Part II: Specifics


II§1. Despite the attestation of Greek as far back in time as the second millennium BCE, a chronological chasm remains between this language and the proto-language reconstructed as Indo-European. What I just said is most apparent on the phonological level. The necessity for positing a multitude of phases and patterns of phonology which must have been operative at some early point in prehistoric Greek but which reveal only residual traces in extant Greek is well illustrated by such Indo-European studies as those of Kuryłowicz (1956) and Szemerényi (1964).
II§2. Granted, the discovery that Linear В was a writing system for an early form of Greek that was spoken in the second millennium BCE brings us closer to Common Greek. And in some respects, the Greek of the Linear B texts, known to us as Mycenaean Greek, resembles the Common Greek reconstructed by linguists more than it resembles the attested forms of Greek in the first millennium BCE. For example, Mycenaean Greek still preserves such Common Greek phonological features as labiovelars and intervocalic --. [1] In other respects, however, the phonology of Mycenaean Greek makes us appreciate even more keenly than before the remoteness of Common Greek from all attested stages of Greek, including Mycenaean.
II§3. In some ways, Mycenaean Greek complicates our reconstruction of unattested phases of Greek, in that it presents idiosyncrasies that could only be suspected at best from the evidence of extant post-Mycenaean Greek. On the phonological level, for instance, the multiple and varied Mycenaean reflexes of the Common Greek syllabic sonorants *ṛ *ḷ *ṃ *ṇ point to more complications than we could ever have imagined from the reflexes attested in the first millennium BCE. The evidence of Mycenaean Greek now forces us to reckon with the conditioning of these sonorants on the following levels: (1) purely phonological, (2) morphological, (3) dialectal, or (4) any combination of the preceding three.
II§4. Nor is the linguistic evidence from the second millennium BCE appreciably more helpful than the corresponding evidence from the first millennium when it comes to elucidating the numerous problems involved in the Greek transmission of the Indo-European consonantal series known as the laryngeals. {33|34} {Such difficulties on the phonological level are only to be expected, since the primary evidence for reconstructing laryngeals is not phonological but morphological.}
II§5. Occasionally, the evidence of an archaizing medium like Homeric poetry, as attested in the first millennium BCE, reveals more about prehistoric Greek phonology than the evidence of Mycenaean Greek as attested earlier, in the second millennium BCE, by way of the Linear B tablets. For example, we have already seen in Part I the metrical traces of word-initial *s + sonorant in Homeric diction. There is even at least one instance where epic has preserved a phonological pattern traceable all the way back to a pre-Greek phase of the prototypical language that we reconstruct as Indo-European. The pattern in question can be described as the morphophonemic principle of Dehnungsgesetz. ◊My approach in what follows diverges from that of Wyatt (1969) and converges with that of Blanc (2008). [2]
II§6. The basic formulation of Dehnungsgesetz goes back to Wackernagel (1889/1953), who used the comparative linguistic evidence of Greek and Indo-Iranian combined. Basically, Dehnungsgesetz can be defined this way: when two vowels come together as the final and initial elements of two compound-formants, the resulting contraction will entail the elision of the first vowel (V1) and the lengthening of the second vowel (V2):
-V1 + V2- = -V̄2-
II§7. A case in point is *στρατο + αγός, which becomes *στρατᾱγός > στρατηγός. A reflex like στρατηγός illustrates the anteriority of such a contraction to the type of contraction resulting from the Greek innovation whereby intervocalic σ is lost: here *οα becomes ω rather than η, as in θάττ-ω (< *…-οσα).
II§8. But the Dehnungsgesetz becomes extended, in that an initial vowel of the second compound-formant becomes lengthened no matter what the final element in the first formant may be. A case in point is *κυν + άγος, which becomes *κυνᾱγός > κυνηγός. Here we see the removal of one constraint in the natural language. But in the artificial language of the epic, there are further extensions of Dehnungsgesetz, as lengthening of initial vowel spreads from the second compound-formant to simplex nominals as well, and ultimately to any word. [3] One locus of diffusion is probably from compounds where the first constituent was an adverb: after the adverb evolves into a preposition (as well as a preverb) there is an opportunity for transition of artificial Dehnungsgesetz from (a) compound consisting of adverb + noun to (b) preposition + object of preposition. [4] Finally, even the constraint of word-initial vowel is removed. That is, not only does -VC+ ‖ V- become -VC ‖ V̄- but also, by extension, -V+ ‖ СV- becomes -V ‖ CV-. (In this formulation, the sign “‖” stands for word-boundary.) To put it another way, any word-initial syllable becomes subject to lengthening in the epic language. What results is a set of artificial vs. non-artificial pairs such as ἀ̄νήρ (II 553 / II 701, etc., with / without elision) vs. ἀνήρ (II 673, etc.), Οὔλυμπον (I 221 / I 497, etc., with / without elision) vs. Ὄλυμπον (I 402, etc.); also μείλανι (XXIV 79) vs. μέλανα (VII 265, etc.), Πουλυ-δάμας {34|35} (ΧΙΙ 60, etc.) vs. Πολύ-φημος (ix 407, etc.).
II§9. This artificial Dehnungsgesetz of epic was operative even after the Ionic phonological change whereby ᾱ became η had ceased. So we see ἀ̄νήρ rather than *ἠνήρ in II 553, vs. the compound-formant -ήνωρ showing the old product of natural Dehnungsgesetz and stemming from a period of Ionic when the phonological change ᾱ > η was still operative. [5] As for the type ἠνορέη (VI 156, etc.: instead of *ἀ̄νορέη), the artificial Dehnungsgesetz which produced it goes back to the same archaic period when the change ᾱ > η was operative; the ultimate morphological extinction of the configuration *ἀ̄νορέη in spoken Ionic has precluded the recent poetic creation of *ἀ̄νορέη, whence the survival of inherited and residual ἠνορέη. For more on ᾱ / η in Homer, this time with ᾱ-samples representing not an innovation but an old formation that is even older than than η-samples, see below under dialectology.
II§10. So much for instances where the obsolescence of a phonological rule in the natural language permits the ultimate extension of the same rule beyond its etymological confines, in the retentive poetic language of the epic. Retention can also be static, however, and subject to ultimate attrition. For example, let us consider the early loss of ϝ (= * or “digamma”) in a prehistoric phase of Ionic, the last major dialectal phase of Homeric diction. As Milman Parry points out, ϝ was lost in epic diction “neither sooner nor later than it was lost in the daily speech, but the singers who had to compose in a rigorous and therefore highly conservative verse-form, still used the old phrases and verses because that was their way of making poetry, because to have given up the traditional phrase wherever the loss of the digamma now caused hiatus or failure to make position, would have been to destroy the diction almost entirely.” [6] Thus in contrast to roughly 300 Homeric cases of elision despite digamma, there are still roughly 2,000 cases of non-elision because of digamma. [7] As Parry said, directly challenging Richard Bentley, “Homer’s language has traces of the digamma, but not the digamma itself.” [8] But although the formulaic language of the epic is an admirable preservative of traditional patterns dating back to a time when digamma was still extant, new patterns ignoring the etymological digamma eventually emerge—sometimes even in the most overtly formulaic expressions. To quote Parry again: “Just as we can show the metrical usefulness of the older phrase, and the fixed place which it holds in the diction, so can we do for phraseology with newer forms.” [9] For instance, before loss of ϝ, the following verse-type could refer only to a masculine speaker:
καί μιν φωνήσας (ϝ)έπεα πτερόεντα προσηύδα
‘and addressing him he spoke winged words’.
II§11. In the Iliad and Odyssey, there are 30 occurrences of this verse; but there are also 9 others where the speaker is feminine, and the necessitated elision is possible only without ϝ:
καί μιν φωνήσασ’ (ϝ)έπεα πτερόεντα προσηύδα. {35|36}
II§12. At times the same formulaic verse will entail both the presence and the absence of the digamma-factor in the metrical pattern, as in the following type of Homeric verse (IV 403, XVII 90, XVIII 5, etc.):
ὀχθήσας δ’ἄρα (ϝ)εἶπε προς (-)ὃν μεγαλήτορα θυμόν
‘angered, he said tο his great-hearted thumos’.
II§13. In sum, we have to contrast “the stability of the diction as a whole” with “the fluidity of the diction in the grouping of its elements” (the expressions I am using here come directly from Parry). [10]
II§14. There have been thorough synoptic treatments of Greek phonology from both the diachronic and synchronic points of view. Especially useful are the books of Lejeune (1955) and Allen (1968a) respectively.
II§15. In the case of vowels, there has been a noteworthy contribution to our understanding of their system by Ruipérez (1956), who has applied the approaches of Martinet (1964) in interpreting the consecutive stages of the Attic-Ionic vowel-system with reference to the dynamics of pressure and counterpressure in that system. Bartoněk (1966) has provided further details, and Szemerényi (1968a), further refinements. In particular, Szemerényi has revised the relative scheme for the fronting of ū to ǖ and for an Attic Rückverwandlung of ǟ to ā.
II§16. As for consonants, efforts to establish their prehistory are adequately represented by the works of Allen (1958) on palatalization and of Stang (1957) as well as Diver (1958) on the prehistoric phenomenon of gemination before *--. Bartoněk (1961) has attempted reconstructions of the consonantal systems of several well-known dialects at several (including late) developmental stages.
II§17. Occasionally, certain latent features of a consonantal series become overt only within the framework of specialized orthographic conventions. For instance, π and κ are written regularly φ and χ immediately before θ as in the aorist passive constructions with -θη-: τέρπω vs. ἐτάρφθην, δέρκομαι vs. ἐδέρχθην, and so on. But the backwards-assimilation here is probably not from non-aspirated to aspirated (and in this sense the writing here is imprecise) but from tense (π κ) to lax (φ χ), and the feature of laxness in the Greek series of aspirated stops would not become apparent if it were not for this convention in the spelling of consonantal clusters. [11]
II§18. In any case, we may posit as a general principle that the constraints and licenses of the various Greek orthographic systems are the primary key to evaluating the nuances of the underlying phonological systems.
II§19. Conversely, the evolution of Greek writing-systems is inextricably linked with the {36|37} evolution of the Greek phonological system itself. Here I highlight the important contemporary trend of recognizing the extent to which linguistic conditions affect graphic conventions. We can illustrate this relationship by observing the representation of
ę̄ (resulting from a collapsed opposition of ǟ vs. ę̄)
by way of
in the post-Euclidian Attic alphabet, as opposed to the cumbersome representation of all three by <ε> in the pre-Euclidian Attic alphabet. The basic motivation for this eventual orthographic reform was not the genius of some εὑρετής, but rather, the accidental convergence of (1) the studious but mechanical application of the acrophonic principle of the alphabet and (2) phonological shifts in the Attic-Ionic vowel-system. Of the two Semitic aspirate-signs ḥēth and the former was apparently the closer approximation to the single Greek spiritus asper, whence the original generalization of Semitic ḥēth = <η> as representing Greek h- (or -, on the acrophonic principle). Consequently, there was no need for another aspirate-sign, and the initial element of is viewed simply as ē from a Hellenic standpoint: hence the original generalization of Semitic = <ε> as representing both short-e and long-e in Greek. With the onset of psilosis (loss of spiritus asper) in East Ionic dialects, however, the initial element of ḥeth / <η> becomes viewed as ǟ (which later loses its distinction from ę̄), whence now the restriction of <ε> to representation of e and ę̄ only, vs. the generalization of <η> for ę̄ (< ǟ).
II§20. There are Central Ionic dialects where ǟ and ę̄ are still distinct and still represented by <η> and <ε> respectively. [12] A case in point is archaic Naxian; in this case, there is the added complication that the orthographic systems of psilotic (East Ionic) and non-psilotic (Central Ionic) converge: hence <η> is used for writing both h- and ǟ respectively.
II§21. Finally, the Attic-Ionic monophthongization of *ei to ē in around the fifth century BCE leads to the opportunity of representing all instances of ē, whether etymologically diphthongal or not, with <ει>.
II§22. What, then, if this Attic-Ionic monophthongization had occurred, say, two centuries earlier? (Such was the situation in Corinthian. [13] ) In that case, the putative sophistication of a separate graph <ει> for ę̄ might have been regularized that much earlier in the Attic script as well.
II§23. I should add that the very introduction of a vocalic system into the Greek alphabet was no original invention but rather the generalization of a Semitic mechanism founded on the principle of matres lectionis. [14] We must acknowledge, however, that the practical improvement resulting from the separate representation of vowels and consonants is remarkable: it amounts to a shift from syllabary to alphabet, that is, a movement away from the constraint of representing only phonetically concrete segments toward the admission of such phonetically abstract segments as p t k (which cannot be produced in isolation, that is, without adjacent vowels or aspiration). [15] {37|38}
II§24. Next we turn to the realm of accentuation, where we have to distinguish between intonation and stress. [16] It is intonation, not stress, that dominates the testimony of the ancient grammatical tradition. And it is intonation that preserves patterns of Indo-European accentuation. Jakobson (1937) has produced a most elegant synchronic description based on the interrelation between mora-structure and accent-limitation, and Kuryłowicz (1958) has added diachronic perspective. [17] Allen (1967) has added still further perspective on the mora-structure, indirectly raising the question whether recessive accentuation should be connected with generalization of intonational variants not in pausa. [18] For an approach to intonation using the techniques of generative phonology, I cite Kiparsky (1967). As for attempts to recreate ancient Greek intonation by training in pronunciation, I agree with Allen’s reaction: “A primary stumbling-block […] is not so much the inability to handle pitch, but rather to divorce lexical pitch from attitudinal intonation.” [19]


II§25. The present sequence of treating morphology immediately after phonology is appropriate because of the frequent interrelations between these two levels. It can even be formulated typologically, without specific reference to Greek: where phonology and morphology come into conflict, the resolution is generally in favor of the morphological level. An ideal example in Greek is the morphological restoration of intervocalic s in the future tense (hence λύσω), despite its prehistoric loss on the phonological level. [20] The key to such restoration is the factor of the morpheme-boundary. For the future tense, the morpheme-boundary is between verbal base and future-marker s. Here it is simply a matter of extension from base-final consonant + s to base-final vowel + s. What results is the re-establishment of s between vowels. Where the base always ends in a vowel, on the other hand, s cannot be restored: hence optative -οι-ο, -αι-ο (not *-οι-σο, *-αι-σο); elsewhere, the generalization of s from consonantal base-final to vocalic base-final is still attested in a state of transition, that is, morphological restoration of s has not yet leveled out the phonological reflex without it. Thus besides the morphological innovation μέμνη-σαι ‘уоu remember’ (XXIII 648), Homeric Greek also preserves the old phonological reflex μέμνη-αι ‘уоu remember’ (XXI 442). [21]
II§26. Next I turn to morphophonemic rules, which I define as those phonological rules that apply only within a restricted morphological framework. In terms of this definition, I find that there are markedly few instances {38|39} of morphophonemic rules in Greek (or even in Indo-European), and this dearth is especially striking when we compare other languages of the world.
II§27. Among the few morphophonemic rules that can be singled out in Greek is the generation of ν ἐφελκυστικόν after word-final -ι or -ε, as in the dative plural -σι-ν (but not in dative singular -ι), in the 3rd singular or plural -σι-ν / -τι-ν, in the 3rd singular -ε-ν, and in a few isolated forms such as εἴκοσι(ν) ‘twenty’, πέρυσι(ν) ‘last year,’ ἄμμι-ν / ὔμμι-ν ‘for us’ / ‘for you’. [22]
II§28. Another morphophonemic rule is the predetermination of -ό- or -ώ- before the comparative formant -τερος, depending on whether the syllable preceding -ό-τερος or -ώ-τερος is long or short. [23]
II§29. Much research remains to be done on such phenomena, especially from the diachronic standpoint, but I make room here for some preliminary observations. In the case of the rule concerning -ό-τερος vs. -ώ-τερος after long vs. short syllables, I suggest that it can be connected with the avoidance of a sequence of three short syllables in various archaic Greek meters. Such a metrical pattern of avoidance may be a reflex of a purely phonemic rule that was once operative in the Greek language—a rule that later becomes specialized as a morphophonemic rule. Such a posited specialization from a phonemic to a morphophonemic rule may be described as an alternative to other types of specialization that are purely morphemic.
