Contributions to a relative chronology of Persian The non-change of postconsonantal y and w in Middle Persian in context

Old Persian shows a change of postconsonantal y , w to iy , uw , respectively. However, if one applies (pre-)Middle Persian sound changes to the Old Persian forms, the result is at variance with certain Middle Persian forms. If one were to assume a syncope revers-ing the Old Persian change of y , w to iy , uw , this would also affect old cases of iy , uw and likewise yield incorrect results for Middle Persian. The Old Persian change can thus not have operated in the prehistory of Middle Persian, and there is a dialectal difference between attested Old Persian and the later stages of the language, which is to be added to those already noted. The paper also discusses some sound changes that are connected to the Old Persian change in one way or the other. Cases in point are the processes called Epenthesis and Umlaut in previous scholarship, which this article suggests to interpret as occurring in different contexts and in different periods.The former is limited to V ry , which yields V ir and feeds into a monophthongisation that, as shown by some late Old Persian word forms, occurred within Achaemenid times, giving ēr and īr from ary and əry . Epenthesis did not occur in the prehistory of Parthian, whereas the monophthongisationdid.TheAppendixpresentsatentativesequenceof theprocesses discussed in this article, which is intended as a contribution to the relative chronology of Persian historical phonology.


Introduction
As argued by Hoffmann (1976: 636-38), Old Persian (op) shows a change of postconsonantal y, w to iy, uw, respectively. The present article discusses evidence that this change did not operate in the dialect(s) on which Middle Persian is based; this difference between Middle Persian (mp) and attested Old Persian is thus to be added to those already identified.
This article is part of ongoing research on the chronology of Persian sound changes, which intends to proceed differently from what has been done so far: Traditionally, a "black box" method has been applied in studies on the historical phonology of Iranian, i.e., apart from isolated notes that a certain change X has to be earlier than change Y, they have limited themselves to contrasting an input-Proto-Iranian (Ir.), or even Proto-Indo-European (pie)-with an output (Old, Middle, or New Persian) without examining what happens in between. Conversely, the "glass box" approach tries to determine in what sequence the various processes take place, while at the same time also taking into account how they interact.1 This article will show examples of sound changes that yield incorrect results if one disregards how they interact.
The present attempt at presenting a relative chronology of some Persian sound changes is hypothetical in many details-necessarily so for lack of data in some parts-and awaits further refinement in others. It does try to establish what seems reasonably certain, and to offer additional points for subsequent discussion, laying out at each step the evidence (or lack of the same) on which the proposal is based. The suggestions systematically attempt to follow the principle of economy in terms of hypotheses: where two explanations seem equally viable, the approach which explains the data in the most straightforward way will be preferred. The Appendix (Section 5.) lists a preliminary tentative chronology of the sound changes suggested here; the numbers cited throughout the article refer to the items on this list, which, however, will be limited to changes discussed in this article. In the formulaic form of the sound changes, asterisks are omitted.
Dialectal differences between Old, Middle, and New Persian (np) are of central importance to this article. I will use the terms "Proto-Old Persian", "Proto-Middle Persian", and "Proto-New Persian" to refer to the dialect(s) from which attested Old, Middle, and New Persian, respectively, derive.2 For rea-sons of economy and also to avoid ex nihilo arguments, the assumption will be that Proto-mp is identical to attested Old Persian unless there is evidence to assume a difference (and similarly for Proto-np). Similarly, I will assume that Elamite renderings of Ir. word forms reflect Old Persian unless they show features implying a dialectal difference, and that Armenian has borrowed from Middle Persian and Parthian (Pth.) unless there is evidence to the contrary.3 The term "Classical Old Persian" will be used to refer to what seems to be the "normal" form of attested Old Persian, differentiating it from other dialects and from late stages of Old Persian.4 As in any discussion of Middle Iranian phonology, the evidence of Ir. loanwords in Armenian (Arm.) is of crucial importance. Expressions such as "Armenian shows …" are meant to refer to the Ir. items found in the Arm. lexicon, more precisely the older layer of (Proto-)Middle Iranian loanwords (showing inherited p, t, k in postvocalic position),5 which is most relevant for the present discussion.6 3 Some Armenian forms require the assumption of a third Western Iranian language as source, see Korn & Olsen (2012). 4 Brust (2018: 1) uses the term "Achaemenid Old Persian", but this has the unfortunate consequence that variations within the extant inscriptions such as ahmiy 'I am' in Xerxes' inscription in Persepolis instead of op amiy elsewhere (see 1.8) would qualify the former inscription as "non-Achaemenid". 5 See, e.g., Hübschmann (1897: 12-14) on this matter. 6 In the transliteration of Old Persian, I follow Schmitt (2009) and others by noting the potentially a-inherent signs as C (e.g., ⟨t⟩ instead of ⟨ta⟩ or ⟨ta⟩) and the signs with inherent i and u as ⟨Ci⟩, ⟨Cu⟩. I transliterate ⟨c⟩, ⟨j⟩, ⟨v⟩ following the tradition, but prefer č, ǰ, and w for the transcription; the latter seems a more likely pronunciation, as also shown by the processes to be discussed in this article. For practical purposes, I also use y, w instead of i̯ , u̯ for Proto-Iranian. I follow Werba (1993: 142-43) in noting the diphthongs as ai, au (see also n. 8), and Hoffmann, Mayrhofer, etc. in noting the Proto-Ir. outputs of pie *ḱ, *ǵ(h) as *ś, ź, respectively (Proto-Indo-Ir. *ć, j(h)). For discussion of Ir. *r̥ > op ǝr (which other authors transcribe ar or ạr), see 1.4. Unless otherwise specified, forms with asterisk (*) refer to Proto-Iranian and/or the forms underlying Old Iranian, abstracting from the specificities of Old Persian and Avestan (Av.). A list of abbreviations is in the Appendix.
