Aspects of the Early History of Romani

We owe to Ralph Lilley Turner the correct classification of Romani as originating from a central or inner form of Indo-Aryan. Turner also clarified that the “Dardic” elements in Romani have been borrowed into early Romani after its speakers had left their original home and reached the north-west of South Asia where they stayed for several hundred years before finally leaving the subcontinent. Until now, the extent of the “Dardic” influence on early Romani was poorly understood. In the present article much data has been put together which shows that this impact indeed is considerable. But it is intelligible only if we accept Turner’s hypothesis of a long stopover in north-western South Asia. The data presented below will also show that the notion of “Dardic” is too narrow in this context: the impact on early Romani, in fact, comprises linguistic elements and features found in Nuristani, Dardic and West Pahāṛī.


Introduction
I have been working for some time on languages within the three Indo-Aryan subgroups 1 Nuristani, Dardic and West Pahāṛī. 2 They are spoken in the mountainous tracts between the Yamuna valley in the south-east and eastern Afghanistan in the north-west. Recently I have also been working with speakers of Norwegian Romani in a project financed by the Norwegian Research Council-to whom I want to express here my gratitude. It lent itself for me to combine the two fields of work and reconsider the question of the impact of languages of north-western South Asia on early Romani. 3 It has been known for a long time that some influence does exist, but no detailed study has been done. The present article attempts to do this and I will show that not only Dardic but also Nuristani and West Pahāṛī need to be kept in view. My conclusions will be that their impact on early Romani is much more extensive than previously assumed.
The article begins with a succinct reevaluation of some questions concerning the early history of Indo-Aryan. It then recapitulates the most important arguments for the most likely geographical area from where the speakers of Romani must have originated. It will be seen that I agree to a large extent with the thesis suggested by R. L. Turner (1927), namely that Romani originally belongs to Central Indo-Aryan, respectively to the so-1 The structuring of these subgroups is discussed in section 2. 2 The presentation of the language names follows the system found in R. L.
Turner's A Comparative Dictionary of the Indo-Aryan Languages. There are, however, additional languages not found in Turner for which I have used simplified transliterations. For the varieties of Romani I have used own abbreviations. All language names, their linguistic affiliations and the abbreviations are found at the end of the article before the bibliographical references. Note also that if a word is said by me to be found, e.g. in Norwegian Romani this does not exclude the possibility that it may also be found in other varieties. 3 Throughout this article I use the notion '(languages of the) north-west' for the geographical area of the mountainous tract between the Yamuna valley in the south-east and eastern Afghanistan in the north-west. This covers the linguistic area of West Pahāṛī, Dardic and Nuristani. The notion 'west' I use for the catchment area of the Indus River.
called inner branch of Indo-Aryan (see section 3). By clarifying that Romani originated in Central Indo-Aryan and not in the area of the Dard languages in north-western South Asia, Turner rectified previous views held for instance by F. Miklosich, R. Pischel, and G. A. Grierson (see literature). I will support Turner's view with much data which prove a considerable influence on early Romani by Indo-Aryan languages which are found today in the mountains between the valley of the Yamuna and eastern Afghanistan.

Early Indo-Aryan
Instead of summarising the complex debate on the migration of speakers of Indo-Aryan into South Asia (see for instance Bronkhorst and Deshpande, Masica, Erdörsy, Renfrew, Marcantonio) I want to point out here that the whole discussion (including by authors who reject such an immigration) usually stops short with the beginning of Old Indo-Aryan in north-western South Asia. This gives the impression as if the further development of Indo-Aryan would not be of interest for this debate. But this is not the case. In a nutshell: (a) the fact that the most archaic forms of Indo-Aryan are found in north-western South Asia, and (b) the fact that it is possible to distinguish branches of Indo-Aryan whose vanishing point directs exactly to that north-western area are additional arguments that make it impossible to assume that the original home of Indo-Aryan is located elsewhere in South Asia.

The debate on inner and outer languages
Mainly in the fifth and sixth chapters of Linguistic Archaeology in South Asia the author Franklin C. Southworth discusses a controversial hypothesis of George Abraham Grierson. Grierson had suggested dividing New Indo-Aryan into three subgroupings which he called midland languages, intermediate languages and outer languages (Southworth 2005: 130). According to this model, West Pahāṛī would be an inner (or midland) and Kashmiri, generally classified as Dardic, would be an outer language (see Masica 1991: 451 for a diagram). Grierson's hypothesis was not widely accepted because it was argued that there is too much diffusion and overlaying between the different IA languages so that no clear picture can emerge. 4 It would be too long-winded to present here all the details of how Southworth not only defends Grierson's hypothesis but, in fact, modifies and places it on safer ground with the help of additional data. The main arguments for distinguishing between inner and outer languages (plus a transitional zone) are the following. All or most of the outer languages have (in the transitional zone the evidence is ambiguous) the following features which are missing with the inner languages (only the most relevant are quoted here): past forms in -l-; gerundive, nominal and future forms based on OIA ˗(i)tavya-; particular behaviours of the OIA vowels r̥, i and u; lexical evidence. Discussing questions of early dialect variations Southworth says (2005: 155): "In interpreting earlier evidence it is important to note that some scholars, for instance Chatterji, have assumed (tacitly or explicitly) that Pali and the Prakrits represent a stage intermediate between the earliest Indo-Aryan and the modern spoken languages. Others take the position that, from the Vedic period onwards, there were varieties of Indo-Aryan which were outside the "high" tradition … If this was true in Vedic times, it would have been even more true during the MIA period when the Indo-Aryan languages were spread over a much larger territory. Thus it is reasonable to assume that along with the attested literary Prakrits there were also "colloquial Prakrits" which never appeared in writing." Relevant here is that with regard to the Ashokan dialects, Southworth arrives at the following conclusion (2005: 167f.): "Bloch's three-way division of the Ashokan dialects (Center-East, Northwest, West) can be resolved into an earlier two-way division between the Northwest and the remaining dialects." North-west is reflected in the inscriptions of Shahbazgarhi and Mansehra, and the remaining dialects are classified by Southworth as West (Girnar, Sopara), Midland (Kalsi) and East (Dhauli, Jaugada). It is a well-known fact that the Dard languages are modern descendants of Prakrits more or less close to the Ashokan Northwest dialect (that is, Gāndhārī). Southworth, however, does not include Dardic in his model. Instead, he gives several reasons in the fifth chapter of his book (p. 149, footnote 9) why he excludes the Dard languages from his further investigations: fragmentary knowledge about past forms in -l-in Dardic; inadequacy of the descriptive material; questionableness of Dardic being a genuine subgroup of Indo-Aryan. The author has already been criticized for this (see the review of Kulikov) and a look at his diagram on page 168 is revealing. The diagram illustrates, with the help of a series of isoglosses, the above-mentioned division between the north-west and the remaining dialects. So what he does here is to almost exclude from his modernized inner-outer languages model MIA Gāndhārī and NIA Dardic because they share so few isoglosses with the rest. On page 169f., Southworth summarizes: "Evidence for the existence of two distinct sociolinguistic regions, inner (North-Central) and outer (South-Eastern) Indo-Aryan was presented … The totality of the evidence points to the existence of two sociolinguistic regions, each showing some internal uniformity vis-à-vis the other, which however were probably in at least intermittent contact throughout most of their history … The evidence of Vedic dialects … does not conflict with, and possibly supports, the inner-outer group hypothesis, in that the major dialect division in the late Vedic period is between a midland dialect and an eastern-southern dialect, with a transitional dialect in the area of Kosala, the modern Avadh-exactly where Grierson placed his intermediate group. The northern Panjab, less active at this time in terms of text production, forms a separate dialect area." Southworth's goal is not just to present additional evidence for Grierson's original thesis but he also suggests (2005: 181ff.) "[a] reconstruction of the prehistory of outer Indo-Aryan." Here follows a very short synopsis. On page 181 he says in the section on Indo-Aryan in the Indus valley: "By 1500 BCE, when the first hymns of the Rigveda are believed to have been composed, that portion of the Indo-Aryan speech community which was associated with the OIA texts was located in the upper Indus Valley … Given the archaeological evidence for intrusive Central Asian elements on the lower Indus … it can only be assumed that OIA speakers also occupied this area by the end of the second millenium BCE … as the OIA 'mainstream' society expanded eastward across the Indo-Gangetic divide … its counterpart in Sindh probably did the same, following the route mentioned here, leading to Malwa, Gujarat, and the Deccan." According to Southworth's model, the southern migration movement turned south and ultimately east towards the eastern limits of the subcontinent, and both (language) movements met and mixed in the transitional area of Kosala (Avadh).

