i. Consonants

13

a. Back Consonants k, kh, g, gh, ṅ. All these stops are always rendered with Sogdian k (which could be pronounced as /k/ or /g/): knkmwny – Kanakamuni; kntrβ – gandharva; ’’m(’)wkp’š – Amoghapāśa.

14

The groups ṅk, ṅg are rendered as nk: rnk’/lnk’ – Laṅkā; knky(h) – Gaṅgā⁸. Internal k is often geminated in writing: ckkr – cakra; rkš, rkkš – rakṣā.

15

In a few cases we see that Indian suffixal k is recorded as (’)y, which interchanges with spelling with (’)k(k): ’wp’s’y, wp’s’k, ’wβ’s’k – upāsaka; βyr’wr’y, βyr’wt’kk, βr’wr’k – Virūḍhaka; maybe nysrky – niḥsargika. Similarly an expected k can be omitted after long ā: pr’’, pr’kh, Pl. pr’tt’ – paṭāka. We return to this problem later (see 2.s; 3.b).

16

One should notice explicitly the single occurrence of writing back stops as x or γ so far in the corpus: ’n’wγt’r’ s’m myγ s’m pwδ’y – anuttarasamyaksaṃbodhi.⁹ This has long been considered a Chinese loan (Gauthiot and Pelliot 1926, 2:66; see also under 3. g.).

17

b. Palatal consonants: c, ch, j, jh, ñ. Usually c and j are rendered as c: cynt’’mny – cintāmaṇi; ’’c’r’y – ācārya; βcrp’n – Vajrapāṇi. I have not found cases of jh in the borrowings. The only case of aspirate ch is in geminated cch and is rendered as šc: ’yšcyntyt, ’yšc’tyk – icchantika or c: kycrwnmyl – kr̥cchronmīla.¹⁰

18

It is not uncommon to see j rendered as t: p’štsyn – Bhaiṣajyasena; rwk’yntr r’t – Lokendra rājan.¹¹ There are no cases of j simplified into y recognized so far, on kršny’n see 3. a, 3. i. below.

19

ñ is rendered as n: kncnsr – Kāñcanasāra and sometimes as ny: ’’tny’tkwtyn – Ājñātakauṇḍinya; pr’tny’p’rmyt, prtnyh p’rmyt – prajñāpāramitā. The initial group jñ appears as ny in ny’ncynt – Jñānacinta (the name of translator, not the character in doctrinal text; compare see Pāli ñāṇa, Gāndhārī ñāna = Jñāna, Toch. B. Ñāna- in personal names).

20

In one special case j is rendered as both c and z: r’zβwrt, r’cβrt – Rājyavardhana (in the same text)¹² and in one case probably as š: ’ykr’šn - ekarājan.

21

c. Retroflex ṭ ṭh ḍ ḍh are rendered through t and r: For ṭ, the variant t seems to prevail: kr’ytkwt – Gr̥dhakūṭa; šr’ykwty – Śrīkūṭa, with the exceptions of n’r(kr’k) – nāṭa; pr’’ – paṭāka. I do not have reliable examples for ṭh. ḍ is either t or r: kr’wr – Garuḍa and βcrkr’wt – Vajragaruḍa; cwrypnt’kk – Cūḍā̆panthaka and mncwty – Maṇicūḍa. The same pattern holds for the rarer ḍh: βyr’wr’y, βyr’wt’kk, βr’wr’k – Virūḍhaka. There is no attested usage of δ for retroflexes and, moreover, there are no cases of spelling Indian retroflexes through /l/ (which is recoreded in some Sogdian texts through a newly invented letter or variation of r and δ), unlike in Khotanese and Tocharian B. A special case is kwrt(t)y, kwtty – koṭi.¹³

22

ṇ is normally rendered as n: β’yšrβn – Vaiśravaṇa; nyrβ’n – nirvāṇa: βynwβn – Veṇuvana.¹⁴

23

The group ṇḍ is often rendered as ntr: tntr’k – Daṇḍaka;¹⁵ mntr – maṇḍa. But it remains nt in cnt’’r – Caṇḍāla; kntyk – Ghaṇṭikā and is simplified in ’’tny’tkwtyn – Ājñātakauṇḍinya. The rendering is unclear in pwntr’yk – puṇḍarika.

24

Some irregular changes can be the result of elimination of haplology, such as the loss of ṇ in ’’tny’tkwtyn – Ājñātakauṇḍinya, and the loss of ṭ in snk’swtr – Saṃghāṭasūtra (see 3. a).¹⁶

25

d. Dental t, th, d, dh, n. The voiceless stops are usually rendered as t, and often as tt in non-initial position: tr’ymwkt – Trimukti; pr’ttymwkš – prā̆timokṣa; ’yry’’pt – īryāpatha; p’rmyt – pāramīta. An Indian geminate often remains the same: pt(t)r – pattra (the word seems to be restricted to Turfan texts), δyβδtt – Devadatta. T(h) transcribed δ(δ) is limited to non-initial position but fairly common (see. under 3. e): k’wδ’m, k’wt’m – Gautama; ’’ry’βrwkδyšβr – Ārya Avalokiteśvara. T in the cluster tp is lost in (’)wpδ(’)y – utpala (compare Gāndhārī upala, Pāli uppala, Toch. B. uppāl, A oppal), and in the final position in twškr–duṣkr̥ta.

26

d can be rendered as t and δ in any position, and for dh the latter prevails. Initial d is δ: δyβdtt – Devadatta. Initial d is t: twškr – duṣkr̥ta. Non-initial d is t: mx’tyβ – mahādeva. d is δ: šrγw swδ’s – Siṃhasaudāsa. Initial dh is δ: δy’(’)n(y) – dhyāna. Initial dh is t: t’rn’y – dhāraṇī. Non-initial dh is δ: s’δw s’δw – sadhu sadhu. Non-initial dh is t: syt(t) – siddhi, srβ’’rtt sytt – Sarvārthasiddha. Variation occurs in βytty’tr / βyty’δr – vidyādhara. D after n is always spelled t: cntn – candana.

27

Geminate ddh results once in nt in šnt’wδn – Śuddhodana, and r renders dh in s’m’’r – samādhi¹⁷; we consider these forms irregular; one sees a diacritic sign in kr’ẉt’ – krodha (invocation).

28

Indian n is always n in Sogdian: n’t’y k’š’yp’ – Nadī Kāśyapa; mx’y’n[y] – Mahāyana. N is lost in δ’p’t – dānapati and d is simplified to y in k’wy’ny, kwy’n’ – Godānīya (see under 3. g).

29

e. Labials p, ph, b, bh, m. All stops are regularly rendered by p in all positions: pr’pt – Prāpta; pyms’r – Bimba/isāra; ptrp’r – Bhadrapāla; mx’kp(’)yn – Mahākapphiṇa; ’wrpyrβ’ k’š’yp’ – Urubilvā (Uruvilvā) Kāśyapa; kwp’yr – Kubera (invocation); βyšβ’pw – Viśvabhu.

30

m usually appears as m: mnc’wšyry – Mañjuśrī; c’δysm’r – Jātismara, kr’’m – grāma; but before p, n can be written: rnpyh – ḍomba; s’δynp’y’ – sālambha.

31

Sogdian β for the stops is rare and is known in a few cases in non-initial position: ’wβ’s’k¹⁸, ’wβ’s’nc ’wp’s’k – upāsaka, upāsikā; prβr’c – Prabhārāja.¹⁹ In ṗykšw – bhikṣu the diacritic writing <ṗ> probably indicates /b/ (Sims-Williams apud MacKenzie 1976, ii 9, note. 37).