II§30. Relevant to such patterns of specialization is a phenomenon known as Kuryłowicz’s “fourth law of analogy.” This “law” predicts one of several things that can happen when a new form B and an old form A come into conflict for possession of the same function. In terms of this “law,” the new form B will acquire the primary function while the old form A will be relegated to a secondary function. [24]
II§31. Clearly, such a “law” predicts only one of several options possible after the new form B and the old form A converge upon the same function. For example, the old form A may lose its status as a functional subordinate of the new form B and become completely disconnected from its old function, which can become exclusively owned by the new form B; or the old form A may be ousted altogether by the new form B, so that A does not even survive; or again, forms A and В may both survive and remain equivalent, with their complementary distribution becoming determined not by functional factors (as is the case in Kuryłowicz’s fourth law of analogy) but rather by the factor of morphophonemic complementarity. {A striking example of the last of the possibilities here listed is the Hittite treatment of the Indo-European proclitic / enclitic morphemes *-o- and * -i̯o which show functionally distinct reflexes in several Indo-European languages; in Hittite, however, the reflexes of the two are equivalent (both meaning ‘and’) but morphophonemically conditioned, with -a (< *-o) in final postconsonantal position and -ya (< * -i̯o) in final postvocalic position respectively. [25] }
II§32. Similar explanations, then, might be developed for morphophonemic rules of Greek. In one case, the historically extant phases themselves suggest the broad outlines of an explanation. The case in point is the morphophonemic conditioning of the preconsonantal augment in Modern Demotic Greek: the general rule is that the augment occurs only where the accent would fall upon it: so ἔ-γραψα ‘I wrote’ vs. γράψαμε ‘we wrote’. Since the past-tense endings of and by themselves are functionally sufficient for the Demotic Greek verb, the augment is superfluous. That is why, I propose, the augment is susceptible to morphophonemic redistribution. But here the historical background is also relevant: in the earliest attested phases of Greek, attachment of the augment in past tenses was still optional, {39|40} by virtue of its function as a segmentable syntactical connective. [26] The regularization of augment as obligatory marker of past tenses in classical Attic-Ionic can be considered to be only transitional, as we can see from the newly-segmentable augment of Demotic past tenses, albeit caused this time by morphophonemic rather than syntactic factors.
II§33. {Deserving of special mention are the general works about Greek morphology by Chantraine (1961) and Risch (1937). In some specific works, the chosen basis of inquiry is a given suffix, the attested distribution of which is then thoroughly investigated, leading to diachronic conclusions often extending in relevance even to Indo-European. Distinguished examples of this genre are Lejeune (1939) on -θεν, Holt (1940) on -σις, Redard (1949) on -της / -τις, Prévot (1935) on -θη-. Where the suffix in question is specifically derivational and not inflectional, many aspects of such studies may overlap with the factors of vocabulary (I will return to this point under the heading etymology / vocabulary). For a comprehensive treatise where this overlapping is continually illustrated, I cite Chantraine 1933 on the taxingly broad subject of Greek nominal derivation.}
II§34. {Specially to be noted for its cohesiveness in the simultaneous treatment of morphology, vocabulary, and syntax is Benveniste’s (1948) examination of nomina agentis / actionis and ordinals in Indo-European languages, with pivotal evidence adduced from Greek.}
II§35. {As for an exemplary treatment of suffixal formations that are inflectional rather than derivational, I cite Chantraine’s history of the Greek perfect (1926). For diachronic analysis of inflectional categories, the evidence of heteroclitics provides particularly valuable material, as demonstrated by Meillet 1926. [27] }
II§36. {Implications of a dichotomy between inflectional and derivational categories, as well as several other morphological typologies viewed principally from the standpoint of Indo-European linguistics, have been outlined by Kuryłowicz (1964), often with instructive references to Greek examples.}
II§37. For originality in conception and format, a book that stands out is Kastner’s investigation (1967) of simplex adjectives in -ος / -ον (instead of -oς / -η / -ον). In the compound adjectives of a dialect like Attic, the prevalent split from an originally unitary animate -ος into a binary masculine / feminine opposition -ος / -η has been repressed by an obligatory rule, which thereby preserves the archaism of an old animate / inanimate opposition -ος / -η inherited from Indo-European. In the simplex adjectives as distinct from the compound, by contrast, the new tripartite rearrangement into masculine / feminine / neuter = -oς / -η / -ον has prevailed; but the older arrangement -ος / -ον has been sporadically preserved for simplex adjectives as well, because of the archaism of certain inherited combinations of o-stem adjective + feminine substantive. It is these sporadic reflexes that Kastner {40|41} undertakes to examine in their attested collocations. His discovery procedure requires an intensive survey οf archaic expressions within archaic contexts, and this survey gives us a veritable panorama of archaisms in the social and cultural heritage of the Greek-speaking world. At the same time, this panorama is neatly delimited in scope by a single morphological factor.
II§38. Kastner’s indices are a revelation: in one of them (Die wichtigsten Wortfelder), for instance, we find a guide to the various semantic spheres in which the old combination of o-stem adjective + feminine substantive may be found. Among the rubrics of this index are: religion (sacrifice, hearth, festivals, oath, oracles), government and law, heredity, epichoric features, and so on. [28]
II§39. The novelty of Kastner’s approach, then, is that besides morphological archaisms themselves, the combinatory features of these archaisms are also taken into consideration; and this amounts to a survey of old contexts as well as old forms.
II§40. {A notable example of a straightforward study on evanescent vs. incipient morphological mechanisms is the treatment of the Greek comparatives by Seiler (1950). Schwyzer’s critical comparison of Greek -άζω with Gothic -atja (1937), on the other hand, is an ideal illustration of the procedural need to examine both the derivational distribution and the functional exponents of a Greek suffix in terms of Greek itself before any attempt is made to compare it with an apparent formal cognate from the Indo-European standpoint. A similar lesson may be derived from the detailed analysis by Cowgill (1964) of two difficult forms attested in the Cypriote Edalion Bronze.}
II§41. On the morphological as also on other linguistic levels, the Homeric corpus frequently preserves archaisms lost elsewhere in extant Greek. For example, such Homeric expressions as ἱερὸν μένος Ἀλκινόοιο or ἱερὴ ἲς Τηλεμάχοιο, which at first seem like stylistically-motivated periphrastic constructs, actually turn out to be reflexes from a prehistoric period when the so-called Caland’s rule was still operative. This rule is of Indo-European origin and essentially entails the following distribution: suffix *-i- for adjective-root when it is the first constituent of a compound, vs. suffix *-ro- replacing *-i- when the root forms a simplex adjective and is not in compound formation (Caland 1893 and Wackernagel 1897/1953). Thus *῾Ιερο-μενο- or *ἱερο-ϝι- would be violations of Caland’s rule, since ἱε-ρο- is the non-compound variant; on the other hand, the phonological reflexes of the morphologically predictable *isǝ i -(meno-) or *isǝ i -(ṷ i -) would disrupt any overt formal connection with non-compound ἱερός (that is, *ἱει- or the like could no longer be perceived as related to ἱερό-). Hence the circumvention resulting in ἱερὸν μένος or ἱερὴ ἴς + genitive. But the very occurrence of this circumvention indicates that Caland’s rule was still operative at the time when these expressions became embedded in the epic formulaic system. [29]
II§42. Homeric diction, in fact, is replete with morphological phenomena long since replaced by new counterparts in alphabetic Greek. For example, let us consider the phenomenon, well-attested in alphabetic Greek, of n-infix present-tense formations generated from the old thematic (“2nd”) aorist: examples include
II§43. In many instances, the new n-infix present has ousted the older variant altogether: hence μα-ν-θ-άνω from ἔ-μαθ-ον (zero-grade root *mn̥ th-), with only a trace of the older present (full-grade root *menth-) in the nominal derivative μενθήρη· φροντίς (Hesychius). Elsewhere, the old present survives, because it was appropriated by the formulaic system of Homeric diction: hence Homeric present πεύθομαι, aorist ἐ-πυθ-όμην -> “classical” present πυ-ν-θ-άνομαι. Significantly, forms of “classical” πυνθάνομαι occur only twice in the Homeric corpus, vs. 16 instances of “pre-classical” πεύθομαι.


II§44. In the area of comparative syntactical analysis, the Greek language has provided raw material for some new strides. For example, the technique of employing the analytical criterion marked / unmarked (from the functional standpoint) has been applied by Kiparsky (1968) to specific phenomena of Greek syntax (see also Jakobson 1932/1964 on the Russian verb), with explanations deriving these phenomena as reflexes of an Indo-European process that he calls conjunctional reduction. In effect, his main discussion concerns reflexes of the so-called injunctive in Greek and other Indo-European cognate languages; this injunctive essentially entails a neutralization of oppositions in tense and mood, wherein the verbal forms after the ensemble of first verb + conjunction revert to the unmarked correlate of the oppositions in tense and mood, namely to the injunctive. Especially interesting is Kiparsky’s suggestion [32] that from the Indo-European standpoint tense and mood were originally adverbial constituents in the deep structure [33] and that they were in complementary distribution with certain functionally related classes of adverbs. For example, in the prohibitive constructions of Greek, μή is combined with the aorist subjunctive—qua reflex of the injunctive (= zero) mood—because μή itself is “the realization of the mood constituent.” [34]
II§45. In Greek, the principle applied by Kiparsky to tense and mood can be applied to the verbal constituent of aspect as well. We read in a familiar Greek grammar this definition of the aorist: “The aorist expresses the mere occurrence of an action in the past. The action is regarded as an event or single fact without reference to the length of time it occupied.” [35] This statement by itself provides an adequate description of the aspectually unmarked past tense, as opposed to the marked past tense called the imperfective, but it is contradicted by another statement found further on in the book, where the claim is made that the aorist can express the marked imperfective aspect as well: under the rubric empiric aorist, we read: “With adverbs signifying often, always, sometimes, already, not yet, never, etc., the aorist {42|43} expressly denotes a fact of experience (ἐμπειρίᾱ).” [36] A plausible way of solving this contradiction is to equate (1) often + unmarked aorist with (2) the marked imperfective without an overt adverbial adjunct but with an underlying adverbial implication of often by virtue of the aspectual markedness of the imperfect.
II§46. As for the gnomic aorist in Greek (expressing a general truth without implication of tense), it is an old remnant from a prehistoric phase when the primary function of the aorist was not yet temporal but still aspectual (specifically, zero-aspect vs. marked imperfective aspect). By the time we reach the period of alphabetic Greek, however, a primary opposition in tense had developed, namely, present vs. past, and the basic aspectual dichotomy had become a subdivision within the temporal framework: imperfective present vs. imperfect / aorist past; or, in traditional terms: present vs. past, with past subdivided into imperfect (marked) vs. aorist (unmarked).
II§47. {Alternatively, it can be argued that the aorist indicative is marked for aspect, while the imperfect is unmarked; both are marked for tense.}
II§48. Kiparsky applies the principle of conjunctional reduction to the declension of nouns as well as to the conjugation of verbs. [37] Given that the vocative is a marked nominative (see Part III below), Kiparsky adduces several Vedic instances of vocative + conjunction followed by unmarked nominative. Since the conjunction -ca is enclitic, the actual Vedic realization of this conjunctional reduction is Voc Nom + -ca. As for Greek, the cognate of -ca is enclitic τε, and, significantly, the same ancient construct as in Vedic is attested in Greek. The attestation in Greek, however, is limited to the Homeric corpus, and there too we find it only once:
Ζεῦ πάτερ, ῎Ιδηθεν μεδέων, κύδιστε μέγιστε,
Ἠέλιός θ᾽, ὅς πάντ᾽ ἐφορᾷς καὶ πάντ᾽ ἐπακούεις

‘O Father Zeus, ruler of Ida, most renowned and greatest,
and O Helios (Sun), who oversees and hears all.’
Iliad III 276–277
II§49. That the type Ζεῦ πάτερ … Ἠελιός θ᾽ should still be extant at all is testimony to the preservative force of epic. As Wackernagel puts it, “So enthüllt sich in dieser minimalen Kleinigkeit die Macht der Gewohnheit und der Einfluss der Vererbung.” [38]
II§50. Aside from its preservation of such Indo-European archaisms as the one just discussed, Greek syntax is replete with phenomena appropriate for the synchronic study of linguistic universals. For example, let us consider Benveniste’s typological formulation of the 3rd person singular as the zero-person functionally. [39] Because of its function as zero-person, the base + 3rd singular ending is subject to formal reinterpretation as base + zero ending, with the old ending absorbed into the base, whence the constitution of new conjunctional paradigms on the now new base. That is, as Watkins points out, the productive endings of the other persons can now be added directly to what used to be base + 3rd person ending but which has become reinterpreted as pure base + zero ending. [40] Such a paradigmatic shift, based on a formal {43|44} resegmentation, is obvious in the paradigm of the verb ‘be’ in Polish (with the exemption of only the 3rd plural):
  singular plural
1st jestem jesteśmy
2nd jesteś jesteście
3rd jest
II§51. The same sort of paradigmatic shift is evidenced by Modern Demotic Greek, in the 2nd and 3rd singular of contract verbs like ρωτάεις, ρωτάει extant in Epirus, Central Greece, Ionic Islands, Peloponnesus. What we see here is the addition of the productive 2nd singular and 3rd singular endings -εις and -ει to the old 3rd singular form (ἐ)ρωτᾷ, still preserved elsewhere in Greece along with 2nd singular ρωτᾷς (the iota subscript in the preceding transcriptions is of course merely an orthographic archaism). [41]
II§52. Brief lists of noteworthy Einzelschriften on Greek syntax are readily available in a work by Löfstedt—despite his primary concern with Latin. [42] Singled out here as examples of syntactical treatises on Greek are the following: Denniston (1954), Burguière (1960), Guiraud (1962), Monteil (1963). Particularly illuminating insights into Greek syntax (as well as Latin and Germanic) are to be found in Wackernagel’s Vorlesungen über Syntax (1926, 1928). ⊛I predict that this collection of observations will remain useful for countless succeeding generations of linguistic scholarship.⊛ The book gives a precious collection of typologies, straightforward exposition, and intuitively appealing concepts. One example that stands out is Wackernagel’s discussion of the 3rd singular in terms of die unpersönlichste Form (with the adducing of forms like σαλπίζει ‘the trumpet is blown’). [43] We see here a forerunner of the definitive treatise on the 3rd singular by Benveniste (1946).
II§53. Aside from such synoptic works as Wackernagel’s, it is important to note that there are numerous trend-setting observations on syntax in such compendia as the Griechische Grammatik vol. 2 of Schwyzer and Debrunner (1950). What now follows is an illustrative selection of such observations, with commentary, from the first 150 pages of this volume.
II§54. Certain case-functions no longer current might still survive embedded in onomastic compounds, which are prone to special archaisms. One example is the genitive in Διόσδοτος ‘given from Zeus’, to be compared with the German plant-name Vergissmeinnicht ‘forget-me-not’ (with residual genitive mein) vs. the current syntagma vergiss mich nicht. [44]
II§55. Just as the French genitive de + noun springs from Latin + noun [in ablative], a construction restricted in classical Latin to just one area of functional overlapping with the genitive, namely the partitive, so also many of the Greek case-forms inherited from Indo-European might have originally borne a similarly restricted function, later subsumed into the broader functional categories of the evolving case-system. [45] In other words, just as French de + noun betrays a systematic removal of functional {44|45} constraints from the historical standpoint of Latin (inasmuch as the Latin equivalence of partitive noun [in genitive] with partitive + noun [in ablative] leads to the latter’s extension into the functions of possessive originally restricted to the former, and then to the ultimate displacement of the former by the latter), so also in Greek a case like the dative must have an Indo-European antecedent that is far more restricted functionally.