The spellings of op words are cited from Brust (2018), with the adaptations just mentioned. Middle Persian is quoted from MacKenzie (1986) and/or Durkin-Meisterernst (2004) unless otherwise noted, while (classical) New Persian follows Steingass (1892) for the vowels. Quotes from works in languages other than English are my translations. 'you (acc.sg)' ⟨θ-u-v-a-m⟩ θuwām θwām dvar-'door' ⟨du-u-v-r-⟩ duwara-dwara-dar √cyu 'move' ⟨š-i-y-v-⟩ šiyaw-šyaw-šaw-*čyāti-(Latin quiēs) 'joy' ⟨š-i-y-a-t-i-⟩ šiyāti-šyāti-šād satyá-'truth' ⟨h-š-i-y-⟩ hašiya-hašyaanyá-'other' ⟨a-n-i-y-⟩ aniya-anya-any a Hoffmann (1976: 636-38). b Meillet & Benveniste (1931: 88-89, §144, 146). one would expect it to be written †⟨h-r-v-⟩; one would hardly write ⟨h-ru-u-v-⟩ with a ⟨Cu-u⟩ "representing a phonetic nothing"; the same argument could be made about ⟨du-u-v-r-⟩, which could just as well be written †⟨d-v-r-⟩ (or †⟨du-vr-⟩) if a form /dwara-/ was meant.12 12 One might be tempted to compare the Avestan ii and uu which render *y and *w in intervocalic and postconsonantal position. Hoffmann & Forssman (1996: 84) suggest that this orthography might be due to the influence of Old Persian and thus belongs to the transmission of the Av. texts, not to the Av. language while it was spoken. At any rate, metrical and other details show that the difference between historic iy vs. y (both written ii) and uw korn 10.1163/22125892-bja10009 | Indo-European Linguistics (2021) 1-43 Hoffmann (1976: 638) further argues that both inherited -iya-and the output of Ir. *ya are written ⟨Ci-ya-⟩ in Elamite, and old and secondary uwa are both rendered by ⟨Cu-ma-⟩ or ⟨ma-⟩ (cf. Mayrhofer 1973: 33, 85, 93).13 Moreover, both inherited and secondary iya are contracted in later Old Persian according to Schmitt (1989: 71, see 1.7, 2.1). Potential examples are op marīkā (voc.sg) ⟨m-ri-k-a-⟩14 'lower-rank / younger man'15 from *maryaka-(cf. OInd. maryaká-) via *mariyaka-16 and several possible cases of -y-a-> -iy-a-> -ī-where the contraction seems to operate across the morpheme boundary of the preverbs ni-, abiand the augment a-(*niy-a-> nī-; *abiy-a-> abī-from *ny-a-, *aby-a-as seen in OInd. parallel forms): vs. w (both written uu) is preserved in Avestan (Hoffmann & Narten 1989: 39-49). We can conclude that there are the Av. phonemes /y/ and /w/, which have word-initial allophones written with the signs transcribed ⟨y⟩, ⟨v⟩ and word-internal allophones written ii, uu.-See also n. 53. 13 ⟨m⟩ is the regular Elamite rendering of any op w in any position of the word (see the list in Mayrhofer 1973: 85-87). 14 None of the three instances (DNb 50, 55, 57) is fully preserved; the fact that the text is structured in parallel sentences may confirm the restorations. The rock shows both old cracks that seem to antedate the inscription and later ones, which does not facilitate the interpretation (see also the details in Schweiger 1998/i: 74 and ii: 215-21). The comparatively best instance is the one in line 50. Schmitt (2000: 34) reads ⟨m-r-i-•-k-a⟩, which has a lacuna in the middle (cf. the photo in Plate 18 of Schmitt 2000). It seems to me that the crack is old, and that no sign is lost (which is what Schmitt's ⟨•⟩ implies; thus also Sims-Williams 1981b: 3); also, the ⟨k⟩ is rather narrow, as if compensating for the space lost by the crack, and in the lines below, the crack continues in a way coherent with line 50, and no signs appear to be lost. As for the individual signs, I think that ⟨k-a⟩ are beyond doubt, ⟨m⟩ and ⟨i⟩ fairly certain, while ⟨r⟩ is not. 15 Schmitt (1999b: 129-31) argues in favour of marīka-meaning 'young man' (cf. mp mērag 'young man, husband') rather than 'servant' (which is the meaning of Bactrian μαρηγο). For cognates from other ie languages meaning 'boy / girl' and pointing to a possible root *mer(H) 'young' , see Milanova (2021: Section ii. 2.3.1). In the Aramaic version (cf. Sims-Williams 1981b), the equivalent passages are not preserved. 16 The argument by Brust (2018: 25-26, 187, 283) is circular: According to him, the middle signs in op marīka-are not readable (this is not entirely accurate, see n. 14), from which he concludes that there is no change iya > ī in Old Persian. From the supposed absence of this change he concludes (p. 187) that ⟨a-b-i-j-a-v-y-m⟩ cannot be an imperfect abī-ǰāwayam < *abiy-a-ǰāwayam (according to him, ⟨n-i-š-t-a-y⟩, ⟨n-i-š-t-a-y-m⟩ and ⟨n-i-š-a-d-y-m⟩, are injunctives as well). While ⟨a-b-i-j-a-v-y-m⟩ might indeed be an injunctive (abi-ǰāwayam), even if it would be the only injunctive in this inscription (among many imperfects and some aorist forms), a lack of readability of marīka-would not prove that there is no change iya > ī. Also, and independent from the details of marīka-and the interpretation of the cited verb forms, the change iya > ī clearly did take place at some point in the prehistory of Middle Persian and also in numerous other Ir. languages (see 1.7, 2.1); it thus seems to have occurred rather early.
I suggest that this change and the one discussed in 1.1 are in fact a single process, which can be interpreted as a general tendency for both i, u and y, w to be excluded from certain positions in the word. Note that interconsonantal i, u (including i, u occurring in the diphthongs ai, au) are not affected.17 Similar processes sporadically operate in other contexts. While y and w mostly appear as such between vowels (V_V), there are cases of a change to iy, uw even in this position, e.g., adāra(i)ya ⟨a-d-a-r-(i-)y⟩ 'he held' (OInd. adhārayat), ba(u)watiy ⟨b-(u-)v-t-i-y⟩ 'becomes' (OInd. bhávati), etc.18 However, these sporadic cases are not on the same level as the systematic ("regelmäßig", Schmitt 1989: 69) change of postvocalic y, w to iy, uw. Likewise apparently related is, according to Hoffmann (1976: 635), the phenomenon of ⟨u-v⟩ for expected ū, but it needs to be kept in mind that this 17 Word-internal i and u are normally written ⟨Ci-i⟩, ⟨Cu-u⟩ where a Ci / u character is available (⟨Ci / u⟩ alone is also found), otherwise ⟨C-i/u-⟩, thus identical to ai, au. Word-initial ai and au are written ⟨a-i/u-⟩, and i-, u-⟨i-⟩, ⟨u-⟩ (Kent 1953: 13-14). 18 Examples from Hoffmann (1976: 63524). Cf. also Kent (1953: 21 is limited to isolated cases, it is thus different from the systematic change in (9). Relevant cases include gen.pl dahyuwnām ⟨d-h-y-u-v-n-a-m⟩ beside dahyūnām ⟨d-h-y-u-n-a-m⟩ 'countries' (OInd. dásyūnām) and uwnara-⟨u-v-nr-⟩ 'skill' (OInd. sūnára-). The analysis on the part of Old Persian scribes of ū as a combination of u+w is noteworthy from a phonological point of view (Hoffmann 1976: 636).19 Some changes preceding op y > iy, w > uw The op change (9) can be used as point of reference for a relative chronology of some other changes.