A third branch in Indo-Aryan
In a forthcoming book I will demonstrate in great detail the existence of a third branch of Indo-Aryan (besides the inner and the outer branch) which I call the north-western branch. It includes Nuristani, Dardic and West Pahāṛī. The inclusion of Nuristani seems to contradict the widely shared opinion that Nuristani constitutes a separate branch within Indo-Iranian. It is certainly true that only Nuristani reflects a pre-Old Indo-Aryan stage; however, Nuristani has to be included in the north-western branch due to the overwhelming number of features it shares with Dardic and West Pahāṛī in grammar, vocabulary and a common cultural heritage of which traces can be found at different places. Nuristani branched off at the time of Proto-Aryan, but its present geographical adjacency to Dardic must be very old. In fact, I believe that Nuristani never got spatially much separated from Dardic. If I may employ here an image: Nuristani, Dardic and West Pahāṛī are like three siblings, Nuristani being the eldest, Dardic the middle and West Pahāṛī perhaps the youngest. This image works reasonably well only with regard to the preservation of archaisms. But to be regarded as a separate north-western branch this is not sufficient. It has to be shown that these three language groups also share innovations which are not found in the other Indo-Aryan languages. I do this in my forthcoming monograph; here it has to suffice to just list some of these innovations. The most important are: (a) preservation of a three-or two-stepped system of sibilants (e.g. s, š, [ṣ]) which facilitated the innovation of a three-or two-stepped system of affricates (e.g. dz, dž, [dẓ]) as a result of depalatalization which had an impact on all three language groups over a long period of many centuries; (b) converbs (absolutives) and past forms built with an element -t-; (c) use of a non-aspirated auxiliary tū 'is; was' which historically derives < OIA sthitá-'standing, settled'. In addition to these innovations there are a number of other innovations whose geographical extension within Nuristani, Dardic and West Pahāṛī is not as comprehensive as the three quoted elements; still they too are only found there. Pointing out that the three language groups share a common vocabulary is, at first sight, an argumentum e silentium; however, it is the sheer amount of shared vocabulary which is necessarily convincing. Romani participates in this, as I will show in this article, and there are even a few instances of the common north-western cultural heritage that come out in Roma traditions. Thus I argue that Romani belonged originally to the inner group of Indo-Aryan, but it has been influenced quite strongly by languages of the north-western group. This can only be explained by assuming that the speakers of early Romani stayed, after having left their original home, over a long time (perhaps several centuries) in the north-west of South Asia.

Romani
Romani belongs originally neither to the outer nor to the northwestern branch of Indo-Aryan. This means that the speakers of Romani originated from the area of the inner branch. Their original home could have been somewhere in the area where today Hindi is spoken. 5 In order to substantiate this I summarize here Turner's most important arguments (1927) for allocating Proto-Romani to the inner group: 6 Early innovations 1. Syllabic OIA r̥ got changed at an early stage into a, i, or u; however, there are geographical differences: it changed into a in the south-west and south, elsewhere into i or u, and in the north-west it was partially preserved as ri. In Romani, r̥ got changed into i or u: kišlo 'thin' < OIA kr̥śá-'lean, thin', bukko 'intestines' < OIA vr̥kká-'kidneys'. Turner says (1927: 8) that Romani ric 'bear' (OIA r̥ḱṣa-'bear') is, like Hindi rīch, a loan from a Hill language. However, very similar forms are found in many other modern Indo-Aryan languages, and there is Prakrit riccha-. It is therefore unclear when and where the speakers of early Romani borrowed this word. 2. The consonant group OIA r̥t led to retroflexion both in the east and in the north-west, whereas the dental was preserved in the central and south-western languages (although there are exceptions like Hindi and Panjabi maṛak 'plague' < OIA mr̥takka-). Romani agrees with the central and south-western languages: mulo 'dead' < OIA mr̥tá-'dead' where the -l-goes back to older -t-and not -ṛ-(which would have resulted in ˗r˗). 3. The OIA sound kṣ developed either into ch (c̣h) or kh, but the geographical picture is quite confusing. In the north-west there is a strong tendency for ch (c̣h), but both Hindi and Romani usually show kh: Romani jakh and Hindi ākh both 'eye' < OIA ákṣi-'eye'. Turner lists four Romani words where OIA kṣ appears as č(h), but it also does so in central languages like Hindi: rič ' bear' (see above 1.), čhar 'glowing ashes' < OIA kṣārá-'corrosive' (Hindi chār 'alkali, ashes'), čhuri 'knife' < OIA kṣurá-'razor' (Hindi churā 'dagger, razor'), čulo 'a little' < OIA kṣulla-'small' (Hindi chullū 'childish').
4. The OIA consonant groups śm, ṣm, sm later on developed into sp, ss (śś), mh, pph, mbh. Again it is difficult to draw a clear picture. The change to sp and ss (śś) is largely limited to the north-west (but see the wide dispersal of OIA raśmí-'rope' as raśśi, rassi etc.); there is only little evidence for pph and mbh; the most frequent and widespread is mh which also applies for Romani: ame 'we' < OIA *asme. Also here Romani agrees with the central languages like Hindi, but disagrees with the north-western languages including Sindhi, Lahnda and Panjabi. 5. For the development of the OIA consonant groups tv, dv, tm there are only very few cases. Thus it suffices to say that for OIA tm Romani has p as in po or pes (oblique) 'self' < OIA tmán-'one's own person'. According to Turner (1927: 14) here Romani differs from Dardic. However, there is Dardic Torwali pāe, Kashmiri pān 'self', West Pahāṛī Pŏgulī panun 'own' and Sindhi pāṇa 'reflexive pronoun'. So the Romani word po is probably a loan word from the north-west. 6. The change of OIA initial y-to j-is very widespread in Indo-Aryan but has not (completely) taken place in Dardic, Sindhi and Sinhalese. Romani belongs to the majority group: džov 'oats' < OIA yava-'barley'. 7. Intervocalic OIA -m-has been preserved in Dardic, Sinhalese and, to a certain extent, in West Pahāṛī. Otherwise it changed into a nasalized vowel and in Romani the nasalisation got subsequently lost: kovlo 'soft' < OIA komalá-'tender, soft'.