II§56. “Das Mundartstudium, das besonders fruchtbar ist, wenn der es betreibende Gelehrte selbst noch eine Mundart spricht, ist auch für die Syntax als Kontrolle und Anregung unentbehrlich.” [46]
II§57. A list of syntactical innovations in Greek, juxtaposed with those of other Indo-European languages, implicitly illustrates the principle of common innovation. [47]
II§58. Modern Greek compensates for loss of the dual by “die Anschauung der Paarigkeit” in nominal composition: e.g. ἀνδρόγυνον ‘married couple’, γυναικόπαιδα ‘wife and children’, etc. [48]
II§59. {Syntactically most versatile, from the Indo-European / Greek points of view, is the substantive: it may function as subject, object, attribute and apposition, predicate, adverb; the adjective is restricted to the functions of attribute, predicate, adverb; the verb serves as predicate. [49] }
II§60. In epic, the formal archaisms τοί / ταί (masculine / feminine plurals of ὁ ἡ τό etc.) are attested with the obligatory function of demonstratives: this is also the oldest function of ὁ ἡ τό etc. that we can reconstruct from Indo-European; on the other hand, the formal innovations οἱ αἱ (modeled on singular ὁ [ἁ̄ >] ἡ respectively) are only optionally demonstratives and are also attested in the function of definite article. [50] Since it is the latter function of ὁ ἡ τό etc. which ultimately prevails in classical Ionic, the pre-classical antecedent of which represents the last and most important major dialectal phase of epic, it is crucial that the very language of epic thus betrays, through its hierarchy of functional constraints, the functional progression of ὁ ἡ τό etc. from demonstrative to definite article in Ionic.
II§61. Likewise with τοῖο τά̄ων τοῖσι(ν): none of these has been retained in classical Ionic, and each is attested exclusively with the old demonstrative function in epic; by contrast, the corresponding forms of classical Ionic, τοῦ τῶν τοῖς, are attested in epic both as demonstrative and as definite article.
II§62. Nor are the constraints just functional; another sign of the archaism of τοῖο and τά̄ων is positional: τοῖο occurs almost exclusively in the first or fifth foot of the dactylic hexameter, while τά̄ων is in absolute verse-initial position everywhere in the Homeric corpus except XVI 833 (for more on the archaism of τά̄ων, see below under dialectology). No such positional restrictions hold for τοῦ and τῶν, which are the forms that have prevailed in classical Ionic, including of course Attic. [51]
II§63. It is such collocational evidence, by virtue of its cumulative impact, which has led to the following operational principle in formulaic analysis (as practiced e.g. by Ruijgh 1957 and Hoekstra 1965): the narrower the range of positional variation for any given word or phrase in the dactylic hexameter of epic, the greater the archaism involved.
II§64. The poetry of epic, of course, is not the sole repository for such archaisms; an equally potent preservative, mutatis mutandis, is the poetry of {45|46} “lyric.” Thus for example the poetry of Archilochos reveals a restriction of ὁ ἡ τό etc. to the pronominal usage. [52] Or again, in Aristophanes’ Thesmophoriazousai 100 ff, the absence of articles in Agathon’s lyrical outpourings is noticeable in contrast to their presence in the adjacent dialogue. [53] Legal formulae too provide an ideal context for petrified demonstrative usage of ὁ ἡ τό etc., as in the expression τῇ καὶ τῇ δὲ ἀτιμία (Plato Laws 4.721b). [54] Even ordinary Attic prose has sporadically preserved the demonstrative function, embedded in such phrases as ὁ μὲν … ὁ δὲ, τὸν καὶ τὸν, καὶ τὸν εἰπεῖν, etc.; a typological parallel is German dér Art, French de lá sorte. [55] In certain archaic expressions, precedent for the insertion of the functionally recent definite article has been consistently wanting: hence such inherited collocations as πόλιν καὶ οἰκίαν, παῖδες καὶ γυναῖκες, δεξιὰν διδόναι, etc.; [56] in some cases the collocation can be more precisely defined in terms of syntactical components: e.g. prepositional phrases like ἐπὶ θήραν, εἰς χεῖρας ἰέναι, etc. and possessive constructions like τέρμα τοῦ βίου (instead of τὸ τοῦ βίου τέρμα), περὶ φυλακῆς τῆς χώρας, etc. [57]
II§65. Finally, as an extension of Kiparsky’s already-mentioned theories on conjunctional reduction (1968), Ι cite the cancellation of the second article: a(b+c) instead of ab+ac; e.g. τὴν μὲν γῆν καἰ οἰκίας ἀφεῖναι, τῆς δὲ θαλάσσης καὶ πόλεως φυλακὴν ἔχειν (Thucydides 1.143.5). [58] It is significant that the marked exponent ac is the innovation and the unmarked c, the archaism. Likewise in e.g. Vedic, the unmarked injunctive (c) is formally older than any of the four moods (ac) from which it is conjunctionally reduced (from a synchronic point of view).
II§66. {Residually-attested suffixless cases such as αἰϝέν / αἰϝές are explained as reflexes of a caseless period in Indo-European. [59] }
II§67. In Attic, ὦ + vocative becomes the unmarked correlate of the plain vocative, which takes on the marked function of indicating “die Bedeutung des streng Sachlichen, Zurücksetzenden, Würdigen, Kalten, auch Unwilligen, Formlosen, Verächtlichen”; hence e.g. in de Corona, Demosthenes addresses his opponent exclusively as Αἰσχίνη, not ὦ Αἰσχίνη. [60] Or again, in Plato’s Symposium ὦ occurs 70 times with proper names and is missing only 8 times, while the Protagoras yields exclusively ὦ + proper name, ca. 100 times. [61]
II§68. The vocative is examined as a functionally restricted compartment of the nominative, from the diachronic point of view; thus a formal split between nominative / vocative may result from an earlier unitary nominative. Certain petrified expressions dating back to a period before this formal split may actually preserve the archaism of nominatives in vocative function and resist the imposition of a specific vocative form: hence e.g. the formulaically-preserved φίλος for φίλε, in φίλος ‖ ὦ Μενέλαε # (IV 189, etc.). [62]
II§69. Here is a most valuable statement on methodology in syntactical analysis: “Dem Sprachpraktiker, der in erster Linie dem Verständnis und der Übersetzung der Texte zu dienen hat, steht das Häufige und Eigentümliche im Vordergrund (so der Akkusativ des Objekts und der accusativus Graecus); dem Sprachhistoriker müssen {46|47} vielleicht Verwendungen zum Ausgangspunkt werden, die dem Praktiker nur Unregelmässigkeiten oder erratische Blocke sind (der Akkusativ der Richtung).” [63] Of course, the locus of diffusion (Ausgangspunkt) is often elusive as an operative mechanism, simply because the given grammatical category may no longer be productive, and such a condition leads to atrophy of old boundaries. Vestigial features in turn count as irregularities from the synchronic point of view, whence the difficulty in effecting an adequate diachronic perspective. Therefore it seems justified to modify the claim that “Jede Anordnung—und dies gilt nicht nur für die Kasuslehre—ist ein Kompromiss.” [64] The inevitable compromise may simply be the after-effect of jamming together the synchronic and diachronic perspectives. After all, Schwyzer himself succeeds in deriving the accusative of the object from the accusative of goal / direction. [65] Once this is diachronically achieved, a synchronic analysis may still justifiably be expected to induce new and different perspectives.
II§70. Certain transitive verb-formants which evolved within Greek, such as aorist in -σα-, aspirated perfect, and perfect in -κ-, may represent formal termini ante quem for genesis of the functionally objective accusative. [66]
II§71. On the syntagma known as σχῆμα καθ᾽ ὅλον καὶ μέρος, where a double-accusative construction specifies both person and part of body, e.g., τόν ῥ᾽ ᾽Οδυσεὺς … βάλε δουρὶ | κόρσην ‘Odysseus aimed and hit him with a spear on the temple’ (IV 501–502): “Die Entstehung des Schemas durch Zusammenziehung von zwei Sätzen ist im Musterbeispiel angedeutet.” [67] In essence, what has been contrasted here is “deep” vs. “surface” structure. [68]
II§72. {There are distinctions, in functional load, between various morphemes originating from Indo-European and ultimately belonging to the genitive case. [69] For example, *-ot / d had once been exclusively ablatival, not partitive; or again, *-om had once been exclusively partitive and possessive. [70] Such an accumulation of distinctions establishes in broad outline a relative chronology for the Entstehungsgeschichte of the genitive case as attested in alphabetic Greek.}
II§73. A functional opposition between genitive and ablative, despite their formal merger in Greek, can nonetheless be formally preserved—by combination. [71] Thus for example the formal contrast of comparative vs. superlative, combined with the formal genitive, makes overt an opposition between functional ablative and functional genitive respectively: σοφώτερος πάντων ‘wisest’ = ‘wiser than all’ (ablatival) vs. σοφώτατος πάντων ‘wisest’ = ‘wisest among all’ (genitival, specifically partitive).
II§74. That the genitivus auctoris of Greek was not originally ablative is readily demonstrable from other Indo-European languages: e.g. the type Διόσδοτος is paralleled in Sanskrit by the regular construction of genitive (indicating agent) + verbal adjective in -ta- ; since Sanskrit still preserves a formal distinction between genitive and ablative, its syntactical testimony is decisive. [72]
II§75. Despite formal collapse of the instrumental, locative, and dative in favor of the last in Greek, the alphabetic Greek dative never really became a syntactically integral entity: “vom echten Dativ und unter sich bleiben einzelne Anwendungen des Lokativs und {47|48} Instrumentals syntaktisch stets geschieden, und die Unterschiede werden auch wieder sichtbar durch den Präpositionsgebrauch (im allgemeinen ἐν u.a. bei lokativischem, σύν bei instrumentalem Dativ, Präpositionslosigkeit beim echten Dativ).” [73] Such lack of syntactical consolidation must have contributed to the formal loss of the dative in Modern Demotic Greek; even in Demotic, the instrumental function of the dative has gone on its own way, so to speak, from the formal point of view: “seinen eigenen Weg geht.” [74] Hence μετά / μέ + accusative, vs. εἰς + accusative for the functions of locative and genuine dative. Also to be consulted is Humbert (1930) on the loss of the dative in Greek.

Etymology and Vocabulary

II§76. A basic prerequisite of etymological studies in general, as Benveniste has pointed out, is simply “common sense”:
Mais, en matière de sens, on n’a pour guide qu’une certaine vraisemblance, fondée sur le ‘bon sens’, sur l’appréciation personnelle du linguiste, sur les parallèles qu’il peut citer. Le problème est toujours, à tous les niveaux de l’analyse, à l’intérieur d’une même langue ou aux différentes étapes d’une reconstruction comparative, de déterminer si et comment deux morphèmes formellement identiques ou comparables peuvent être identifiés par leur sens. [75]
II§77. But the “sens” of a linguistic form must be viewed in the entire ensemble of its distribution. One of Benveniste’s most striking illustrations involves the Greek word πόντος ‘sea’ and its formal cognates in other Indo-European languages: Latin pōns ‘bridge’, Armenian hun ‘ford’, Old Church Slavonic рǫtǐ and Old Prussian pintis ‘path’, Sanskrit pánthāḥ, and Avestan pantå ‘path’. The problem is to bridge the semantic gulf between e.g. Greek πόντος and Latin pōns. Benveniste maintains that the key to the solution is to discover which, if any, of the cognates preserves the primary meaning, the least common denominator. [76] The secondary meanings of the other cognates could then be motivated as divergences from (or modifications of) the primary meaning. After arguing that the semantic spheres of e.g. hun ‘ford’, πόντος ‘sea’, and pōns ‘bridge’ must be secondary because they are mutually irreconcilable, Benveniste shows that the semantic common denominator survives in Indo-Iranian, most clearly seen in the Vedic usages of pánthāḥ, commonly glossed as ‘path’, ‘chemin’:
Ce qui caractérise le pánthāḥ est qu’il n’est pas simplement le chemin en tant qu’espace à parcourir d’un point à un autre. Il implique peine, incertitude et danger, il a des détours imprévus, il peut varier avec celui qui le parcourt, et d’ailleurs il n’est pas seulement terrestre, les oiseaux ont le leur, les fleuves aussi. Le pánthāḥ n’est donc pas tracé à l’avance ni foulé régulièrement. C’est bien plutôt un ‘franchissement’ tenté à travers une région inconnue et souvent hostile, une voie ouverte par les dieux à la ruée des eaux, une traversée d’obstacles naturels, ou la route qu’inventent les oiseaux dans l’espace, somme toute un chemin dans une région interdite au passage normal, un moyen de parcourir une étendue périlleuse ou accidentée. {48|49} L’équivalent le plus approché sera plutôt ‘franchissement’ que ‘chemin’, et c’est bien ce sens qui explique la diversité des variantes attestés. [77]
II§78. Thus it is from the basic notion of’ ‘chemin’ that the ultimate context of πόντος has developed, and it is this same semantic sphere which gave rise to such epic expressions as ὑγρὰ κέλευθα ‘watery pathways’. [78] As for the still more basic notion of ‘franchissement’, it is still preserved in the compound Ἑλλήσ-ποντος. [79]
II§79. The notion of ‘une étendue périlleuse ou accidentée’ is still latent in Homeric collocations of πόντος with the harmless-looking epithet ἰχθυόεις ‘swarming with fish’:
πόντον ἐπ’ ἰχθυόεντα φέρεν βαρέα στενάχοντα
‘[The squall] carried him, heavily groaning, over the ikhthuoeis pontos.’
Odyssey iv 516, xxiii 317; cf. also v 430
ἠμὲν ὅσ’ ἐν πόντφ πάθετ᾽ ἄλγεα ἰχουόεντι
‘and how much suffering you underwent in the ikhthuoeis pοntos.’
Odyssey x 458
II§80. (There are other collocations of πόντος with πάσχειν, perhaps likewise relevant to pánthāḥ from the comparative point of view: Odyssey i 4, ii 370, v 377.) The original selection of ἰχθυόεις was probably motivated not by a striving for fanciful descriptions of the sea, but rather, by the implication of dangers lurking underneath the ship:
ἢ τόν γ’ ἐν πόντῳ φάγον ἰχθύες …
‘or the fish devoured him in the pοntos
Odyssey xiv 135
ἠέ που ἐν πόντῳ φάγον ἰχθύες …
‘or perhaps the fish devoured him in the pontos’.
Odyssey xxiv 291
II§81. Of course it can happen that in a given set of cognates, the least common denominator of the semantic sphere is no longer extant in any of the Indo-European languages with relevant lexical evidence. For example, despite the formal correspondence between Greek δίκη and Sanskrit diśā, there is a functional anomaly between the two, in that neither can be motivated semantically in terms of the other: the Sanskrit word means ‘direction, celestial district’ vs. the basic notion ‘rule, ruling’ inherent in the Greek word. The ā-stem in both is irrelevant to this anomaly: it is just a formal renovation of the root-stem *dik-, and the latter is actually attested in Sanskrit diś-, with the same meaning as that of diśā; so also in Latin dic-is (causa) ‘(for the sake of) judicial form’, which has basically the same meaning as that of δίκη. But we may seek a more general functional relevance in the morphology—specifically, in the expansion of the root-stem by ā-stem here. It is significant that the ā-stem is an inherited formant of deverbative nomina actionis. Thus the nominal root-stem *dik- was also a nomen actionis until its replacement by *dik-ā-. And the founding verb is still reflected in Greek: it is δείκνυμι ‘designate’. Then, from the original root-stem nomen actionis *dik- ‘designation’, the specific notions ‘direction, celestial district’ (diś- / diśā-) and ‘rule, ruling’ (δίκη) can evolve. [80] And since there is no indication that a nomen actionis with the configuration *dik- was not already inherited from Indo-European, we may by extension call diśā and δίκη cognates. {49|50}
II§82. Often, however, the etymology given as the least common denominator of a word’s meaning may seem like nothing more than a conjecture, if the only criterion is the formal pairing of a given Greek word with an apparent cognate from some other Indo-European language. A case in point is the matching of Greek θέλγω ‘enchant’ with Lithuanian žvelgiù ‘look’. Here we might reconstruct the prototypical meaning of these two words, from the standpoint of Indo-European linguistics, as ‘enchant by looking at’. [81] Such a reconstructed meaning is fitting as a semantic least common denominator, in that the current meaning of Lithuanian žvelgiù could then be explained as involving the loss of an otherworldly connotation and the retention of a sensory connotation; as for Greek θέλγω, conversely, there would be a loss of the sensory and a retention of the otherworldly. Such an explanation is resisted by some (I note the doubts expressed by Frisk about the interpretation ‘Bezauberung durch den bösen Blick’). [82] But there is corroborating evidence available from the language of Homeric diction.