1.3
As pointed out by Meillet (1915: 75 §144, 77 §147),20 the change of *čy and *θy to *šy, yielding instances such as ⟨š-i-y-v-⟩ šiyaw-'move, proceed' (OInd. √cyu); ⟨š-i-y-a-t-i-⟩ šiyāti-'joy' (Latin quiēs); ⟨h-š-i-y-⟩ hašiya-'true' (Av. haiθiia-, OInd. satyá-), presupposes contact of the two consonants; it thus predates the change of Cy/w to Ciy / Cuw. It may be noted in passing that Avestan shows šíi for *čy as well (but not for *θy). However, the agreement of the two Old Ir. languages should not be taken to imply that *čy > šy is a Proto-Iranian change. Some later Ir. languages have an affricate in the relevant words (Khotanese ts, Ossetic c, Kurdish č),21 so it is more likely that the change *čy > šy occurred independently in those Ir. languages that show it (in fact, most Ir. languages)-aided perhaps by contact among some of them, but not as part of a common inheritance.
If, then, the op change *čy > šy is independent of the parallel change in Avestan, this offers the possibility to assume (as is generally done), if only for reasons of economy, that the specifically Persian change *θy > šy occurred together with the parallel change *čy > šy.

Change (3) Ir. čy, θy > op šy
It would be interesting for the purposes of relative chronology if the fact that op θ is of several origins-pie *t preceding a consonant, *th2 and *ḱ (> Ir. *ś)- 19 See Werba (1993: 142-43) for an alternative interpretation: He suggest that there was a neutralisation of vowel quantity in word-final i, u in the same way as there clearly is in the case of -a/ā, which both yield op -ā (see 1.5); he interprets ⟨-i-y⟩ [ij] as an allophone of /ī/, implying a lengthening of -i, -u > op -ī, -ū. 20 Note the change in opinion by Meillet;Meillet & Benveniste (1931: 87) consider as "sûrement graphique" the i and u in op Cy/w (see n. 9). 21 Thus also the argument by Cheung (2007: 41) for the root *čyaw, cf. also MacKenzie (1961: 71-72 provided an element of relative chronology for the subsequent processes, but unfortunately there is no good example nor a compelling counterexample of whether or not op θ deriving from pie *ḱ is included in the change *θy > šy. It has generally been assumed that it is (Brandenstein & Mayrhofer 1964: 135, following Kent 1953.22 If this is so, *θy > šy would postdate the op merger of the output of Ir. *ś with Ir. θ from *t in preconsonantal position (and, by extension, the merger of Ir. * ź with *d).

22
The question merits a separate study, which is beyond the scope of this article. 23 See Tremblay (2005: 674-85) for an argument that the change is "Common Iranian", i.e., shared by the individual Ir. languages, but not necessarily dating back to Proto-Iranian.

Input op form Schmitt My reading mp form Meaning
While at least the latter seems a valid point considering cf. OInd. kriyá-, mriyá-, it is difficult chronologically: since k changing to x (in fact, the change of any stop to fricative) in pre-consonantal position (1) surely is Proto-Iranian (see 1.3), this would date *kr̥ y > *kriy > *xriy-to Proto-Iranian. However, the corresponding Avestan forms are unambiguous and demonstrate that this scenario is unlikely: the perfect stem is čāxr-and the passive kiriia-; the latter is obviously not the full grade, but from *kr̥ ya-and does not yield **xriya-. Pointing in the same direction is Manichean mp (mpm) ⟨qyryh-⟩ 'be done' and inscriptional mp (mpi) ⟨kylyt, klyty⟩ 'is being done' , according to Skjaervø (1997: 161) "the direct descendant" of the op form,24 thus militating against Schmitt's assumption of a full-grade passive in Old Persian. čaxriyā is then likely to contain the prevocalic variant ča-xr-V-which was generalised as the only perfect stem, including in the position preceding -y-, thus showing the same process as Avestan čāxr-.
Similarly, as pointed out by Gippert (2001: 15), Av. miriia-and mp/np mīr-'die' speak against Schmitt's assumption of a full-grade op passive /mariya-/ replacing the expected *mr̥ ya-, and in favour of an op form *mərya-with consonantal r necessary here as well to yield məriya-.25 In fact, in view of the rigorous method otherwise applied by these authors, it is surprising that Brandenstein & Mayrhofer (1964: 33, 35), Mayrhofer (1989: 10), Hoffmann (1976: 631), Schmitt (various works), and Brust (2018: 13) are somewhat unclear on the status of *r̥ ; assuming that there was "a phoneme r̥ that was pronounced [ǝr]", as it were (thus explicitly Schmitt 1989: 69). However, "it is difficult to motivate the development in op if one assumes a real r̥ " (Gippert 24 See 2.1, 2.4 for the further development of kərya-. 25 nb that a scenario Proto-op *marya-> pmp *maira-(by the change discussed in 2.1, 2.5) > mp †mēr-> np mīr-is impossible: except for the position preceding a nasal, the merger of mp ē and ī is limited to postclassical standard New Persian, while the two vowels are kept distinct in the classical literature and in dialects and varieties such as Afghan Persian (see Korn & Olsen 2012: 209-1034; the vocalism of mīr-for classical np (and by extension for mp) is thus beyond doubt. See also 2.4 for further discussion of mərya-.
Indo-European Linguistics (2021) 1-43 | 10.1163/22125892-bja10009 2001: 15), and the idea that the question of the phonemic and phonetic reality does not matter (thus explicitly, e.g., Brust (2018: 13), following others) is linguistically odd. One cannot help but feel that the approach aims at "explaining away" evidence against the preservation of *r̥ in Old Persian, and at positing a maximum of regularity for the op verbal system (thus also Schmitt's argument for other verbal categories in his 1967 article).
Concerning the vocalism of the op output of *r̥ , the op orthography does not distinguish ar and the output of *r̥ (both are written ⟨a-r⟩ word-initially and ⟨C-r⟩ word-internally), but the Elamite rendering of *r̥ is ir, e.g., ir-ta-for op ⟨a-r-t-⟩ ǝrta-'justice' and Birdiya-(personal name) for op ⟨b-r-di-i-ya-⟩ (cf. Mayrhofer 1973: 25). Conversely, had *r̥ given op ir generally as Elamite might suggest, one would not expect both ir and ur as mp/np outputs of *r̥ (cf. mp burd 'carried'). Taken together, this makes a pronunciation ǝr for the op result of *r̥ likely26 and establishes ə as a phoneme for Old Persian.27 While op provides direct evidence for *r̥ yielding a vowel plus consonantal r (ǝr) only for *r̥ preceding y (*r̥ y yielding *əry being shown by op əriy), the Elamite rendering of *r̥ as ir shows that *r̥ > ər was a general process.

