The Pronunciation of Classical Attic Zeta
2023.07.25 | By George A. Sheets
Introduction
This paper seeks to reopen consideration of a question that has been widely taken as settled: what was the normative pronunciation of classical Attic zeta? Most modern authorities claim that Attic <ζ>[1] in the fifth century BCE had a phonetic value similar to that of the medial cluster in, e.g., the English word “wisdom”—i.e. [zd] vel sim.[2] Newer primers of ancient Greek reflect this consensus,[3] as do the standard handbooks on Greek pronunciation.[4] Dissenting views have been relatively infrequent.[5]
The modern consensus in favor of a value of [zd] for classical Attic <ζ> echoes the view of the great humanist Erasmus in the 16th century,[6] who invoked the authority of ancient grammarians as evidence. Erasmus was not concerned that the testimonia he relied upon date from the Hellenistic and Roman periods, centuries after the phonetic realization of <ζ> as [z] had already become normative in the Greek Koine.[7] Yet the anachronism of the ancient sources is significant, since it effectively renders the probative value of their testimony no greater than that of any other “restored” pronunciation.
In the 19th century Friedrich Blass produced a number of philological observations in support of a phonetic realization as [zd].[8] These continue to constitute most of the arguments that are commonly rehearsed in connection with this issue, including the arguments of Sturtevant and Allen. Additionally, inscriptional evidence from various epichoric dialects of different periods[9] is often referenced in the literature, although scholars generally do not agree on its interpretation in specific cases.
The concepts of structural linguistics, especially that of the phoneme and its functional characteristics within a phonological system, had not yet been developed when Blass wrote his influential study. Although these concepts have long since become fundamental in all models of phonological analysis, they have had surprisingly little impact on the consensus view of classical Attic <ζ>. Even the most basic question of whether <ζ> expresses a unitary phoneme or a cluster of phonemes is rarely addressed directly in modern literature on the subject. Generally, however, the literature presumes that classical Attic <ζ> masked a cluster of /s/ + /d/ with a phonetic value of [zd].[10] Indeed a cluster analysis is probably unavoidable in any account of the emergence of <ζ> that invokes notions of “metathesis” or “assimilation” of the segment’s phonetic constituents. This can be illustrated by a summary of Allen’s analysis:[11]
“[A]t a time when the semitic alphabet was adopted by Greek, the ‘zayin‘ symbol was at first applied to a still existing affricate type of combination; for it is difficult to see why a sequence of [zd] should not have been represented by σδ instead of by a special sign; … it has also been suggested that the affricated combination was at this early period a single phoneme and so preferably represented by a single symbol. … The value of ζ as an affricate may also have survived in some of the Greek dialects. … However, the metathesis of [dz] to [zd] must have occurred at an early date in Attic and most other dialects. … Later in the 4th c. … there begin to be some [epigraphic] confusions between ζ and σ [indicating] the change to the modern Greek value of [z] was already taking place.” (footnotes omitted)
Certainly phonemes can evolve into clusters and the reverse is also true, although structurally such transitions involve phonological isolation, emergence, and merger, not sporadic phonetic phenomena like metathesis and assimilation. The issue is not merely terminological, as Allen would of course acknowledge. In fact he states that the source of the metathesis he posits was “the sequence [dz] [which] would … have been peculiarly isolated in Greek when it possessed neither any other affricates [ts] nor an independent /z/ phoneme” (1987: 57). But to invoke “metathesis” as an intermediate stage is to blur the line between phonetic and phonological analysis. If [d] and [z] were separable sounds capable of metathesis, then they were already separate phonemes (/d/ and /z/), and the second of them was already “an independent /z/ phoneme.” Yet Allen here posits an affricate phoneme /dz/ evolving into a cluster of /z+d/ by “metathesis” and then later evolving into an independent phoneme /z/.[12] The intermediate stage in this circuitous development is not based on phonological argument but on the need to accommodate an a priori assumption that attributes a phonetic value of [zd] (vel sim.) to classical Attic <ζ>. That assumption and the arguments that flow from it are ultimately based on the same testimonia invoked by Blass.
