Language background

This illustration deals with one of the several Romani varieties spoken in Greece. Romani is an Indo-Aryan language whose speakers were migrants from India, probably belonging to service-providing castes (Matras Reference Matras2002). They first arrived in Greece during the period of the Byzantine Empire, around the 10th century ad. Romani was considerably influenced by Greek during this period particularly at the levels of lexicon and syntax. At the end of the Byzantine era, some groups migrated toward western and northern Europe and influences on Romani from additional contact languages began. Today Romani is spoken in several European countries as well as in the Americas (Adamou Reference Adamou2013), in Asia and in Australia, showing substantial dialectal variation.

Recent statistics indicate that approximately 150,000–200,000 Roma live in Greece (Bakker Reference Bakker, Extra and Gorder2001).Footnote ¹ The Romani dialects currently spoken in Greece belong to the Balkan and Vlax branches (for the classification of Romani dialects see Boretzky, Cech & Igla Reference Boretzky, Cech and Igla2008, Matras Reference Matras2010). The presence of Balkan Romani speakers in Greece has been continuous since Byzantine times (see Matras Reference Matras2004, for a description of the Parakalamos variety and Sechidou Reference Sechidou2011, for a description of the Ajios Athanasios variety). In contrast, Vlax Romani groups migrated to Greece between the end of the 19th century and until the Treaty of Lausanne in 1923 (see Igla Reference Igla1996, for a description of the Vlax Romani variety of Ajia Varvara in Athens). ‘Mixed Romani’ varieties are also spoken in Greece, combining Romani lexicon and Greek morphology (see Sechidou Reference Sechidou2005, for a short description of the mixed Greek-Romani variety of Finikas in Thessaloniki).

Here we focus on two closely related varieties of Xoraxane Romane (Turkish Romani) spoken in Greek Thrace. One of these varieties is spoken in the neighborhood of Anahoma in the town of Komotini and the other in the neighborhood of Drosero in the town of Xanthi 50 km from Komotini (see map in Figure 1). The variety spoken by the approximately 300 inhabitants of Anahoma is a Vlax variety though due to intermarriage and other connections, speakers show elements of other varieties of Romani in their speech. In particular, as is often the case with Roma communities, these Komotini Roma have mixed origins, for example, the older ones have relatives in various places in Greece and Bulgaria. At present, the group is closely connected to and intermarries with other Roma groups living in various neighborhoods of Komotini or in the nearby city of Xanthi. As noted, data were also collected in Drosero, a Roma settlement on the outskirts of Xanthi. The speakers in this settlement are also of various origins and, importantly, their variety shows a greater mixture of features from various Balkan and Vlax Romani varieties than that spoken in the smaller community of Anahoma.

Figure 1 Map of the area of Thrace, Greece (shown also in darker shade in the inset map of Greece). The recordings took place in the towns of Xanthi and Komotini.

In both communities, most speakers are trilingual in Romani, Turkish, and Greek, usually in this order of acquisition. The speakers themselves distinguish between ‘pure Romani’ and Xoraxane ‘Turkish’ on the one hand, and their own Xoraxane Romane, or ‘Turkish-Romani’, on the other. These mixed varieties used by the communities discussed here are heavily influenced by contact with Turkish since the Ottoman times, as can be seen in both the lexicon and grammar (Adamou Reference Adamou2010). At this point it is not possible to talk of simple borrowings from Turkish, as this implies full integration into the Romani system of phonology and morphology. Although such integration does apply to nouns as far as morphology is concerned, it is not the case for verbs which transparently retain Turkish morphology when used in Xoraxane Romane. Further, borrowed Turkish words are largely not adapted phonologically; instead, new phonemes borrowed from Turkish are now part of the phonological system (though other aspects of Turkish phonology, such as vowel harmony, have not become part of Xoraxane Romane, as shown in more detail below). At present, the influence of Turkish is so strong that these varieties of Romani are endangered due to a shift of the speakers to Turkish.

The data presented in this paper were collected during five fieldwork visits carried out by the first author between 2005 and 2010. Our corpus includes data from 12 speakers producing speech in various styles. In particular, the corpus includes the following: (a) natural dialogs involving eight female speakers and one male speaker; (b) story-telling by two male and one female speaker; (c) elicitation data using the questionnaire of the Intercontinental Dictionary Series (Ritchie Key & Comrie, http://lingweb.eva.mpg.de/ids/) and elicited from one male speaker; (d) elicitation data using the QUIS questionnaire (Skopeteas et al. Reference Skopeteas, Fiedler, Hellmuth, Schwarz, Stoel, Fanselow, Féry and Krifka2006) and elicited from two female and three male speakers; (e) recordings of words in isolation elicited specifically for this illustration from one female speaker. With the exception of one male and one female speaker who were teenagers and one female speaker in her fifties, our speakers were in their twenties or thirties (below speakers are referred to by gender and number, e.g. ‘MS1’ stands for Male Speaker 1). Due to the living conditions of the community some of the recordings, especially those excerpted from natural dialogues and story-telling, are quite noisy.