32

f. Sonorants r, l, y, v do not follow a unitary rule. R is always rendered as r: rkkš – rakṣā; ’’s’wr – Asura; βrδ’(m)[ – Vardhamānamati (?). When several R-like sounds are in the source, one r sometimes drops out: kr’ytkwt – Gr̥dhrakūṭa or Gr̥ddhakūṭa; trytr’št – Dhr̥tarāṣṭra; βympx’r – Vimalaprabhāsa, but š’rypwtr – Śāriputra. A non-etymological r appears in βrx’r – vihāra (see 4. d), as well as in kr’ẓ’k (nγ’wδn) – kāṣāya, for which numerous Central Asian variations have been documented (Bailey 1949, 130).²⁰

33

l can be treated in three ways, as the newly shaped letter l (ṛ) in some relatively late texts, as r or (somewhat less commonly) as δ: lnk’, rnk’ – Laṅkā(vatāra-sūtra); δwk’ – loka; kδp klp krp-h – kalpa; ptr klp(-) – bhadrakalpa; p’r’wr – balūla; stwl(’nc) – Sthūlā(tyaya). Yutaka Yoshida (2010, 90) noticed that the spelling with δ is common among well-established Sogdian terms. Manichaean writing δwk- (and not lwk) might indicate a fricative /δ/, and not a sonorant pronunciation, compare also prδwk’ – paraloka; but rwkδ’t – lokadhātu.

34

In two instances, postvocalic l and r are written as n: ’ncn(δst) – añjali²¹ and p’n nyrβ’n parinirvāṇa (see under 3. g).

35

Y is rendered as y: ywcn – yojana; βyny – vinaya; p’ytyk – pāyattika, pātayantika; βy’y’m – vyāyāma; kš’try – kṣatriya.

36

Quite often, post-consonantal y is lost: cwδ’yk’ c’wtyšk’ – Jyotiṣka; p’štsyn, p’tsyn – Bhaiṣajyasena; ’’tny’tkwtyn – Ājñātakauṇḍinya; ’’s’nk(y) – asaṃkhyeya. One can explain in the same way š’kmwn from š’kymwn – Śakyamūni. The first form appears in Manichaean Sogdian and has been explained as Prakritic (Sims-Williams (1983), 137; š’kymwn in a Manichaen text is largely a restoration, see Lurje (2010), 365). Consequent simplifications are observed in mwtkr’’y’n, mwtklyn – Maudgalyāyana.

37

Another common feature is metathesis of non-initial ya, ye into ’y (pronounced as /e/?): ’’c’r’y – Ācārya; k’š’yp – Kāśyapa; prt’ykpwt(t) – pratyekabuddha. See under 2. m. for the opposite case.²²

38

Indian v is always rendered as β in Sogdian: β’yšrβn – Vaiśravaṇa; β’swmytr – Vasumitra.

39

In few exceptional cases it is recorded as p: srp’šwr as variant for srβšwr – Sarvaśūra in the Saṅghāṭasūtra; pwrpδ’yš – *Purvadeśa and synt’’p – saindhava.²³ I did not notice any case of Sogdian w for Sanskrit v.

40

g. Sibilants ś ṣ s h. The first two are regularly rendered by š: š’ky– Śākya; β’yš’ly – Vaiśālī; wšn’yš – uṣṇīṣa; (p)wrβ’š’(t) – Pūrvāṣāḍhā;²⁴ snk’’βšyš – Saṃghāvaśeṣa; š’str – śāstra.

41

Twice in the corpus ṣ appears as ẓ /ž/: twẓ’yt βγyst’n – Tuṣita-Heaven (but sntwšʾyt – Santuṣita) and rz’y (usually rš’’k) – r̥ṣi.²⁵ The group kṣ simplifies into š in š’ymnkr (alias kš’ymnkr cf. inscriptional Tocharian B Ṣemaṅkar-) Kṣemaṃkara. Ṣ is lost in p’tsyn – Bhaiṣajyasena. All these cases are either archaic, dialectal or clerical errors.

42

S is also rendered as s in a straightforward way: srβ’stβ’t – sarvāstivāda. There is one case of r /l?/ for Sanskrit s in βympx’r – Vimalaprabhāsa.²⁶

43

Indian h can be rendered through x or less commonly with no sign in Sogdian (no cases of initial h recognized so far): pr’xmn / pr’m(’)n – brāhmaṇa; ’r’x’n, rx’nt- (Manichaean rhnd!) – arhat; r’ckr(’)y – Rājagr̥ha. Skt. mahā (when not translated) is always rendered as mx’-: mx’r’c – mahārāja; mx’stβ – mahasattva; mx’yšβr – Maheśvara.

44

h. Visarga (ḥ), anusvāra (ṃ) Since endings will be treated separately, and these signs have limited usage in non-final position, the documentation is scarce. Visarga is not recorded in nysrky[ – niḥsargika, and anusvāra appears as n (k)š’ymnkr – Kṣemaṃkara; snk’ – saṃgha; snkr’m – saṃghārāma; typ’nkr – Dīpaṃkara ( all before back consonants) or m: smyk’ smpwtt – samyaksaṃbuddha (before labial), smrk – saṃrāga; the common snks’r – saṃsāra is unusual; recently N. Sims-Williams (2021, 34–35) proposed to see here a contamination with saṃskāra. In final position, it is m or n: ’wm – oṃ; pk’’β’n (in Skt invocation; see 3. g.) – Bhagavant, nom. Bhagavāṅ; ’’ry’nc – āryaṃ ca (invocation: before c).

ii. Vowels

45

i. Short a is rendered as single or double aleph in the beginning and single aleph or more often no sign in the middle of a word: ’’m’(w)kp’š – Amoghapāśa; ’pyδrm – abhidharma; βyp’š – Vipaśyin; knkmwny – Kanakamuni ; m’k(’)t – Magadha.

46

Fronting to y, ’y sometimes occurs: prsn’ycy – Prasenajit; ’yšcyntyt, ’yšc’tyk – icchantika; βymyrkr’yt (and βymrkyrt) – Vimalakīrti; k’r’ynt knδh, k’r’ynk’ – Kalandaka /Karaṇḍa / Kaliṅga; s’δynp’y – sālambha. Initial a can be lost in ’psm’r / psm’r – apasmāra.

47

Long ā is always double aleph initially and single or double medially: ’’ry’βrwkδyšβr – Ārya āvalokiteśvara, cnt’’r – Caṇḍāla, βyr’wp’kš – Virūpākṣa, p’tr – pātra, šr’’βk – śrāvaka. Shortenings are uncommon: tn(’)pt – dānapati; kncnsr – Kāñcanasāra²⁷, prβr’c – Prabhārāja. On the probable articulation of single and double initial aleph in loanwords in Sogdian see Sims-Williams (1983, 138–39).

48

j. In initial position i and ī are usually rendered as ’y: ’yšcyntyt, ’yšc’tyk – icchantika; ’yšβr – īśvara, once as y: yntr’y – indrāya (Skt Dative). In medial position, the most common is y: βykn βyn’ywkh – vighnavināyaka; kntyk – ghaṇṭikā, but ’y or no sign are attested rather widely, too: kpl[β]st, kp’yrβst – Kapilavastu; ’’βcy, ’’βycy, ’’βyc – avīci; kwmp’yr – kumbhīra; kšytkrp – Kṣitikalpa.

49

Metathesis is visible in βymyrkr’yt – Vimalakīrti; the group śr is often changed into šyr (a result of reetymologization, see 3. a. below): mnc’wšyry – Mañjuśrī (also BHS Mañjuśiri); kwm’ršyr – Kumāraśri; šyr’’βsth – Śrāvastī.