Ancient Testimonia and Arguments Based Upon Them
The presumption that the sound of <ζ> must at some point have been [zd] stems from the claims of Atticist grammarians, particularly Dionysius Thrax (mid-2nd cent. B.C.E.), who writes:
“Furthermore there exist three double consonants, ζ, ξ, ψ. They are called ‘double’ because each one of them is composed of two consonants, the ζ of the σ and δ, the ξ of the κ and σ, and the ψ of the π and σ.”[13]
As noted previously, this statement amounts to a restored pronunciation, not a contemporary vernacular one. This can be shown by reference to an emended passage from Aristotle Met. 993a 4-7:
“In the case of certain syllables one might dispute [what they consist of]. For some say that the ζα is from the ς and δ and α [cdd. σμα ἐκ τοῦ ς καὶ μ καὶ α], while others say it is a different sound and not one of the known ones.”[14]
If the text as emended by Bonitz is correct, the passage indicates that the phonetic character of <ζ> was in dispute long before the orthodoxy of Atticist grammarians had prevailed. Allen and others would reconcile the Atticist description with Aristotle by invoking the seemingly circular explanation that Aristotle was referring to the sound of <ζ> when “the change to the modern Greek value of [z] was already taking place,” while the Atticists are refering to the <ζ> of a still earlier period. Possibly so, but it seems equally plausible that Aristotle was referring to the <ζ> he knew and heard, while the Atticists are referring to an orthodox systematization of the Greek στοιχεῖα.[15] Nevertheless it is the description of the Atticists that created the presumption that Blass and others have sought to prove: namely that <ζ> had a phonetic realization as [zd].
In support of the Atticist description of ζ three additional types of evidence are commonly adduced: (1) etymology, (2) Near-Eastern loans in Greek, and (3) the morphophonemic reduction of /n/ to zero when followed by <ζ>. Of these the first is perhaps the strongest. Blass cites the adverbs Ἀθήναζε and θύραζε, presumably derived from *Ἀθήνασ-δε and *θύρασ-δε respectively, as proof that <ζ> expressed the sound [zd].[16] The etymologies are unimpeachable, but it is surely significant that with only a single very late exception[17] neither of these common words is ever found without the <ζ> spelling . The essentially total absence of the digraph suggests that the original morpheme boundary in this adverbial formant was no longer perceptible. In other words, -αζε is not evidence of a synchronic cluster, and the adverbial formant in these words is not -δε but is instead -αζε (or less likely -ζε), as further demonstrated by the extension of the same formant to other stems: e.g., Μουνυχίαζε, Ὀλυμπίαζε, including non-toponyms: e.g., Hom. χαμάζε, ἔραζε, Hesiod μέταζε. Thus –(α)ζε had become its own morpheme already in the pre-alphabetic period. Also pre-alphabetic is the development of ἵζειν ‘to seat’ < *si-sd-.[18] The same PIE root (*sed-) is apparently also present in ὄζος ‘offshoot’ (ὔσδος in Lesbian poetry), apparently < *h3e-sd- (cf. German Ast, Hittite hasd-). Again, the pre-alphabetic date of these mergers should be emphasized, because it effectively negates the possibility that <ζ> in any of them expressed a synchronic cluster in the alphabetic period. The merger of the cluster */sd/ into the segment or segments that would later be written as <ζ> is therefore a prehistoric event that has little relevance to how <ζ> was perceived or pronounced in classical Attic.
The spelling of Near-Eastern loan-words in Greek is the second type of evidence commonly cited to support a cluster interpretation of <ζ> as [zd]. The only classical-period examples cited by Blass and/or Allen respectively are Ὠρομάζης (Auramazda), the Genitive of which is found in Plato, Ἀρτάοζος (Artavazda) in Xenophon, and Ἄζωτος (Ashdod) in Herodotus. These have been taken to demonstrate that the cluster which occurs in the source-words is phonetically the same as the sound of Attic <ζ>. While this could be true, a less tendentious conclusion is equally plausible: namely that <ζ> was as close as Attic orthography could get to the medial cluster in the source-words. Examples from numerous living languages demonstrate that the spellings of loan-words are often inexact approximations, and the same was undoubtedly true of Attic Greek (e.g., Xerxes for Persian Khshayarsa).