Change (8) *r̥ > ǝr
As for relative chronology, *r̥ > *ǝr (8) needs to precede y > iy, w > uw (9) as argued above. I am grouping *r̥ > *ǝr next to y, w > iy, uw (thus after *θy > šy (3) discussed in 1.3), assuming that they follow the same tendency of insertion of anaptyctic vowels, although there is no direct evidence for this specific relative chronology.28 26 Thus also Gippert (2001: 15). Bartholomae (1925: 48) states: "The opinion I voiced earlier (…) about Proto-Indo-Iranian r̥ being still pronounced as a sonant [i.e. vocalic *r̥ ] in Proto-Iranian is wrong." He also points out (1925: 17-19) that the vowel can neither have been a nor i or u. He thus notes ɒ r for Old Persian and (54) ə r for Proto-Iranian. 27 See also Werba (1993: 142), who postulates (for slightly different reasons) that the op output of *r̥ was a biphonemic combination of a reduced vowel and r, and that the former is thus to be added to the inventory of op vowels. He follows Hoffmann (1976) in transcribing "ạr", but interprets it as /ər/. 28 Some authors (e.g., Werba 1993: 142) have dated *r̥ > ər to Proto-Iranian (see also n. 26). While the question surely needs more research, I am sceptical of such an early date, seeing that the result of *r̥ is treated differently from combinations of vowels with r in Sogdian: It is lost between voiceless consonants (e.g., as in *kr̥ taka-> kte 'done' , Schwartz 2005: 165) and otherwise yields a "light syllable" (Sims-Williams 1984: 210). For an argument along similar lines, see Tremblay (2005: 680-81). See also n. 97. korn 10.1163/22125892-bja10009 | Indo-European Linguistics (2021) 1-43 1.5 The change i, u > iy, uw also needs to postdate the loss of word-final t/d, as shown by *čit > =čiy and *nait > naiy (see 1.2). The change is usually formulated as affecting word-final t, but in fact this includes -d, because there is no voice opposition of word-final stops in pie and Proto-Iranian, and the voice feature depends on that of the initial sound of the next word (thus, e.g., *čit/d).29 It may safely be assumed that the loss of other word-final consonants is part of the same change, so that the only consonants permitted in word-final position are y, w, m, r, and š (Brandenstein & Mayrhofer 1964: 45). This produces the coalescence of many verbal and nominal endings in -a (< Ir. *-ah, -an(h), -a(n)t/d, -ats).30 As pointed out by Hoffmann (1976: 633-34), this -a does not coalesce with inherited word-final *-a because the latter was lengthened at a previous stage. That this lengthening was likewise a real (not only graphic) phenomenon is shown by the consistent spellings as well as by the absence of the lengthening if a clitic follows (e.g. manā I.gen/dat vs. mana=čā with =čā 'and'). The change is shared by Proto-Middle Persian, as shown by some mpm fossilised phrases where -ā became phrase-internal later on and thus treated as one word with the following item, cf. mpm ⟨psʾc⟩ pasāz (cf. op pasā 'afterwards') with the mp clitic =(i)z 'also' (see 1.6), mpm ⟨gwʾm⟩ < *gaubā=mai 'tell me!' and several such instances in the Psalter (Sims-Williams 1981a: 174).
Conversely, Classical Old Persian shows =čiy, which can only be from *čid, not from *čim, because -m is not among the word-final consonants which were lost (cf. (7)). The substitution *čid → *čim is another dialectal contrast between Classical Old Persian and Proto-Middle Persian.
If one assumes that the op loss of word-final t/d etc. in (7) also operated in Proto-Middle Persian, the substitution of *čim for *čid needs to precede this change: the element *=či(y) probably no longer functioned as an independent element and was not perceived as a pronominal nom/acc anymore; it thus was not a candidate for being replaced by *čim.
The substitution *čid → *čim probably also precedes the change *sč > s (seen, e.g., in *pas-ča > op pasā, mp/np pas 'after'):32 Had the latter change happened first, *kas=čid and *čis=čid would have yielded †kasid and †čisid, arguably morphologically obscure and unlikely to be replaced by †kasim, †čisim. This yields a relative chronology *kas/čis=čid → *kas/čis=čim > *kasim, *čisim > *kasu, *čisu. I do not see particular evidence for a relative chronology of this change and of the loss of word-final t/d, n, h and the changes in 1.3-1.4. Perhaps both could be part of a tendency favouring open syllables that could be seen in 1.2, so as a preliminary hypothesis, I place sč > s near y, w > iy, uw (9), but leave *r̥ > ǝr (8) and the loss of word-final consonants next to y > iy, w > uw as possibly even more closely related phenomena.
Change (5) Ir. sč > op s Some changes following op y > iy, w > uw Conversely to the cases just discussed, the op change of postconsonantal y, w to iy, uw is presupposed by some other changes.