Conservations
1. Turner states (1927: 17): "Romani preserves -t-, probably -d-, perhaps -th-, and less certainly -dh-, under the form l in the European and Armenian dialects and r in the Syrian." For instance gili 'song' < OIA gīti-'singing', len 'river' < OIA nadī́ 'river' (with metathesis Turner concludes from these observations (1927: 22f.) that at the time of the Ashoka inscriptions the clusters with sibilants had already been changed except in Girnar (south-west) and Shahbazgarhi (north-west). Since Romani cannot be associated with either, he concludes-in my eyes absolutely correctly-that the speakers of Romani must have left their original home already before the time of the Ashoka inscriptions. He further observes that at about 250 AD -d-and perhaps -t-still turned up in Kharoṣṭhī documents of Khotan where also the clusters with sibilants and clusters with stops and -r-survived. Then he concludes that the speakers of Romani, which possessed the above features at the time of the departure from their original home, could preserve them in the north-west (where they stayed a considerable time) when the features disappeared in their original home.

Later innovations
1. The first innovation here Turner discusses is this (1927: 24): "A breathed consonant preceded by a nasal has been voiced in Sindhi, Lahnda, Panjabi, the whole Dard group (except perhaps Gawar-Bati; and not in Kafiri), and all the Pahari dialects as far as, and including, Nepāli (except for a few small enclaves). This is the normal treatment of Romani." For example: dand 'tooth' < OIA dánta-'tooth', bango 'crooked' < OIA vaṅka-*'bent, crooked'. However, Masica (1991: 203) points out that there are several West Pahāṛī languages where this process has not taken place, and for instance in Bangani there are a number of doublets with unvoiced and voiced stops. We are dealing here with an incomplete phonological process. According to Turner, this sound change is not found in the Ashoka inscriptions but only by the time of the Kharoṣṭhī documents. He takes this as a possible hint that the speakers of Romani had reached the north-west of South Asia before Ashoka, because then "… it is not surprising that they should have shared subsequent innovations of that linguistic area" (1927: 24). Then follows the first example for this claim: 2. Metathesis of r. Turner states (loc. cit.): "In Sindhi, Lahnda, Dardic, and West Pahari, when the group r + consonant or consonant + r occurs in the middle of a word, the r is transposed (after the accompanying consonant has been doubled) and pronounced after the first consonant of the word. Thus Si. ḍrigho 'tall' (dīrgháḥ > dīrggháḥ > drīgha) …" 8 An example from Romani: tradel 'to drag, drive away' < OIA tárdati 'sets free'. Another example, where there is, however, metathesis without a "supporting" consonant, is brivel 'to comb wool' < OIA *vivarati 'uncovers'. 9 3. Initial OIA v-has become b-in the central and eastern groups as well as in Pahāṛī, Ḍogrī and many Dard languages probably at a relatively late time. The original approximant has been preserved in the west and various parts of the north-west. This inconsistent picture seems to be reflected in the fact that we find b-in European Romani but v-in Syrian and Armenian Romani: European berš but Syrian vars both 'year' < OIA varṣá-'rain'. 4. The Turner article contains a small section on morphology. An interesting point made here by him is the fact that in European Romani the nominative singular masculine ends in -o because this may be a hint that Romani did not belong to the Māgadhī area where the ending is -e.
Summarizing the above points, Turner concludes (1927: 31) that there was "… an original connection with the Central group, and a subsequent migration to the North-west group." 10 Then Turner continues to substantiate his claim with regard to vocabulary where he demonstrates that the core vocabulary of Romani belongs to the central group. He continues on p. 32: "… it would not be surprising to find that the Gypsies had borrowed some words from the North-western languages, among which they must have lived for several centuries after leaving the Central group." He shows that the words for 'four' and 'six' come from the north-west (the former seems to be connected with Nuristani).
The few examples given by Turner will be supplemented in this article with many more words which further corroborate his view.

Words already known (or suspected) to be of IA origin and which (may) have been borrowed from the North-western branch
Here the main sources of information are Turner's Comparative Dictionary of the Indo-Aryan Languages, Norbert Boretzky and Birgit Igla's Wörterbuch Romani -Deutsch -Englisch für den südosteuropäischen Raum, and Mathias Metzger's 'Etymological glossary of Indic words in Romani' (besides the sources quoted in the Literature section). Note that in the following section the known borrowed words are not just listed, but usually supplemented with additional information. It will also be seen that it is not always clear whether a Romani word is inherited or borrowed. Concerning the above-mentioned sound shift of dental stops to -lboth in Romani and in Dardic I assume that Romani words displaying it are all borrowings from the north-west because: (a) there is no evidence for a parallel development to this in the inner languages; (b) European Romani shows parallels with Dardic regarding this sound shift in case of medial dental stops (initial stops were not affected in European Romani and Dardic); (c) Armenian Romani shows parallels with some varieties of Nuristani and some East Iranian languages because initial dental stops also underwent this change there. The following words are ordered according to the North Indian alphabets. 11

Rom. arakhel 'to protect; to find'
Compare OIA ārakṣati 'watches over, defends' (1298). 12 Modern descendants of the verb are only found in Waigali (arac̣-'schützen') and Sinhalese; and of the noun OIA ārakṣa-'protection' (1297) again only in Waigali and Sinhalese. Since, however, the Romani word has kh for OIA kṣ and not an affricate as Waigali, it is not clear whether it is inherited or borrowed.

Rom. avdive(s) 'today; in the course of time'
This is a compound connected with OIA *ā nūnam 'up to now' (1180) + divasá-'day' (6333). Sub 1180 see Ashkun yanū̃ and Waigali önū́ both 'today'. Forms without prefixed nūnam 'now' 11 I have cautiously standardized the many different transliterations and transcriptions used in the various quoted sources. Care should be taken that in Romani transcription the sound j is an approximant, whereas the same letter is a palatal voiced affricate in languages spoken in South Asia; the Romani sibilant š is equal to the Indo-Aryan sibilant ś, Romani c corresponds to Indo-Aryan ċ, whereas č corresponds to c and ć.

Rom. ukljel 'to climb, ascend'
Compare OIA *utkalati 'goes out or up', utkalita-'rising, prosperous', utkalyate, utkālayati 'drives out' (1716). Modern descendants are found in many Indo-Aryan languages but not in Nuristani, Dardic and Sinhalese. However, the meaning 'to climb, ascend' is only shared with Pahāṛī languages: West Pahāṛī Koṭgaṛhī ukəḷno 'to climb, ascend' and Bangani ukḷɛṇɔ 'to mount (as a bull a cow)', Gaṛhwālī ukaḷnu 'to climb, ride' and Nepāli uklanu 'to ascend'. Still it cannot be said with certainty whether the Romani word is inherited or a borrowing.

