. Linguist, 63–71. 172 Acta Linguistica Lithuanica LXXIV Kiekybinis diferencialinių tarminių požymių tyrimas: rytų aukštaičių panevėžiškių ir širvintiškių atvejis Shackleton Robert George Jr. 2007: Phonetic variation in the traditional English dialects. A computational analysis. – Journal of English Linguistics 35, 30–102. Spruit Marco René, Heeringa Wilbert, Nerbonne John 2009: Associations among linguistic levels. – Lingua 119(11), 1624–1642. Stanford James N. 2012: One size fits all? Dialectometry in a small clan-based indigenous society. – Language Variation and Change 24, 247–78. Szmrecsanyi Benedikt 2011: Corpus-based dialectometry: a methodological sketch. – Corpora 6(1), 45–76. Urbanavičiūtė-Markevičienė Žaneta 1994: Dar kartą apie rytų aukštaičių tar-mių ribas. – Kalbotyra 43(1), 85–87. Vaitkevičius Raimundas, Saudargienė Aušra 2010: Psichologinių tyrimų duomenų analizė. Praktikos darbai. Kaunas: VDU leidykla. Zinkevičius Zigmas 1966: Lietuvių dialektologija. Vilnius: Mintis. Zinkevičius Zigmas 1994: Lietuvių kalbos dialektologija. Vilnius: MEL. Zinkevičius Zigmas 1997: Dėl rytų aukštaičių tarmės skirstymo. – Lietuvių kalbotyros klausimai 38, 237–241. Quantitative Research of Differential Dialectal Features: Case of Eastern Aukštaitians of Panevėžys and Širvintos SUMMARY The article presents the first-time use of the quantitative methods of dialectometry to analyse Lithuanian dialects. After quantifying the differential features of the Eastern Aukštaitians of Panevėžys and Širvintos subdialects, it aims to identify the dialectal dependence of Lithuanian Language Atlas (LLA) points under analysis – Stumbřiškis (171), Šilai (338), Kavárskas (340), Taujėnai (374), Vidiškiai (375 a), Deltuva (409), Jakūtiškiai (446), Pabaiskas (475), Mančiūšėnai (500), Čiobiškis (520). The research material consists of 12 hours of audio recordings of the Eastern Aukštaitians of Panevėžys and Širvintos. Differential dialectal features were selected from it: lengthening of the stressed shorts; shortening of the longs; pronunciation of the stressed short a, e, i, u in the ending; o instead of unstressed a; short e instead of ė; pronunciation Straipsniai / Articles 173 AGNĖ ČEPAITIENĖ of inarticulate first stressed / unstressed component; retraction of accent; reduction of endings. They are analysed by the web application Gabmap based on dialectometry methods. It is designed for the research of linguistic (and geographical) variability. On the basis of phonetic transcriptions of words, the Levenshtein distance is calculated between LLA points, the statistical computations are performed and the differential dialectal features of the Eastern Aukštaitians of Panevėžys and Širvintos are determined. The results suggest that the dialectal features of the traditional Eastern Aukštaitians of Panevėžys and Širvintos coincide from the linguistic and linguistic-geographical points of view. The closest locations by the Levenshtein distance measure are Taujėnai and Kavárs-kas (0,15), Jakūtiškiai and Pabaiskas (0,23), Taujėnai and Mančiūšėnai (0,31), Jakūtiškiai and Mančiūšėnai (0,31). All areas under analysis are linked by significant phonological similarities. Statistically equal dialectal features are frequently used not in the same LLA points of subdialects but in the localities inhabited by the speakers of separate subdialects under analysis. The potential differential features of clustered LLA points under analysis found in the study are common to both subdialects and statistically justify the similarity of the Eastern Aukštaitians of Panevėžys and Širvintos. Įteikta 2015 m. lapkričio 19 d. AGNĖ ČEPAITIENĖ Lietuvių kalbos institutas Petro Vileišio g. 5, LT-10308 Vilnius, Lietuva [email protected] 174 Acta Linguistica Lithuanica LXXIV