1.7
As pointed out by Hoffmann (1976: 637), instances of the contraction iya > ī occurring within Old Persian show that both the inherited sequence iya as well 32 The change *sč > s is not seen in the op compounds containing *=čid, which have the "Median" form with šč (e.g., kaščiy "someone"); see Korn (2013: 83) for further discussion. as iya arising from Cy undergo this change (see 1.1). One obvious effect is that the frequent suffix combinations *-iya-ka-and *-ya-ka-both yield -īka-. Indeed, a contraction iya > ī is widespread in Iranian, and *-iya-ka-yields -īk/g in much of Middle and New Iranian, which makes it likely that the change is rather old.33 In the Arm. Bible, -ik occurs in "adjectives of appurtenance, mostly of Iranian origin" and reflects both Ir. *-(i)yaka-and *-ika-according to Olsen (1999: 454-55); obviously the suffix was borrowed into Armenian before the change of postvocalic Ir. p, t, k (for which see 2.7).-As I will argue in 2.1, however, the merger of *-iya-ka-and *-ya-ka-is a specifically Old Persian phenomenon, and *-ya-ka-does not necessarily yield -īg in Middle Persian.
Quite probably there was a parallel change *uwa > ū (Schmitt 1990: 18-20). This appears likely a priori, and is also suggested by some Elamite evidence: It seems that a certain op place name is rendered both by Si-qa-hu-ut-ti-iš /Θikayūtiš/ and Ši-ik-ki-ú-ma-ti-iš reflecting Ir. /Sikayuwatiš/ (also showing the dialectal difference op θ vs. other Ir. s).34 There are also some rather convincing mp examples: xūb 'good' may be derived from *hu-Hapah-(via *huwapah-, cf. Av. huuapah-) 35 and d(u) However, there are also cases of uwa yielding mp ō rather than ū, e.g., padrōd 'farewell' .36 The change uwa > mp ō could be comparatively late, which would mean that some cases of uwa escaped the change to ū. The contrast of d(u)rūd vs. padrōd, both containing a cognate of Av. druuatāt-'health' , makes it unlikely that the phonetic context plays a role. It is likewise unclear whether the accent might play a role. One might assume that úwa could have yielded ū while uwá (and/or unstressed uwa) changed to ō only later,37 but this assumption is rendered difficult by the accent of OInd. s u v-ápas-(EWAia i: 84) vs. np xūb. While 33 Rastorgueva & Molčanova (1981: 53, 69), Durkin-Meisterernst (2014: 160). A similar change is seen in Bactrian (-ιγο, Sims-Williams 2007: 217) and Sogdian (-yk /-īk/, Gershevitch 1954: 31 §202, 148 §977). 34 Schmitt (1990: 18-20). For Elamite uma rendering op uwa see 1.1. 35 A derivation from *hu-Hpah-seems less likely because of the Av. and OInd. forms, both pointing to Proto-Indo-Ir. *s u w-ápas-. Also, Ir. compounds with hu-are productive and transparent in Old Iranian (hence one expects the independent form of the word as second member). As pointed out by de Vaan (2003: 565-68, 575), hu-is even frequently restored in front of vowels in Avestan rather than changed to xv-as would be regular for inherited *hwV-. 36 Hübschmann (1895: 169) has drōd (thus also read in other sources) and padrūd, but the rhymes speak in favour of the reverse distribution of vowels in the Shāhnāma (Horn 1899: 182-84) and also elsewhere (Chams Bernard, p.c.).

37
This is what I thought at some point (noted in Bernard 2019: 48).
the details of these changes are not quite clear at present, the development of uwa does not seem to be entirely parallel to that of op iya.
(conditions of change unclear) 1.8 Another change that needs to be later than Cy/w > Ciy / Cuw is the op loss of h preceding u, e.g., ukāra-⟨u-k-a-r-⟩ 'having good people' < *hu-kara-, Aura-⟨a-u-r-⟩ < *ahura-(god's name). Since hu > u includes huw that comes from *hw , it needs to postdate (9) y > iy, w > uw (Hoffmann 1976: 641-42). That this is a real change in Classical Old Persian again (not only graphical) is shown by the consistent absence of h preceding u. Also, Elamite renderings mostly have u in op words with *hu, and this in spite of the fact that there is an Elamite sign hu available. This sign is of course used in Elamite words containing hu, and also in op words containing hǝyu, which is changed to yhu in Elamite (thus Elamite da-ay-hu-na-um renders op dahəyūnām). The fact that this sign is normally not used for cases of Ir. *hu confirms the change *hu > op u (Hoffmann 1976: 639-42).
However, some Elamite words do show preserved Ir. *hu, providing evidence for the existence of a dialect that preserved hu (Hoffmann 1976: 639, see also 2.1, 2.2). This dialect would agree with Middle Persian, where *h preceding u and w was not lost, but yields hu-/xu-and xw-.38 It is generally assumed that the loss of h in other contexts is part of the same process. Hoffmann (1976: 63933) thus suggests that hm > m is regular for Old Persian, which means that ahmiy ⟨a-h-mi-i-y⟩ 'I am' in XPl (several attestations) beside more common amiy ⟨a-mi-i-y⟩ differs from Classical Old Persian just as regular hu > u vs. dialectal hu reflected in Elamite. Brandenstein & Mayrhofer (1964: 42-43) treat the op loss of h preceding r (rautah-, nom/acc.sg ⟨r-u-t⟩ 'river' < Ir. *hrautah-) as part of the same change, but whether this is really so is not quite clear.39 Synchronically, y following h is an exception to (9) in that it shows no trace of an i, i.e., *hy does not appear to yield op hiy. Werba (1993: 142), following 38 That Middle and New Persian differ from "the dialect of the [op] cuneiform inscriptions" in the treatment of *hwa-was already seen by Hübschmann (1895: 218). 39 Hoffmann (1976: 63933) appears to regard loss of h other than that preceding u and m as sporadic and due to op h being "weakly articulated" in general. The loss of word-final h is part of the change (7) and summarising Hoffmann (1976: 642-43), suggests that y after h did yield iy together with all other instances of postconsonantal y and that *hiy (including the output of word-final -hi) then changed to həy (i.e., op ⟨h-y⟩ is to be read /həy/).40 Sporadic cases of ⟨h-i-y⟩ hiy for ⟨h-y⟩ həy in Xerxes' inscriptions support the reality of a vowel in the op result of *hy (Hoffmann 1976: 643) and may reflect a memory of hiy being the older form. It seems possible to me that op h at least in hə was pronounced in a specific way (e.g., like the German "ich"-fricative, ipa [ç]),41 and that Elamite ⟨yh⟩ mentioned above might be an attempt to render this sound.
In view of proto-mp not sharing the op loss of h preceding u, one might expect that it does not share the loss of h preceding consonants either. However, whether this is indeed so does not seem quite clear to me. It also seems possible that the loss of h preceding m is not on the same level as that of hu > u: cases of preserved hm are found in the op corpus itself, while preserved hu is not found in the op corpus and is reflected only in Elamite. For the time being, I prefer to separate the case of hu, for which the developments to op u and to mp xu are equally well established, from other op losses of h.42 Change (10) h > ∅ /_u, m (does not operate in pmp) 40 The op spelling of the name of Anāhitā as ⟨a-n-h-t⟩ in A2Sa (several copies) and A2Ha (beside ⟨a-n-h-i-t⟩ in A2Sd) might point in the same direction (see also 3.4).

41
Thomas Jügel (p.c.). Note that Avestan xíi likewise indicates a specific pronunciation of the result of *h(i)y. 42 For the np loss of h in word-initial ha-, see Hübschmann (1895: 215).

Evidence against the op change y > iy, w > uw in Middle Persian
As pointed out by Hoffmann and Risch (see 1.1), it is a mistake to think that a change of postconsonantal y, w to op iy, uw would be disfavoured by the mp developments presupposing šy and dw: iy, uw could have reverted to y, w in the same way as it demonstrably was in other languages. Still, a process y > iy, w > uw being possible a priori is obviously no proof that it did take place in the dialect(s) on which mp is based. Indeed, I argue that there is evidence against Cy/w > Ciy / Cuw having operated in Proto-Middle Persian, and that this difference to Classical Old Persian is to be added to the dialectal differences already established.