Rom.G. and Rom.Germ. ulo 'born'
The word derives < OIA bhūtá-'become, been, past' (9552); cf. also Rom.H. ūló 'was' and other Romani varieties ulō '(he) became'. Besides the -l-, which reflects OIA -t-and which is thus an indication for borrowing from the north-west, it is also the semantics which points into the same direction. There is Bangani ɔṇɔ 'to be born' and uaṇɔ 'to give birth', Deogari uaɔṇɔ 'to give birth' and Khaśdhari hui 'delivered (said about a baby)'. Even though these three languages are spoken in the same area and I am not aware of further semantic parallels in the north-west, it is quite likely that the Romani words are borrowings from the north-west.

Rom. kiral 'cheese'
See OIA kilāṭa-'inspissated milk' (3181); modern descendants are only found in Nuristani and Dardic; the Romani form is thus a borrowing (it underwent metathesis in the second syllable).

Rom. koř 'neck'
The word derives < OIA kroḍa-'breast, bosom' (3607) but has semantic parallels with the meaning 'neck' in modern languages only in Dardic and Kashmiri: Kalasha kŕuŕa, Phalūṛā kirṓṛ, Kashmiri koru all 'neck'. This semantic change is apparently due to overlap with phonologically similar derivations < OIA kr̥kāṭikā-'joint of neck' (3419) which are limited to Nuristani, Dardic and West Pahāṛī. Thus the Romani word is a borrowing from the north-west. Note, however, that Rom.T. has, besides korri 'neck', also the word kirko 'throat'. This resembles e.g. Bangani kerkɔ 'neck' and Waigali kir'ik 'Nacken' (both derive < 3419) so that the Rom.T. word may have changed its original meaning through the influence of korri. Thus, also this word is a borrowing from the north-west.

Rom. xanřudel 'to scratch, scrape'
The word belongs to OIA kaṇḍū-'itching, the itch' (2688) + the verb del. 13 The derived OIA lemma kaṇḍūyáti ʻscratches' (2689) has several modern descendants with initial aspirated stops in the west and north-west: Shina of Gilgit khaṇóĭki̯, Sindhi khanhaṇu 'to scratch', khanvaṇi 'scratching', Panjabi khanūhṇā 'to itch', the Kocī variety of West Pahāṛī khəṇamiṇe 'itching', khəṇamiṇo 'to itch', the Koṭgaṛhī variety of West Pahāṛī khəṇeuṇo. According to Boretzky and Igla, the word was perhaps also influenced by Persian xeneš 'Jucken' and perhaps even by Armenian xanʣ-el 'versengen; anbrennen'. And according to Turner (Addenda and Corrigenda), the aspirated forms of the IA words are influenced by OIA kharju-¹ 'itching, scratching, scab' (3827) or khára 'hard, sharp, pungent' (3819). This, together with the fact that the lemma is not found in inner NIA, suggests that the Romani word is a borrowing from the north-west or west, but got perhaps also influenced by words from other languages.

Rom. gili 'song'
See Indus Kohistani gī̀l 'song', Šāṭōṭī gī̀lⁱ and Savi gíli 'song' which are < OIA gīti-'singing' (4168). According to Turner the Dard forms have an extension -l-, but this is untenable because of the final high front vowel in Šāṭōṭī and Savi, and because it would mean that the considerable number of words e.g. found in Indus Kohistani with -l-going back to a dental stop would all be borrowings from an unknown Dard language. Turner quotes also Rom.Syr. gref and asks "whence f?" Compare this, however, with Rom.N. jilipa 'song' and Rom.S. gijepa 'song' both of which employ the well-known suffix -(i)pa.
In connection with 'song' there are also interesting words for 'musician': The common Romani term is bašaldo, 14 but there is also Rom.T. bašado 'violinist' and Rom.S. bàši-moš 'gårdsmusikant'. Rom.S. has in addition a word ghildo, translated as 'fest, party'. This, however, cannot be the original meaning which must have been 'musician'. Compare the designation Ghilabari for a Roma group living in Romania who are professional musicians (Berger 1985: 779); their name corresponds to that of the South European Gitanes, which also means 'musicians'.
The Rom.S. word ghildo has an exact correspondence in Indus Kohistani gildṑ 'singer'. The aspiration in ghildo and in Ghilabari is spontaneous (more on this see below in section 11) and has correspondences in West Pahāṛī Chinali ghit 'song' and ghitāru 'singer', and in Paṅgwālī ghīt 'song'. The first element of bašaldo is related to bašalel 'to play (an instrument)' which in turn is related to bašel 'to bark, roar, howl' which derives < OIA vā́śyate 'roars, howls, bellows, lows, bleats, sings (of birds)' (11589). The word bašalel probably contains an -l-transitive/causative suffix. Modern descendants of OIA vā́śyate are limited to Nuristani, Dardic, West Pahāṛī, Kumaoni, Nepāli and Gujarati; it is thus not a central language word.
The words bašaldo, ghildo and gildṑ are compounds with a second element -do. This element derives < OIA dádhāti 'places, lays on, gives, seizes' (6145), but regarding the exact meaning one needs to consider the meanings given for the precursor PIE *dheh₁-, namely 'to put, lay down, sit down, produce, make, speak, say, bring back'. The OIA verb has (almost) exclusively survived in Nuristani and Dardic whereas in the other Indo-Aryan languages it was displaced by the very similar OIA dádāti 'gives' (6141). That we are indeed dealing with OIA dádhāti 'places' is corroborated by Indus Kohistani gīlmā̀r 'singer' the second component of which comes from mārʌ́ṽ 'to kill (< OIA māráyati¹ 'kills' [10066]) which also means, e.g. in Hindi mārnā 'to perform an action with vigour'. Thus, the underlying meaning of gildṑ, ghildo, gīlmā̀r is something like "one who produces/belts out/performs a song". It seems that we are dealing here with a very old compound. And indeed, the compound has a striking parallel compound building in the Celtic word bard which goes back to PIE *gʷr̥h₂-dʰh₁-o-'praise-maker' (West 2007: 27). Whereas the first components of the compounds obviously have several different words associated with 'song' (in case of the Celtic word it is 'praise', in case of the Indic words it is 'sing'), it is remarkable that the second component in the PIE reconstruction is exactly the same verb which we identified above: PIE *dheh₁-'to put, produce, speak etc.'. Of course I am not in a position to say whether we deal here with a common Indo-European heritage or whether this is a matter of two independent developments at the two ends of the Indo-European world.