It was Blass who first called attention to the morphophonemic (not his term) phenomenon whereby the preverb συν- drops its <ν> before <ζ> in the same way as happens with συν- before <σ>. Thus he compares συζῆν (‘live together’) < συν-ζῆν, and σύζυγος (‘yoke mate’) < συν-ζυγος, with συστέλλω < συν-στέλλω, συστήσω < συν-στήσω, etc.[19] In contrast, the final -ν of συν is not lost before other obstruents, including especially <δ> (e.g., σύνδικος, ἔνδικος, συγγραφή, σύνθεσις, etc.). From this contrast Blass inferred that <ζ> will have been pronunced [sd] rather than [ds]. Surprisingly many scholars have repeated this argument without questioning its validity.[20] Lejeune (p. 114) relies on it alone for an alleged pan-Greek pronunciation of [zd] in place of [dz]. Teodorsson, who argues that the [zd] pronunciation once existed only in Attic and never elsewhere, considers “the case of σύζυγος [to be] definitive” (1974: 313). Yet while Blass’s inference might be definitive if we could be sure that <ζ> masked the cluster /s/+/d/, it tells us nothing about whether the same morphophonemic reduction might not have obtained before an affricate or assibilate phoneme /dz/. In fact the morphophonemic reduction of /dz/ to [z] in the sequence /n+dz/ would seem entirely natural, leaving us with the phonetic sequence [nz],[21] passing then to [z] by operation of the very rule that Blass points to. It is only if we accept the a priori assumption that <ζ> masked the cluster /sd/ that Blass’s argument becomes “definitive.”
We have seen that the Atticists claim <ζ> had a complex articulation with two perceptible constituents, one of which was identified with sigma and the other with delta, apparently in that order. If the underlying segment were actually /dz/ rather than /sd/, why would the Atticists have put the two sounds in the opposite order? One reason is that Greek morphophonemically reduced all dental obstruents – voiced, voiceless, and aspirated – to zero before /s/: e.g., Nom. Sing. ἄσπις < *ἄσπιδ-ς, DP ἄσπισι < *ἄσπιδ-σι; aorist inf. ἐλπίσαι < *ἐλπιδ-σαι; perfect middle inf. πεφράσθαι < *πεφραδ-σθαι, etc.[22] A sequence of delta followed by sigma would therefore have hardly seemed possible in the ears of a native speaker, even if the sounds of <δ> and <ς> were perceived as both present in a phoneme /dz/. How else, then, to refer to the collocation of these two sounds except by ordering the letters in the only sequence that was phonotactically possible — i.e, sigma followed by delta? Additionally, even assuming for argument that there is some measure of phonetic accuracy in the Atticists’ perception that the constituents of <ζ> were sigma followed by delta, one must recall that in the Hellenistic period the transition of /d/ from a plosive [d] to a fricative [ð], the value it retains in modern Greek, was already in progress. Therefore the Atticists’ could have been describing a sound they perceived as [zð], which is a reasonable approximation of [z] while still maintaining the essence of its “double” character.
To conclude this section of my argument, the ancient testimonia are certainly suggestive, but they are not definitive—principally because, whatever their precise meaning, they express a restored pronunciation. Secondly, they have created the a priori assumption in modern scholarship that <ζ> masked a cluster of /sd/, and that assumption undergirds essentially all of the arguments that have been marshalled in support of it. As I shall now attempt to demonstrate, the assumption itself is wrong, because the synchronic behavior of <ζ> is not consistent with a cluster analysis.
Unitary Phoneme or Cluster?