2.2
Having seen that the op forms with iy, uw cannot be the input for Middle Persian, the alternative would be to follow Hoffmann and Risch and assume a syncope to revert šiy, duw, etc. to šy, dw, etc. to yield mp šād 'joy' , dar 'door' from Ir. *šyāta-, *dwara-, etc. (see 1.1 and 2.3). However, this would likewise produce wrong results in some cases.
For instance, a syncope of Cuw would yield Ir. *hu-Hapah > op *huwapah > *hwapa-> mp †xwab (cf. *hwat-'self' > mp xwad) instead of the attested 48 This form would be from an op dialect that conserved word-initial hu-(see 1.8). This can only mean that the contrast between Ir. word-initial Ciy / Cuw vs. Cy/w was preserved in Proto-Middle Persian, whereas the distinction was lost in Old Persian. In view of all this, the conclusion seems unavoidable that Proto-Middle Persian goes back to a dialect that, differently from attested Old Persian, did not show Cy/w > Ciy / Cuw.

2.3
The change of word-initial dw-to d-mentioned in 1.1 and 2.1-2.2 and seen in examples such as dar 'door'49 has generally been discussed together with the cluster simplification of šy to š, which suggests that the two changes are contemporary. It is not clear to me whether this really is so, since *dw-> d-is specifically Persian (in Parthian, the result of *dw-is b-) while *šy > š also happened in other Ir. languages such as Parthian and Balochi and *šy yields š́in Young Avestan. On the other hand, *šy > š is surely natural 49 For this change, see Hübschmann (1895Hübschmann ( : 1162 enough (a palatal pronunciation of š "absorbing" a following y, as it were) for it to occur in several languages independently. There is evidence of *šy to š happening in late Old Persian times, since A3Pa shows ⟨š-a-y-t-a-m⟩ (apparent šāyatām) for expected (and attested) ⟨š-i-y-a-ti-m⟩ šiyātim 'joy' , which appears to imply a pronunciation š-rather than šiy-. Given that ⟨š-a-y-t-a-m⟩ also shows the effects of vowel-neutralisation in wordfinal syllables (see 3.4), I tentatively place the change šy > š next to these processes. Elamite and Greek forms likewise reflect šāt- (Schmitt 1999: 96-98). On the other hand, Armenian dahlič 'chamber' (np dihlīz, mpm ⟨dhryz⟩50 vs. op duwarθi-'hall, portico') shows that *dw-> d-precedes the change of postvocalic č to z (26) discussed in 2.7 and seen in examples such as mp rōz 'day' (op raučah-) and =čim > =(i)z 'also' (see 1.6).

3).53
Differently from the argument just made, Skjaervø (1997: 161, followed by Durkin-Meisterernst 2014: 229) suggests kirīh-for the mpm passive ⟨qyryh-⟩ 'is done' (apparently for reasons of the mpi variants). However, New Persian forms deriving from *r̥ y unambiguously show īr, not only in mīr-mentioned in 1.4, 50 See Korn (2009: 205) for discussion of the mp word. 51 Note, however, that they include under Epenthesis also forms that in my view rather show Umlaut, see 2.5, 3.1. I use Epenthesis and Umlaut with initial capitals to refer to the mp sound changes under discussion here, differentiating them from similar processes in other languages whose conditions may be different. 52 Cf. Bartholomae (1925: 362), who notes " ə ri̯ " changing to īr. 53 The Avestan forms that Hübschmann cites at various points, showing "airii" or "irii", should not be held to imply that we are looking at the same phenomenon; these orthographies probably indicate a palatal pronunciation of the consonant (Martínez & de Vaan 2004: 17-18), parallel to, e.g., the orthography of Irish, and might even be only a feature of the transmission of the Avestan text, whereas the pmp change *ary > *air is a phonological process, leading as it does to a long vowel ēr.
In the light of this evidence, it seems preferable to regard mp mēnōg/y as an Avestan loanword, which is particularly likely for a Zoroastrian concept.61 The other items are better explained as containing a (short) -e-arising from the Middle Persian "Umlaut" discussed in 3.1 as suggested by MacKenzie (1986). Crucially, Ir. *any > mp en yields an, e.g. in np dušman 'enemy' (not †dušmēn), without any epenthesis and monophthongisation that would give a long vowel.