50

k. Initially u is rendered as ’w or w: ’wp’k’ – Upāka; ’wβ’s’k, wp’s’k – upāsaka; wpʾtyʾy – upādhyāya. In medial position u and ū are mostly rendered as w or ’w as well: ’’s’wr – asura; pwšpcwty – *Puṣpacūḍa; kwm’r – kumara; δ’wt’ – dhūta; m’ywr – mayūra; n’ywt – nayuta; pwr’wš – Puruṣa. Unusual is sywpwδ’y – Subhūti;²⁸ šnt’wδn – Śuddhodana.

51

l. Skt r̥ is attested only in rz’y, rš’’k – r̥ṣaya in the beginning and more often in non-initial position. It is rendered as r or r(’)y: kr’ytkwt – Gr̥dhakūṭa; r’ckr – Rājagr̥ha; trytr’št – Dhr̥tarāṣṭra; twškr – duṣkr̥ta; m’trk š’str – mātr̥kā-śāstra-. The vowel l̥ is absent within the corpus I could collect.

52

m. Initially e is attested once and rendered as ’y: ’ykr’šn – ekarājan. In medial position, e is ’y, y: δyβδtt – Devadatta; nymyš – nimeṣa; pr’yt – preta; rarely y’: ’rny’m / ’rn’ym – Araṇemi (see under 2f), pwtkšy’tr / pwt(’)kš(’)ytr – Buddhakṣetra. It is aleph in Sanskrit invocation: cnt βcrp’n’y mx’’k’y s’n’pt’y – Caṇḍavajrapāṇaye mahākāye senāpate.

53

Medial ai can be rendered as ’y, y or even no sign: β’yš’ly – Vaiśālī; βyr’wcn – Vairocana; pyš’ckwr – Bhaiṣajyagūru; cntrβrwcn – candravairocana. Note aleph for ai in p’štsyn, p’tsyn – Bhaiṣajyasena.

54

n. Initially, o is once rendered as ’w: ’wm – oṃ. In medial position, o and au are usually ’w or w: cwδ’yk’, c’wtyšk’ – Jyotiṣka; mwkš – mokṣa; k’wšyk’ – kauśika; mwtklyn, mwtkr’’y’n – Maudgalyāyana; k’w šwt – semi-translating Gocara. Unusual are rnpyh – ḍomba and ’’m’kp’š – Amoghapāśa (var. ’’m’wkp’š).²⁹

55

o. On non-etymological prothetic vowels, see 3. b.

iii. Endings

56

p. As it is well known, in Sogdian ‘light’ and ‘heavy’ stems are distinguished. The former are ones with no long vowels or diphthongs, the latter have at least one long vowel or diphthong. In the nominal inflection in singular, ‘light’ stems differentiate gender and six cases while ‘heavy’ stems have only direct and oblique cases. The major part of the Indian loanwords in Sogdian are ‘heavy’ stems that have zero ending in Rect. Sg., /-i/ in Obl. Sg, /-t/ in Rect. Pl and /-ti/ in Obl. Pl as well as vocative in /-a/.

57

The Indian loans that follow, with some reservations, the ‘light’ stem pattern are pwty – Buddha; rtn- – ratṇa; šmn- – śramana; smwtr- – samudra; ykš- – yakṣa; ’st’wp- – stūpa.³⁰ The rendering of Gaṅgā, nom. knk’ (see n. 8) and gen. knky(h) might represent feminine ‘light’ stem.

58

What concerns us here more is the rendering of Skt endings in Sogdian inasmuch as we can detect them within the declension of the different language.

59

q. The thematic -a ending is naturally most common among the borrowed lexemes. It is lost in the majority of cases: ’’n’nt – Ānanda;³¹’’kš’r – akṣara (?). Sometimes it is rendered as ’: mx’k’š’yp’ – Mahākāśyapa;³² s’l’ ryzkry xwβw – Sāla-Īśvararāja. Elio Provasi (2013, 277–78) supposes that the loans with aleph for final -a are to be placed in time between the earliest Prakritic forms and the standard transliterations from Sanskrit, as in Chinese Buddhist borrowings, but I do not find definitive proof. The recording of final -a as aleph seems to be more common after k, probably in order to avoid reading /-e/: snk’ – saṃgha; δwk’ – loka; smyk’ smpwtt – samyaksaṃbuddha; ’wp’k’ – Upāka; šl’wk, šr’wk(’) – śloka. It is lost in declentional forms: n’k’, pl. n’kt – nāga (also Manichaean); pwtr’ky (obl) – Potalaka.

60

In a few isolated cases we see final a as y or -’y: mncwty – maṇicūḍa; rtncwty – Ratnacūda; swβrncwty – Suvarnacūda; swttršny – Sudarśaṇa; pwrn’y – Pūraṇa.³³ The final Sogdian h, which can be purely graphical marker or convey -ā̆ is sometimes attested too: kδp klp krp-h – kalpa.³⁴

61

r. Feminine -ā is usually rendered as ’, ’h or h: mx’m’yh – mahāmāyā; yš’wδrh – Yaśodharā; ’’š’h – Āšā; ’wrpyrβ’ k’š’yp’ – Urbilvā Kāśyapa; k’y’’ k’š’yp’ – Gayā Kāśyapa; βc’ – vacā ‘Acorus calamus’. This writing might indicate a final /-ā̆/. A zero ending is probably attested in kntyk if it is ghaṇṭikā, mwtr – mudrā; note variants šβ’kwšh, šβk’wš, šyβkwš, šyβkwšh etc. – Śivaghoṣā.

62

s. The masculine -i and feminine -ī is either rendered with -y or not recorded: ’’βcy, ’’βycy, ’’βyc – avīci; š’r’βst, šyr’’βsth – Śravastī; syt(t) – siddhi; m’ytr – maitrī (or maitrā, also in Manichaean); β’yš’ly – Vaiśālī, rtnkyrt – Ratnakīrti; swk’βty – Sukhāvatī. Digraph ’y appears in šβ’y, šβ’’y – Śivi; t’rn’y – dhāraṇī. In rz’y, rš’’k – r̥ṣi, BHS r̥ṣaya it alternates with ’’k which would indicate spelling /rišē/, see 3. b.

63

t. Similarly, -u can be recorded with or without final w or ’w: s’δw s’δw – sadhu sadhu; r’xw – Rāhu; rwkδ’t – lokadhātu; m’δ’w – Madu; kpl[β]st, kp’yrβst – Kapilavastu. One can notice that the final y and w are more common in monosyllabic bases. The form pykš’k (Manich. pykšy) – bhikṣu, unlike pykšw, is quite unusual.

64

u. The final n of athematic endings is often not preserved and the preceding vowel may or may not be preserved (see Kasai 2015, 405, 413 for Uyghur): ’’n’k’my – anāgāmin, nom. anāgāmī; βyp’š – Vipaśyin, Vipaśśin (L, P), ckkrβrt – cakravartin, nom. cakravartī; rwk’yntr r’t – Lokendra rājan, nom. rājā; kwm’rβ’s – Kumāravāsin; it is preserved in rtnšykyn – Ratnaśikhin, but šyky – Śikhin; ’ykr’šn – ekarājan.

65

v. Final t is not recorded in prsn’ycy – Prasenajit, compare Pali Pasenadi, Toch A. Prasenaji, B Prasenaci, Old Turkic Prasaniči. In -nt stems we see ending n or nt³⁵, ’r’x’n, rx’nt (also in Manichaean rhnd) – arhant, nom. arhāṇ.

66

w. In general, Sogdians tended to borrow words in the nominative singular form rather than the base. Sometimes we notice Indian cases other than the nominative, especially in dhāraṇis: kry’n ’sy’ pykšw – Kalyaṇasya bhikšoḥ (genitive); yntr’y – Indrāya (dative); βyr’wkt’yn – vilokitāyāṃ (locative).

