1.
According to the classical treatments by Tedesco (Reference Tedesco1921: 199 f.) and Henning (Reference Henning1958: 96 f.), Proto-Ir. *θw (PIE *t
) gives Middle Persian h (MP čahār ‘four’ vs. Avestan čaϑβārō, OInd. catv
raḥ; MP čihil ‘forty’ vs. Avestan čaϑβar əsat-, OInd. catvāriṃśát-; abstract suffix MP -īh < *-iya-θwa-), but f in Parthian, as in čafār ‘four’, čafrast ‘forty’, and the abstract suffix -īf in inscriptional Parthian.
This interpretation of the Parthian data needs to rely on explaining the additional -t seen in -īft, the variant of the abstract suffix seen in Manichean Parthian, as an additional suffix (Tedesco Reference Tedesco1921: 200 suggests a derivation from *-iya-θwa-tā-). However, this approach does not offer an explanation for the Pth. verb (present stem) <nydfʾr-> niδfār- / (past stem) <nydfwrd> niδfurd ‘to hurry (intr. and tr.)’, which is likely to derive from *ni-θwāraya- (cf. OInd. √tvar), and the noun <nydfʾr> niδfār ‘haste’ (Henning Reference Henning1958: 97 n. 2).Footnote 2 While Henning's etymology is certainly convincing, his further suggestions are less so: he assumes that the word-internal result of Proto-Ir. *θw is Pth. -f-, while δf in ni-δfār- would show the result in word-initial position (for which there is no other example), and that δf would have been adopted from the (unattested) simplex *δfār-. This scenario is improbable not only because it implies the unlikely assumption that a cluster that is reduced to -f- in word-internal position would be retained word-initially, but also because the parallel consonant cluster PIE *d > Proto-Ir. *δw is reduced to Pth. b- word-initially (Sims-Williams Reference Sims-Williams and Penney2004: 540).
Sims-Williams (Reference Sims-Williams and Penney2004: 540, 545) thus suggests the alternative solution that δf is the regular result of *θw in word-internal position. For čafār ‘four’ he assumes a dissimilatory loss of the dental elements of the consonant cluster (*[tšaδfār] > [tšafār]), a development that also occurred in this word in other Ir. languages (e.g. Bactrian σοϕαρο ‘four’ vs. regular λϕ < *δf in αλϕανζ- ‘attain’ < *θwanǰa-, abstract suffix -ιλ(α)φο, Sims-Williams Reference Sims-Williams and Penney2004: 542). For the word-final position, he posits a dialectal difference in the further development of *-δf > -f for inscriptional vs. -ft for Manichean Parthian (Sims-Williams Reference Sims-Williams and Penney2004: 543, 546).
2.
This set of changes is so far based on one example of each, but there seems to be additional evidence confirming Sims-Williams' assumption that *θw gives Manich. Pth. -ft, also implying that the abstract suffix -īft does not contain an additional suffix.
2.1.
The word <pwrt> ‘bridge’ occurs in the Pth. hymn cycle Angad Rōšnān VI 57b.Footnote 3 Although this is a hapax legomenon, its reading and meaning are reasonably clear. The existence of such a word in North-West Iranian is also confirmed by Gilaki purt, purd and Zazaki pırd ‘bridge’.Footnote 4 Etymologically it is obviously related to Avestan pərətu- (cf. Boyce Reference Boyce1954: 194: “< *pərətu-”). However, a derivation from Proto-Ir. *p
tu- would raise a problem on the phonological side: Proto-Ir. *t following a labial otherwise, and expectedly,Footnote 5 gives Pth. <wrd> -urd, e.g. <bwrd> burd < *bta- (past stem of <br-> ‘carry’), <mwrd> murd < *mta- (past stem of <myr-> ‘die’). Proto-Ir. *ptu- should thus have given †<pwrd> purd.
So it is worth considering whether Pth. <pwrt> could derive from the oblique stem *pθw-, i.e. from the form that has always been seen as underlying the MP cognate puhl (*pθw- > *purh > puhl, Hübschmann Reference Hübschmann1895: 195, 207, Hoffmann Reference Hoffmann, Schmitt and Skjærvø1986: 171, 181 n. 20).Footnote 6 The application of the change suggested by Sims-Williams for Manich. Parthian (see Section 1) yields *pθw- > *purδf > *purft. Since a consonant cluster -rft is not permitted by Pth. phonotactics,Footnote 7*purft could have been reduced to purt by a dissimilation vs. the initial p- that is not unlike that in čafār.