2.6
As cautiously suggested by Hübschmann (1895: 131, 146), pmp *r̥ wy does seem to be treated in the same way as *r̥ y, both giving mp īr. Examples for the former include pīr 'old' (Ir. *pr̥ wya-62 > *pǝrwya-> *pǝrya-) as well as gīr-'seize' and mp padīr-, np pazīr-'accept' from *(pati-)gr̥ bya-; for the latter there is mīr-'die' (see 1.4) and the examples discussed in 2.4. Whether or not one should assume that the pmp forms of gīr-and padīr-contained Old Ir. *b (thus *gr̥ bya-)63 or rather a lenited sound as in Av. gǝuruuaiia-is not clear to me. In the former case, Ir. *b > pmp *β could have coalesced with Ir. *w at least in this context.64 without *-ya-, but the Ir. data rather point to a present stem *man-ya-with the possible exception of some Avestan forms (see Cheung 2007: 262-63). 59 Hoffmann (1976: 63726, following Bartholomae 1906 takes this as another argument in favour of the op change of postconsonantal *y, w to iy, uw as being real, stating that the mp cognates of op aniya-'other' and haruwa-'every' (see Table 1), mpm ⟨hrw⟩, mpz ⟨hl⟩, "are hardly likely to have been pronounced ani̯ and haru̯ , but anī(…) and haru." He goes on to explain the difference in the development of -iya-in "ani, haru" vs. that of *martya-> mp ⟨m(y)rd⟩ (not †⟨mrdy⟩ mardi) as being due to the light syllable of the former (see also n. 91 for this word). However, this approach has not been upheld in recent scholarship; MacKenzie 1986 reads har(w) for mpz, Durkin-Meisterernst 2004 harw for mpm (np har), both read any. np nīz 'also' can also be derived from the combination of mp any and the clitic =(i)z mentioned in 1.6. Hübschmann (1895: 104) suggests a derivation from *anid=čid (op *aniščiy), thus likewise assuming the presence of the same element that gives mp =(i)z. 60 This problem was already seen by Bartholomae (1906: 62-633), who adduces unlikely analogies to account for the form, which he reads anī. 61 Thus also the suggestion by Skjaervø (2002: 307). 62 There seems to be no trace of the laryngeal seen in Av. paoiriia-, op paruwiya-(ta) and OInd. pūrvyá-, all from *pr̥ Hwya-(cf. Fischer 1998). 63 Old Persian has a different present stem for this verb, viz. gərbāya-(corresponding to OInd. gr̥ bhāyá-). 64 For pmp *β, δ, γ as results of Ir. postvocalic *b, d, g, see 2.8.
These facts suggest that Parthian does not share the Epenthesis of *ary > *air, and that, by extension, *ǝry does not change to *əir either. We therefore do not expect a long vowel in Pth. ⟨myr-⟩, ⟨(pd)gyrw-⟩. The Pth. output of (*r̥ y, *r̥ wy >) *əry, ərwy could thus be read ir, irw.70 This in turn might mean that Balochi, which shows short vowels in these words (mir-, gir-), might not have experienced Epenthesis either, in which case any instance of ēr from *ary would be a Persian loanword.
The op monophthongisation 2.7 It seems obvious that the result of (14) ary > air (see 2.1) feeds into the wellknown and amply evidenced monophthongisation of ai to mp ē, i.e., that ai in Ir. *air from *ary changes to ē at the same time as old ai. It is highly likely and has generally been assumed that the parallel monophthongisation of *au to ō (as, e.g., in mp pērōz mentioned in 2.1) is part of the same change.71 As mentioned in 2.4, I suggest that *ǝi > ī is part of the monophthongisation of ai and au.
The op inscriptions indicate that the monophthongisation occurred within Old Persian, and that it included *aya and *awa (as is probable in any case). One relevant example is ⟨p-r-d-y-d-a-m⟩ 'palace' (A2Sd) for what would be *paridaidam *⟨p-r-i-d-i-d-m⟩ in Classical Old Persian. By a hypercorrect application of the following logic: "Where we say ē, the older, i.e., better / more official forms have aya"-generalised from the (correct) observation that older forms with 70 Thus the suggestion in Korn & Durkin-Meisterernst (2009: 12) for reasons of the Balochi evidence. 71 For ai, au > mp ē, ō see Hübschmann (1895: 141-42).
aya and ai were pronounced with ē at the time of writing-, the scribe (and/or the author) "restored" an apparent par(a)dayadām from the contemporary pronunciation pardēd-.72 It is entirely possible, but not a necessary assumption for the purposes of this paper, that aya, awa were first "syncopated" to ai, au73 which then yielded ē, ō generally, and that there was an intermediary stage ei, ou.74 This sets a point of absolute chronology, as the inscription A2Sd is from the reign of Artaxerxes ii (first half of the 4th c. bc), and the monophthongisation had happened by this date, but the scribe must still have remembered a register (or words from older inscriptions) that had ai / aya for their ē.75 By extension, all preceding changes are from the Achaemenid period as well. Where we do not have direct evidence in the op inscriptions for this being the case, this can be due to the limited size of the corpus. More crucially, any orthography tends to be stable once it is installed and to mask changes ongoing in the language-particularly in royal inscriptions, which will certainly aim at a formal and conservative standard. All the more remarkable are those instances that do deviate from the traditional spellings and thus demonstrate a change in the language. The Armenian loanwords likewise reflect ē from Ir. *aya and *ai (yielding Arm. ē, later e), and ō from *awa and *au (Arm. oi, later o),76 and this in combination with preserved postvocalic voiceless stops, e.g. kapēnkʿ 'dowry' (mp kābēn), hro(r)t-icʿ (gen.pl, month name) < *frawarti-(mp Fraward-īn), kapoit 72 Thus Schmitt (1999: 82-84, with in-depth discussion of the various interpretations of this word). For discussion of the final syllable, see 3.4. 73 Cf. Werba (2006: 279-83) and Schmitt (2014: 225) about ⟨p-r-d-y-d-m⟩. 74 Thus Werba (2006: 283), adducing further evidence for ai > ei; he assumes that ⟨p-r-dy-d-m⟩ reflects "an allegro form /pardeida/ that was on its way to monophthongisation" (281). Further evidence for -ei-comes from Elamite ik-še-ir-ša for Xerxes (op ⟨x-š-y-a-r-š-⟩, cf. Schmitt 1999: 114) and from Greek renderings e.g. of Artaxerxes (cf. Korn frthc.). Werba's word-final -a seems unlikely to me, however (see 3.4).

75
As pointed out by Hoffmann (1976: 643-45, contra Brandenstein &Mayrhofer 1964: 29 andothers), the Elamite data militate against this change having happened already in Classical Old Persian, as they render the op diphthongs as well as can be expected of a language that does not have diphthongs. 76 For the Arm. outputs, see Hübschmann (1897: 14). In nonfinal syllables, ē, oi yield i, u. 'blue' (mp kabōd).77 This shows that late op forms with p, t, k, č were not mere historical spellings, and that they were preserved at the time when the monophthongisation had already taken place. As likewise shown by Armenian ( ǰatuk vs. Ir. *yātuka-'magic'; zatik vs. *ǰatika 'sacrifice'), the voicing of the voiceless stops is also preceded by the change of Ir. *ǰ to z, and by that of word-initial *y-to ǰ (both specific to Persian), of which the former must precede the latter since otherwise the new ǰ-would also have yielded z. The use of etymologically unjustified ⟨c⟩ for Ir. *z and *ǰ in the early Sasanian inscriptions shows that č had coalesced with z from other sources (i.e. from postvocalic č and from non-Persian z < Ir. *ź) by that time.78 Indo-European Linguistics (2021) 1-43 | 10.1163/22125892-bja10009 mon phenomenon cross-linguistically, it is not surprising that there are several instances of ai > ē in the prehistory of Persian.

Middle Persian
The Middle Persian vowel assimilation 3.1 I argue that the mp sound change called "Umlaut" by Hübschmann (1895: 129) and others is not part of the process called Epenthesis (14), although various authors have included some or all cases of Umlaut as Epenthesis. As argued in 2.1, 2.4, the Middle Persian Epenthesis is rather early and yields long ē, ī that are preserved as such in Middle and New Persian. Conversely, Umlaut is later and appears to yield a short vowel (e, o) in Middle Persian, as indicated by the spellings, which oscillate between forms with and without ⟨y/w⟩ (see 2.5).86 Moreover, this vowel has frequently "reverted to /a/" in New Persian (MacKenzie 1967: 24),87 which rather confirms that the vowel is short. An alternative interpretation would be that Umlaut did not happen in the mp variant(s) which is/are the predecessor of New Persian (Proto-New Persian); in this case as well, Umlaut would be different from Epenthesis since the latter did occur in Proto-New Persian. Instances of Umlaut yielding mp e include mp men-'think' (Ir. *man-ya-) and dušmen 'enemy' (Ir. *duš-man-yu-, np dušman)88 as well as mp deh 'land, vil-84 Thus also Hübschmann (1893: 181, 198, 247). Old Persian thus would have resembled Spanish with respect to this feature. 85 Examples from Hübschmann (1895: 182, 201). There is no sound γ in Old Armenian, so that Middle Ir. γ is rendered by Arm. g (Hübschmann 1895: 247); Armenian thus does not provide evidence for g > γ. 86 See n. 51. 87 This statement refers to the mp vowel assimilation in general.