67

As we see, the larger part of Indian Buddhist loanwords in Sogdian follows relatively clear rules of transmission with limited variation which can be put in the following conspectus. In chapter 3. h. I argue that the source of these standard borrowings is Buddhist Hybrid Sanskrit as perceived by Sogdians. The loans that follow the rules of this table, which we consider regular, constitute 62 percent of the material at hand.

Conspectus: Standard Buddhist Sogdian Renderings of Sanskrit

a. Contaminations

68

As it is evidenced from the conspectus, Sogdian writing did not possess the means for rendering all Indo-Aryan phonological detail. In some parts it was triggered by limitations in Sogdian alphabet or phonetic system, but in others (differentiation of aspirate and non-aspirate consonants and long and short vowels other than a, ā) I could not notice any attemtp of Sogdian scribes to record the distinction. So sometimes the spellings of different words merged: mntr comes from mandala, mantra, and maṇḍa in different contexts³⁷, and perhaps non-etymological n in the name mntr’yh Madrī was triggered by one of these loans (note also the inclusion on n in Chinese, Uyghur and Tocharian forms of the name, Provasi 2013, 230–31; Ogihara 2018, e41).

69

The last example highlights a feature we see rather often in these borrowings: re-etymologizations on the base of Sogdian or Indian language. We already saw the inclusion of y in the group śr >šyr: kwm’ršyr – Kumāraśri; šyr’’βsth – Śrāvastī; mnc’wšyry – Mañjuśrī.³⁸ In this, an association with S šyr “good, nice” probably played a role. A doublet form of šyr’’βsth: š’r’βst might be affected by Western Middle Iranian word šahr “town, province,” note š’ryst’n in Sogdian Manichaean church history. The spelling cntrsyn for the river Citrasena is probably affected by very common Indian candra “beauty, moon,” while swttrsyn implies virtual Sūtrasena, an unattested(?) but quite Buddhist looking name “army of Sūtras,” based on swttr – sūtra. snks’r – saṃsāra and maybe snk’swtr Saṃghāṭasūtra might be influenced by snk’ saṃgha (differently N. Sims-Williams, see under 2. h). See also discussions on kršny’n – Kr̥ṣṇājinā styled as Sogdian personal name “Boon of Krishna”; ’ykr’šn – *Ēkarājan below.

70

In the word βykn βyn’ywk’ – vighna vināyaka one can explain w at the end as contamination with S ywk (ywk’ in the Sūtra of Causes and Effects 59) “teaching.” Sogdian sym’βntt, sym βynt’y, if from sīmābandha might be affected by Sogdian βnt “bundle,” βynt- “to bind” with symh “fear” in the beginning. Nārāyaṇa is constantly (in four different texts) rendered in S with aleph in the final syllable, n’r’y’n and not *n’r’yn, as if in contamination with Sogdian names in y’n “boon.” Compare also cytβnt, kyn(n)tr under 3.b; ’yšcyntyt, ’yšc’tyk – icchantika could be influenced by Sogdian ’šcy’n’k “worthy.”

b. Simplifications

71

The long consonant groups in Sanskrit, especially in compound words, were hardly articulated by a foreigner, so various shortenings took place. We do not witness the insertion of epenthetic (svarabhakti) vowels in the consonant clusters (with possible exception of twr’w(n)[wtn if for Droṇodana), although prothesis or loss of the initial short vowel—in accord with native Sogdian development—is documented: ’knšk – Kaniṣka; ’st’wp- – stūpa; ’psm’r, psm’r – apasamara; ’pr’tyk’ pwt, prt’ykpwt – pratyekabuddha;’kwšty, Brāhmi kuṣṭ(h) – kuṣṭha.

72

In respect of consonant clusters, ṇ and y are lost in ’’tny’tkwtyn – Ājñātakauṇḍinya (note Gāndhārī A(ṃ)ñadako(ṃ)ḍi(ṃ)ña, Pāli Aññātakoṇḍañña), the syllables na and va probably fall in cyttr’n – Cittanavarāṇa (?), n in kycrwmyl = kr̥cchronmīla, subsequent simplifications we attest in p’štsyn, p’tsyn – Bhaiṣajyasena, r is lost in trytr’št – Dhr̥tarāṣṭra, t falls in sm’nptr, var. sm’nt pwtr, sm’ntpttr – Samantabhadra. These simplifications could result either from Sogdian processes or from a mediating tradition, be it Indic, Iranian, or Chinese. The shortening and metathesis of pwtystβ, pwtsβ, pwδysβt etc. bodhisattva (Yoshida 2008, 347–48) goes along the same lines. The Sogdian tendency to a labial articulation of the vowel between P and D (Sims-Williams 1985, 61, n. to 24R) is visible in sm’nt pwtr – Samantabhadra.

73

The shortening or shift of vowels in long words may sometimes result from mediation in transmission and may sometimes be a Sogdian development or even inaccuracy in the rendering of long words: cntrβrwcn – Candravairocana; swryβrwcwn – Suryavairocana; prsn’ycy – Prasenajit (see 2. i); prytpkwpt – pratyekabuddha. Abbreviation probably occurred in ’’m’wk – Amogha(vajra?) and rwkδyšβ’r – (Aryāva)lokiteśvara which may be contaminated with Lokeśvara “lord of the world,” which was often used as an epithet of the former Buddha (Brough 1982, 68).

74

Non-etymological consonants r, n or t, sometimes appear, too: cytβnt – Jetavana forest;³⁹ kyn(n)tr – Kiṃnara;⁴⁰ šnt’wδn – Śuddhodana.⁴¹ Most often discussed in this context is βrx’r – vihāra, Old Turkic vrxar (Gauthiot 1911, 52–59; Gershevitch 1954, 54, compare Sogdian βrywr from Old Iranian *baivar- “10 000”), probably a Sogdian development, on which see also below 4. d, kr’ẓ’kh – kāṣāya (see n. 25). The change of r or l into n has been discussed above (under 2. f).

75

The Sogdian sound-change of *-aka- stems into /-e/ and *-ākā- into /-ā/, which is evident by the interchange of historical spellings with k and synchronic without k, and by Manichaean and Christian writing, can be also documented in Buddhist loans: ’wp’s’y, ’wβ’s’k – upāsaka; pr’’, pr’kh, Pl. pr’tt’ – paṭāka (see 2. a); one can explain the spellings rš’’k and mytr’k, mytr’y either as non-standard loans or as Sogdian attempts at pseudo historic spelling. The aleph written after k in the borrowings like snk’ – saṃgha (see 2. q) can be explained as an attempt to record articulated /-k/.

76

In the index below, 55 items, or 18 percent of the material in question, can be considered deviations explainable through Sogdian development without any mediation in transmission.

c. Mediations in Transmission

77

This problem has been analyzed more than all the above ones, and the role of a Middle Indic source of loanwords and a Parthian, Tocharian, or Chinese transmission has been recognized in a number of words. Some of them will be reviewed below.

d. Parthian and Manichaean Transmission and the Word Buddha

78

The Indian loans in Parthian and non-Buddhist Sogdian texts were the subject of a detailed article of Nicholas Sims-Williams (1983). He showed that Indian loans started to appear in the earliest texts in Manichaean Parthian and were common in the later texts as well.⁴² A number of conventions among these forty or so loanwords, such as the simplification of some consonant clusters, the development of kṣ > xš, the spelling of g for Sanskrit y, the change of v into b and of ḍ into l were convincingly ascribed to the North-Western Prakrit (Gāndhārī), while metathesis and the ś > c shift are Parthian (and also Bactrian, see. n. 44) developments. The voicing of postvocalic consonants could have taken place either in Indian or Iranian context (Sims-Williams 1983, 132–35).