As a phoneme /z/ cannot be reconstructed for PIE. In all of the Indo-European dialects where a /z/ phoneme exists, the phoneme has emerged from a secondary development within the history or pre-history of the language in question. In the case of Greek specifically, what ultimately became the phoneme /z/ with a normative value of [z] in modern Greek can be traced to a complex restructuring of the inherited phonological system in proto-Greek.[23] Foremost among the initial changes was the palatalization of inherited obstruents concurrently with the loss of inherited consonantal /i/ (i.e., *[y]). This restructuring resulted in the emergence of a new set of palatalized phonemes, of which all the voiceless palatalized plosives merged into */ty/, and all the voiced palatalized plosives into */dy/.[24] The alphabetic reflexes of these two phonemes varied across time and dialect, but since they ultimately resulted in the Koine graphemes of <σσ> and <ζ> respectively, it is reasonable to infer that the latest stage of their pre-alphabetic evolution entailed assibilation to */ts/ and */dz/ respectively.[25] The same inference is supported by epichoric graphemic conventions, all of which involve some selection of dental and/or sibilant graphemes. Presumably, therefore, a unitary phoneme /dz/ underlay the sound whose further evolution we are concerned with. To be clear, then, in tracing the history of Attic <ζ> we are concerned with the development of a unitary phoneme /dz/ (vel sim.) into the Koine unitary phoneme /z/.[26]
A change of /dz/ to /z/, or a replacement leading to the same result, would seem to be a straightforward phonetic or phonological development that is attested in numerous other languages. The structural motivation in Greek would have been isolation of the articulatorily anomalous /dz/, leading to its eventual resituation as a voiced sibilant standing in structural balance with the voiceless sibilant /s/, itself appearing medially as the geminate <σσ> when derived from /ts/.
Attic literary texts of all periods express a presumptive phoneme /dz/ with the grapheme <ζ>. Epigraphically too the same convention is normal, although when medial it can appear as <σζ> or <ζζ>.[27] In verse <ζ> usually “makes position,” a fact that has been taken to support a cluster analysis of <ζ>, although it can equally well support the affricate/assibilate analysis.
If one looks for evidence of any synchronic process in classical Attic that might support a cluster analysis, none is to be found. There is no clear example of a morphologically productive cluster being written with <ζ>. So, for example, the patronymic suffix -δης when added to an s-stem noun does not result in a suffix of *-ζης. Instead, the stem-final /s/ is replaced with <-ι->, as in Θουκυδί-δης, Ἐυγενί-δης, Ἱπποκλεί-δης (< *-kudes-de:s, *-genes-de:s. *klewes-de:s).[28] Likewise πρὸς and εἰς/ἐς when compounded with verbs that begin with /d-/ never generate <ζ> in place of the cluster <σδ>: e.g., προσδέω, εἰσδύομαι. Among the countless sigmatic case-forms of demonstrative pronouns and adjectives with suffixed -δε (τῆσδε, τοῖσδε, τούσδε, τάσδε, etc.), there is apparently only a single example of an outcome of <σζ> rather than <σδ>.[29] This lack of synchronic productivity indicates that <ζ> did not have the phonological status of a cluster. Conversely, <ζ> does behave like a unitary phoneme, since it forms “minimal pairs” both intially and medially with other phonemes: e.g., ζῶον vs. σῶον; ζῶστρα vs. σῶστρα; ζῶμα vs. σῶμα; ζωρός vs. σωρός; πεζός vs. πέσος, future and sigmatic aorist stems of most verbs in ‑ζω (δικάζειν vs. δικάσειν; ἐλπίζειν vs. ἐλπίσειν). Therefore based upon its synchronic behavior classical Attic zeta gives every appearance of being a unitary phoneme, not a cluster of two phonemes. As a unitary phoneme it possessed the same phonological status that has been reconstructed for its ancestor at the beginning of the alphabetic period and the same phonological status that it retained in the Koine.
Phonetic Realization
If we are not misled by the unwarranted assumption that <ζ> had to mask a cluster of /sd/, and if, further, we observe that the synchronic behavior of <ζ> is that of a phoneme and not a cluster, then we are ready to examine what the phonetic realization of that phoneme most likely was. We have seen that the Atticists seem to restore a sound like [zd] to zeta, in which case the phoneme should be something like */zd/ or */zd/. Based upon the phoneme’s origin, however, and upon its eventual outcome in Koine [z], /dz/ would make better sense despite the Atticist description. Not only is /dz/ historically more likely but, as I will now attempt to show on the grounds of phonological typology, it is the only realistic possibility for a unitary phoneme with the requisite co-articulatory elements.