3.3
It seems possible to me that the "colouring" of the op vowel ə, which arises by the change (8)  It seems a priori likely that ə arising from other processes is "coloured" at the same time. A possible candidate is *ə in word-final *-əm, which, as argued in Korn (2013) is the pmp result of Old Ir. -am and of other word-final syllables and then changes to *-um (see 3.4). Differently from mp e, o, the products of the pmp Umlaut, however, these coloured vowels are not changed to a in New Persian, but are treated like inherited i, u.
Change (23)  See also Bartholomae (1925)  Note that this more detailed account removes a problem that would arise if one were to posit a development of mp ur/ir from *r̥ directly: Had *r̥ yielded ur in labial environments and ir elsewhere, instances such as *mr̥ ya-'die' and *gr̥ wya-'seize' could be expected to show ur as well. Under the present account, w in the latter example is lost at an early stage (13), and Epenthesis and the subsequent monophthongisation *əry > əir (14) > īr (15) likewise took place quite early (in Achaemenid times, see 2.7). The relevant examples thus get their ī from changes independent of the phonetic contexts long before the contextdependent colouring takes place.

Word-final phenomena again
3.4 The argument just made adds some perspective to the sequence of changes in op/mp word-finals discussed in Korn (2013): Certain op word forms show that op vowels in final syllables experienced a "Quantitätenkollaps" ("collapse of [vowel] quantities")99 by which -ām and -am (the accusative of the most frequent f. and m. stem classes, and the nominative/accusative of the most common neuter class) would have coalesced; similarly, -im, -īm and -um, -ūm would have coalesced to -im and -um, respectively. This was followed by a neutralisation of the vowel, the just mentioned word-finals thus all yielding *-əm. The assimilation of *-əm > *-um (cf. 3.3) could be part of the mp vowel assimilation; *-um < -am etc. is thus parallel to -dom < *-tama-etc. mentioned in 3.

2.
If this account is on the right track, it would refine the relative chronology suggested in Korn (2013: 85) in dating the merger of final nominal syllables to *əm before Umlaut. Since op ⟨p-r-d-y-d-a-m⟩ 'palace' (implying -ām instead of expected -am) in an inscription from the reign of Artaxerxes ii (see 2.7) reflects both monophthongisation and the neutralisation of vowel quantity, I tentatively place the "Quantitätenkollaps" next to the op monophthongisation ((15), see 2.7).100 The neutralisation of vowel quality is likely to have been later than the neutralisation of vowel quantity because a neutralisation of vowel quality seems more probable in a short vowel. This sequence may be confirmed by the fact 99 I borrow the term "Quantitätenkollaps" from Romance linguistics, where it refers to the loss of vowel length distinctions in popular / late Latin, which is followed by a loss of some distinctions of vowel quality depending on the individual Romance language (see, e.g., Lausberg 1969: 144 for the term and Banfi 1996 for a survey, specifically p. 166). 100 As pointed out in Korn (2013: 85), a weakening in successive stages seems more likely to me a priori than a loss without trace of word-final syllables. The mp and Arm. data as well as the mp elements mentioned in 1.6 likewise militate in favour of intermediary stages.
'palace' (2.7), both containing *pari-, as well as mp ēr 'below' if this derives from Ir. *adari > pmp *aδr-(2.8), suggesting a syncope (at least) in open middle syllables. However, we also find unsyncopated forms such as Arm. aparankʿ 'palace' (2.8) and Spandaramet (3.1). It seems possible that there are several types of syncope in the prehistory of New Persian,105 one of them possibly in Achaemenid times for reasons of the missing ⟨-i-⟩ in ⟨p-r-d-y-d-a-m⟩, and (as shown by the Arm. examples just mentioned) earlier than the voicing of postvocalic voiceless stops, possibly close to the op monophthongisation (15) and to the neutralisation of vowel quantity in word-final syllables (16) as potentially related processes of vowel reduction. Their study is reserved for future research.

Conclusion
While one cannot help but be amazed at the insights obtained by the pioneers of Iranian historical linguistics such as those mentioned throughout this paper, it is clear that the data available today permit a refinement of their research results. This is not only a question of the quantity of data and results, but also one of method. I argue that it is possible to go beyond the "black box" approach applied until now, which compared an input (such as Proto-Iranian or Proto-Indo-European) with an output (Middle or New Persian), largely resulting in an unsorted list of sound changes, as it were, disregarding their sequence and the ways in which they interact. The present contribution attempts to demonstrate the results at which one arrives by putting into practice the converse approach called "glass box". This implies the study of a given sound change (such as postconsonantal y to iy and w to uw) within the context of somehow connected processes. In our case, this concerns, on the one hand, other phenomena affecting syllable structure such as the change of vocalic r (*r̥ ) to ǝr, by which some instances of y come to be in postconsonantal position, as well as processes in which y and w are involved. Examples of the latter are changes affecting vowel quality, called Epenthesis and Umlaut in previous scholarship. The present approach permits us to view them as two distinct processes, which moreover took place in different periods.
The output is, I think, twofold. On the one hand, the study of these sound changes yields the relative chronology presented in the Appendix below (Sec-korn 10.1163/22125892-bja10009 | Indo-European Linguistics (2021) 1-43 tion 5), which is a sorted list resulting from the application of the glass box approach. An absolute chronology is possible to some extent since certain word forms in Old Persian inscriptions attest a terminus ante quem, i.e., they show that a given change must have happened by the date the inscription was composed. All this also contributes to the philological interpretation of Old and Middle Persian word forms by suggesting, e.g., that the present stem of 'die' and the identically structured passive of 'do' are to be read mp mīr-(as always assumed), mp kīr-and op məriya-, kəriya-, respectively.
At the same time, the "glass box" approach highlights an element of dialectology by showing that the op sound changes under study cannot be brought into line with the Middle Persian varieties that have come down to us. The op change of postconsonantal y and w, which, as I argue, did not happen in Proto-Middle Persian, thus constitutes a dialectal difference between Old and Middle Persian to be added to those already noted. I am sure that more such cases can be found where a look into the "glass box" allows to refine both the sequence of changes and their dialectal status.