79

Among ten Indian loanwords in Christian Sogdian and forty in Manichaean Sogdian, Sims-Williams identified the ones borrowed via Parthian (’’k’c – ākāśa, ’xšn- – kṣaṇa⁴³, p’š – bhāṣ-, cxš’pt/δ – śikṣapada) and those from Middle Indic, North-Western Prakrit in particular, noting that Old Indian was not its principal source (Sims-Williams 1983, 137). Five loans appear in the Sogdian ‘Ancient letters’, mirroring the trade vocabulary. Only one (dubious, see n. 17) example shows s in place of ś, ṣ, which is typical for most of the Prākrits as well as Pāli, but not the North-Western Prākrit (Gāndhārī). The Khotanese and Tocharian Brāhmī systems also keep s, ś, ṣ distinct, but not Bactrian in Greek script!⁴⁴

80

A case to be mentioned here specially is the word for Buddha himself, Sogdian pwt-, rarely pwtt-. In principle, it follows the ‘standard’ rendering of BHS in Sogdian, but a priori it cannot be separated from the form bwt- in Manichaean texts.

81

Thanks to the work of Iris Colditz on the onomasticon of Iranian Manichaean texts, we have a contextually organized list of appearances of the Buddha in the compositions of the Manichaeans (Colditz 2018, 264–68, No. 170). It appears that the form bwt or bwt is predominant in the corpus: it appears in Manichaean Parthian, Middle Persian (including the translation of Mani’s ‘Book of Giants’), Bactrian and, with final vowel -y, in Sogdian in Manichaean script. The similar form βοτο appears once in inscriptional Bactrian⁴⁵, Book Pahlavi bwtꞋ and New Persian but “idol” clearly belong here too (Ḥasandūst 1393, 1:408–9). All these forms share the same phonetic value /but/ with the last consonant voiceless and as a rule not geminated.⁴⁶

82

Werner Sundermann (1991, 427–30, 2001437–40) analyzed this form in great detail and proposed that the Parthian language (as well as Middle Persian) preferred unvoiced geminate stops after vowels: pattabag “splendor” (< *pati-tapaka-, spelt ptbg), appar “predatory” (< *apa-bar-, spelled ’pr). In a postscript of 2001 (p. 450), he considered the devoicing after simplification of bisyllabic words into monosyllabic: dat < dahat “er gibt,” nēk < nēwak “gut, schön,” Bāt < Bagdāt etc as the “einfachste Erklärung” for bwt. Sundermann’s second explanation (which is accepted by Colditz) does not look satisfactory, since it requires a bisyllabic prototype with final consonant.⁴⁷

83

As for the first possibility, some additional comparanda can be supplemented: N. Sims-Williams postulated for Bactrian the rule that the comparative suffix is normally δαρο, but with a final d of an adjective base it turns into ταρο: þαδο “happy” þαταρο “happier” *οαδο “bad” οαταρο “worse, worst” (Sims-Williams and Tucker 2005, 591–92). Parallel cases are observed in Manichaean Western Middle Iranian wtr “worse” and Khotanese battara “less”. Compare also Middle Persian jwtr “different, otherwise” < jwd(y) + dr, kbwtr “pigeon” < *kbwd “blue” + dr, Parthian p’twg “punishment” < *pati-taug, pt’w “stay, suffer” < *pati-taw-. The suffix which is normally spelt -gr, obtains the form -kr after bases ending in -g (as zyntkr “redeemer,” see Durkin-Meisterenst 2014, 168–69). It is true that in the cases listed above historically one or both consonants were voiceless, so wtr e. g. comes from *wata-tara, but still I think that this process, namely the devoicing of postvocalic geminates was the main reason for the peculiar form but in Middle and New Western Iranian, wherefrom it was borrowed into Sogdian.

84

It is important to notice here that there are no Indic loans that passed through Parthian mediation into Sogdian that are not also documented in Manichaean texts and limited to Buddhist ones. No Buddhist loans with Bactrian mediation have been attested in Sogdian so far, despite the fact of close contacts between the two neighboring regions.⁴⁸

e. Prakrit Forms in Buddhist Sogdian

85

For the Buddhist Sogdian texts, Sims-Williams suggests that the majority of loans are of Prakrit origin but notes that “however there are also many spellings which reflect the increasing prestige of Buddhist Sanskrit” (1983, 137). One of the features he considered Prakritic was the rendering of t as δ, thus k’wt’m would be from Sanskrit and k’wδ’m from Prakrit. I do not know cases of δ for Sanskrit t(h) in the beginning of a word, while in postconsonantal position it appears quite often: c’δysm’r – jātismara; c’δyšrwn – Jātiśroṇa; tδ’ktswm – Tathagatasoma. Yoshida noticed that the same convention appears in the name ’’ry’βr’wkδyšβ’r – Āryavalokiteśvara attested only in the late, Tantric sutras that cannot have had a Middle Indic prototype and thinks that the irregularity of δ/t is an inner Sogdian phenomenon (Yoshida 2008, 348 with n. 26). Yutaka Yoshida kindly reminded me of variance of Buddhist mx’pwδy and Manichaean mx’pwtty for Mahāboddhi. One can further adduce the same convention in transcribing Sanskrit invocation: š’kymwn tδδ’kt’w ’r’x’n smyk’ smpwtt – Śākyamumi tathagato arahan (!) samyaksambuddha (Dhyāna 358, MacKenzie 1976, 74–75), showing that this was a perception of Sanskritic postvocal t(h) by the translator.⁴⁹

86

Among the Middle Indic loans we should distinguish ones that have Prakritic features surviving in Buddhist Hybrid Sanskrit as well as occasional shortenings: krkswn(t)[⁵⁰ – Krakucchanda, Turfan Sanskrit Krakasunda; βcyr- – vajira; perhaps ’yšcyntyt, ’yšc’tyk – icchantika (see n. 10); βr’yšmn – Vaiśramaṇa;⁵¹ ’yry’’pt – īryāpatha; p’ytk – pātayantika, pāyattika (if not Tocharian, Yoshida 2008, 340), k’l’y xwt’w – half-translated Kalirājan, Kaliṅgarājan (Khot. Kalä rri, Uyghur Kali bäg and Chinese Jiali 迦利).

87

What remains are several ‘cultural’ words: k’rt’k⁵² – gr̥hastha (Khot. ggāṭhaa, Toch. B. kattāke); kr’ẓ’k – kāṣāya, Toch. A kāṣār(i), Niya Prākrit kaṣara, Old Turkic k(’)r’z̤’ (Bailey 1949, 130); pwrsnk – bhikṣusaṃgha with Khotanese bilsaṃgga, Toch A pis-saṅk, Uyghur pursang (Bailey 1967, 242), trywr – tripuṭā, Khotanese ttrola (Maue and Nicholas 1991, 491, n. 38, although one cannot exclude error for *trypr, Manichaean trypl, Brāhmi tr̥phāl – triphala), kwn’k’r – kūṭāgāra (Middle Indic kūḍāgāra, Toch A k^urekar etc, Sims-Williams 1983, 137) the names of continents (see n. 56).