II§83. The formulaic system of Homeric poetry has insured the preservation of grammatical patterns stemming from such varied diachronic phases as to span about a millennium. In Homeric poetry, the following phrases are relevant to the etymology of θέλγω:
          ‖ ὄμματα θέλγει #
Iliad XXIV 343; Odyssey v 47, xxiv 3
# θέλξας ὄσσε φαεινά ‖
Iliad XIII 435
II§84. I have indicated here the metrical positions of the relevant phrases (‖ = ‘caesura’, # = ‘absolute verse-initial / final position’). In terms of their placement within the verse, both phrases belong to the formulaic system of Homeric diction. The phrase ὄμματα θέλγει occurs exclusively in the metrical space – ⏑  – situated between the bucolic diaeresis (‖) and absolute verse-final position (#). This space accommodates the so-called Adonic sequence, featuring a metrical pattern – ⏑  – that matches the pattern we find in the celebrated phrase of Sappho, ὦ τὸν Ἄδωνιν. As for the phrase ὄσσε φαεινά, it occurs in the metrical space – ⏑  – situated between the end of the first foot and the trochaic caesura (‖). As a rule, this space – ⏑  – allows interchange of its phraseology with phraseology in the space – ⏑  – situated between the bucolic diaeresis (‖) and absolute verse-final position (#). In fact, just such an interchange is attested for the phrase # (θέλξας) ὄσσε φαεινά ‖. A case in point is ‖ ὄσσε φαεινώ # in XIII 3, etc. (For the variation of innovative φαεινά vs. archaizing φαεινώ, I compare the use of ἧμιν before the trochaic caesura vs. ἡμῖν in absolute verse-final position, as discussed below under dialectology.) Given these formulaic connections between the word θέλγω and words for ‘eye(s)’, I argue that the system of traditional epic diction preserved the idea of ‘look’ originally inherent in the meaning of θέλγω.
II§85. At times the collocational patterns of a given word in epic may suggest an etymological connection with another Greek word, even without the additional aid of any comparative Indo-European evidence as in the case of θέλγω. An example is ἥρως. From the internal evidence of Greek, it is possible to compare the feminine proper names Πατρώ Μητρώ ‘Ηρώ with the masculine substantives πάτρως μήτρως ἥρως. [83] But beyond this point it is difficult to make further morphological {50|51} generalizations. In epic meter, the archaism of ἥρως is apparent from the highly restricted positional range of e.g. its nominative. Although the dactylic hexameter could have theoretically allowed eleven positions for this form, ἥρως is actually found in only three positions: (1) absolute verse-initial, (2) absolute verse-final, (3) paired with the preceding word γέρων, after the trochaic caesura. In one of these positions (2), there is an interesting precedent for substitution: whereas the formula αὐτὰρ ὅ γ᾽ ἥρως occurs after the bucolic diaeresis 7 times in the Homeric corpus, there is also one instance of αὐτὰρ ὄ γ᾽ Ἥρην (XXI 367) attested in the same metrical position. [84] According to one explanation, this formulaic interchange between Ἥρην and ἥρως was motivated by “l’association des sons.” [85] I argue that there is more to it. We may consider another instance of such substitution in absolute verse-final position: ἤγαγεν Ἥρη in V 731 vs. ἄγεν ἥρως in X 179; here too the explanation of “l’association des sons” could be invoked, but now we will see other instances of this association of ἥρως and ῞Ηρη that must have resulted from a deeper motivation. For example, the absolute verse-final ἤλυθεν ἥρως of iii 415 is matched by the common formula ἤλυθεν Ἠώς ‘dawn came’ of x 541, etc. To explain this match as a mere “sound-association” is an oversimplification, since there are attested further matchings that have nothing to do with sound-association, such as ἤλυθε μήτηρ ‘the mother came’ (VI 251) and ἤλυθεν ὄρνις ‘the bird came’ (VIII 251, xx 242) in the same metrical position. It is essential to note that there is a latent contextual link connecting these words Ἥρη, Ἠώς, μήτηρ, and ὄρνις:
like Ἥρη, Ἠώς is a goddess
V 721, etc.: # Ἥρη, πρέσβα θεά ‖
II 48: # Ἠώς μεν ῥα θεά ‖
Hymn to Aphrodite 223, 230: ‖ πότνια ‘Ηώς #
iv 513, etc.: ‖ πότνια Ηρη #
μήτηρ is a regular title of goddesses in verse-final position:
I 357, etc.: ‖ πότνια μήτηρ #
a standard epiphany of goddesses is in the form of an ὄρνις ‘bird’: [86]
… ἀπέβη γλαυκῶπις Ἀθήνη,
ὄρνις δ᾽ ὣς ἀνοπαῖα διέπτατο. τῷ δ᾽ ἐνὶ θυμῷ
θῆκε μένος καὶ θάρσος, ὑπέμνησέν τέ ἑ πατρὸς
μᾶλλον ἔτ᾽ ἢ τὸ πάροιθεν.

‘… owl-vision Athena went away
and like a bird she flew up, and into his thumos
she put strength and daring, and she reminded him of his father
even more than before’.
Odyssey 319–322
II§86. In the passage from the Odyssey, the attributes of a hero are being conferred on Telemakhos by the goddess appearing as a bird. So also in the Iliad, Hera and Athena appear in a joint epiphany as birds for the sake of helping the Achaeans:
αἱ δὲ βάτην τρήρωσι πελειάσιν ἴθμαθ᾽ ὁμοῖαι,
ἀνδράσιν Ἀργείοισιν ἀλεξέμεναι μεμαυῖαι.

‘the two of them went, like fluttering doves,
eager to protect the Argive men.’
Iliad V 778–779
II§87. Thus even contextually as well as formulaically, the ἥρως is correlated with goddesses. And the fact that ῞Ηρη as the mother-goddess par excellence (even Athena is her surrogate: e.g. I 194–195), is included in these correlations with ἥρως now takes on a {51|52} formal significance, which is this: the language of epic betrays traces of an early period when the masculine configuration *hērōs was still synchronically motivated by a feminine *hērā. There may even be traces of stylistic juxtaposition, we see in these verses:
ἡρώων οἶσίν τε κοτέσσεται ὀβριμοπάτρη
Ἥρη δὲ μάστιγι θοῶς ἐπεμαίετ᾽ ἄρ᾽ ἴππους.
Iliad V 747–748 = VIII 391–392
II§88. It is essential to note that the Hellenic ἥρως ‘hero’ par excellence is none other than Ἡρακλῆς or Herakles. If there is in fact a connection between ἥρως ‘hero’ and Ἥρη ‘Hera’ in the language of epic, then we see here a new corroboration of the theories of Nilsson (1921) about the evolution of the Mycenaean mother-goddess in general and of Athena in particular. On the basis of non-linguistic evidence, Nilsson posited that the Minoan mother-goddess had been a guardian-deity residing in the king’s palace; but then, with the advent of Mycenaean hegemony and the concomitant displacement of Minoan institutions, the goddess took to the field, as it were, in support of the warlike Mycenaean princes now in her charge: “Sie ist die Beschützerin und persönliche Helferin der Helden.” [87] While the god Zeus is the patron of a king like Agamemnon from a legal or even moral point of view, a goddess like Athena is a hero’s patroness from a personal point of view. The hero is the protégé of the goddess, and this relationship can be inherited from one hero to another, as in the case of Telemakhos, who inherits the relationship of Odysseus to the goddess Athena. In the light of the additional evidence provided by the language of epic, we can now see that the status of the *hērōs as a personal protégé of his divine patroness *Hērā is grammatical as well as contextual.
II§89. In view of these examples of establishing etymologies with the help of collocational evidence from archaic Greek poetry, it is important to add that collocational evidence from the archaic poetry of other Indo-European languages also occasionally helps settle the etymology of a Greek word. For instance, the derivation of νέκταρ ‘nectar’, a most ancient word owing its preservation in alphabetic Greek perhaps solely to its transmission in the language of epic, can be resolved only with reference to Vedic poetry. The central etymological problem in the word νέκταρ is the semantic connection between its components. There is little difficulty with the initial νεκ-: it is what Benveniste used to call “theme II” of the root, *h2nek- ‘death’, also seen in e.g. νεκ-ρός, νέκ-υς, νέκ-ες (·νεκροί: Hesychius), Latin nex, ē-nec-tus, nοxa, noceō, etc.; Hittite ḫenk-an ‘pestilence, death’ is an example of radical “theme I” *h2enk-. [88] As for the final segment -ταρ, however, Benveniste’s explanation in terms of a suffixal formation leads to semantic problems, since the Homeric contexts of νέκταρ / ἀμβροσία are associated not with death but with the negation of death. [89] Thieme proposed a solution to the problem: that -ταρ is not a suffix but rather the second constituent of a compound, from the root *tr̥h2- as seen in the Sanskrit verb tárati ‘overcome’: in terms of this proposal, the two components of this word would be reconstructed as *h2nek- and *trh2-, with prevocalic external sandhi-generalization of the zero-grade *trh2- into -ταρ. [90] The ideal corroboration of this proposed etymology would be the Indic attestation of a syntagma involving ‘death’ + *trh2- corresponding to νέκταρ, that is, corresponding to the Greek attestation of a compound originally motivated by this syntagma. Thieme could find no such combination in the Rig-Veda. [91] But his efforts were not in vain. Schmitt succeeded {52|53} in finding the combination in the Atharva-Veda. [92] In the refrain of a song of praise to the odaná-, the ‘rice-mess’ of the Brahmans, we read:
ténaudanénā́ti tarāṇi mṛtyúm
‘by that rice-mess let me overcome death’
Atharva-Veda 4.35.1d-6d
Likewise elsewhere in the same hymn:
yénā́taran bhūtakṛ́tó ‘ti mṛtyúm
‘by which [rice-mess] the being-makers overcame death’
Atharva-Veda 4.35.2a
Likewise in another source:
vināśéna mṛtyúm tīrtvā́ sámbhūtyāmṛtam aśnute
‘after having crossed death by destruction, he reaches immortality by becoming …’ [93]
Vājasaneyi-Saṃhitā 40.14 / Īsopaniṣad 14
The Homeric word νέκταρ, then, is a faint vestige of a whole nexus of related ritualistic terminology stemming from the indogermanische Dichtersprache.
II§90. There are instances where an etymological solution is achieved without direct use of the comparative method but rather with internal analysis of the relevant Greek morphology and syntax. In the case of the Attic-Ionic particle ἄν, attempts to connect it with Latin an and Gothic an have proved unsuccessful, simply because the formal plausibility of this connection is not matched by the functional. Neither the Latin nor the Gothic an is used as a potential particle. The problem is described well by Palmer (1963):
This identification implies that an Indo-European an of unknown function persisted through the common proto-Greek period and was preserved solely by the speakers of Attic-Ionic and Arcadian (not Cypriot!), who employed it to differentiate the prospective subjunctive and the potential optative, whereas the other groups of Greeks, though making essentially the same syntactical differentiation (this is the essential common feature of all the Greek dialects), used another word. Such procedure violates the first law of etymology, which has been phrased ‘Look for Latin etymologies first on the Tiber’. [94]
Attic-Ionic ἄν, then, must be compared with its equivalents in the other dialects: Arcadian (κ)αν, Doric κα, Aeolic and Cypriote κε(ν). Το repeat, all these particles are syntactically equivalent to each other.
II§91. Now the zero-grade of a full-grade κεν would be *kn̥ > κα (in preconsonantal position) or καν (in prevocalic position). Then we can set up a proportion:
καν is to κα as κεν is to new κε. [95]
As for Doric κᾱ (vs. κα), it can be explained as a metrically-conditioned variant. [96] At this point, only ἄν remains to be motivated. The solution of Forbes (1958) is that ἄν is a new positive to a negative οὐκ ἄν. [97] The etymologically false division of οὐκαν as οὐκ ἄν instead of οὐ καν must have been triggered by the morphophonemic alternation of prevocalic οὐκ vs. preconsonantal οὐ. The implications go further:
It is remarkably interesting, if this solution is correct, that ἄν in Homer is found most frequently in the phrase οὐκ ἄν, commonly thought to be an Ionicism from Aeolic oὐ κεν. {53|54} Ιt is hardly to be wondered at that *οὐ κεν was not interpreted as *οὐκ ἐν where *ἐν would be homonymous with the preposition. Likewise οὐ κα would scarcely give rise to *οὐκ ἀ because there were no Greek words apart from certain forms of the article and the verb ‘to be’, which belonged to clear semantic groups, consisting only of a single vowel, except for the interjection ὦ (where also the vowel is long). [98]
II§92. In sum, internal analysis forces us to connect Attic-Ionic ἄν not with Latin and Gothic an, but with e.g. Sanskrit ká m and Hittite kan, however unlikely this equation seems at first, on the surface. Final corroboration comes from such collocational matchings as Homeric νύ κεν vs. Sanskrit nú kám vs. Hittite nu-kan. [99]
II§93. As a conclusion to the discussion of trends in Greek etymology, it seems appropriate to cite some perceptive comments from the preface to Chantraine’s dictionary (1968).
II§94. On the reasons for the last three words in the title, Dictionnaire étymologique de la langue grecque, histoire des mots (modeled on the title of the famous Latin etymological dictionary by Ernout and Meillet, 1959): “… l’étymologie devrait être l’histoire complète du vocabulaire, reflet de l’histoire tout court, que je me suis donné le plus de peine.” [100]
II§95. On the application of structuralist methodology: “Pour qu’une étymologie soit irréfutable, il est nécessaire d’une part que la structure du mot envisagé s’insère de manière évidente dans le système des alternances et de la morphologie indo-européenne, de l’autre que l’on trouve des correspondants nets dans plusieurs langues indo-européennes bien attestées.” [101]
II§96. On the controversial theories of a pre-Hellenic substratum sometimes designated as “Pelasgian”: “Le pélasgique est pour l’instant une vue de l’esprit et son cas diffère essentiellement de celui de l’indo-européen. L’indo-européen n’est pas attesté, mais c’est un système cohérent défini par les lois rigoureuses. Ce n’est pas le cas du pélasgique et cela ne le sera peut-être jamais.” [102] {For an analysis of the “Pelasgian” controversy: Hester 1965.}
II§97. On words stemming from the pre-Hellenic non-Indo-European substratum: “Il faut toutefois prendre garde que l’hypothèse de l’emprunt à une langue inconnue est une solution paresseuse et qu’il faut tâcher de tirer parti du témoignage des langues plus ou moins mal connues qui bordent les rives de la Méditerranée.” [103] {For exemplars in methodology: Heubeck 1961 and Neumann 1961.}
II§98. On the etymology of words traceable to Indo-European: “elle a donné naissance à une bibliographie accablante: examiner les multiples hypothèses qui sont venues à l’idée de savants d’ailleurs honorables et bien informés, c’est parcourir le plus souvent, comme on l’a dit, un cimetière d’enfants mort-nés.” [104]
II§99. In the study of vocabulary, there has been some movement in the direction of internal investigation of word-formations and away from an earlier emphasis on Indo-European origins. {Representative of works about vocabulary which rely mainly on internal analysis are those of Fournier (1946), Laroche (1949), Trümpy (1950), Redard (1953), Chantraine (1956), van Brock (1959), Bader (1965), Corlu (1966), Casabona (1967), {54|55} Latacz (1967).}
II§100. Of course the research still to be done on Greek vocabulary from the Indo-European standpoint is boundless. Benveniste’s “Don et échange dans le vocabulaire indo-européen” (1951/1966) illustrates how rewarding such an approach can be. For example, the apparent semantic clash between Greek δω- ‘give’ and Hittite - ‘take’ does not invalidate the certainty that they are formal cognates, simply because both can be explained as verbal manifestations of a social institution called reciprocity, well-documented in anthropological fieldwork and in this instance deriving from an original Indo-European setting: the root *- actually means ‘hold in order to engage in a social transaction’, and its reflexes in Hittite and Greek merely show the generalization of one transactional option or the other, give or take. [105] Besides several other illustrations from Greek showing reflexes of this phenomenon of reciprocity (e.g. νέμω, δαπάνη, ἀλφάνω, etc.), Benveniste’s repertory of examples could even be expanded further: worth studying in this context, for example, is the usage of ὠνεῖν, which means ‘buy’ in most Greek dialects, in the sense of πωλεῖν ‘sell’ in the Cretan dialect of the city of Gortyn (cf. Willetts 1965).