2.2.
A derivation of <pwrt> purt from *purft < *pθw- suggests a parallel explanation for Pth. <mwrt> murt ‘death’Footnote 8 from *murft < *murδf < *mθw- (Nicholas Sims-Williams, personal communication). A dissimilatory loss of f in *murft is surely as motivated as it is in *purft. On the other hand, *mθw- would be the oblique stem of an as yet unknown Ir. stem *mtu- besides the otherwise attested *mθyu- (Avestan mərəϑiiu-, Old Persian (uv
-)məršiyu-,Footnote 9 OInd. mtyú-), but a stem *mtu- / *mθw- ‘death’ is indeed reflected in Sogdian mwrδw /murθú/.Footnote 10 This is likely to derive from the nominative and accusative forms *mθuš and *mθum Footnote 11 while a derivation from *mθyu- should effect a palatalization of the vowel (Sims-Williams, personal communication). Similarly, the derivation of Pth. <mwrt> from Proto-Ir. *mti- suggested by Henning (Reference Henning1937: 85) should probably give †<myrd>, cf. *kta- > <kyrd> (past stem of <kr-> ‘do’), *mya- > <myr-> ‘die’.Footnote 12
So far as the existence of *mθw- in Sogdian is concerned, the word is found in B pyšmwrδw Footnote 13 ‘after death’ and in the phrase M zʾδmwrδw ‘birth-death’, B zʾt (ʾt) mwrδw ‘birth (and) death’ (i.e. circle of reincarnation, saṃsāra), where zʾδ- shows a change of Old Ir. *-t that otherwise does not occur in Sogdian. So Benveniste (Reference Benveniste1940: 216) assumes a Pth. origin while the original Sogdian phrase would be ʾʾzy myry ‘birth-death’.Footnote 14 Indeed, Pth. <zʾdmwrd> zādmurd is quite well attested, and Pth. influence in the Sogdian Buddhist lexicon has been noted for other words as well.Footnote 15 However, while Pth. influence in the use of Sogdian (°)mwrδw and in the formation of zʾδmwrδw is possible, the assumption of a direct borrowing is faced with the difficulty that the attested Pth. forms are in fact <mwrt> murt and <zʾdmwrd> zādmurd.Footnote 16 It would also be unlikely that Sogdian borrowed mwrδw from the stage of Pth. *murδf, since one would expect Pth. δf to be rendered by Sogdian <δβ>. Such an output may be seen in Sogd. pwtysδβ ‘Bodhisattva’ (besides variants such as pwt/δystβ), which could owe its <δβ> to Pth. <bwd(y)sdf> bōdisaδf (thus Sims-Williams Reference Sims-Williams and Penney2004: 544 f., see also Section 3).Footnote 17
If mwrδw is thus an inherited Sogdian form, Sogdian would show several words for ‘death’ (cf. mrc, B mwrtk(y)). The stem *mθu- underlying Sogdian mwrδw would derive from a paradigm *mtu- / *mθw-, the oblique stem of which yields Pth. <mwrt>.Footnote 18
3.
Another item to be considered in the discussion of the Pth. result of Proto-Ir. *θw is the inscriptional Pth. form <nytprywt> ‘hurried’, corresponding to Manich. Pth. <nydfʾr> etc.Footnote 19 It seems that the most straightforward interpretation of the <-tp-> is tf. In this case, one might consider a modification of the changes noted in Section 1.
Proto-Ir. *θw could have yielded Pth. tf first, which would be shown by inscriptional <nytpr-> nitfr-. In ‘four’, a dissimilation *[tšatfār] > [tšafār] <cfʾr> would have taken place.Footnote 20 Word-final *-tf would have undergone a metathesis to -ft in Manich. Parthian and a reduction to -f in the dialect of the Pth. inscriptions, thence the abstract suffix Manich. -īft <-yft>, inscriptional -īf <-py>. The output of Proto-Ir. *pθw- and *mθw- would have been reduced to <pwrt> purt and <mwrt> murt by the phonotactic ban on tautosyllabic clusters of three consonants (cf. note 7), either at the stage of *purtf and *murtf or in the metathized stage of *purft and *murft.