88

Among the Indian loans that are found outside Buddhist texts, the majority come from Middle Indic (Sims-Williams 1983), but some of them show certain older features: Manichaean krm “(evil) action” – karman (Gāndhārī kaṃma but sometimes spelt kram-, karm-, Pāli kamma), Manichaean (?) (’ynt’wk)δyβr’c – Devarājā, r’’t – rājā (Gāndhāri raya, but sometimes spelt raja, Manichaean Middle Persian mh’r’c, Persian rāja).

f. Tocharian

89

Yoshida (2008, 338–40) noticed several Indic loans in Sogdian that witness vowel and consonant shifts typical for Tocharian B: synt’’p, Sanskrit saindhava, Tocharian B sintāp “stone salt,” tn’pt, Sanskrit dānapati, Tocharian B tanāpate,⁵³ pkc’n, Sanskrit *upagacchana⁵⁴, Tocharian B pakaccāṃ, A pākäccāṃ; p’ytk – pāyattika, pātayantika, Tocharian B pāyti. The same is supposed for the name kncns’r Sanskrit Kāñcanasāra with Tocharian initial shortening, the name Kañcanasāre is now attested in Tocharian B, and the closely related form is Uyghur kancanasare (Sundermann 2006, 719–20; Yoshida 2019a, 155 with n. 45). Note also ny’ncynt Jñānacinta (name of the monk who translated the Intoxication Sutra from ‘Indian’ into Sogdian) under 2. b., explainable as Middle Indic or Tocharian.

g. Chinese

90

Despite the fact that Chinese was the direct source of most of the Sogdian Buddhist texts, very few of the Indic loanwords have traces of Chinese mediation. Elio Provasi examined almost 70 loans and compared them to the wide range of Chinese renderings in the Tripiṭaka. He reached the conclusion that Chinese transmission can be detected in very few cases (2013, 280–81).

91

The phonetics of the Chinese language is fairly distant from Sogdian (like any other any Indo-European) even if in the Tang period it possessed more means to record postvocalic consonants than in modern Mandarin.

92

Chinese loans in Sogdian texts (such as the names of eras) are often detectable by their ‘atomistic’ outlook: monosyllabic characters are written as separate words that are usually shorter than Sogdian lexemes (see Chinese in Sogdian transcription in Yoshida 2013, 169ff., pl. I).

93

So, it is easy to see this convention in ’n’wxt’r’ s’m myγ s’m pwδ’y – anuttarasaṃyaksaṃbodhi which passed through Chinese a-nou-duo-luo san-miao san-pu-ti 阿耨多羅三藐三菩提 (EMCh ʔa-nəw^h-ta-la sam-mjiaw’ sam-bɔ-dεj; see already Gauthiot and Pelliot 1926, 2:66)⁵⁵ as compared to ’nwtr’y’n sm(’)yk smpwδ’y (loc. sg. anuttarāyāṃ, see Yoshida 2010, 90), which is attested in other texts and was borrowed from Buddhist Hybrid Sanskrit directly. The Chinese mediation here is documented not only with rendering n’wx for nəw^h and myγ for mauk, but also with the separate writing of the prefix sam-.

94

The names of four continents in the Sogdian Mahāparinirvāṇasūtra: cympwδ’y –Jambudvipa, ’wt’nwr – Uttaravati, k’wy’n’ – Godānīya, pwšp’δ’y – Pūrvavideha are difficult and at least two were borrowed from Chinese⁵⁶: Yanfuti (閻浮提 EMCh jiam-buw-dεj), Yudanyue (欝單越 ʔut-tan-wuat), see extensively Provasi (2013, 217–19, 264–65). The Sogdian forms, however, do not look like Chinese loans; perhaps the translator of the Mahāparinirvāṇasūtra did not know the Indian forms and tried to reconstruct them from the Chinese transcription.⁵⁷ The same development might be attested in kyδy’ – Jetavana, Chinese Qidia (祇陀, gji-da, see Provasi 2013, 216–17). The form cwδ’yk’ – Jyotiṣka, Chinese Shutiqie (樹提伽, EMCh. ʥuə̆ʼ dεj gɨa), unlike the doublet c’wtyšk’ might be considered Chinese, too (Sims-Williams 1983, 138).

95

A special case is Sogdian ’’myt’, which is much closer to Chinese Amituo (阿彌陀 ʔa-mji-da) than to Sanskrit Amitābha, Amitāyus (Yoshida 2010, 88–89; Provasi 2013, 199). As for Chinese Amituo, this form did not appear from nowhere, and many suggestions have been made on the shorter Indic prototype;⁵⁸ the recent suggestion of Jan Nattier (2006, 2007) that a North-Western Prakrit form *Āmitā’a from Amitābha was the source of Lokakṣema’s translation should be mentioned. Moreover, the figure of Amitābha/Amitāyus, and his paradise in Sukhāvatī, has been considered an Iranian plot which was incorporated in Mahāyāna, and the early translators of sūtras on Amitābha were of Parthian, Bactrian, Sogdian⁵⁹ and Khotanese origin (Scott 1990, 68–71). With these reservations, the Chinese source of Sogdian ’’myt’ cannot be taken for granted.

96

In the remaining cases, we notice Chinese mediation only in a vowel shift or other minor phonetic features, the transcription appears to be combined from both Indian and Chinese sources: sywpwδ’y, sypwty – Subhūti, Chinese xu-pu-ti (須菩提 EMCh. syou-bou-dei, see Sims-Williams 1983, 138); maybe twr’w(n)[wtn if for Droṇodana, Chinese Tulünuotanna (途慮諾檀那 EMCh dɔ-liə̆^h-nak-dan-na’). One can compare pn nyrβ’n parinirvāṇa with Chinese (da)ban nie pan (大般涅槃 EMCh pan-nεt-pan), see Sundermann (2010, 77). The dissimilation of r – r could however appear in an Iranian context as well (under 2. f). Characteristic is pyms’r – Bimbisāra, where the loss of the second b agrees with the Chinese Pingsha (洴沙EMCh. bεjŋ-ʂai/ʂε:), while the second syllable follows Sanskrit well. Sims-Williams (1983, 138; Yoshida 2010, 90) explains Sogdian pk’β’m as a loan from Chinese Pojiafan (婆伽梵 EMCh ba-kia-buam^h)⁶⁰. This term for Buddha is most common in Chinese texts and appears twice in Sogdian (once in the transcription of Skt invocation). Much more common is the epithet βγ’n bxtm, which is similar to other Central Asian traditions: Skt. Devatideva, Gāndhārī devadideva, Khotanese: gyastānu gyastä balysä, Manichaean Parthian bg’n bgdwm, Tocharian B ñakteṃts ñakte pūdñäkte (pañäkte), Uyghur täŋri täŋgrisi burxan.

97

We have to bear in mind however that many names and technical terms of Chinese Buddhist texts were translated and not transcribed, and in some cases the source of translation is detectable: Sanskrit kleṣa is translated as wytxwy (’t) sryβt’m “suffering (and) pain,” rendering Chinese fannao 煩悩, “vexation – annoyance” (Yoshida 2015, 838) and bodhisattva ’’k’c ptc’’n “concealer of empty space” stands for Chinese Xukongzang, 虚空藏, “empty-space-conceal” (Lurje 2010, 68), while pw px’rš pw nm’n’k βγpyδr’k “prince without retreat without repentance” is translated from Chinese butuizhuan tianti (不退轉天子), “no-retreat-return-god’s-son” and not from Sanskrit Devaputra Avaivartika (Lurje 2010, 312). Eight of 32 names of Buddhas and 16 of 42 names of Bodhisattvas attested in Sogdian are translations and not transcriptions (Lurje 2010, 523–24).

h. Buddhist Hybrid Sanskrit as the Source of the Main Body of the Borrowings

98

Theoretically, in the case of many standard borrowings there is a possibility that they entered Sogdian from a mediating language: e.g. Sogdian knkmwny – BHS Kanakamuni could be borrowed from Tocharian A Kanakamuni, Khotanese Kanakamāmṇa,⁶¹ Old Turkic Kanakamuni or Chinese Jianuojiamouni (迦諾迦牟尼, MCh Kia-nak-kia-muw-nri, see Provasi 2013, 218–21). None of these forms is at odds with Sogdian rendering.