Consonants

Xoraxane Romane has in the native part of its vocabulary voiced, voiceless and aspirated stops and affricates, voiced and voiceless fricatives, and also nasals, laterals, a palatal approximant and an alveolar tap. Most consonants present substantial variation in pronunciation. This variation is often due to context-dependent allophony, but in several instances no context-dependent reasons for the variation could be identified. This applies particularly to the series of aspirated obstruents, especially the affricates, which can be produced with or without aspiration even within the same utterance. In addition, variation is common across speakers. Such variation applies not only to the pronunciation of particular segments but also to the exact form of words and even to word choice, a point about which speakers can feel quite strongly. For example, FS1, who recorded the words used here for illustration purposes, refused to use the word /ˈkavako/ ‘tree’ (a borrowing from Turkish) and offered instead the form /koˈpatʃi/ which is of Romanian origin. Similarly, while MS1 pronounced ‘wind’ as [baɫˈbaɫ]] and ‘brother’ as [p ^h al], FS1 consistently uses [baɫˈvaɫ] and [p ^h ral], respectively. Such differences abound in the corpus.

Obstruents are lenited in various ways, particularly word-finally. Voiced plosives and fricatives become devoiced word-finally, e.g. [daˈdesko] ‘dad-gen’ but [dat] ‘dad.nom’, [aˈlevi] ‘flame.f’ (with Romani inflection for feminine gender) but [aˈlef] ‘flame’. Aspirated plosives are likely to be pronounced without significant VOT duration utterance-finally (though they are released), e.g. [maˈk ^h a] ‘flies’ but [mak] ‘fly’, [boˈk ^h av] ‘I am hungry’ but [bok] ‘hunger’. In addition, the VOT of word-medial aspirated plosives and of the affricate /tʃ^h/ can vary quite substantially across speakers and tokens (even within the speech of one speaker).

Voiceless unaspirated plosives become voiced before nasal consonants, e.g. /ˈlen-ke/ > [ˈlenɟe] ‘to them (3pl-dat)’ (compare /ˈtu-ke/ > [ˈtuce] ‘to you (2sg-dat)’). More generally, voiced plosives may be pronounced with some prenasalization used as a mechanism for sustaining voicing. On the other hand, voiced plosives can be lenited to fricatives intervocalically, e.g. [p ^h aˈga] or [p ^h aˈɣa] ‘armpits’.

Velar plosives are produced as palatals when followed by the front vowels, /i/ and /e/ (and front vowels borrowed from Turkish; see below), e.g. [manˈgav] ‘want.1sg’ vs. [manˈɟesa] ‘want.2sg’, [c ^h eɾ] ‘house’, [ceɾ] ‘do.imp.2sg’, [ɟiˈve] ‘day’. However, morphological alternations lead on occasion to forms in which palatal plosives are (on the surface) followed by back vowels as well, e.g. [buˈca] ‘jobs’ ([buˈci] ‘job’).Footnote ²

Postalveolar fricatives also show some variation in quality, with the pronunciation of the same speaker alternating between postalveolar and more fronted articulations. This variation applies also to affricates, which can be treated as fronted postalveolars or retracted alveolars. The uvular fricative /χ/ also shows variation, often being pronounced as a velar. This variation appears to be speaker-specific and could be related to the use of Greek (which has velar but not uvular fricatives). Thus, we find that some speakers, e.g. one of our male speakers, are more likely to adopt a more fronted articulation while others use the uvular pronunciation more systematically, e.g. MS4 [ˈχurdo ˈχeɾ] ‘small donkey’, FS1 [aˈɾaχni] ‘spider’, MS1 [xuaˈɾav] ‘drill.1sg’.

In addition to its pronunciation as a tap, /ɾ/ is often pronounced as a short alveolar trill [r] showing two to three cycles. This often happens before and after stops (e.g. [oˈpre] ‘above, up’, [drak] ‘raisins’, [marˈno] ‘bread’), but it can also be the result of emphasis particularly word- and utterance-initially, e.g. [rat] ‘blood’. Intervocalically, the tap can also be lenited to a short approximant, while it is largely devoiced utterance-finally.