Rom. gošni 'cowdung'
The word is a compound going back to OIA gó-'cow' (4255) plus *śakana-'dung' (12238). The latter is, according to the information in Turner, limited to Nuristani and Dardic (but a sideform *chakana-is found in Marāṭhī and Koṅkaṇī). There is also, but with a deviating semantics, Bangani goṣṇi 15 'outdoor fireplace during monsoon in which dried cowdung is burnt in order to keep away biting flies' which is an extension of gośu 'dried cowdung' which itself derives < OIA gośakr̥t-'cowdung' (4333). This lemma is found in the north-west but also in Hindi. The likelihood that this is a north-western borrowing is perhaps enhanced by the fact that similar compounds tend to be typically found in the north-west. In any case, a direct parallel is Ashkun ga-sā̃ 'cowdung' (with trace of a nasal consonant), and semantically comparable is Khashi kuster 'dunghill' which goes back to OIA *go-stara-(see 13685) (devoicing of mediae is quite common in Khashi).

Rom. čhela 'smallpox'
The word is a combination of elements of a derivation < śītalā-'(goddess of) smallpox' (12490) and < śītala-'cold' (12487). 16 Only in Dardic languages has this lemma semantically split into 'cold' and 'fever', compare Gawar-Bati šal 'fever' but šalá, šolɔ́ 'cold', Savi šal 'fever' but šalo 'cold'. In the other modern IA languages only the meaning 'cold' is found. The original meaning 'cold(ness)' is preserved in Romani šil which is < OIA śīta-'cold' (12485). A parallel to the affricatized Romani word is found in Kalasha and Burushaski čilá 'cold, cold season'. For the Romani word Boretzky and Igla suggest irregular phonological development in a taboo word, but now we see that these irregularities are geographically located in the north-west from where this Romani word must have been borrowed. I am not aware of a śītalā cult beyond Kashmir, but ideas concerning smallpoxthe goddess both sends out and heals the disease-may have been more widespread. When she disperses smallpox this is accompanied with fever; and if she heals the disease she does so with cold water.

Rom. džamutro 'son-in-law'
The word derives < OIA jā́mātr̥-'daughter's husband' (5198). It looks like a north-western borrowing because Romani does not preserve medial -m-; however, also in some other languages which usually have lost OIA -m-the nasal consonant has been preserved as in Hindi jamāī. Thus, the exact origin of the Romani word is unclear.

Rom. thaj 'and, also', Rom.T. the 'and'
The word may derive < OIA tathāpi 'even so, nevertheless' (5647), but Turner is doubtful about the whole lemma. Still, what speaks in favour of a borrowing is first the fact that the meaning 'and' is limited to the north-west; second the Romani word possibly displays aspiration fronting (more on this below in section 10), which is a characteristic feature of Dardic and West Pahāṛī.

Rom. dad 'father'
Compare OIA *dādda-'father or other elderly relative' (6261). The word might be a north-western loanword since the meaning 'father' is limited to that region.

Rom. nilaj 'summer'
Compare OIA nidāghá-'hot season' (7193) and nidāghakāla-'heat of summer' (7194). Turner is here slightly confusing as very similar Romani forms are quoted under the two lemmata. The first lemma is basically limited to Nuristani and Dardic, but it is also found in Oriya, however with a different meaning. I want to add here that a modern derivation of nidāghakāla-is also found in West Pahāṛī Khashi (or Khashali?) nēlā 'summer' (see Kaul 2006 I: 335). Thus the Romani word is most likely a borrowing from the north-west.
muṣṭīke 'fist' and in the Shina of Gures and Kohistan mŭṣṭăk 'fist'. The Romani word thus appears to be a borrowing from the northwest.

Rom. mol 'wine'
The word may derive < OIA mádhu-'honey, mead' (9784). A direct parallel is found in Prasun mū́lu 'wine' and there is Burushaski mel 'Wein (aus Trauben)', but there is also Persian mul 'wine'. So it is not clear whether the word was borrowed in the north-west or in Iran.

Rom. lima 'mucus, phlegm'
The word derives < OIA śleṣmán-'mucus, phlegm' (12744); similar looking derivations are only found in the west and northwest: Khetrānī lim 'marrow', Ḍumāki līma 'mucus from nose', Lahnda lim 'phlegm, mucus from nose', and West Pahāṛī Khashi and Bhadrawāhī limm 'mucus of nose'. Even though the phonological change ṣm > m can be inherited, the geographical limitation of lima forms to the west and north-west makes it likely that this lemma is a borrowing.

Rom. šerand 'pillow'
The word derives < OIA *śiraanta 'head-end' (12448). It could be inherited from a phonological point of view; however, the fact that modern descendants are only found in Sindhi, Lahnda, Panjabi and West Pahāṛī makes also this word a strong borrowing candidate.

Rom. sasto 'healthy'
The word derives < OIA svastha-'well, healthy' (13917). It could be inherited from a phonological point of view; however, the fact that modern descendants are only found in Dardic makes also this word a strong borrowing candidate.

Rom.G. seli 'bran'
Compare OIA satīná-'the pea Pisum avense' (13116). The OIA lemma was so far not known to have a modern descendant besides Romani. But there are Bhadrawāhī setu and Bhalesi sete both meaning 'bran' (Kaul 2006 I: 327). Both words have preserved the original -t-(which is quite rare in West Pahāṛī, but there are other cases as well, and see the preceding entry). The background of this lemma is complicated by Rom. šelī (also Rom.T. šelja) 'bran' and Rudhari śeli 'grain, bran (of maize)' (Kaul: 2006 II: 274). Turner derives the Romani form < OIA śadaka-'unhusked corn' (12287), but that cannot be the origin of the Rudhari word if one does not want to assume an isolated borrowing from an unknown Dard language since West Pahāṛī does not know a historical change -t-, ˗d-> -l-. On the other hand, the Romani word šelī/šelja does seem to have been borrowed from an unknown Dard language. So the matter is really unclear. Note, however, that the -e-in the words seli, setu, sete, śeli, šelī/šelja possibly resulted from the same phonological process, namely a so-called epenthesis (which is found in many areas where Dardic and West Pahāṛī is spoken; it is discussed below in section 13).

Words not yet known to be of IA origin and also borrowed from north-western languages
This section runs the risk of being characterized of using the argumentum e silentium. Indeed, this danger cannot be completely ruled out as it is always possible that new evidences for words at unexpected places come up. However, it is unlikely that all of the words presented below would finally be found to be known also in the inner group. This is simply also not possible because some of the words have undergone sound changes which definitely have not occurred in the inner group. Also here I include words from all branches of Romani.

Rom.Arm. akli 'a lie; untrue; insufficient'
Because of the preservation of the -k-I assume a borrowing from a north-western language even though the word is also found in the inner group. It derives < OIA alīká-'unpleasing; untrue; a little' (718).