99

In very many other instances, however, these parallel renderings are significantly different from Sogdian. From a general point of view (according to Research in the Social Scientific Study of Religion or Occam’s razor), it is advisable, however, to formulate a simple scheme, namely to derive, as much as possible, the material at hand as coming from the single source, and to explain eventual deviations as much as possible as features of the Sogdian language (see 3. a., 3. b). Only if the shape of the loan does not fit this explanation, or if there are significant cultural or philological considerations,⁶² can we adduce mediation of a tertiary language or tradition (3. c. – 3. g).

100

The features of Western Iranian, Prakrit, Tocharian, or Chinese transmission of the borrowings in question are present in a limited number of cases, and each one received special attention above.

101

Returning to the main body of material in standard transmission (in the conspectus), we should evaluate arguments contra ascribing them to any of the neighboring languages of the Buddhist teaching.

102

These borrowings cannot come from a ‘regular’ Prākrit (or Pāli) since they do not show coincidence of sibilants or simplification of consonant clusters. They cannot come from the Gāndhārī (North-Western Prakrit) because of the preservation of consonant clusters or the absence of lenition of postvocalic consonants (such as j > y). These borrowings did not pass through Parthian or Bactrian because of the different rendering of sibilants (3d, n. 44), the absence of such features as changes in consonant clusters, restoration of v into b, or the fricativization of non-initial stops.

103

Tocharian A or B cannot be the source of these borrowings because of the absence of vowel quantity and quality shifts, the preservation of -v as -β and not -p, and the few detectable traces of the voiced articulation of stops (see under 2. e.). We do not see in these borrowings the shift of vowels, or the lenitions typical for Khotanese. Uyghur Buddhist texts cannot be the source of these borrowings for obvious chronological grounds, being later than Sogdian ones, let alone the absence of vowel quantity in Uyghur.

104

We do not see Chinese imprints in vowel change or the loss of postvocalic consonants. There are no cases of syllable-by-syllable spellings save for one example.

105

In the majority of cases there is no contradiction to the assumption that Buddhist Hybrid Sanskrit was the main source of these borrowings (see Yoshida 2009, 291; Provasi 2013, 277).

106

However, there are three features that are notable exceptions from a straightforward rendering: the transmission of retroflexes as r; the rendering of j as t and the representation of the four postvocalic dentals as t or δ (2c, 2b, 2d). The first is explainable through the absence of retroflexes in Sogdian and the proximity of their articulation place to the sonants l, r. A similar development is visible in Khotanese and Chinese rendering (as in Sogdian βyr’wr’y, Khotanese värūlei, vīrrulai, Chinese Piliule, 毗留勒 EMCh bji-luw^h-lək) and in numerous Indic varieties (e. g. Pālī vēḷuriya from vaiḍurya). However, we do not encounter rendering of retroflexes through /l/ in Sogdian. The spelling of j as t /d/ or /t/, as if reducing the affricate /dž/ to its first element, is unusual. One can notice that a voiced affricate did not exist in the Sogdian phonetic system. It is more common to render a foreign /dž/ as c or ž in loanwords.⁶³ For the representation of postvocalic t, th, d, dh as either <t> or <δ> see the discussion under 3. e.

107

It is evident of course that the words in question were borrowed from the recorded literary language. However, the literary language was pronounced in some way which was more or less close to the written text as we perceive it now. When writing down loanwords, the Sogdian translators, using their own script, had nothing else to do but to record the pronunciation as they heard or learned it, since they could not just copy the writing (as is commonly the case of Tocharian or Khotanese written in Brahmi script). With more experience, however, the translators worked out certain rules of rendering, but one cannot expect a fully systematic scheme.

108

I suppose that the source of the major part of the borrowings is Buddhist Hybrid Sanskrit, pronounced in the articulation of Central Asian monks (for none of them Sanskrit was mother tongue!), and recorded according to the perception of foreign sounds by the Sogdians. However, I cannot offer definitive explanations for the peculiarities discussed earlier in this section.

109

As is well known, for Buddhist loans in various languages there is a feature of reversed stratigraphy: in the earlier records we observe more advanced Middle Indic forms, but in the later texts, more archaic standardized Sanskrit ones are predominant (e. g. Karashima 2006 for Chinese; Salomon 2002 for Gāndhārī; Sims-Williams 1983, 137 for Sogdian).

110

I propose to explain the predominance of BHS loans in Sogdian as an indication of a relatively late date (after the seventh century) of the bulk of the translations. These loans are thus later than those of the Indian words that entered common Sogdian and Manichaean usages and form an older layer of Indic loans attested in Buddhist texts.

i. Incomprehensible Loans and Distribution of Non-Standard Loanwords within Sogdian Buddhist Texts

111

Although the great majority of the names in Buddhist Sogdian texts have been explained satisfactorily, there remain some that withstand any attempts. One of them is the name of the fish king in the story in the P2 manuscript, pr’wxy (L). In the Sogdian version of the Śukasūtra (L-93) we encounter a number of incomprehensible names: c’wš’r, k’wšr’t, šwr’t, mx’wr without Indian or any other satisfactory explanation at hand;⁶⁴ the other names from this text š’ymnkr Kṣemaṃkara (cf. 2. g) and pyms’r Bimbisāra are irregular, too (Rosenberg 1920).

112

Some other texts are characterized by irregular, somewhat artificial names. One of them is in the Berlin manuscript of the Saṃghāṭasūtra with the names ’ykr’šn, βympx’r, srpšwr, p’tsyn. They have been explained as prakritisms and pseudo-Sanskrit retrospections (Yakubovich and Yoshida 2005; Yakubovich 2013, 42–45).

113

In the Sogdian Vessantara Jātaka we see many re-etymologized names: mntryh, kršny’n, r’zβwrt. For the name of the main protagonist swδ’’šn foreign explanations, Parthian and Gāndhārī, have been proposed.⁶⁵

114

In general, it is safe to assume that non-standard loans appear in certain Buddhist Sogdian texts of rather popular nature, with free narration, lacking footprints of verbatim translation from Chinese, and having even an elaborate style in the case of well-preserved Sogdian Vessantara Jātaka (Yoshida 2015, 840 with literature).

a. The Perception of the Buddha by Lay Sogdians

115

The Buddhist Sogdian texts were of course written and read by a limited circle of educated people, principally monks, but rich laymen acted as sponsors. Even if we can presume a relatively high degree of literacy among the Sogdians, the beautiful sūtra script of these texts indicates professional scribes. A large number of Sogdian settlers in the Turfan and Dunhuang area and elsewhere continued to profess their native religion (Grenet and Zhang 1998). Others were Buddhists, Manichaeans, or Christians. The colophons to Sogdian Buddhist texts (P-8 and the Ōtani fragment) as well as the recently discovered Sino-Sogdian funerary inscription (Sims-Williams and Bi 2020) show that within one and the same family of patrons of Buddhism, various persons had Buddhist and non-Buddhist names. In the mentioned epitaph, the An (Bukhara) clan of the sponsor was comprised of possessors of Sogdian names wys’k, w’ywš/wy’ws, srδm’n, ’’δprn, while only the third son of the deceased laywoman had the name pwtyδβ’r “given by Buddha.” Some theophoric names based on Buddha appear in other colophons and inscriptions in Sogdian⁶⁶, and many more in Chinese documents from the Turfan and Dunhuang areas, which were collected by Yoshida (2017, 47, elaborating earlier versions; see also Wang Ding [王丁] 2019, 192–95). We put them side by side with standard Sogdian theophoric names based on the popular goddess Nanaia and the Moon-god Mākh to show the similarity of the formation.