A putative phoneme */zd/ or */zd/[30] would exhibit an articulatory structure that is highly anomalous and perhaps unique among the as yet described languages of the world. For example, among the approximately 700 languages inventoried by Merritt Ruhlen[31] in his review of the Language Universals Project of Stanford University, not a single one was reported to utilize such a phone. Likewise among the 451 languages included in the UCLA Phonological Segment Inventory Database (UPSID) none presents such a phone.[32]
The typological anomaly of a putative */zd/ or */zd/ is further suggested by an analysis of its coarticulatory constituents. The coarticulatory features of complex phonemes (i.e., phonemes with one or more secondary articulatory features) comprise a set of recurrent patterns among natural languages. One finds, for example, palatal and labial offglides and assibilates among all classes of obstruents: e.g., /kw, tsj, dz, zj, nw/. Only with the more restricted set of plosives, however, are nasal onsets in evidence: e.g., /nk, nt, ndz, mb, mbv/. Stops with double closure, though relatively rare, appear among Ruhlen’s data (e.g., /kp, gb, pt, bd/. What apparently fails to occur anywhere, however, is a stop with fricative onset—i.e., the type of coarticulation involved in */zd/ or */zd/. Thus Ruhlen reports no */vb/ nor */hk/ nor */st/ nor any other complex phoneme comparable to */zd/ or */zd/. This absence suggests that the coarticulatory structure of the putative */zd/ or */zd/ violates a “language universal.”
Still a further anomaly of */zd/ or */zd/ is revealed when one considers it in the context of the rest of the phonological system of Ancient Greek. Specifically, */zd/ or */zd/ would be a voiced obstruent standing unopposed to a corresponding voiceless obstruent: there is no */st/ or */st/ in Greek. From the perspective of structural economy this is unnatural. Normally if a language utilizes the correlation of [±voice] to define phonological oppositions, then obstruents that are unopposed in that feature are [-voice]. In other words, the voiceless member of the opposition is the “unmarked” or default one.[33]
These typological anomalies are sufficient, I believe, to exclude the possibility that a monophonemic <ζ> was realized as [zd] in any period of Greek.
Conclusion
In attempting to recover how any segment was pronounced in a language with no surviving native speakers, one must reckon with an infinity of synchronic and diachronic variables. Even limiting the scope to Athens in the fifth century, there will be no control for the numberless variety of speech contexts, different social registers, and personal idiolects that must have obtained there, as in any living speech community. What is it, then, that the search for a socially “correct” pronunciation seeks? Presumably it must be an elite normative pronunciation – the pronunciation advocated by ancient γραμματικοί perhaps, or by modern classicists. It is not, however, the actual pronunciation of every speaker at every time in every social context of a living speech community. In the case of a phoneme, its normative pronunciation is sometimes called its “elsewhere allophone”—i.e., the value it would possess absent any syntagmatic factors. In this paper I have attempted to demonstrate that <ζ> in classical Attic usage was indeed a phoneme with the structure of a voiced alveolar assibilate /dz/ and an elsewhere allophone of [dz] vel sim.
Minneapolis, 2023
[1] In this paper I follow the standard convention of using angled brackets < > to express graphemes, square brackets [ ] to express phonetic values, and oblique slashes / / to express phonemes. Where the discussion refers simply to an alphabetic letter, as when quoting an ancient grammarian, no brackets are employed.
[2] E.g., Kühner, R. and Fr. Blass. Ausführliche Grammatik der Griechischen Sprache, 3rd ed., I.1, Hannover, Hahnsche Buchhandlung, 1890, p. 57-58; Schwyzer, E. Griechische Grammatik, 4th ed., I, München, Beck Verlag, 1953, p. 329-332; Lejeune, M. Phonetique historique du Mycénien et du Grec ancien, Paris, Klincksieck, 1972, pp. 113-116, 138-139; Rix, H. Historische Grammatik des Griechischen, Darmstadt, Wiss. Buchgesellschaft, 1976, p. 92 et passim; L. Lupas Phonologie du grec Attique, The Hague, Mouton, 1972, p. 27 (“plus probable”); S. Teodorsson, The Phonemic System of the Attic Dialect 400-340 B.C., Studia Graeca et Latin Gothoburgensia xxxii, Lund, Institute of Classical Studies of the University of Göteborg. 1974, p. 225-27; Palmer, L. R. The Greek Language, New Jersey, Humanities Press, 1980, p. 210-211; L Threatte, The Grammar of Attic Inscriptions I: Phonology, Berlin; New York, De Gruyter, 1980, p. 546 ff; Bubenik, V. The Phonological Interpretation of Ancient Greek: a Pandialectal Analysis, Toronto, University Press, 1983, p. 81-82. A. L. Sihler, New Comparative Grammar of Greek and Latin, Oxford (1995) p. 194, who follows C. D. Buck, Comparative Grammar of Greek and Latin, Chicago, University Press, 1933, p. 142-42 in adding the qualification: “the most straightforward reflex of [dy] and [gy] would have been something like [dž] or [dz], and such a pronunciation must in fact have been current in some parts; it was with this value that the letter <z> was carried to Italy, where it was used to represent [ts].”