Rom.Arm. anlēš 'paradise'
There is no straightforward etymology for this word, but it may belong to OIA anudeśa-in the more literal sense of *'adjoining land'. On the one hand, OIA anudeśa-has only the technical meaning 'reference to something prior'. On the other hand there is, however, Shina õṍšo (Turner: ōśŭ) '(male) guest' which Turner derives < OIA *apadeśya-'foreign' (427) and which has been borrowed into Burushaski as oóśin 'Besucher, Gast'. However, a derivation < OIA anudeśin-'residing at the same place'-but here with a suggested meaning *'belonging to an adjoining area'-is more convincing also with regard to the initial nasalization of the Shina word (in the north-west dēś frequently means 'village (and adjoining area)'). Admittedly, the palatal sibilant in anlēš cannot be the same as the OIA one and is probably an Armenian suffix (cf. Rom.Arm. lehi, leji 'village' and for possible suffixes Finck 1907: 50; and cf. Pašaī dē, Khowar deh and Rom.Syr. dē all 'village'). I may also point out here that the word deś is (or rather was) used by the Prasun people in Nuristan in "urdesh" 'heaven' and "yurdesh" 'paradise' and 'hell' (Jettmar 1975: 51f.).

Rom.Burg. eklik 'a little'
The word consists of two elements: ek-'a; one' (< OIA *ēkka 'one' [2462]), and regarding the second element compare Indus Kohistani lʌk hu 'small; (a) little', Iranian Saka laka 'little, small amount', Burushaski luk 'a little', Woṭa-pūrī lukúṛ 'small'. In case of Bashkarīk lúkuṭ 'shorter, younger', Khashi lokuch 'smaller' and Kashmiri lŏkoṭ ͧ 'small, shorter, younger' comparative suffixes have been added. According to Turner < OIA *lukka-¹ 'defective' (11072), but this is implausible because the geographical distribution of this word with the meaning 'small' is limited to the north-west and to Saka. This makes it clear that it cannot belong to 11072, whatever its origin. Thus the Rom.Burg. word may be a borrowing from the north-west.

Rom. kermuso 'mouse, rat'
According to Turner, the first syllable is < OIA ghara-'house'. But compare Pašaī kavar-mūs 'rat' and Persian karmūš 'muskrat'. It is unclear whether here also belongs Rom.T. maskariš 'mouse', but the forms suggest a compound word of unclear derivation but borrowed from somewhere in the north-west.

Rom. čičalo₂ 'meat'
This word must have a different origin than the preceding one even though the two have been put together by Boretzky and Igla, and even though it also is of onomatopoetic character. It belongs to a fairly large group with examples known to me in the area between Ossetic and West Pahāṛī. Thus it seems to be a Wanderwort of unknown origin: Bangani cĩcaũ 'meat' (children's language), Deogari cī̃cī̃ 'piece of meat' (children's language), Koṭgaṛhī cī̃ 'meat, cooked meat', Indus Kohistani čʌčā̀ 'meat' (children's language), Burushaski and Shina ćáća 'Fleisch' (Kindersprache), Ossetic dzidza 'Fleisch' (Babysprache).

Rom. čhungar 'spit, saliva'
There may be a connection with Indus Kohistani čurùk ͪ karʌṽ 'to spit by pressing the saliva between the teeth in order to create a whizzing sound', perhaps also ċʌrʌk ͪ karʌṽ 'to squirt', and Bangani curuk-curuk 'sound of spitting and verbal curkaṇɔ 'to spit'. But a derivation is unknown.

Rom.Arm. santhu 'oven'
Compare Bangani sɔndaṇɔ 'to warm or heat up (e.g., an aching limb through a hot compress)', and (poetic) sɔdaɔ and sɔdar 'funeral pile'. Turner postulates sáṁdahati 'burns up' (12899a) with one modern descendant in Maldivian. The Bangani forms, however, rather derive < OIA sáṁdāhayati 'to cause to burn up'. The Rom.Arm. word could be inherited, but the scarce evidence just from the fringes of Indo-Aryan makes borrowing not unlikely.

Rom.Arm. samli karel 'to fabricate, prepare, make, build'
OIA saṁmāti 'construct, grant' (12975) has the past participle sáṁmita-'of the same measure' (Pali sammita-'measured'). The Armenian form must derive from the past participle. Regarding the meaning, cf. Phalūṛā samū́m 'I build, arrange (house, bed)' (sub 12975) and Bangani sɔminuaṇɔ 'to manufacture, make; to repair, join (together), fit together (as carpenter)'. The Bangani word is a synonym verbal compound with independent nuaṇɔ meaning 'to bend'. The Armenian form is certainly a loanword since the meaning 'construct' is limited to the north-west and the -l-seems to reflect an old -ta-.
Rom.Arm. sol, söl 'loud; voice, word, narration, news'; plus lel 'to take' it means 'to revile'; plus ǵrel 'to call, shout, sing' (which also has this meaning); soli plus karel 'to do' means 'to speak, narrate' The forms are connected with OIA ślóka-'sound, hymn of praise' (12748). Since modern descendants are limited to the north-west this must be a loanword.

On the original home of Romani
As has been pointed out by Turner (see section 3), Romani originally belongs to what he calls the Central group of Indo-Aryan. This is perhaps not exactly the same as Grierson's division of New Indo-Aryan into three subgroupings which he called midland or inner languages, intermediate languages and outer languages, but it is also not very different. We have already seen that Romani has borrowed a substantial amount of words from the north-western branch, and below more data will be furnished. However, I want to stress in this section that Romani contains a substantial amount of words which are apparently limited to the inner languages. This might be taken as additional corroboration for the assumption regarding Romani belonging originally to the inner group. The following words appear to me as potential candidates for the inner group as none of them has descendants in the outer languages: 20

Rom. čunř/čunřa 'tress, plait'
One finds here some closely related OIA forms-cū́ḍa-¹, 'protuberance on brick; topknot on head', *coṇḍa-, *coṭṭa-, *cunda-(4883)-with a main meaning '(knot of) hair'. According to Turner (who quotes Mayrhofer) this is a Dravidian lemma; but the problem is that its alloforms have quite different geographical distributions. This suggests overlap of more than one lemma: The relatively early attested cū́ḍa-¹ is also the most widespread allomorph with modern descendants found in the triangle between Nuristani, Assamese and Maldivian. The allomorph *coṇḍa-, from which the Romani forms derive, is largely limited to the central languages (and Dardic).

Rom. bilal v.i. 'to melt, thaw' and bilavel v.t 'to melt'
The words derive < OIA vilīyate 'is dissolved, melts' and vilāpayati² 'dissolves, melts' (11906). Modern descendants are predominantly found in Nuristani, Dardic, Central and Eastern 20 It is very likely that the inner (or central) languages exerted a heavy influence on the Pahāṛī languages and large parts of the north-western languages. Influence in the other direction was certainly much less pervasive. Thus, as long as there are no clear markers identifying a word as belonging to the north-west, the following examples having such parallels can be either borrowings from the central area into the north-west or they belong to a common heritage.
Pahari, and Central NIA.; but there are no descendants in the outer languages. 21

Rom.Ger. daro 'tree'
Not mentioned by Turner, 22 < OIA dā́ru-'piece of wood' (6298). Modern descendants are found in Dardic, Pahāṛī, Central NIA and Sinhalese. Thus this is a word shared by the inner and the northwestern languages, but is absent in the outer languages.