116

This table shows that in the eyes of ordinary lay Sogdians, Buddha was considered a kind of benevolent deity worthy of devotion of a newborn’s name after him. Of course this view was not restricted to Sogdians and such names are attested in many other people’s versions of Buddhism. In many cases, as said, these names are mingled with ones based on Sogdian deities.

b. A Newly Discovered Buddhist Wood-Carving from Panjakent

117

Much literature has been devoted to the remains of Buddhist art in Sogdiana. The surveys of Tigran Mkrtychev (2002) and Matteo Compareti (2008) are in general up-to-date and reliable sources on this issue. They are now supplemented with the collective volume Religions of Central Asia and Azerbaijan. III. Buddhism (Voyakin et al. 2019). Within the vast range of late antique Sogdian art, the Buddhist remains are very few. Apart from dubious architectural remains and one mural painting from a secondary level in Panjakent, Buddhist artefacts of Sogdiana are limited to a few sculptures in terracotta and metal (the latter being a Chinese import). In some other murals the Buddhist subjects can be considered sources of inspiration for the non-Buddhist messages. On the next pages I add several little known additions to this corpus, one of them being doubtful.

118

In 2016 the Panjakent expedition found a burned wooden plaque at area XXVI in the eastern block of the city (Kurbanov and Lurje [Lur’e] 2017, fig. 1). The stratigraphic considerations suggest that it was abandoned before the fire of 722 (although the traces of fire in this part of the city are not visible). The semi-oval shape of the plaque 58 x 53 cm, tentatively suggests that once it decorated a lunette above a door. Its sides however did not survive. The central figure (of which the upper quarter is lost) is sitting cross-legged on what one can considered a stylized lotus flower and is raising his right hand. His dress with a triangular collar resembles garments on Panjakent murals and at the same time the ‘Buddha parè’ of Fundukistan in Bactria (e. g. Rowland 1961, 20–24). A smaller-sized donor on bent knees is offering something under the lotus on the right-hand side, and traces of another donor can be detected on the opposite side, and maybe another smaller lotus was further to the right.

119

This panel is the first detected example of Sogdian wood carving with a Buddhist scene, and probably represents a well-known scene of adoration. The iconography of the panel, however, witnesses Sogdian style (as in the case of painted and terracotta Buddhas from Panjakent): the lotus flower features typically Sogdian petals and the Buddha has thin, elongated legs with small feet, like the characters in the murals of Panjakent and unlike the heavy limbs of early Buddhist style.

c. Sealings from Kafir-Kala and Some New Plastic Finds

120

The site of Kafir-kala near Samarkand, probably the countryside palace of rulers of Sogdiana, ikhshids, was the spot of numerous finds of well-preserved clay bullae which were discovered in the fire layer of 711. There is a set of offprints of one seal with a figure in full face sitting cross-legged on an ornamented throne (fig. 2); Sogdian inscription appears on two sides of the character. The excavators proposed to see here a depiction of a Buddhist character with the features of Avalokiteśvara, Amitābha, and Padmapāṇi (Berdimuradov et al. 2016; Bogomolov 2019, 339–45). An alternate interpretation may be preferred.

121

In the inscription the Turkic word x’ttwn “lady, wife of the kaghan” is clearly written in Sogdian script and on its other side there is a less comprehensible Turkic name that includes ’p’ “elder” (Berdimuradov et al. 2016, 53). It is logical to assume that a Khatun was depicted on the seal. Very peculiar is the headgear (or halo, as the authors suggest) with three tips. This kind of headdress is common among the depictions of Old Turkic noble women. The best example is apparently the petroglyph on the lost funerary statue from Kudyrge in Highland Altai, where the lady is depicted as well (Azbelev 2010 with literature; Yatsenko 2013, 75). The double portrait coins from Tashkent oasis are particularly close in date, place, and iconography (Shagalov and Kuznetsov 2006, 75–86, 308). One of the types is inscribed by Ton Yabghu Kaghan, the leader of the Western Turks in the early seventh century. Recently, the observations linking the sealing with noble Turkic ladies were also expressed in Japanese by A. Begmatov (2017, 208–9) and in Uzbek by G. Boboyorov (2020, with an untenable re-reading of the inscription); Yutaka Yoshida informed me that he reached a similar conclusion in an unpublished lecture he held in Stockholm in 2018.

122

Other features also agree with the interpretation of the seal with Turkic Khatun. Her right hand does not show a vitarka mūdra but holds a small open bowl with two fingers (which was mistaken as a collar), in agreement with stone statues of the Turks (Sher 1963). The crescent to the right is probably the lip of a flask with two handles that can be seen as well. The ribbons from her shoulders must be regarded as a Sasanian cultural practice that was widely followed in Sogdiana. All in all, in my opinion, this seal is an important indication of the role of the Turks in pre-Islamic Sogdiana and the position of noblewomen within their society, but has little to do with Buddhism.

123

One should add here that a fragment of a bronze figurine of the Buddha enthroned (probably an importation, or rather a somewhat careless recasting of an import) was found recently in Kanka in the Chach (Tashkent) area (Bogomolov and Musakaeva 2020, 288–90). The Buddhist attribution of several minor bronze plastics from Fergana (Bogomolov 2019, 346–48) remains unproven, in my opinion.

d. Toponymic Data

124

Both Xuanzang and Huichao (see under 1) mentioned one or two Buddhist monasteries in Samarkand, albeit in weak condition. There are very detailed descriptions of early Muslim Samarkand (and other cities of Sogdiana) and among the gates of the city one finds Nawbahār gates. Persian Nawbahār (Nava Vihāra) is the name of a Buddhist monastery in Balkh, famous for its richness, from where the mighty Abbasid viziers of the Barmakī family originated (De la Vaissière 2010, with literature).

125

The Nawbahār gates were located in the western wall of the city (madīna) of Samarkand and in the eastern wall of the inner suburb (rabaḍ) of Bukhara,⁷³ with two settlements nearby (Bol’shakov 1973, 221–25, 243, 252–3, with literature).

126

From what is said above one can suppose that the Buddhist monasteries were located near the Nawbahār gates in Samarkand and Bukhara, as in fact was once proposed by W. Barthold (1928, 102 n. 4). Following the same logic, Richard Bulliet (1976) considered all the places named Nawbahār of medieval (in Beihaq in Khorasan and near Rayy close to Tehran) and modern Iran to be places of Buddhist structures.

127

In later Persian literature, the word Nawbahār had the meaning ‘beauty, beloved’ since the “idol” of the Balkh temple was considered an ideal. It was easily contaminated with Persian naw bahār “early spring,” the time of great outdoor festivities of Iranians. Consequently, not every place called Nawbahār will have a Buddhist origin.

128

The proof that vihāra is reflected in the Nawbahār of the names of gates of Samarkand and Bukhara can be found in the gate-names of less Persianized cities within the Sogdian sphere of influence in the early Islamic period, where the Sogdian word for vihāra, βrx’r, survived (see Lurje [Lur’e] 2013, 229, 250 n):

129

Among the gates of Isfījāb city on the Arys river (Sayram in modern southern Kazakhstan) there was a gate which can be reconstructed as Farxār (in Arabic script, f was commonly used for the labiovelar β which did not have a parallel in the Arabic phonetic system).⁷⁴

130

Similarly, one of the gates of Binkaθ, the capital of the Tashkent oasis inside the modern city, had the gates which we restore as Farxār from defective writing in the Muslim manuscripts.⁷⁵

131

Of course, the etymology of Farxār / modern Parxor in the north-east of historical Bactria (southern Tajikistan, on the Panj river, above the confluence with Vakhsh) is the same.⁷⁶ Near Parkhor, the late antique and medieval site of Zoli Zard is located (Jakubov and Dovudi 2012). An accidental find of a small limestone head of Buddha from Zoli Zard has been reported recently: https://www.ozodi.org/a/30635395.html (last accessed 03.07.2020); one can recognize Bactrian βοδο in archaic rhombic-shape lapidary script to the right of the face.⁷⁷

a. Standard Borrowings