[3] E.g., Joint Association of Classical Teachers. Reading Greek, Cambridge, England, Cambridge University Press, 1978, p. 1; Balme M. and G. Lawall. Athenaze, 2nd. ed., vol. I, New York, Oxford, Oxford University Press, 2003, p. xii; Mastronarde, D. Introduction to Attic Greek, University of California Press, Berkeley, 1993, p. 10; Groton, A. From Alpha to Omega, Newbury Port, MA, Focus Information Group, 1995, p. 3—Groton also includes [dz] as a possible pronunciation.
[4] Sturtevant, E. The Pronunciation of Greek and Latin, 2nd ed., Linguistic society of America, University of Pennsylvania, Philadelphia, 1940, p. 91-92; and Allen, W. S. Vox Graeca, 3rd ed., Cambridge, Cambridge University Press, 1987, p. 56 ff.
[5] Meisterhans, K. Grammatik der attischen Inschriften, 2. verm. und verb. Aufl., besorgt von E. Schwyzer, Berlin, Weidmann, 1900, p 71: “Die Frage nach dem Lautwert von Zeta (ds oder sd?) in der Zeit vor 350 v. Ch. muß einstweilen noch eine offene bleiben.” Mathews, W. K. “The Pronunciation of Attic Greek ζ in the Sixth and Fifth Centuries B.C.,” Lingua 71.1 (1954-55) 63-80—arguing for a pronunciation of zeta as [dz] vel sim., becoming [z] by the fourth century “[a]s all investigators are agreed” (p. 64); Arena, R. “Il valore di ζ nei vari dialetti greci,” Istituto Lombardo (Rend. Lett.) 94 (1960) 513-553, at 549-553. Bailey, Charles-James N. “The Pronunciation of ‘Zeta’ in Ancient Greek,” Papers from the Fourth Regional Meeting, Chicago Linguistic Society, April 19-20 (1968) 177-193—arguing for a pronunciation of “[žďž] = [žǰ]” (= IPA [ʒj]?), becoming [žž] “in the Hellenistic period,” and ultimately [z]. Steriade, Donca. Greek prosodies and the nature of syllabification, diss. M.I.T., 1982, 1984. pp. 262-269—using the Mycenaean z-series (why me-zo and not *me-do, etc.) and Homeric metrics to question the sometime existence of any metathesis rule transforming /dz/ into [zd]. Rohlfs, G. “Die Aussprache des z (ζ) im Altgriechischen,” Das Altertum 8.1 (1962) 3-8, who argues inter alia that the affricate value of Modern Italian <z> is itself a vestige of the ancient phonetic value of Greek zeta.
[6] De Recta Latini Graecique Sermonis Pronuntiatione, 1528.
[7] That Koine <ζ> was phonetically realized as [z] is not in dispute. In any event Erasmus’ focus was on the “correct” pronunciation, not a vernacular one. See Mathew Dillon, “The Erasmian Pronunciation of Ancient Greek: A New Perspective,” CW 94.4 (2001) 323-334.
[8] Über die Aussprache des Griechischen, 3rd ed., Berlin, Weidmann, 1888, p. 111-122.