II§101. Social implications derivative from the Indo-European perspective are invaluable in the study of such Greek tabu-constructs as λαγώς ‘hare’ < λαγ-ωός, diachronically traceable to the original meaning ‘floppy-ears’. [106] For a valuable diachronic study of a single Greek word, I single out the investigation of ᾿Ηλύσιον / Ēlusion (= Elysium) by Burkert (1960–61).
II§102. One of the most arduous tasks in the study of Greek vocabulary is to confront the elusive problem of early—even prehistoric—borrowings. There is an exemplary study by Émilia Masson (1967) on the oldest phases of Greek borrowings from Semitic languages. As for borrowings from the neighboring Anatolian languages, an important advance has been Benveniste’s establishing the ultimate provenience of ὄβρυζα ‘crucible’ from Hurrian-Hittite h̬ubrušh̬i (DUG h̬u-u-ub-ru-uš-h̬i) ‘vase de terre, terrine’. [107] Such links, aside from their linguistic value, are of profound significance from the viewpoint of cultural history as well. Accordingly, when factors revealing an ultimate Anatolian derivation appear in Greek epic itself, their import is all the more to be emphasized.
II§103. In this light, I draw special attention to an article of van Brock (1959) on Hittite ritual-substitution and on its relevance to the Iliad. The Hittite ritual in question is meant to transfer a miasma from the king to a substitute, preferably to one that is alive: le transfert du mal. The Hittite word for this substitute is tarpalli-. As for the conditioning on the selection of the surrogate, it is extraneously motivated, whence the range (or perhaps evolution) of the victim from relative to friend to stranger to criminal to animal to some symbolic object. But the basic principle is that the closer the surrogate is to the king, the more pleasing the sacrifice becomes to the gods. “Le tarpalli- est un autre soi-même, une projection de l’individu sur laquelle sont transférées par la magie du verbe toutes les souillures dont on veut se débarrasser.” [108] Now there are several by-forms of tarpalli-, among them tarpašša- and tarpanalli-; and van Brock has convincingly connected the forms tarpassa- / tarpan- as the ultimate sources {55|56} for the Greek forms θέραψ / θεράπων. [109]
II§104. ◊Janko (1992) resists the connection of Hittite tarpan- and Greek θεράπων, saying that he doubts that these two forms are “cognates.” [110] But the argument here, as is evident from what follows, is not that Greek θεράπων is a “cognate” of Hittite tarpan-; what is being argued, rather, is that the Greek word is a borrowing from the Hittite language.◊
II§105. Most relevant is the fact that the θεράπων par excellence in the Iliad is the hero Patroklos, who is killed wearing the armor of Achilles himself.
ὄφρ᾽ ἠὺς θεράπων Πηληϊάδεω Ἀχιλῆος
‘until the good therapōn of Achilles, son of Peleus’
(context: Zeus ponders the death of Patroklos = the therapōn)
Iliad XVI 653
τοίου γὰρ θεράπων πέφατ᾽ ἀνέρος, ὃς μέγ᾽ ἄριστος
‘killed was the therapōn of such a man who is by far the best’
(context: the Trojans ponder what to do with the corpse of Patroklos)
Iliad XVII 164
οὐδέ κε Πάτροκλόν περ ἐϋκνήμιδες Ἀχαιοὶ
ἐκ βελέων ἐρύσαντο νέκυν, θεράποντ᾽ Ἀχιλῆος

‘nor could they, well-greaved Achaeans though they were,
save from the missiles the corpse, the therapōn of Achilles’
Iliad XVIII 151–152
II§106. From the internal evidence of the Iliad itself and with an approach completely different from that of van Brock, Whitman (1958) has also noticed the same sort of surrogate-motif in the rôle of Patroklos, “who represents the human side of Achilles”; [111] “Achilles is indispensable, but Patroclus is dead.” [112] At the climactic moments of the onslaught of Patroklos who is taking the place of Achilles, even the epithet of Patroklos is switched from just the patronymic (Μενοιτιάδης) or qualifier (ἱππεύς ‘charioteer’) to the title δαίμονι ἶσος ‘equal to a god’ (XVI 705), even at the moment when Patroklos confronts Apollo himself (XVI 786), just before the god destroys him. Significantly, the epithet δαίμονι ἶσος is later applied to Achilles too (XX 447), whose onslaught is mirrored by the earlier onslaught of Patroklos. We can see in this mirroring a reference to an identity in rôles. [113] Whitman explains: [114]
There can be little doubt that the change in Patroclus’ character and characteristic epithets is not due simply to his presence in a battle scene. A kind of double image, as in surrealistic painting, is involved. Patroclus is playing the role of Achilles. For the moment, he has become Achilles, and acts much more like the great hero than like himself. When Achilles prays to Zeus for Patroclus’ safety, he seems to ask, indirectly, whether his friend can play his role adequately or not:
… Give him glory, far-sighted Zeus,
Strengthen the heart in his breast, even that Hector
May learn whether this companion of ours
Knows how to wage the war, or if only his hands
Rage resistless, when I myself go to the moil of Ares.
Iliad XVI 241–245
The actual Greek word for ‘companion’ in Whitman’s translation of XVI 243 is θεράπων.
II§107. ◊In my original analysis, I considered at this point whether the provenience of the god Apollo is Anatolian. [115] In Nagy 1994, I reconsider the etymology of the name of Apollo.◊ {56|57}
II§108. Whitman continues: [116]
Nowhere else in the Iliad [except in the slaying of Patroklos by Apollo] does a god directly, with his own hand, overcome a hero. The passage where Apollo approaches Patroclus like a mist is one of the most unholy terror, a blinding vision of the identity of glory and death. But it belongs to Achilles more than to the man who has only for the moment assumed the tragic mask.…the bitterest poignancy in Achilles’ tragedy lies not in his own death, but in that of the friend who was so far a part of himself that he played his mortality for him. Achilles accepted Patroclus as his proxy as a means of being at the same time above all other men and yet one of them, and this was, of course, impossible and incompatible with life.
I bring to an end here my analysis of the Anatolian provenience of the word θεράπων, as corroborated by the linguistic evidence of Homeric poetry. ◊In later work (especially Nagy 1983), I have investigated the Antatolian provenience of other words attested in Homeric poetry.◊
II§109. There are also traces of actual Greek-Anatolian contacts to be found in Homeric poetry. A case in point is the following passage from the Odyssey:
τοῖος ἐὼν οἷός ποτ᾽ ἐϋκτιμένῃ, ἐνὶ Λέσβῳ
ἐξ ἔριδος Φιλομηλεΐδῃ ἐπάλαισεν ἀναστάς,
κὰδ᾽ δ᾽ ἔβαλε κρατερῶς, κεχάροντο δὲ πάντες Ἀχαιοί.

‘being such a man as the one who [i.e. Odysseus], in well-founded Lesbos,
in rivalry stood up and wrestled Philomeleides
and threw him down mightily, and all the Achaeans were glad.᾽
Odyssey iv 342–344 = xvii 133–135
II§110. Let us focus here on the association of the island of Lesbos with the figure of Philomeleides. In a Hittite document, we find the place-name Lazpaš mentioned in a context involving a figure named Piyamaraduš, who is described as a political renegade in that context. [117] The pairing of Lazpaš / Piyamaraduš in the Hittite document may be distantly related to the pairing of Λέσβος / Φιλομηλεΐδης in the Homeric passage I just quoted. Even if this match is just a coincidence, the point is still worth making that any comparison of Hittite and Greek forms must be backed up by comparing the contexts in which these forms are embedded. Another related point is worth making here: even if the forms Piyamaraduš and Φιλομηλεΐδης are in fact related, it does not necessarily follow that the Hittite name Piyamaraduš is some sort of deformation of a genuine Greek construct: the Anatolian name Piyamaraduš could have been deformed into morphologically recognizable elements if it was taken over into Greek. [118]
II§111. In the study of Greek vocabulary, morphology and syntax are often contributing factors, as we see in an article of Benveniste (1964): basing his observations especially on Greek designations for eating and drinking, he offers valuable typological insights into the interaction between lexical and suffixal factors. In the same work, {57|58} Benveniste notes a disequilibrium in the nominal derivation with -τυς from the root for ‘eat’ *ed- (ἐδητύς, with normalized base-enlargement) vs. the nominal derivation with -σις from the root for ‘drink’ *pō̆- (πόσις). [119] Such disequilibrium in these nomina actionis is not of Indo-European origin, however: the suffixes -τυς and -σις had simply shifted to other bases meaning ‘drink’ and ‘eat’ respectively.
II§112. There is a broader issue here: a functional distinction between the actual derivational suffixes -τυς and -σις is indeed of Indo-European origin, as demonstrated by Benveniste himself in Noms d’agent et noms d’action en indo-européen (1948). Besides the exposition of nomina actionis as well as nomina agentis, he also includes a treatment of the comparative, superlative, and the ordinal from the Indo-European standpoint—but with a relevance to Greek syntax and vocabulary. [120]
II§113. {Here I mentioned the supplement to the ninth edition of the Greek-English Lexicon of Liddell and Scott (Barber et al. 1968); also the supplements of Renehan (1968); also the Lexilogus zu Homer by Bechtel (1914). As for the onomastic aspect of Greek, I mentioned the work of Bechtel on historical personal names (1917), of Strömberg on plant-names (1940), and of Risch on ethnika (1957).}
II§114. Under the rubric of lexicography we must also include the valuable research surviving from the ancient world; foremost is the Alexandrian lexicographical tradition, the prime exponent of which is a cumulative dictionary attributed to Hesychius.
II§115. The reliability of this Alexandrian dictionary, in however abbreviated form it has been transmitted, had been convincingly defended. [121] Even from the standpoint of Indo-European linguistics, Hesychius is sometimes the only source for a Greek word corresponding to cognates well attested in other Indo-European languages. For example, let us consider these glosses in Hesychius: ἔορ· θυγάτηρ, ἀνεψιός (‘daughter, cousin’), ἔορες· προσήκοντες, συγγενεῖς (‘those akin, relatives’). We see here in the form ἔορ the Greek counterpart to Latin soror ‘sister’ < *sṷe-sor, literally ‘female belonging to self’. Especially in view of the etymology involving *sṷe- ‘self’, the Hesychian ἔορ corresponding to soror proves to be a most significant lexicographical contribution. Nowhere else is the formal cognate of soror to be found in attested Greek, and this instance of hapax-attestation as well as numerous others corroborates all the more the lexicographical value of Hesychius. [122]


II§116. ⊛In the analysis that follows, I rely especially on the criterion of shared innovation for the purpose of establishing the affinities of the attested ancient Greek dialects. Such a criterion is I think of prime interest for those who are interested in developing more effective methodologies for analyzing these dialects.⊛ [123]
II§117. ⊛Viewing the dialectal situation by starting from the earliest attested times and proceeding to later times, we can see three kinds of evidence:
  1. the attested texts of the Linear В tablets in the second half of the second millennium BCE. As we have seen, the Greek language as written in the Linear B script is conventionally called Mycenaean. {58|59}
  2. the attested texts of the first millennium BCE, written in distinct dialects. These dialects show distinctions that can be reconstructed as far back as the second millennium BCE. In terms of these distinctions, the prototypical dialects of the second millennium BCE are Arcado-Cypriote, Aeolic, Ionic (or, more broadly, Attic-Ionic), and Doric or West Greek. The term West Greek will be explained at a later point.
  3. archaic poetry, especially epic, as it evolved in the first millennium BCE.⊛
II§118. The third of these three categories is particularly complicated, especially in the case of epic. In the language of epic, as represented by Homeric poetry, there are at least three main groupings of dialects involved: Arcado-Cypriote, Aeolic, and Ionic. The existence of forms that are typical of these three dialects in epic can be viewed in terms of three successive layers, as we see in the analysis by Meillet (1935), among others. [124] Such a three-layer theory is most clearly articulated by Parry (1932).