The alveolar lateral approximant /l/ is clear when followed by the front vowels /e/ and /i/ but heavily velarized when preceded or followed by the back vowels [a o u] (similar to Turkish; Zimmer & Orgun Reference Zimmer and Orgun1999: 155), e.g. [baɫˈvaɫ] ‘wind’, [zuɾaˈɫo] ‘strong’, [p ^h raɫ] ‘brother’. The same applies to loanwords from Turkish, e.g. [buˈɫuto] ‘cloud’, [jɯɫˈdɯzi] ‘star’. However, there is variation among speakers: some speakers do not show this allophonic alternation at all and use a clear [l] in all cases, as in Standard Greek, others show evidence of harmony, avoiding velarization when there is a clear [l] in the same word; consider [ɫoˈɫo] ‘red.m’, but [loˈli] ‘red.f’.

Xoraxane Romane also shows variation in the realization of /j/. The most typical form of /j/ is as a palatal approximant, but on occasion it can be lenited to the point that it becomes barely audible. On the other hand, there are instances in which it is produced as a palatal fricative instead. As a result, a word like /jek/ ‘a, one’ can show extensive variation depending on prosodic position and speaker: utterance-initially it is most likely to be pronounced [ʝek] or [jek] (although [ek] is also attested utterance-initially in our corpus); phrase-medially it is more likely to be [ ek] or [ek]. At the same time, /j/ can be used as a dummy onset for onsetless loan words, such as Turkish /ˈama/ rendered either [ˈama] or [ˈjama] ‘but’).Footnote ³

Vowels

Xoraxane Romane has the traditional vowels of Romani, five monophthongs, /i e a o u/. Of these five monophthongs, /a/ is a central somewhat backed low vowel, /i/ is a high front vowel, while /e/ and /o/ are rather close in quality; /u/, on the other hand, is rather centralized and relatively low. Data from one of our speakers (MS1), based on between four and six tokens per vowel and stress condition, are shown in Figure 2.

Figure 2 Mean F1 and F2 for the five monophthongs of Xoraxane Romane based on the speech of one male speaker (MS1). Symbols in black represent vowels in stressed positions; symbols in gray represent vowels in unstressed positions.

Consonants and vowels in borrowed vocabulary

The above inventory pertains largely to the Indo-Aryan lexicon and the numerous integrated borrowings from earlier contact languages, including Persian (e.g. /amˈbɾol/ ‘pear’), Armenian (e.g. /gɾas/ ‘horse’), Byzantine-era Greek (e.g. /ˈpapu/ ‘grandfather’), Romanian (e.g. /pleˈtʃinta/ ‘pie’), and Ottoman-era Balkan Turkish (e.g. /maaˈla/ ‘neighborhood’). As noted, however, these Romani speakers also use contemporary Istanbul Turkish extensively in their everyday speech, including a significant number of non-integrated Turkish verbs, as well as Modern Greek vocabulary (for more details see Adamou Reference Adamou2010).

Interestingly, the speakers neither fully adapt these code-switch insertions to native phonology nor do they strictly follow the phonology of the original language. Thus, the Turkish vowels, [y ɯ œ] which do not belong to the traditional system of Romani are used in words of Turkish origin, e.g. [kɯr ɟyn] ‘forty days’ (from Turkish kırk gün), [saˈɾɯ] ‘yellow’ (from Turkish sarı). However, the speakers preserve vowel harmony only if the words are not adapted to Romani morphology; if they are, harmony is lost. For example, from Turkish yıldız ‘star’, Xoraxane Romane obtains [jɯɫˈdɯzi] ‘star.f’ (with feminine suffix -i), which contains both front and back vowels. Similarly, [ˈœɾymˌdʒava] ‘spider’ (from Turkish örümcek) and [œkˈsyska] ‘orphan.F’ (compare [œkˈsyzɯ] ‘orphan.m’, from Turkish öksüz) contain vowels that do not agree either in roundness or backness.

On the other hand, Turkish /h/ has not been adopted in Xoraxane Romane either in early or late borrowings from Turkish. Thus, Turkish mahal becomes /maaˈla/ ‘neighborhood’, hep is /ep/ ‘always’, hap is /ˈapo/ ‘pill’, and hayvan becomes [ajvaˈnato] ‘animal’. In addition, some Turkish clusters show metathesis under borrowing, e.g. Turkish anlamayacak becomes [alˈnamaˌdʒak] ‘he/she will not understand’.