[9] Particularly in West Greek the segment(s) spelled with Attic <ζ> are most commonly rendered with <δ> or <τ> (in Crete) word-initially and <δδ> or <ττ> medially. Sporadically elsewhere the segment corresponding to Attic <δ> is sometimes rendered as <ζ>, which in early Arcadian is also used to spell the segment corresponding to the secondary <τ> of Attic-Ionic τις/ὅστις. Buck, C. D. The Greek Dialects: Grammar, Selected Inscriptions, Glossary, Chicago University Press, 1955, p. 71-72. Interestingly <σδ> is normally found epigraphically only in Hellenistic-era Lesbian, where <ζ> has been redeployed to express palatalized /d/ in ζά (for διά). The use of <σδ> in mss. of the Aeolic poets is generally understood to reflect the intervention of Hellenistic φιλόλογοι, as also in Aeolicizing poetry (principally Theocritus) of the period.
[10] Sturtevant (1940), Allen (1987), Teodorsson (1974), Threatte (1980), Bubenik (1983) and others explicitly adopt a cluster analysis. Schwyzer (1953), Lejeune (1972), Palmer (1980), Horrocks (2010) and others do so implicitly. Lupas (1972) seems to treat <ζ> as both an allophone of /s/ and as a cluster of /sd/.
[11] Allen (1987: 57-58).
[12] Schwyzer, Lejeune and others suggest that the penultimate stage in this process of phonetic evolution involved a voicing of [s] to [z] before [d] in the cluster [sd], and then a change of the [zd] cluster to [zz] by “assimilation,” and finally a change to [z] by “degemination.” These steps are not in themselves implausible, but the starting point, which would necessarily be the cluster /sd/ is merely an assumption.
[13] Ἔτι δὲ τῶν συμφώνων διπλᾶ μέν ἐστι τρία, ζ, ξ, ψ. διπλᾶ δὲ εἴρηται, ὅτι ἓν ἓκαστον ὐτῶν ἐκ δύο συμφώνων σύγκειται, τὸ μὲν ζ ἐκ τοῦ σ καὶ δ, τὸ δὲ ξ ἐκ τοῦ κ καὶ σ, τὸ δὲ ψ ἐκ τοῦ π καὶ σ. Ars Grammatica 14.4-6 Uhlig. Essentially the same observation is repeated by Dionysius of Halicarnassus a century later at De Compositione Verborum 53.1-7.
[14] Ἀμφισβητήσειε γὰρ ἄν τις ὥσπερ καὶ περὶ ἐνίας [5] συλλαβάς: οἱ μὲν γὰρ τὸ ζα ἐκ τοῦ ς καὶ δ καὶ α φασὶν εἶναι [cdd. σμα ἐκ τοῦ ς καὶ μ καὶ α], οἱ δέ τινες ἕτερον φθόγγον φασὶν εἶναι καὶ οὐθένα τῶν γνωρίμων.
[15] ζ along with ξ and ψ were classified as διπλᾶ, the common feature of which was thought to be the presence of ς. The διπλᾶ in turn fell within the class of ἡμίφωνa, which comprised ς and what we would call “sonorants.” The ἡμίφωνa themselves formed a subset of the σύμφωνα (Lat. consonantes), the rest of which we would call “plosives”. The feature that we call “voice” was not distinguished as such either analytically or teminologically.
[16] Additional examples are Βήσαζε and Θήβαζε,
[17] The TLG disk C reports the single occurrence of Ἀθήνασδε in a certain Theodosius Grammaticus (7.16) of Constantinian date. There are no epigraphic examples in IG or elsewhere in the PHI epigraphic database.
[18] A spelling of this verb with -σδ- is found only in Hellenistic-era texts of the Aeolic poets. The emergence of stem-final <ζ>/<σδ > may have been due to analogy with its intransitive cognate ἕζομαι < *sed-yo-mai.
[19] Blass (1988: 114). As noted by Allen and others, the same reduction occurs in the present stem of verbs with stem-final <ζ>: e.g., pres. πλάζω, σαλπίζω, but aor. ἔπλαγξα, ἐσάλπιγξα.
[20] Allen (1987: 56); Sturtevant (1940: 92); Bubenik (1983: 81); Sihler (1995:194). It appears that only Bartonek has disputed the inference, but only as evidence for a sometime pan-dialectal value of [zd]. Bartonek nevertheless does maintain that “in Attic and Ionic there positively existed z+d … changing into z(z) form the 4th cent. B.C. onward.” Bartonek, Antonin (1961). Vývoj konsonantického systému v reckých dialektech. Praha: Státní Pedagogické. English Appendix, 150-151.