II§119. ◊In retrospect (2008), I now distance myself from speaking of successive dialectal “layers” in epic. In general, I am persuaded by the argumentation of Horrocks (1997) in criticizing various current “layer theories.” [125] Instead of speaking of earlier and later dialectal “layers” in epic, I will hereafter speak of earlier and later dialectal “phases,” since the term phases allows for an overlap and even a coexistence of relatively earlier and later dialectal forms at any given time in the evolution of epic. To the extent that the term layer may not allow for such overlap or coexistence, it seems to me preferable not to use it. In general, my current thinking about the dialectal mix of epic is closest to that of Wachter (2000). [126] In the case of coexisting Aeolic and Ionic forms in epic, I must add, it is essential to distinguish between optional and obligatory Aeolic forms.◊
II§120. ⊛Of the three dialects that shape the diction of epic, which I have listed as Arcado-Cypriote, Aeolic, and Ionic, the Arcado-Cypriote component must have extended all the way back to the Mycenaean period in the second millennium BCE. Implicit in the term Arcado-Cypriote is a reconstruction of Arcadian and Cypriote, as they existed in the first millennium BCE, back to a common dialect that existed in the second millennium BCE. The term Arcado-Cypriote, as a unified heading, is apt. The Arcadian and the Cypriote dialects, as Chadwick points out, “display an astonishing similarity, for at the time they are recorded (fifth to fourth centuries [BCE]) they had certainly been out of touch for at least five centuries.” [127] And what would be the most apt term for the prototypcial Arcado-Cypriote that existed five or more centuries earlier? Arguably, that term would be Achaean (Ruijgh 1957; see also Janko 1992:11n10). Such a term evokes the idea of a prototypical cultural unity in the second millennium, when speakers of “Mycenaean” were most likely to have called themselves “Achaeans.” Unity was followed by fragmentation in the first millennium. By that time, Arcadian was an enclave-dialect, the only significant non-Doric dialect in the Peloponnesus, while Cypriote was a frontier-dialect, studiously archaizing and ostentatiously self-conscious of its Achaean legacy (the elites of this insular culture still retained the custom of chariot-fighting and the practice of syllabic writing, using a scribal system that is evidently cognate with the Linear A and Linear B systems).⊛
II§121. On the basis of the mutual similarities between Arcadian and Cypriote, Chadwick concludes: “Historically these facts are only explicable if these two dialects are the remnants of a widespread dialect which was elsewhere displaced by West Greek; this implies that Mycenaean Greek should also belong to the same group, and the decipherment of the Linear В script has shown this to be true, though Mycenaean does not show all the features shared by Arcadian and Cypriot.” [128]
II§122. ◊I need to point out that the term West Greek, as Chadwick uses it here, corresponds to what I have been describing as Doric. I also need to point out that Chadwick, in studying the “mutual similarities” of Arcado-Cypriote, does not separate cases of shared innovation from cases of shared retention. I made this point only indirectly in the original 1972 version of my text. Now in 2008 I say it directly. As we will see later on, cases of shared innovation are more significant than cases of shared retention.◊
II§123. ◊Viewing Mycenaean Greek as a dialect, we encounter an important complication. As Risch (1968) has shown, the Greek of the Linear В script was written by scribes who spoke two different dialects. One of these dialects was the standard language—standard, that is, for the scribes—while the other was substandard. Making use of studies identifying individual scribes by way of their handwriting, Risch demonstrated that scribes who spoke the substandard dialect were inconsistent in the spelling of words that they pronounced differently from the scribes who spoke the standard dialect and who spelled such words consistently. In another work, I produced a detailed study of standard and substandard Mycenaean as reflected in the scribal hands at Pylos (Nagy 1968). More recently, the study of standard and substandard Mycenaean has been extended from the scribal hands at Pylos to the scribal hands at Knossos (Woodard 1986).◊
II§124. ◊I offer here a test case, with reference to this phonological rule: in standard Mycenaean, vocalic * becomes o next to a bilabial, while in substandard Mycenaean it becomes a. Here is an example: the common Greek word for ‘seed’, reconstructed as *spermṇ, becomes spermo in standard Mycenaean, spelled pe-mo in Linear B, while it becomes sperma in substandard Mycenaean, which can be spelled pe-ma in Linear B. I say “can be spelled” not “is spelled” because scribes who spoke the substandard dialect could be inconsistent in their spelling, writing either pe-mo or pe-ma in free variation, while only the scribes who spoke the standard dialect would consistently write pe-mo. I offer further analysis in my article on standard and substandard Mycenaean (Nagy 1968).◊
II§125. ◊We find residual survivals of standard Mycenaean in the first millennium BCE. Risch cites two examples of such survivals: ἵππος and ἁρμόττω. [129] The meanings of these two words are relevant to their survival, as we are about to see.◊
II§126. ◊I start with the second word, the verb ἁρμόττω (secondarily ἁρμόζω), which means ‘fit, join’ with reference to the work of a joiner, that is, of a master carpenter. We see here a parallel to a form we have already seen, which is standard Mycenaean spermo as opposed to the substandard Mycenaean sperma. The verb ἁρμόττω is derived from the standard Mycenaean form harmo ‘chariot-wheel’, spelled a-mo in the Linear B tablets. Just as the meaning of spermo / σπέρμα as ‘seed’ derives from its etymology as an action-noun *spermṇ, which refers to a ‘sowing’ and which in turn derives from the root of the verb attested as σπείρω ‘sow’ in alphabetic Greek, so also the meaning of harmo / ἅρμα as ‘chariot-wheel’ derives from its etymology as an action-noun *arsmṇ, which refers to a ‘fitting’ and which in turn derives from the root of the verb attested as ἀραρίσκω ‘fit, join’ in alphabetic Greek. Just as the meaning of spermo / σπέρμα shifts from the abstract sense of ‘sowing’ to the concrete sense of ‘seed’, so also the meaning of harmo shifts from the abstract sense of ‘fitting’ to the concrete sense of a ‘fitting’ for a chariot-frame: such a ‘fitting’ for a chariot-frame is the chariot-wheel itself (in Linear B tablets, the perfect participle ararmotmeno- of what becomes the verb ἁρμόττω ‘fit’ in alphabetic Greek refers to the fitting of wheels to a chariot-frame). Whereas the standard Mycenaean form harmo means ‘chariot-wheel’ in the Linear B tablets, the substandard form *harma survives in alphabetic Greek as ἅρμα, meaning ‘chariot’ (just as German Rad, which means ‘wheel’ etymologically, becomes the word for ‘bicycle’). A point of special interest here is the fact that even the Linear B scribes who are speakers of substandard Mycenaean consistently write a-mo and not *a-ma with reference to chariot-wheels. It appears that the speakers of the substandard dialect pay relatively closer attention to the standard spelling of words that have to do with social prestige.◊
II§127. ◊Having dealt with ἁρμόττω, which is the second of the two residual standard Mycenaean words isolated by Risch, I now turn to the first, which is the noun ἵππος, meaning ‘horse’. The corresponding Mycenaean form is hi(k)kṷos ‘horse’, spelled i-qo in the Linear B tablets. From the standpoint of Indo-European linguistics, we would have expected the common Greek form to be *hekkṷos or, without expressive gemination, *hekṷos, since it is cognate with Latin equus ‘horse’. (The reconstructed alternation *hekkṷos / *hekṷos would be parallel to the alternation we see in Latin vacca / vaca ‘cow’, respectively with and without expressive gemination, as reflected in derivative Romance languages.) But the attested Mycenaean form is not *he(k)kṷos but hi(k)kṷos, following a linguistic rule that distinguishes standard Mycenaean from substandard Mycenaean. The rule can be formulated as follows: e is raised to i next to a bilabial. This rule is one of the three rules that Risch has highlighted as criteria for distinguishing standard from substandard Mycenaean in the Linear В tablets. [130] And, of those three rules, only this one is clearly definable as a linguistic innovation.◊
II§128. ◊In my study of standard and substandard Mycenaean (Nagy 1968), I have analyzed examples of standard Mycenaean forms that follow this rule, which says that e is raised to i next to a bilabial. And I have also analyzed corresponding examples of substandard Mycenaean forms written by scribes who are inconsistent in spelling standard forms. In the case of the word hikkṷos ‘horse’, however, the Linear B scribes who are speakers of substandard Mycenaean consistently write i-qo and not *e-qo with reference to horses. As in the case of the word for ‘chariot-wheel’, it appears that the speakers of the substandard dialect pay relatively closer attention to the standard spelling of words that have to do with social prestige.◊
II§129. ◊Risch (1966) noted a surprising fact about the substandard dialectal forms stemming from the substandard dialect spoken by some of the scribes writing the Linear B script. The characteristics of this substandard dialect as it existed in the era of Linear B writing in the second millennium BCE are normally matched by the same characteristics in all the surviving dialects of the first millennium BCE. For example, all the attested dialects in the first millennium show the form σπέρμα, which as we have seen corresponds to the substandard Mycenaean form sperma, and none of them shows the form *σπέρμο, which would correspond to the standard Mycenaean form. It can be inferred, then, that the standard dialect of Mycenaean Greek become extinct with the collapse of Mycenaean civilization toward the end of the second millennium BCE.◊
II§130. ◊Nevertheless, we have seen that some standard Mycenaean words have survived into the first millennium BCE, and the two examples we have considered are ἵππος and ἁρμόττω. [131] It is no accident, I think, that these two surviving examples of standard Mycenaean, ἵππος and ἁρμόττω, have to do with the elite activities of horsemanship and charioteering.◊
II§131. ◊An analogous point can be made about the elite activity of scribal writing. It has to do with the noun διφθέρα / diptherā, meaning ‘leather, parchment’, which is derived from the verb δέψω / depsō in the sense of ‘tan’—as in the tanning of leather or parchment (cf. δέψα in the Suda, s.v.). The noun διφθέρα passes the phonological test of the standard Mycenaean dialect, showing the linguistic innovation of raising e to i next to a bilabial, whereas the corresponding verb δέψω fails the same test, with its original e left unraised.◊
II§132. ◊In translating διφθέρα as ‘parchment’, I mean ‘parchment for writing’, following Herodotus (5.58), who says that the word διφθέρα was used by Ionians in that sense. Relevant here is the existing archaeological evidence for the use of parchment by the Linear A scribes in the administrative center at Zakro in Crete. [132] Evidently, the procedure of these scribes was to use parchment for their permanent archival records, as opposed to their use of clay tablets for making temporary records. I infer that the Linear B scribes of the Mycenaean era followed an analogous procedure: they would write their temporary records on clay tablets, and these records would then be transferred at the end of a given fiscal year from clay to parchment (the notion of a fiscal year is indicated by references in the Linear B tablets to the current year as opposed to the immediately preceding and following years). There is an irony to be noted here: when the administrative centers of the Mycenaean era were destroyed by fires, the temporary records of the Linear B scribes were made permanent for archaeologists because they were baked and thus preserved by the same fires that must have destroyed the permanent records recorded on parchment.◊
II§133. ◊I argue, then, that the noun διφθέρα is a reflex of standard Mycenaean, referring to the elite activity of scribes writing on parchment, while the corresponding verb δέψω is a reflex of substandard Mycenaean, referring to the non-elite activity of tanners tanning hides—whether or not these hides ever become the parchment of scribes. The use of διφθέρα with reference to the parchment of elite scribes survives in the Cypriote word διφθεραλοιφός, which means literally ‘parchment-painter’. This word is preserved in the ancient dictionary attributed to Hesychius, where it is glossed as γραμματοδιδάσκαλος παρὰ Κυπρίοις ‘teacher of literacy, in Cypriote usage’ [literally, ‘teacher of letters, among Cypriotes’]. This word is relevant to my earlier statement about the studiously archaizing culture of the Cypriotes in the first millennium BCE: “the elites of this insular culture still retained the custom of chariot-fighting and the practice of syllabic writing, using a scribal system that is evidently cognate with the Linear A and Linear B systems.”◊
II§134. Viewing Mycenaean Greek as a dialect, we see further complications. For one, the language as written by the Linear B scribes shares some features with Attic-Ionic to the exclusion of Arcado-Cypriote. [133] The features shared with Attic-Ionic are relevant to an argument developed by Risch (1955), as also by Porzig (1954). In the late second millennium BCE, according to this argument, the ancestor of Attic-Ionic was essentially undifferentiated from Mycenaean.
II§135. There are also features shared by Mycenaean with Aeolic. [134] As for these features, they are adduced by those like Palmer (1963b) who see a close prehistoric affinity between Arcado-Cypriote and Aeolic. [135]
II§136. This view of Palmer {59|60} about Aeolic cannot be reconciled with the view of Risch (1955) when he argues that the dialect of the region known as the Pelasgiotis in East Thessaly, which shows unassibilated -τι, is more representative of prototypical Aeolic than is the dialect of Lesbos, which shows assibilated -σι < *-τι. That is because the Lesbian dialect has supposedly undergone extensive remodeling because of its linguistic contiguity or Sprachbund with the neighboring Ionic dialects. [136] In terms of this argument presented by Risch, the prototypical Aeolic dialect was closer to the prototypical Doric dialect.
II§137. ⊛In broad outlines, as Risch sees them, Aeolic can be understood as a prototypical dialect ancestral to Thessalian and Lesbian. In even broader outlines, Aeolic can be understood as ancestral not only to the Thessalian and the Lesbian dialects of the first millennium BCE but also to the Boeotian dialect of the same era. In the case of Boeotian, however, we have to reckon with a massive Doric or West Greek accretion that may go as far back as the second millennium BCE. [137]
II§138. ⊛Whether or not Risch is right, I agree with him that Mycenaean Greek cannot be viewed as a dialect undifferentiated from prototypical Aeolic. It would be one thing if the distinctive features of Mycenaean Greek that are shared exclusively with the Aeolic dialects were instances of shared innovation. But they are merely instances of shared retention. [138] And the shared retention of forms in Mycenaean and in the Aeolic dialects of the first millennium BCE to the exclusion of the Arcadian dialect and the Cypriote dialect in that same era does not prove that those same forms did not exist in the prototypical Arcado-Cypriote dialect of an earlier era, in the second millennium BCE. It may be that the Arcadian and the Cypriote dialects as they existed in the first millennium BCE had lost certain archaisms that were retained by Aeolic dialects such as Thessalian and Lesbian in the same era. So long as we find no instances of innovation shared by Mycenaean and Aeolic to the exclusion of Arcado-Cypriote, there is no justification for positing a special link between Mycenaean and the prototypical form of Aeolic as spoken in the second millennium BCE.⊛
II§139. A good case can be made, on the other hand, for innovations shared by Mycenaean and Arcado-Cypriote to the exclusion of both Aeolic and Ionic. I offer two examples:
II§140. Although I am not certain about the validity of the first example, I am absolutely certain about the second. So I am ready to argue that Arcado-Cypriote is the only dialectal prototype, established by way of criteria stemming from the evidence of the alphabetic period, that shares exclusively its dialectal innovations with Mycenaean. In terms of this argument, Arcado-Cypriote is the dialectal prototype that comes closest to being identical with Mycenaean (for a similar view, see Lejeune 1968).
II§141. The one major dialectal innovation that Mycenaean shares with Attic-Ionic, that is, the assibilation of -ti to -si, is not decisive, since this innovation is not exclusively shared. The patterns of assibilation that we see being shared by Mycenaean and Attic-Ionic are shared also by Arcado-Cypriote (on the phonological and morphological conditioning of this assibilation, I have much more to say in Nagy 1968).
II§142. ⊛Still, at least in this respect, prototypical Attic-Ionic may be considered closer to Mycenaean than prototypical Aeolic, if the non-assibilation that we find in the dialect of Thessalian Pelasgiotis in the first millennium is really an archaism inherited from the prototypical Aeolic of the second millennium.⊛
II§143. ⊛Even if there was no assibilation in prototypical Aeolic, however, it would be going too far to infer that we can separate Aeolic from Mycenaean altogether—as if the distinction between Aeolic and Mycenaean could be situated at the same remote level of genealogy {60|61} as the distinction between Doric and Mycenaean. The distinction between Doric and Mycenaean is more basic, and the same can be said for the distinction between Doric and all the other dialects as attested in the first millennium BCE. The most basic distinction can be formulated in terms of West Greek and East Greek—if we define West Greek as Doric and East Greek as the common ancestor of Arcado-Cypriote, Attic-Ionic, and Mycenaean.⊛
II§144. According to Buck (1955), “the most fundamental division of the Greek dialects is that into the West Greek and the East Greek dialects.” [144] And the first isogloss to be listed for East vs. West Greek in Buck’s synopsis is the assibilation vs. non-assibilation of -τι. [145] In this regard, it is important to reconsider the ambiguity of Aeolic in the alphabetic era.
II§145. As we have already seen, Risch (1955) posits unassibilated -τι for the ancestral dialect of Thessalian and Lesbian, thereby making Aeolic less close than Attic-Ionic to Mycenaean. [146] Cowgill (1968) reacts this way to the formulation of Risch:
I would guess that the inhabitants of Late Helladic sites in Thessaly used a dialect of Aeolic type, and that if Linear В tablets are ever found in that area they will turn out to exhibit this local dialect (perhaps overlaid by Southern features from an administrative koine).
At least some subdialects of this Aeolic must have preserved -ti unchanged. But it is not easy to suppose that the -si of Lesbian was borrowed in Asia Minor from Ionic; if a feature as central as this had been borrowed, one would expect many others to have been borrowed as well. Perhaps rather Aeolic was already differentiated in this feature in the second millennium, and the speech of the Asiatic colonists reflects a variety of Aeolic into which the wave of assibilation from the South had penetrated, while in Thessaly and Boeotia varieties which had resisted this wave won out (perhaps with help from the neighboring West Greek dialects). [147]
II§146. A further complication is presented by the archaic dialect of Pamphylian, which Buck actually classifies under the heading of Arcado-Cypriote. [148] Despite this classification, Pamphylian shows the non-assibilation of -τι.
II§147. For traces of non-assibilated -τι in Mycenaean and for possible dialectal implications, I refer to the detailed analysis in two works of my own (Nagy 1968, 1970).