The next stage would assimilate the tf to δf. This would have concerned word-internal cases of tf other than ‘four’, thence Manich. <nydfʾr-> niδfār- and derivatives vs. inscriptional nitfr-, as well as borrowed tf, which is likely to be seen in <bwd(y)sdf> bōdisaδf ‘Bodhisattva’ and <sdf> saδf ‘being (sattva-)’.Footnote 21
This approach appears to account for the data in an economic way and motivate the dissimilation in čafār particularly well. A development of word-final *θw > *tf > -f(t) also seems to be more straightforward than *θw > *δf (> *-fδ ?) > -f(t). Pth. *θw > tf is also quite parallel to Sogdian and Khwarezmian *θw > θf (Sims-Williams Reference Sims-Williams and Penney2004: 541, 543), agreeing with these being “closely related languages” (Sims-Williams ibid.), and Bactrian *δf (> λϕ) would correspond to the stage of Manich. Pth. word-internal -δf-.
Alternatively, we could consider an interpretation of both inscriptional <tp> and Manich. <df> as θf (Jost Gippert, personal communication), comparing it to Avestan fəδr- (oblique stem of pitar- ‘father’), which is likely to reflect /fθr-/,Footnote 22 and to the development of word-internal *δw > Pth. <db>, if this is δv as per Sims-Williams (Reference Sims-Williams and Penney2004: 540). However, the assumption implies that one would need to posit word-final developments of *θf > -f; *θf > -ft; plus *-tf > -θf for loanwords to account for <bwd(y)sdf> and <sdf>, a set that is perhaps not altogether compelling. Hence a development *θw > *tf > -f(t) appears to be preferable.
4.
There is another piece of evidence which is incompatible with the classical view of the development of *θw in Western Iranian. MP nixwār- (Manich. <nyxwʾr- >, Pahlavi <nswbʾl->) ‘hurry, hasten, incite’ is obviously a cognate of Pth. niδfār-, but čahār ‘four’ and čihil ‘forty’ would lead one to expect MP †nihār-.
In view of the discussion above and of the laconic note by Sims-Williams (Reference Sims-Williams and Penney2004: 540) “[nixwa:r-] < [nihwa:r-]”, one may wonder whether the MP development of *θw is not as traditionally assumed either, and could posit the assumption that nixwār- < *nihwār- < *niθwār- shows the regular MP result of *θw in word-internal position. The reduction seen in čahār and čihil would then need to be due to a specific development here as well, which could have operated at the stage of *hw. A reduction of the consonant cluster would seem particularly likely in the multiple clusters arising in *čaθwθatam (cf. Av. čaθβar əsatəm) > *čahwirhat Footnote 23 (> via *čihwihl or *čihird ?) > čihil ‘forty’, whose -h- would have been transferred also to *čahwār > čahār. In word-final position one would need to assume a reduction *θw > *-hw > -h, which would operate in the abstract suffix -īh (< *-iya-θwa-) and in *pθw- ‘bridge’ > *purh > puhl.Footnote 24 The adverbial suffix -īhā would need to have generalized h by paradigmatic levelling from -īh.Footnote 25
This approach implies ad hoc assumptions for čahār, čihil and -īhā, but accounts for nixwār-, which is otherwise left without explanation.Footnote 26 Moreover, a development *θw > *hw > xw agrees quite well with other MP sound changes: *θ yields MP h generally (e.g. pahn ‘wide, broad’, mēhan ‘home’ vs. Avestan paϑana-, maēϑana-, Hübschmann Reference Hübschmann1895: 203). The sequence *hw < *θw merges with old *hw < PIE *s, both resulting in MP xw.Footnote 27 Also parallel is the development of *fw > MP hw (kahwan ‘old’ < *kafwan, Bailey Reference Bailey1979: 62b, 64b). But this development needs to be later than the change *hw >xw discussed above, as the hw arising from *fw does not yield xw.Footnote 28
5.
5.1.