[21] Consider, for example, the reduction of /d/ in the Engl. homonyms: winds/wins; tends/tens; bands/bans; medially Windsor, etc.
[22] Lejeune (1972:74-75).
[23] For an excellent structural analysis of these phenomena in Greek, including the emergence of <ζ>, see William Diver. “On the Prehistory of Greek Consonanatism,” Word 14 (1958) 1-25.
[24] The sources of <ζ> can therefore be traced not only to */dy / (e.g., Ζεύς < *Dyāus, πεζός ‘foot soldier’ < *ped-yos) but also to */gy/ (e.g., ῥέζω ‘do’ < *wreg-yō, φύζα ‘panic’ < *bhug-ya) and */gwy/ (e.g., ζώω ‘live’ < *gwi-ō, νίζω ‘wash’ < *neigw-yō). The phoneme */dy/ also served as the merger target for certain examples of word-initial consonantal */i/ (ζυγόν ‘yoke’ < *yug-om, ζέω < */yes-ō/).
[25] Palatalization, affrication, and assibilation commonly occur as consecutive stages of phonetic adjustment and phonological evolution.
[26] Possibly the final step was not a “development” but a replacement. See Nagy, G. Greek Dialects and the Transformation of an Indo-European Process. Loeb Classical Monographs. Cambridge, Mass.: Harvard University Press, 1970, p. 127.
[27] Threatte (1980: 546-550). <ζ> is also used sporadically for other segments.
[28] No such constraint blocks suffixes that begin with other dental plosives (Ὀρέσ-της, εὐτελέσ-τατος, κηδέσ-της; verb endings in -σθε, σθαι; verb stems in -σθη- etc.).
[29] ἐ͂ν ἄρα τοῖσζ’ ἀδάμ[αντος ἐνὶ φρεσὶ θυμός] (SEG 10.404), commemorating the fallen who defended Athens against the Persians. Threatte (1980: 546) considers this text to be one of two that “conclusively show the pronunciation of zeta … was the obstruent cluster [zd] in the sixth and fifth centuries B.C,” yet a single example offset by scores, if not hundreds, of counter examples seems very slender evidence. Perhaps this single example is just a nonce spelling of the geminate sigma that is quite commonly found elsewhere, particularly before dental obstruents: e.g., τέλεσι τοῖσς τ[ο͂ν IG I(3) 66.22; Κλοισσθένης IG IV 550. The other “conclusive” text cited by Threatte is the theonym ΖΔΕΥΣ on a black figured vase (ABV p. 96, no. 14), but a reversal of the sibilant element in other dipinti is common. See, e.g., εὐσχαμενος Σχανθος τοσχοφιλος Σχενοκλēς, Πιστοσχενος, σφυχē, ἔγρασφεν. Kretschmer, P. (1897, rep. 1969). Die griechischen Vaseninschriften ihrer Sprache nach untersucht. Hildesheim, New York: G. Olms.
[30] The latter is posited explicitly by Rix (1967: 92) as a transitional stage on the way to a cluster.
[31] Ruhlen, Merritt. A Guide to the Languages of the World, 2nd ed., Stanford, CA: Self-published 1976.
[32] On the relationship between the UPSID database and the Stanford Phonology Archive see Maddieson, Ian. “The Design of the UCLA Phonological Segment Inventory Database (UPSID),” in Patterns of Sounds: Cambridge Studies in Speech Science and Communication, Cambridge, New York: Cambridge University Press 1984, pp. 156-173. In personal correspondence Ian Maddieson has written: “I agree that a unitary element /zd/ is highly improbable, as there is no typological precedent for it that I know among modern languages.” See also Ladefoged, Peter and Ian Maddieson. The sounds of the world’s languages, Oxford, UK ; Cambridge, Mass. : Blackwell 1996.
[33] Martinet, A. Économie des changements phonétiques ; Traité de phonologie diachronique (3rd ed.). Berne: A. Francke., 1970, pp. 99-101.