Some close coincidences between Romani and Ḍumāki
Ḍumāki is an Indo-Aryan language spoken by the Ḍoma in the Hunza valley and some other scattered places in the Northern Areas of Pakistan. Ḍumāki is not a Dardic language-even though it has been heavily influenced by Dardic Shina, by an unknown other Dardic language and by Burushaski-but is related to languages of the North Indian plains. The language shows some peculiar coincidences with Romani which are worth to be quoted here: Notes: (1) In phaaká 'shoulder' the suffix -ká is remarkable as it is otherwise only found-besides Romani-in Kashmiri phyok and Iranian Shughni fyak and some other north-western languages (see above section 4).
(2) The European form for 'ghee' is khil and Shina has gi, but Ḍumāki giríu 'ghee' resembles the form in Ḍomari. 23 (3) The word purṣúm 'flea' is peculiar because of the final ˗m which is the rest of a second word, cf. Torwali "pʸumåsh" 'flea' (Barth and Morgenstierne) and Dameli prāṣū̃ 'fly'. Apparently, either the first sibilant or the second was dissimilated due to syllable contraction.

Remarks on Norwegian and Swedish Romani
The first and the second lemma discussed in this section are related either through common etymology and/or through partly inextricable contaminations and borrowings, and (perhaps repeated) onomatopoetic creations. Yet they form a closely interrelated complex. Note, however, that the first lemma is limited to Scandinavian and German Romani whereas the second one is more widespread. The second lemma is said to derive from Greek and the first either from Turkish, from PIE or even Nostratic. However, both have Wanderwort characteristics. The second lemma has been included here because of its closeness to the first one.
-The signif. given by Su. must be due to a mistake." There is apparently also Turkish kargas 'griffin' and kerkes 'phoenix' (see e.g. Roelof van den Broek and Inez Wolf Seeger 1971: 204). The above evidence shows that the quoted authors are not wrong; however, kakaráška and similar forms are obviously the last ones of a series of repeated borrowings of a widespread and multiform Wanderwort that may, as suggested, have an origin beyond Proto Indo-European.
The case of Yenish (German: Jenisch) 27 čukel 'sauer' (besides šukel 'sauer', see Josef K. von Train) is not quite clear. A derivation < OIA cukrá-'sour, sharp to the taste' (4850) is phonologically possible (compare Ashkun ċukalä́ 'sour, bitter') but would seem to be without parallel in Romani. I therefore assume that also this is a case of affricatization and that the word derives < OIA *śuktala-'become acid or sour' (12504) like Romani šuklo 'sour'. Thus it seems that some of these words underwent affricatization in the north-west and others, much later, perhaps under dialectal German influence.

Fronting of aspiration (h metathesis)
In an article from 1959 (1975) Turner discusses fronting of aspiration in European Romani. He summarizes his conclusions at the end of the paper (1975: 388). The most important conditions for fronting are: • When the initial consonant was g, j, d, b (v), ś, p.
• Apart from two exceptions, fronting of aspiration occurred only between a voiced initial stop and a voiced internal aspirate.
Moreover, Turner shows (1975: 381) that fronting of aspiration preceded the devoicing of voiced aspirates in European Romani. All the above points indicate that the process of fronting of aspiration in Romani occurred when its speakers stayed in the north-west of South-Asia. In other words, h fronting in Romani was caused by the same process as in Dardic (and perhaps West Pahāṛī). Here is further evidence from Romani not yet noted, however without known parallels in NIA: khanči 'nothing, something, anything' < OIA kiṁcid 'anything' (3144) with *-cch-from OIA káścid, cf. Bhalesi kich 'some, a little' (Turner mentions only Rom.Pers. hič ʻnothingʼ, Rom.G. ič, hič ʻsomething, nothingʼ, Rom.Wel. čī ʻanythingʼ).

Fronting parallels in the north-west
Since fronting preceded devoicing, it is not surprising to find in the north-west several parallels of h fronting but (so far) no parallel for h fronting plus devoicing. This may be an indication that devoicing in Romani occurred outside the South Asian language area, namely in Armenia 31 (see e.g. Donald Kenrick 2004: 31). Note, however, that the proto-forms for the following modern Dardic and the Romani words are sometimes not identical but only closely related: Note: Romani khul 'excrement' is < OIA gūtha-'excrement' (4225) and there are a few NIA cases with fronting as in Lahnda ghū̃ and Oṛiyā ghua both 'excrement'. But these appear to be independent developments. The phenomenon of h fronting is much more comprehensive in Dardic and West Pahāṛī than it is in European Romani and it allows fronting of basically all aspirated sounds, single ones and clusters. On the other hand, in Romani it rather appears like an incomplete phonological process. This can only mean that Proto-Romani indeed was affected by h fronting during the period when its speakers stayed in the north-west of South Asia, but apparently the process was in the first millennium CE not as advanced as we find it today. Recalling Turner's observation that apart from two exceptions, fronting of aspiration occurred only between a voiced initial stop and a voiced internal aspirate in Romani it is now clear that this exactly reflects the situation in Dardic and West Pahāṛī where one observes a loose "connection" between h and its associated voiced stops as against the case of aspirated unvoiced stops (see Zoller forthcoming) as e.g. in Indus Kohistani nominative ghùī 'mare' but ergative guhḕ (< OIA ghoṭa-'horse'). Turner therefore rightly concludes (1927: 304) that h fronting in case of aspirated tenuis occurred later than that of aspirated mediae. Exactly this has also to be presumed for Dardic and West Pahāṛī.
Note that Romani has borrowed at least one Iranian word displaying aspiration not justified by etymology: phurd, phurt 'bridge' ← dialectal Iranian phurd 'bridge'.

The historical change a > e
Yaron Matras writes (2003: 34): "Preceding simple consonants, historical /a/ is represented in Romani by /e/, and in some cases by /i/: OIA kar-, Romani ker-'to do'… Historical /a/ is retained however in positions preceding an historical consonant cluster: OIA gharma, Romani kham 'sun'…" Somehow contradicting this statement he quotes on p. 39 OIA varṣa > Romani berš 'rain'. Similar vowel changes a > e are also known from the north-west, however, there they are only occasionally found and in a scattered way. Moreover, it seems that the vowel changes were caused by different factors, e.g. epenthesis or shift of accent. Thus there are quite many cases of epenthesis in Nuristani, Dardic and West Pahāṛī where a final i causes an initial a to be raised to e as e.g. in Ashkun vērī ʻspeech, language, wordʼ, Kati verí, Prasun verī which are < OIA *vari 'speech' (11327), Bhatise ċh ͥ ɛl 'a goat' < OIA chagalikā-'goat' (4963), Bashkarīk ǰēṅg 'shin-bone' < OIA *jaṅghiya ʻbelonging to the shank' (5084), etc. I will therefore not try to formulate alternative rules but simply list the cases where I see possible parallels between Romani and north-western languages. Hence, these parallels suggest that the Romani vowel 48 See http://ling.kfunigraz.ac.at/~rombase/cgibin/art.cgi?src=data/ethn/work/ copper.en.xml change a > e occurred through contact with north-western languages, as it would otherwise fail to explain why this rule did not apply in non-IA borrowings.
ivend 'winter' is usually derived < OIA hemantá 'winter' (14164); however, compare Pašaī (some dialects) emḗn, emen 'winter' which Turner derives < OIA hemānta. But there is no phonological motivation for the e in the second syllable and thus it is not clear whether we have here an accidental similarity.
men 'neck' < OIA maṇi 'hump of camel' (9732) has no comparable modern parallel in South Asia; however, the vowel is probably the outcome of epenthesis.
les 'him' < OIA tá base of nominative singular neuter, genitive masculine, neuter tásya (5612) -compare Kalasha te 'they'; Torwali tes, Bhalesi tes, Bangani tes. Here we have again a clear case of north-western epenthesis. But the change of initial t-> l-in the Romani form is unusual and has only one parallel in len 'river'. But there one would expect the change -d-> -l-to have occurred before the metathesis even though there is no supporting evidence from Dardic. Could the word have been borrowed from Rom.Arm.?
šel 'hundred' < OIA śatá '100' (12278) -compare Bhatise šʸäl, but there are also other NIA forms with an i as in Nepāli, Bihari, Marathi sai, Simhalese siya, etc.; compare above deš 'ten '. 49 There are also some words where the e goes back to older i. Since the number of examples is so small it is, however, an open question whether or not this sound change in Romani was caused by an influence from north-western languages: šero 'head' < OIA śiras 'head' (12452) -cf. Kashmiri hĕri ʻabove, upstream', Siraji of Doda śeri 'head' (Kaul 2006 II: 328).