The interpretation of Pth. <pwrt> purt ‘bridge’ and <mwrt> murt ‘death’ suggested in Section 2 implies that Manich. Pth. <t> and <d> encode two different phonemes also in the position after r. Now there appear to be exceptions exactly in this context: according to Boyce (Reference Boyce1975: 17), <t> otherwise encodes t, but “rarely” also d when in the position “after r (an archaic spelling), e.g. wrt- besides wrd- (ward-)”. This raises the question whether <rt> and <rd> are written indiscriminately and refer to the same pronunciation.Footnote 29 The data are as follows:Footnote 30
• inflectional forms of the verb <wrt-> / <wrd-> wart/d- ‘turn’;
• its derivatives <wrd(g)> ‘prisoner’, <wrdy(y)wn> ‘wagon’;
• its compounds and their derivatives: inflectional forms of <ʾmwrt-> / <ʾmwrd-> am-wart/d- ‘collect’ with <ʾmwrdn> ‘assembly (place)’, <ʾmwrdyšn> ‘collection’, <ʾmwrtʾdnyft> ‘assembly’; one inflectional form of <ʿzwrt-> iz-wart/d- ‘return’ with <ʿzwrdyšn> ‘return’; one inflectional form of <prwrt-> par-wart/d- ‘prevail’Footnote 31 vs. fra-wart/d- in <frwrdg> ‘letter (roll)’;
• <ʾrt> (< *arta-, Avestan a
a-, Old Persian arta°, OInd. tá-) besides <ʾrdʾw> (<*artāan-, cf. Avestan aauuan-, OP artāvan-, OInd. tvan-); both occur only in connection with <prwhr> in a designation of the ether (one of the Manich. elements of light). <ʾrt> could be an archaism of the religious language as is its cognate wrt- /urta-/ in the Sogdian version of the prayer Aəm vohū (Nicholas Sims-Williams, personal communication, cf. Gershevitch Reference Gershevitch1976).
If one explains <ʾrt> as an archaism or a borrowing from an older stage of the language, Pth. wart/d- ‘turn’ with compounds and derivatives is the only case of a variation <d> / <t> in Manich. Pth. orthography.Footnote 32 At the same time, wart/d- is the only instance of Pth. <rt> other than <pwrt> and <mwrt>.Footnote 33 The remaining cases are loanwords or unclear:
• < sʾrt> sārt ‘caravan’ and <s(ʾ)rtwʾ> sartwā ‘caravan leader’ are borrowed from OInd. sārt ha- and sārt havāha- (as is Sogdian sʾrth, Sims-Williams Reference Sims-Williams, Röhrborn and Veenker1983: 133, 135, 140);
• two items are unclear: the hapax <ʾwrtʾdgyft> (thus Sundermann's reading of <(ʾ)wr(t.gy)ft > , cf. DMD 70a), perhaps it belongs to <wrt/d->; and <ʾmrtyn> (twice attested), for which Henning (apud Sundermann Reference Sundermann1973: 115) assumes a connection to Avestan a
a-.Footnote 34
5.2.
The following points may be relevant in evaluating the orthography <rt/d>: Old Ir. t usually gives Pth. <d> post-vocalically and after sonorants, and also after r, e.g. <mrd> mard ‘man’ (Av. marta-), <mrdyft> mardīft ‘manliness’, <srd> sard ‘cold’ (Av. sar əta-), <srdʾg> sardāg ‘cold (noun)’, <wxrd> wxard ‘eaten’ (< *hwar-ta-), <wxrdyg> wxardīg ‘meal’, <nbrd> nibard ‘battle’, <nbrdg> nibardag ‘warlike’, <kyrd> kird ‘done’ (Av. kərəta-), <kyrdgʾr> kirdagār ‘mighty’, <dyrd> dird ‘held’ (Av. dərəta-).Footnote 35 The voiced counterpart, Old Ir. rd, mostly yields Pth. rδ, e.g. <zyrd> zirδ ‘heart’ (< Proto-Ir. *zdaya-).Footnote 36 However, Old Ir. ard gives Pth. ār (Rastorgueva and Molčanova Reference Rastorgueva and Molčanova1981: 162), e.g. <wʾr> wār ‘flower’ (Av. var əδa-), <sʾr> sār ‘year’ (Av. sar əδa-). So there is an opposition between -rd < Old Ir. -rt and -rδ < Old Ir. -rd only for vowels other than a, but no †arδ < ard vs. ard < art.
Connecting the Pth. data to developments in other Ir. languages, one might wonder whether the mixed orthography <rt/d> after a intended to mark a specific pronunciation for which there was no orthographic convention – perhaps voiceless r + t as Durkin-Meisterernst (Reference Durkin-Meisterernst, Emmerick, Sundermann and Zieme2000: 173) assumes. Similarly, Av. <
>, which is the result of rt in certain contexts, has been assumed to represent voiceless r, retroflex ṭ, or a fricative similar to Czech ř (Hoffmann Reference Hoffmann, Schmitt and Skjærvø1986: 173 ff., de Vaan Reference de Vaan2003: 602). Also noteworthy is the occasional lengthening of Av. a preceding < > , e.g. x vāa- ‘food’ < *hwar-ta-.Footnote 37 In Balochi, Old Ir. *art gives ārt and *ard gives āṛ (e.g. wār-t ‘eats’ vs. war- otherwise; gwāṛag ‘blossom’Footnote 38 vs. Av. var əδa-) while rt and rd after other vowels are preserved.Footnote 39 Pashto likewise has retroflex ṛ from Old Ir. rt and rd, but this is independent of the preceding vowel.Footnote 40 So if the Pth. orthography <rt/d> did indicate a specific sound or sound cluster, the result of *art would arrange itself with similar phenomena in other Ir. languages.