II§148. {At this point I mentioned briefly a possible instance of an innovation shared by Mycenaean and Aeolic to the exclusion of Attic-Ionic. [149] }
II§149. In the end, I remain unpersuaded by the argument of Risch (1955) that the prototypical dialect of Attic-Ionic as it existed in the second millennium BCE had really been as close to Mycenaean as was the prototypical dialect of Arcado-Cypriote—or that Attic-Ionic had been even closer to Mycenaean than was Arcado-Cypriote. I invoke here a decisive formulation by Benveniste (1956b):
Pour en venir au problème plus general d’un dialecte ‘méridional’, hypothèse qui a les préférences du rapporteur [i.e., Risch], la seule coïncidence vraiment remarquable entre sa branche arcado-cypriote et sa branche ionienne à date historique est l’assibilation -τι > -σι, que présente déjà le mycénien. Mais, à poser le problème en termes de chronologie absolue, entre nos derniers textes mycéniens et nos premiers textes dialectaux alphabétiques (abstraction faite des poèmes homériques), il s’écoule environ six cents ans. Il faudrait admettre, durant cette période, dans l’hypothèse soutenue par Risch, une remarquable conservation du mycénien dans sa lignée arcado-cypriote, une profonde évolution du mycénien dans sa lignée ionienne. N’est-il pas plus plausible de supposer qu’à l’époque de nos tablettes, l’ionien (qui n’y est pas représenté) était déjà plus ou moins largement différencié? [150]
The most plausible conclusion, then, is that the prehistoric phases of Arcado-Cypriote, Aeolic, and Attic-Ionic were already differentiated in the late second millennium BCE, and that the dialect that comes closest to being identical with Mycenaean is the ancestral Arcado-Cypriote. {61|62} Still, it is unnecessary to posit complete identity, as Palmer points out:
“Arcado-Cypriot” is merely the name given to a group of linguistic features which philologists assign to the dialects of the Mycenaean Peloponnese. It does not imply a completely uniform “Mycenaean” language. The documents of later Cyprus and Arcadia themselves show dialectal differentiation, and we may expect to find in the Linear В tablets forms which differ from Arcado-Cypriot not simply because they are more archaic but because they mirror a variety of “Mycenaean” in some respects different from the direct ancestors of Arcadian and Cypriot. [151]
II§150. Finally, there is evidence for not only dialectal differentiation but also dialectal cross-influence in this early period: for example, there are strong arguments in favor of positing the penetration, in the late Mycenaean era, of Aeolic or North-Mycenaean elements into such South-Mycenaean dialectal areas as the Peloponnesus, [152] even including parts of Arcadia itself. [153]
II§151. I return to the question, posed earlier, concerning different dialectal phases of epic. From what we have seen so far, it is reasonable to infer that the earliest dialectal phase of epic is for all practical purposes the prototypical phase of Arcado-Cypriote in the second millennium BCE—as we reconstruct this phase on the basis of the attested phases of Arcadian and Cypriote in the first millennium BCE. But here we encounter once again the same difficulties we already encountered in looking for a direct affinity between the attested Mycenaean Greek of the Linear В texts and the prototypical counterpart that we reconstruct as Arcado-Cypriote: shared retention is of no probative value. Even if we succeed in detecting archaisms exclusively shared by the attested Greek of the Homeric corpus and by Arcado-Cypriote, which point to the Mycenaean origins of a given epic configuration, the question remains: what kind of Mycenaean Greek are we talking about? In confronting this question, we have to reckon with the possibility that whatever archaisms we detect might have been still extant in the prototypical Aeolic or Ionic coeval with the prototypical Arcado-Cypriote of the Mycenaean era. This possibility is most relevant because Aeolic and Ionic, in that chronological order, are the next two dialectal phases of epic after Arcado-Cypriote.
II§152. In speaking of a distinct Aeolic phase in the evolution of the Homeric language, I recognize that there have been various attempts to disprove the existence of such a phase. [154] Still, I maintain that there is incontrovertible evidence for an Aeolic phase in the form of distinctly Aeolic features embedded in the formulaic system of Homeric diction. I also maintain that these features complement the distinctly Ionic features that are likewise embedded in this formulaic system. In order to prove the existence of such embedded Aeolic and Ionic features, it is essential to locate not their linguistic archaisms but rather their linguistic innovations. Examples of such innovations include the Aeolic perfect participle in -οντ- and the Attic-Ionic particle ἄν.
II§153. If it is true that Arcado-Cypriote represents the oldest dialectal phase of epic, it stands to reason that it will be difficult to find embedded in the formulaic system of Homeric diction any examples of innovations that are exclusive to Arcado-Cypriote. That is because Arcado-Cypriote represents not only the oldest dialectal phase of Homeric diction but also its oldest formulaic phase. And it is inherently difficult to glean new linguistic forms from old linguistic material embedded in old formulaic settings. Still, such gleaning is necessary because it is methodologically insufficient to identify a given epic form as Arcado-Cypriote simply because it is attested in Arcadian or Cypriote texts dating from the first millennium BCE.
II§154. Before we consider such embedded forms, I take this moment to note that I do not mean to underrate the importance of finding correspondences of attested Arcadian and Cypriote words with words attested in Homeric diction. Such correspondences are of interest because the very fact of attestation, either epigraphic or lexicographical, shows the archaism of both Arcadian and Cypriote. {62|63}
II§155. Especially important is the testimony of the Alexandrian lexicographical tradition as represented by a compendium known as the γλῶσσαι κατὰ πόλεις (on which see Latte 1924). As we see from this compendium, the aim of its compilers was to find residual epichoric attestations of poetic words long obsolescent in the general Greek-speaking world. For example, the Greeks of Clitor (Kleitorioi) in Arcadia are credited with the active usage of the following Homeric words that are no longer used in the living language of most other Greeks:
ἀῆται· ἄνεμοι (‘winds’)
αὐδή· φωνή (‘voice’)
δέδορκεν· ὁρᾷ (‘sees’)
ἔνεροι· vεκρoí (‘corpses’)
ἐσθλόν· ἀγαθόν (‘worthy’)
λεύσει· ὁρᾷ (‘sees’)
πάροιθεν· ἔμπροσθεν (‘in front’)
χηλός· κιβωτός (‘coffer’)
ὦκα· ταχέως (‘quickly’)
ὠλέναι· βραχίονες (‘arms’)
II§156. The Clitorian samples we see quoted here from the Alexandrian γλῶσσαι κατὰ πόλεις may have stemmed from a Lokalantiquar who was contemporary with the younger Zenodotus. [155] That so many epic words should survive in the everyday spoken language of a remote community such as Clitor in Arcadia during the historical period shows how much our own understanding of the Greek language is confined to the literary and official languages. In this connection, I draw attention to the attitude underlying the textual regularization in the Clitorian gloss αὐδή· φωνή: it tells us that the Homeric word αὐδή means ‘voice’ and that this meaning is still preserved in Clitorian, but the actual local form, which must have been αὐδά̄, is not even specified. [156]
II§157. On linguistic grounds, yes, we may say that a Homeric form like αὐδή is traceable all the way back to the second millennium BCE. Evidently, it must be a very old form. Still, in terms of Homeric diction, it remains to be proved that Homeric αὐδή is a specifically Arcado-Cypriote form, to the exclusion of parallel forms that may have once existed in Aeolic or Ionic.
II§158. So, there is no decisive proof here, since the relevant evidence is merely lexical. [157] If the argumentation is restricted to lexical evidence, we can expect skepticism even about the relevance of Arcado-Cypriote to Homeric diction. [158] That is why we must look for evidence that goes beyond questions of vocabulary.
II§159. I now turn to a rare example of a linguistic innovation that is exclusively shared by epic and Arcado-Cypriote. This innovation is on the level of morphology, and it concerns the athematization of verbs in -έω into verbs in -ημι (along with a paradigmatic generalization of full-grade -η-). What results in Arcado-Cypriote is a new infinitival ending -ῆναι as opposed to Aeolic -ήμεναι and Ionic -έειν, as in Arcadian ἀπειθηναι, [159] Cypriote ku-me-re-na-i = κυμερῆναι. [160] Similarly in Homeric diction, we see φορῆναι (ΙΙ 107, VII 149, X 270, xvii 224).
II§160. A question remains: is it valid to argue that this Homeric form φορῆναι derives from a prototypical Arcado-Cypriote phase in the evolution of Homeric diction? Risch rejects this argument, counterarguing that φορῆναι is an artificial creation from φορήμεναι (XV 310), on the model of original athematic μιγῆναι {63|64} (VI 161, etc.) vs. μιγήμεναι (IX 133, etc.). [161] But the argument can be defended and the counterargument of Risch can be rejected on the basis of two further arguments.
II§161. First, no other Homeric verbs in original -έω have -ῆναι for infinitive: καλήμεναι (X 125), πενθήμεναι (xviii 174), ποθήμεναι (xii 110), φιλήμεναι (XXII 265). [162] Second, in the case of the original athematic aorist passive pairs like δαμήμεναι / δαμῆναι, δαήμεναι / δαῆναι, μιγήμεναι / μιγῆναι, φανήμεναι / φανῆναι, etc., the type in -ῆναι regularly occurs in verse-final position, or, secondarily, in the position immediately preceding the trochaic caesura; the type in -ήμεναι, on the other hand, occurs as a rule immediately preceding the bucolic diaeresis.
II§162. More needs to be said about the second of these two additional arguments, with specific reference to forms in -ήμεναι that occur before the bucolic diaeresis. In dactylic hexameter, the space before the bucolic diaeresis tends to suit a relatively greater proportion of innovative forms. That is because the 4th foot of the dactylic hexameter, which immediately precedes the bucolic diaeresis, behaves in a way that is different from the metrical behavior of the other feet. In the 4th foot, there is a strong tendency to avoid a trochaic word-ending or caesura (– ⏑ ‖ ) and to substitute instead a dactylic word-ending or bucolic diaeresis (– ⏑ ⏑ ‖). This tendency causes what is called the Verzerrung of many trochee-final words into dactyl-final words, and thus the admission of a wide variety of linguistic innovations before the bucolic diaeresis. [163]
II§163. Here is an example. Besides the 50-odd Homeric cases of old athematic-stem ὦρτο (occurring for the most part immediately after the bucolic diaeresis), the three exceptional cases of innovative thematic-stem ὤρετο are all located immediately before the bucolic diaeresis (ΧΙΙ 279, XIV 397, XXII 102). In other words, the trochee-final archaism ὦρτο had undergone Verzerrung in the 4th foot, and this Verzerrung accommodates the innovation ὤρετο.
II§164. Here is another example. There are 25 Homeric cases of the older form μητρός (cf. also πατρός) vs. 7 corresponding cases of the newer form μητέρος, created by paradigmatic leveling (cf. also πατέρος). [164] Of the 7 cases of μητέρος, only one occurs immediately after the bucolic diaeresis (xxi 110); the rest all occur immediately before the bucolic diaeresis (III 422, XXIV 466, iii 212, xiv 140, xv 432, xviii 267). Prevocalic μητρός (that is, trochee-final μητρός) simply does not occur in the 4th foot. Instead, μητρός undergoes Verzerrung to become the newer form μητέρος, even when it remains in collocation with the older form πατρός: [165]
νόσφι φίλου πατρὸς ‖ καὶ μητέρος ‖ ἀλλὰ καὶ ἔμπης
Iliad XIX 422
καί μιν ὑπὲρ πατρὸς ‖ καὶ μητέρος ‖ ἠυκόμοιο
Iliad XXIV 466
οὐδ᾽ εἴ κεν πατρὸς ‖ καὶ μητέρος ‖ αὖτις ἵκωμαι
Odyssey xiv 140
ὄφρα ἴδῃ πατρὸς ‖ καὶ μητέρος ‖ ὑψερεφὲς δῶ
Odyssey xv 432
μεμνῆσθαι πατρὸς ‖ καὶ μητέρος ‖ ἐν μεγάροισι
Odyssey xviii 267
II§165. Likewise the trochee-final type μιγῆναι (that is, prevocalic μιγῆναι) is avoided in the 4th foot, and what results is Verzerrung, to suit the type μιγήμεναι (acquired from the Aeolic repertoire of formulas), which regularly occurs before the bucolic diaeresis. When we test the new aorist passive in -θη-, which had ultimately replaced the residual aorist passive in -η- (such as μιγ-ή-μεναι / μιγ-ῆ-ναι), the distribution of its infinitives in the verse makes even more apparent which variant, -ήμεναι or -ῆναι, is older from the formulaic point of view: those in -θ-ῆναι that fit into verse-final position still regularly occur there, but the more rare infinitive in -θ-ήμεναι becomes disrupted {64|65} in positional patterns and “donne l’impression d’être assez artificiel.” [166]
II§166. To sum up: the -ῆναι of the aorist passive seems formulaically older than -ήμεναι in both form and position. If therefore the pair-type φορήμεναι / φορήναι is modeled on the metrically-equivalent pair-type μιγήμεναι / μιγῆναι, we would expect the direction of motivation to be from φορῆναι to φορήμεναι, not the reverse. Furthermore, on the evidence of the metrically-equivalent aorist passive infinitives, the occurrence of all five forms καλήμεναι πενθήμεναι ποθήμεναι φιλήμεναι φορήμεναι exclusively before the bucolic diaeresis indicates that these forms are more innovative than the corresponding forms *καλῆναι *πενθῆναι *ποθῆναι *φιλῆναι φορῆναι; and the sole Homeric attestation of the fifth of these forms, φορῆναι, indicates that it is residual and old.
II§167. I note with great interest the consistent association of φορῆναι with a specific verb: δῶκε, δῶκε, δοίης, δῶκ᾽ (VII 149, X 270, xvii 223, II 105). Relevant to this construct is the Mycenaean formula do-ra-qe pe-re po-re-na-qe a-ke, repeated four times in the Pylos tablet Tn 316 (lines 2, v. 2, 5, 8). [167] If we may translate, so to speak, as δῶρα τε φέρει φορῆναί τε ἄγει (without any specification of syntactical boundaries here), then the following significant collocations emerge: (1) a figura etymologica in the juxtaposition of φέρει φορῆναι, (2) a correspondence of Mycenaean δῶρα…φορῆναι with Homeric δῶκε / δοίης…φορῆναι, and (3) a correspondence of Mycenaean φέρει…ἄγει with Homeric φέρειν…ἄγειν.
II§168. The third of these three collocations can best be understood by reading the relevant Homeric verses:
δῶκε δ᾽ ἄγειν ἑτάροισιν ὑπερθύμοισι γυναῖκα
καὶ τρίποδ᾽ ὠτώεντα φέρειν.

‘He gave to his very spirited comrades a woman to lead away
and a tripod (with handles) to carry.’
Iliad XXIII 512–513
II§169. ◊I offered further arguments in later work (Nagy 1994–1995) in support of my interpretation of Pylos tablet Tn 316.◊
II§170. On the phonological level as well, it is possible to find traces of a pre-Aeolic phase in epic. I am about to show two Homeric words that show such traces, coming from a dialectal phase that I described earlier as standard Mycenaean, examples of which survive only sporadically into the alphabetic era. The examples shown by Risch, as we have seen, are the words ἵππος and ἁρμόττω, which survive even in the everyday usage of the classical era, not only in epic. [168] The examples that I am about to show, on the other hand, survive only in epic. My examples are two Homeric words shaped by the same phonological rule that results in a form such as ἵππος. As we have seen, the rule is to be formulated as follows: e is raised to i next to a bilabial. Here are the examples:
πινυτός (i 229, etc.): from a morphological point of view, we expect an original *πενυτός [169]
πίσυρες (v 70, etc.): the original zero-grade * k wetures (vs. full-grade *kwetwores) {65|66} still survives in Lesbian πέσυρες). [170]
II§171. ⊛It is likely that many other non-Ionic and non-Aeolic characteristics of Homeric diction are also derived from a Mycenaean phase in the evolution of this diction. This Mycenaean phase is what I have been calling the Arcado-Cypriote phase, which is older than the Aeolic phase, which in turn is older than the Ionic phase. [171] With reference to the Aeolic and the Ionic phases, I must emphasize again that I see no way to eliminate Aeolic as an intermediate phase following the Mycenaean phase and preceding the Ionic phase in the evolution of epic diction.⊛
II§172. I now turn to examples of pre-Aeolic morphology embedded in Homeric diction, which indicates an Arcado-Cypriote or Mycenaean phase in the evolution of this diction. Following Ruijgh (1957), we may prefer to call this pre-Aeolic phase simply the “Achaean” phase.
II§173. The examples involve 1st-and 2nd-person plural pronouns. While the 1st and 2nd plural nominative pronouns are attested as Aeolic ἄμμες and ὔμμες in Homeric diction, the corresponding Arcado-Cypriote or “Achaean” forms, which we can reconstruct from the phonological evidence of Arcadian as written in the alphabetic era, would be ἁ̄μες and ὑ̄μες. [172] Ruijgh remarks on the form ὑ̄μες: “les Ioniens eussent certainement préféré la forme achéenne [= Arcado-Cypriote] ὑ̄μες, plus proche de la forme ionienne ὑ̄μεῖς, à ὔμμες, s’ils l’eussent connue.” [173] But even if the Arcado-Cypriote forms ἁ̄μες and ὑ̄μες had disappeared from the repertoire of Homeric diction, the fact is that Arcado-Cypriote variants do seem to emerge in Homeric attestations of the possessive adjectives for ἄμμες and ὔμμες, namely ἁ̄μός and ὑ̄μός vs. Ionic ἡμέτερος and ὑ̄μέτερος.