Similarly, in Greek code-switching insertions, some Greek consonants, such as [ɣ θ ð], which do not have obvious correspondences with Romani phonemes are found in Xoraxane Romane, e.g. [dziˈɣaɾe] ‘cigarettes’ (from Greek τσιγάρα [tsiˈɣaɾa]), [pinaˈciða] ‘sign’ (same as Greek πινακίδα), [ˈirθane] ‘came.3pl’ (same as Greek ήρθανε). Others, however, such as the voiceless velar fricative [x] are not systematically used and are often replaced by native phonemes, e.g. Greek λάστιχο [ˈlastixo] ‘hose’ is rendered as [laˈstika] and λέσχη [ˈlesçi] ‘club’ as either [ˈlesci] or [ˈleksi] (with metathesis); on the other hand, αράχνη [aˈɾaxni] becomes [aˈɾaχni] ‘spider’ at least for some speakers.

Word stress

Romani has stress, in that typically one syllable of each word stands out from the others, by means of being longer in duration and showing more extreme vowel quality.Footnote ⁴ Similarly to other conservative Romani dialects (Matras Reference Matras2002: 62–63), Xoraxane Romane has mostly final stress in the native parts of its vocabulary, though the addition of certain suffixes to roots can also lead to forms with penultimate stress, e.g. [maˈnuʃ] ‘human’ vs. [manuˈʃ-es-ki] ‘human-obl-gen’, [man] ‘me.obl’ vs. [ˈman-sa] ‘me-instr’, [ka aˈvel] ‘will come.3sg’ vs. [ka aˈvel-as] ‘will come.3sg-pst’.

Stress can also fall on earlier syllables in non-native vocabulary items up to the fourth syllable from the end of the word, e.g. [asteˈnava] ‘hospital’ (from Turkish hastane), [ˈkavako] ‘tree’ (from Turkish kavak ‘poplar’), [korˈkodilo] ‘crocodile’ (from Greek κροκóδειλος /kroˈkoðilos/), [ˈmejdanuko] ‘valley’ (from Turkish meydan ‘forum’), [ˈkunuluko] ‘land’ (from Turkish, exact origin unknown). Variation in the position of stress in loans can also arise due to the addition of suffixes, e.g. [ˈapo] ‘pill’ but [ˈapoɾa] ‘pills’ (with a Romanian plural suffix). Despite variability in the position of stress, minimal pairs are rare, e.g. [ˈkana] ‘when’ vs. [kaˈna] ‘ears’.

Although in most cases words are produced with only one stress, certain items are systematically produced with both a primary and a secondary stress. This pattern is found in both native and borrowed items, e.g. [ˌkorkoˈri] ‘alone’, [alˈnamaˌdʒak] ‘he/she will not understand’, [ˈœɾymˌdʒava] ‘spider’.

In addition, variation in the location of stress exists for some words. For example, [ˈmami] ‘grandmother’ has penultimate stress for MS1 who is from Komotini and has final stress on most items of the native vocabulary, while the conservative form [maˈmi] is attested among the Xanthi speakers. Similarly, where Xanthi speakers make a distinction based on stress for [χurˈdo] ‘child’ and [ˈχurdo] ‘small’, Komotini speakers encode the difference using distinct lexical items, [χurˈdo] ‘child’ and [sikˈno] ‘small’. In our corpus we have also found many instances in which the position of stress is variable in some words even in consecutive utterances of the same speaker, e.g. the name of one of our speakers is pronounced by FS5 as both [ˈsabia] and [saˈbia] during the course of the same conversation.

In most cases the reasons for this stress shifting are not entirely clear (though they could be related to the mixed dialect background of our speakers). In some instances at least shifts appear to be related to focus and accent, with words that are being focused also showing a shift of their stress to an earlier syllable than the one habitually carrying stress. This is illustrated in the first three examples in Table 1 (see also Arvaniti & Adamou Reference Arvaniti, Adamou, Washburn, McKinney-Bock, Varis, Sawyer and Tomaszewicz2011). However, we have also found examples where a metrical interpretation seems more likely. For example, in /ikaˈlda kaˈtar an jek saˈno bal/ (Table 1), the speaker phrases the adjective /saˈno/ ‘thin’ and the noun /bal/ ‘hair’ separately and accenting both. However, in /ni pukaˈvel laˈke kaj si ˈsano bal/, the adjective and the noun are phrased as one prosodic word and the stress of /saˈno/ shifts to the first syllable though in this case neither ‘thin’ nor ‘hair’ are in focus. In Table 1 we give some examples of these alternations showing the most frequently attested pattern on the left and the shifted pattern on the right.

Table 1 Examples of stress shift.

Transcription of the recorded passage