It is not clear, though, why a variation <rt/d> is only found with the family <wrt/d-> and not with other words containing Old Ir. *art, or why a “specific pronunciation” is only marked for wart/d-.Footnote 41 Perhaps the variation <rt/d> marks the word-internal development, which is exclusively found in the only Pth. present stem with Old Ir. art,Footnote 42 while the word-final position shows the expected <rd> ard. Inflectional forms and derivatives such as <mrdʾn> mardān (plural), <mrdyft> mardīft, etc., were surely related to <mrd> mard ‘man’ by the speakers and thus do not undergo word-internal development, while a present stem mostly occurs with endings. If <rt/d> is the word-internal development, it is perhaps less likely that <rt/d> stands for a devoicing which would not have taken place in word-final position, and a retroflex or fricative output would seem more likely.
6.
Summarizing the argument above, Manich. Pth. <t> and <d> encode two different phonemes also in the position after r, and Pth. <pwrt> purt ‘bridge’ and <mwrt> murt ‘death’ are to be read as purt and murt. These words are likely to go back to *pθw- (the form from which MP puhl also derives) and *mθw- (while Sogdian mwrδw derives from *mθu- with generalized θ). These are the oblique stems of *ptu- and *mtu-, the former familiar from Av. pərətu-, the latter otherwise only found in Sogdian.
Pth. <pwrt> ‘bridge’ and <mwrt> are, then, additional evidence for Sims-Williams' claim that Proto-Ir. *θw does not yield Parthian f as previously assumed, but results in a consonant group, which would be reduced in Pth. *purft and * murft. By the logic suggested here, -ft would be the Pth. word-final outcome of *θw in Manich. Parthian (vs. -f in inscriptional Parthian) vs. -tf- (thus in inscriptional Parthian) > -δf- (Manich.) in word-internal position.
Middle Persian may likewise show a consonant cluster as the result of *θw, yielding *hw > xw. In čahār ‘four’ and čihil ‘forty’, specific processes must then have been at work to effect the simple h; these would be parallel to cluster reductions in these numbers in other Ir. languages.
Table 1 presents the Pth. sound changes of r and * + dental discussed in this paper in comparison with some data of selected Western Ir. languages.
Development of *r, * + dental
Examples for Zazaki include the cognates of Pth. words mentioned above: for *ard: se
‘year’; *d: zei ‘heart’ (Paul Reference Paul and Sims-Williams1998: 169), vılıke ‘flower’; *t: kerd-, berd- (past stems of ‘do’ and ‘carry away’); *art: serd ‘cold’; *θw: pırd ‘bridge’ (cf. Section 2.1). Since *t appears to give erd also in labial context (berd- < *bta-), one could perhaps consider vılıke a loanword (thus Paul Reference Paul and Sims-Williams1998: 169), so that the regular output of *d in labial context could be e or perhaps ı (cf. e.g. pı ‘full’, which at least shows * in labial context although not *t).
In Balochi, the contexts in which * yields ir and ur are not identical to those of MP and Pth. ir, ur. While ir is the result in palatal contexts and ur in labial ones, the neutral context shows Balochi ur, but MP / Pth. <yr> ir, e.g. Balochi turs- vs. MP, Pth. <tyrs-> ‘be afraid’, kurt vs. MP, Pth. <kyrd> ‘done’. Other examples include *d: zird ‘heart’, *ard: gwāṛag ‘blossom’, *art: sārt ‘cold’. Owing to the absence of other examples for the context *θw, it is impossible to decide whether Balochi puhl ‘bridge’ is a MP loanword or not (Korn Reference Korn2005: 143–8, 328, 121).
and 
for Proto-Iranian. As per Iranological tradition, italics represent the transcription (phonemical form) for Parthian and Middle Persian, but the transliteration (graphical form) for Sogdian. Manichean, Middle Persian and Parthian are quoted from and in the form of DMD unless otherwise noted.