The conjunct verb del
In Romani one finds a number of verbal expressions consisting of a noun and the verb del. It seems so that in some of these expressions del goes back to OIA dádhāti 'places, lays on, gives, seizes' (6145) while in other cases it goes back to the very similar OIA dádāti 'gives' (6141). In any case it is not always clear whether a conjunct verb has been inherited or borrowed in the north-west. Yet, a north-western influence seems highly probable in case of several of the following expressions: kandel 'to obey' is probably a conjunct verb built with kan 'ear', cf. Hindi kān denā 'to listen' and Kalasha kọ̃ kárik 'to listen to, to obey' (with kárik < OIA karóti 'does' [2814]).
del čhik or čikdel 'to sneeze' has usually NIA parallels with the verb directly derived from the noun chikkā-'sneeze' (5032) as in Hindi chī̃knā. However, conjunct formations are again found in Dardic as in Gauro čhígi ghō̃́ 'to sneeze' with the verb deriving < OIA gháṭate 'is busy with' (4407).

Magpie
Regarding Romani kakaráška 'magpie', discussed above in section 9, Iversen writes: "Mention may be made of the fact that for the older "travellers" this bird played a special part, as it was a prophetic bird from which they took auguries. A similar superstition is also found among English Gipsies..." However, this must not be something specific to Roma traditions since the magpie has been regarded in Germanic mythology as a messenger of the gods or of the goddess of death Hel (see Wolfgang Epple). Moreover, since the word designates different birds we cannot know whether the Roma word previously also referred to the magpie. It is therefore difficult to say when and where the Roma adopted this mythologem.

Porcupine
According to Hermann Berger (1985: 793), there is a being in Roma mythology called Ḫagrin which is a "Dämonische[s] Wesen in Gestalt eines gelblichen Stachelschweins von ½ m Länge und 1 Spanne Breite. H. quält Tiere im Schlaf, besonders solche, die gerade geworfen haben, indem er sich ihnen auf den Rücken setzt und seinen Urin daran herabfließen lässt, wodurch eiternde Geschwüre entstehen" (my rendering: The Ḫagrin is a demonic being in the shape of a porcupine with a length of ½ m and a breadth of 1 span. H. afflicts sleeping animals, especially those which have just delivered, by sitting on their back and by having its urine running down over them, through which purulent abscesses develop). A parallel is found in Shina and Burushaski hargín 'Drache, Ungeheuer, ensteht aus einer gewöhnlichen Schlange, wenn sie groß und alt wird' ('dragon, monster, evolves from an ordinary snake when it becomes big and old'). The description of the Ḫagrin also resembles strikingly what Karl Jettmar (1975: 285) says about a constrictor in the Hindukush: "Ferner hören wir von einer Riesenschlange mit goldener Mähne, Hargin genannt ..." ('In addition do we hear of a constrictor with golden mane and called Hargin'). Unfortunately Jettmar does not tell us more about this strange creature, but not only do both beings have basically the same name, but the porcupine is of yellowish colour which corresponds with the golden mane of the constrictor. It is thus likely that the ancestors of the Roma acquainted themselves with this myhological being in the northwest of South Asia.

Conclusions
The data presented in this article provide ample support for some important theses of Turner which have been quoted above in section 1: Romani belongs originally to the inner branch of Indo-Aryan; the speakers of early Romani left their original home (probably located somewhere in the Ganges valley) before the time of the Ashoka inscriptions; after leaving their original home the speakers of early Romani came in contact with speakers of Indo-Aryan languages in the north-west of the subcontinent, that is with speakers of Dardic,Nuristani and,perhaps,West Pahāṛī. 50 This contact must have continued over several hundred years before the Roma left South Asia. 51 There have been different views whether or not the languages of the north-west left clear-cut traces in Romani. As pointed out above, Turner was sure that some northwestern words were borrowed into early Romani. Matras, however, is sceptical and finds that the "… lexical evidence remains marginal and largely inconclusive. Noteworthy is the fact that there are hardly any phonological innovations that are shared with the North-western languages…" (2002: 47). However, elsewhere 52 Matras admits north-western influence, and as an example he refers to the Romani pronominal suffixes (as in the Romani past-tense forms kerdjo-m 'I did', kerdja-s 'he/she did' etc.). The geographical extent of languages with pronominal suffixes, however, does not coincide with my postulated north-western branch. The north-western branch comprises Nuristani, Dardic and West Pahāṛī whereas pronominal suffixes are found in Kashmiri, Shina, Lahnda, Sindhi, Pŏgulī and some dialects of Panjabi. In 50 It is self-evident to presume that in the second half of the first millennium the differences between the Dard languages and the varieties of West Pahāṛī were less developed than they are now. 51 It goes without saying that these linguistic matters say nothing about the ethnic, social, cultural etc. features of the speakers. It is thus also impossible to say whether the speakers of Romani at the time of their leaving the subcontinent were the progeny of those who left their original home. Such questions are anyway not part of this article. 52 http://romani.humanities.manchester.ac.uk/downloads/2/Matras-_Rmni_ELL.pdf (second page) and Matras 2009: 11-3ff. other words, it is likely that early Romani adapted this pattern in the north-west, but it is not a pattern characteristic of the northwestern branch of Indo-Aryan. Note also that although early Romani has been in long contact with languages in the north-west there are only few traces of depalatalization-that is phonetically e.g. the change of [ʧ] to [ʦ]-even though depalatalization is widespread and old in that area. However, there are some Nuristan, Dard and West Pahāṛī languages where depalatalization is not found. Thus we may assume that in the first millennium AD this process was less comprehensive than it is today. There is no doubt that the strongest influence on early Romani came through contact with languages of the north-western branch, and the influence is found on all levels of grammar. In addition, there are even a few traces of mythologems originally located in the north-west.