II§174. These Aeolic / Arcado-Cypriote forms ἄμμες / ἁ̄μός and ὔμμες / ὑ̄μός cannot be explained away as the results of arbitrary editorial selection in interpreting some supposedly archetypal spellings αμες / αμος and υμες / υμος. [174] Rather, the neat split in ἄμμες / ἁ̄μός and ὔμμες / ὑ̄μός, as opposed to the unitary ἡμεῖς / ἡμέτερος and ὑ̄μεῖς / ὑ̄μέτερος, points to two distinct dialectal phases, namely Aeolic / Arcado-Cypriote. I draw attention here to the linguistic accuracy we find in the transmission of even suprasegmental features (especially accentual patterns). [175] Presumably there was no Aeolic *ἄμμος and *ὔμμος available to level out a pre-Aeolic ἁ̄μός and ὑ̄μός. That ἁ̄μός and ὑ̄μός did not stem from prehistoric Ionic is shown by the failure of ἁ̄μός to survive as *ἡμός.
II§175. Let us now turn to the Ionic phase in the evolution of Homeric diction. In this phase, which is the setting for the consolidation of the Homeric Iliad and Odyssey as we know these epics, no form was available to level out either ἁ̄μός and ὑ̄μός or ἄμμες and ὔμμες.
II§176. This statement needs to be qualified: in preconsonantal position, ἄμμες and ὔμμες become metrically equivalent to ἡμεῖς and ὑ̄μεῖς, with the result that ἄμμες and ὔμμες are replaced by ἡμεῖς and ὑ̄μεῖς in these positions. So, ἄμμες and ὔμμες are found only in prevocalic position. Here is a further qualification: even where a current Ionic equivalent is available, a dialectal form may nonetheless elude leveling: a case in point is the survival of Aeolic absolute verse-final ἄμμι (e.g. ΧΙΙΙ 379), despite the fact that Ionic ἡμῖν (e.g. i 10) is a metrical equivalent.
II§177. What inhibits the leveling of ἄμμι by ἡμῖν in this case has to do with a common phraseological switch that we see operative in epic: the phraseology we find in the space – ⏑  – between the bucolic diaeresis and the end of the verse can switch into the space – ⏑  – immediately preceding the trochaic caesura, whenever the final syllable of – ⏑  – is short or can be shortened by hiatus. Now verse-final ἄμμι does not prevent this switch (hence ἄμμι before trochaic caesura in e.g. iii 140) whereas verse-final ἡμῖν does prevent it. To put it another way: the form ἡμῖν, when it comes at the end of an expression that fits into the space – ⏑  – between the bucolic diaeresis and the end of the verse, makes it impossible for that expression to shift into the space – ⏑  – before the trochaic caesura. There was a way out, however, if ἡμῖν is replaced by the artificial Ionic form ἦμιν, which we find only before the trochaic caesura (XVII 415, 417; viii 569 = xiii 177; x 563; xi 344; xvii 376; {66|67} xx 272). We may compare also the pair ὄσσε φαεινώ in verse-final position (as in XIII 3) vs. ὄσσε φαεινά before the trochaic caesura (as in XIII 435). In sum, dialectal leveling is sporadic and unpredictable where the grammatical or metrical conditioning is ambiguous.
II§178. Conversely, dialectal leveling is predictable where the grammatical and metrical conditions are themselves equally predictable. For example, let us consider the ᾱ of the frequent Homeric form θυρά̄ων vs. θύρῃσι and θύρηφι. Since inherited Ionic η pervades the paradigm, we begin by assuming that ᾱ survives in θυρά̄ων for the simple reason that Ionic had no corresponding *θυρήων to level it out. And yet, *θυρήων must have once existed in prehistoric Ionic as well, since it is the formal ancestor of native Ionic θυρέων, which is shaped ⏑  – at XXI 191 and which is ultimately contracted into θυρῶν, as we see from the scansion – of θυρέων at xxi 47. But there is still a question: in the metrical space ––, why is the corresponding form a non-Ionic θυρά̄ων and not Ionic *θυρήων? The answer must be this: so long as the phonological mechanism of quantitative / qualitative metathesis (-ηω- > -εω-) was operative in Ionic, the poetic tradition in its Ionic phase rejected the combination -ηω- because of the automatic phonological conversion of -ηω- into -εω- in the natural language. So the only way to fill the inherited position that could no longer be filled by -ηω- was to admit the non-Ionic and metrically equivalent –ᾱω-. This form was probably Aeolic: we may compare the paradigmatically-restored ᾱ of genitive plural –ᾱων, still attested in Thessalian. In general, we see in the Ionic phase of epic a failure to level out ᾱ with η in the genitive plural. To show the extent of the pre-Ionic underpinnings of epic, we note that the older non-Ionic form τά̄ων is more than twice as frequent as the Ionic τῶν. [176] As for the nominative / vocative θεά̄ (e.g. I 1), Ionic failed to level it out with θεή—partly because the Ionic word for ‘goddess’ was (ἡ) θεός.
II§179. The same arguments that I use here to explain the genitive plural in -ά̄ων are also applicable to the genitive singular masculine -ᾱo; but here, even further grammatical and metrical conditioning affects the ultimate distribution: -ᾱо has become restricted to preconsonantal position, even though it could suit the prevocalic position as well (with elision of –ο). Here is an example: Νηληϊάδᾱο γέροντος (VIII 100) vs. Νηληϊάδεω ἀφίκοντο (V 618). Viewing such evidence, Meister asks this rhetorical question: “Haben sie [= die Sänger] etwa archaische Formen nach strengern prosodischen Gesetzen behandelt als moderne?” [177] Meanwhile, the quantitatively / qualitatively metathesized native Ionic -εω (< -ηο) has spread throughout the formulaic structure, from prevocalic position (with the locus of diffusion probably set by -ᾱο V- > -ᾱ᾽ V-) to preconsonantal and even verse-final position as well: the frequency of a verse-final phrase like Κρόνου πάϊς ἀγκυλομήτεω illustrates most effectively the pervasiveness of Ionic. [178]
II§180. Suggestive of relative intensity in the preservation of pre-Ionic mechanisms are the following statistics on the ratio of -ᾱο to -εω: 4.5 to 1 in the Iliad, 2.8 to 1 in the Odyssey, 1 to 1 in Hesiod (Theogony and Works & Days); on the ratio of -ά̄ων to -έων: 8.7 to 1 in the Iliad, 6.5 to 1 in the Odyssey, 3 to 1 in Hesiod. [179]
II§181. With regard to occasional ᾱ / η variations in morphologically {67|68} ambiguous situations, [180] especially interesting is the apparent hypercorrection ἵλᾱος (I 583), despite its etymology *si-slēṷos (> *ἵ̄ληος > ἵ̄λεως); a factor here is the inherited alternation between full-grade *-slēṷ- and zero-grade *-slăṷ- (as in verse-final ἵ̄λαος ἔστω, XIX 178). [181]
II§182. The artificial poetic mechanism of Dehnungsgesetz, discussed earlier under the subsection phonology, can serve as a valuable analytic tool in distinguishing Ionic from generally pre-Ionic dialectal phases of epic. One Attic-Ionic phonological process to be tested is the prehistoric split from one long-e and one long-o into two each, featuring a new opposition of open vs. closed, that is, ę̄ / ǭ vs. ẹ̄ / ọ̄, spelled η / ω vs. ει / ου in the standard post-Euclidian Attic alphabet. Configurations with ę̄ / ǭ correspond etymologically to the / inherited from Indo-European, while those with ẹ̄ / ọ̄ are new creations from a dialectal phase of Greek.
II§183. It is to be expected that the natural morphophonemic mechanism of Dehnungsgesetz, inherited from Indo-European, should produce ę̄ / ǭ from the initial vowel e / o of a compound’s second constituent. Here are some examples: φιλ-ήρετμος ‘fond of the oar’ (i 181, etc.) vs. ἐρετμόν ‘oar’ (xi 77, etc.), δυσ-ώνυμος ‘bearing an ill name’ (XII 116, etc.) vs. ὄνομα ‘name’ (XXII 51, etc.). But with the onset of a split for long-e and long-o in Attic-Ionic, the old short / long opposition switches from ε / η and ο / ω, which is diachronically determined, to ε / ει and ο / ου, which is synchronically determined. Hence the Attic names ει and ου for the letters Ε and Ο (Plato Cratylus 426C, etc.) on the principle, stated by Herodian (2.403), that πᾶν ὄνομα μονοσύλλαβον μακροκαταληκτεῖν θέλει ‘every monosyllabic word tends to be [the equivalent of] a long final syllable’. [182]
II§184. It is also to be expected that, after the period of split long-e and long-o in Attic-Ionic, the artificial Dehnungsgesetz of epic should produce ẹ̄ / ọ̄ from a spoken e / o in the initial syllable of a word. Here are examples: εἱᾰνός (XVI 9) vs. ἐᾰνός ‘fine robe’ (XXI 507, etc.), οὔνομα (vi 194, etc.) vs. ὄνομα ‘name’ (ix 16, etc.). Again we may apply this methodological principle: configurations with morphological mechanisms that had become extinct in Ionic may optionally be preserved by epic in their non-Ionic form, even when an Ionic metrical equivalent is available. For example, besides the common ου-Dehnung of the type οὐλόμενος ‘accursed’ in x 394, etc., we find such rare examples of ω-Dehnung as ὠλεσίκαρποι ‘losing their fruit’, epithet of ἰτέαι ‘willows’ in x 510. Since the compound-formant ὀλεσι- is actually attested in Ionic poetry (ὀλεσ-ήνορας ‘man-destroying’ in Theognis 399, ὀλεσί-θηρος ‘beast-slaying’ in Euripides Phoenician Women 664), we cannot assume that it was unfamiliar to the earlier Ionic tradition. One other possible reason, then, for a constraint against reshaping ὠλεσι- into οὐλεσι- remains: ὠλεσι- could have prevailed by virtue of its provenience from the prestigious tradition of another dialect where no split in long-e and long-o has occurred. Both Aeolic and (apparently) Arcado-Cypriote are such dialects. [183] As for βωτι-ανείρῃ ‘nurse of heroes, man-feeding’ (I 155; vs. βόσι-ν ‘feeding’ in XXI 268), its non-Ionic ω-Dehnung is also relevant from the standpoint of the region for which it serves as epithet. That region is Phthia, homeland of Achilles. The absence of assibilation in the -τι- of βωτι- is also dialectally significant. [184]
II§185. The epic instances of {68|69} ποτί / προτί are likewise probably derived from a dialectal substratum; their preservation is directly ascribable to the failure of [*poti | *proti >] *ποσί / *προσί to survive in spoken Ionic; ποτί is still attested in the alphabetic period, as we see from the evidence of texts written in the Thessalian and Boeotian dialects. [185] In Mycenaean, the corresponding form shows assibilation: po-si. As for προτί, the extent to which it is embedded in the oldest layers of epic is illustrated by the fact that out of its 60 Homeric occurrences, 57 are located before words that originally started with ϝ-. [186] Also relevant are Arcadian πός and Ionic πρός: these forms were probably prehistoric sandhi-variants of ποτί and προτί. [187]
II§186. The traditional language of Aeolic poetry has extended the mechanism of Dehnungsgesetz even further by annexing another artificial mechanism of metrical lengthening: -CV ‖ + R- = -CVRR- (where R stands for sonorant, ‖ for word-break). The precedent for the metrical lengthening of spoken R into RR here was set by the etymologically valid making of position by prehistoric *sR, which became RR in Aeolic. [188] For example, the *sn- in the root *sneigwh- is reflected in ἀγά-ννιφος ‘snowy’ (I 420, etc.) and in the underlying νν- of ὄρεϊ νιφόεντι (⏑  – ⏑  – ) ‘on the snowy mountain’ (ΧΙΙΙ 754). We have already observed in the discussion of epic in Part I that the metrical precedent of word-initial RR- for spoken R- becomes extended beyond its etymological confines, as we see from the underlying νν- in ἅμά δὲ νέφος (⏑  – ⏑ ) ‘and at the same time, a cloud’ (IV 274), even though ν- here had never been *sn-. As for the adaptation of the poetic mechanism -CV ‖ + R- = -CVRR- to the poetic mechanism of Dehnungsgesetz, it involved simply the re-adjustment of word-boundary: -C ‖ + VR- = -CV ̄ RR-, equivalent to Ionic -C ‖ + VR- = -CV ̄ R-. As an example of the Aeolic extension of Dehnungsgesetz, I cite ἐννοσί-γαιος ‘earth-shaker’ (XIII 43, etc.) vs. ἐνοσί-χθων ‘earth-shaker’ (VII 445, etc.) and ἔνοσι-ς ‘shaking, quake’ (e.g. Hesiod Theogony 681, 849).
II§187. Meanwhile, the more simple Ionic mechanism of Dehnungsgesetz produces an εἰνοσι- corresponding to the Aeolic ἐννοσι-, and the former has actually ousted the latter in the epic epithet εἰνοσί-φυλλος ‘with quivering foliage’ (II 632, etc.), vs. the residual ἐννοσί-φυλλος in Simonides 595.1 (epithet for ἀήτα: cf. the Clitorian gloss ἀῆται· ἄνεμοι, as mentioned above). But as for ἐννεσίη ‘suggestion’, found only as dative plural ἐννεσίῃσι(ν) in epic (V 894, etc.), its synchronic morphological motivation (*ἐνεσίη from ἐν-ίημι as ἐξεσίη ‘embassy’ from ἐξ-ίημι) became defunct. [189] In fact, it became defunct so early that the Aeolic ἐνν- could resist displacement by an equivalent Ionic εἰν- (as in Hesychius: εἰνεσίαι· ἐπιστολαί ‘orders’), simply because *ἐνεσίη could not be intuited automatically on the basis of ἐννεσίη. Even beyond the mechanism of Dehnungsgesetz, the Ιonic / Aeolic contrast -εινV- / -εννV- in epic is also apparent on a lexical level, as in the adjectivization of es-stem nouns (type γέν-ος, φά-ος) into *-es-nos > Ionic -εινος (type φαεινός ‘bright’) vs. Aeolic -εννος (type φάεννος ‘bright’). Since the morphological motivation of -εινος / -εννος by -ος is synchronically moribund in the attested periods of both Ionic and Aeolic, the leveling-out of substratal Aeolic forms {69|70} with -εννος can occur only where a corresponding form with -εινος has still survived in Ionic. For example, we see the loss of *φαεννος and *ἀλγεννος to attested Homeric φαεινός and ἀλγεινός, vs. the retention of -εννος in the formula ἐρεβεννὴ νύξ ‘dark night’ (V 659, etc.): “c’est que l’adjectif dérivé de ἔρεβος n’existait pas en ionien.” [190]
II§188. There seem to exist Attic traces (as distinct from Ionic) in Homer. [191] We should draw a distinction in relative values, however, between (1) invocation of supposed factors like μεταχαρακτηρισμός or a Peisistratean Recension and (2) purely linguistic arguments for Attic elements. For an introduction to the controversy about possible Doric elements in epic, I cite Morpurgo Davies 1964; especially telling in this work is the formulation of a textual factor operative on an important Hesiodic dialectal criterion, and it deserves to be singled out for its perceptiveness: “a short accusative in -ας was prosodically odd, but graphically correct [highlighting supplied], while a short accusative in -ος was likely to be automatically corrected either into the equivalent long form in -ους or into a singular in -ον.” [192]
II§189. One of the most thoroughly-reported facets of the Greek language involves the epichoric dialectal inscriptions. I cite the surveys by Bechtel 1921, 1923, 1924; Thumb and Kieckers 1932; Thumb and Scherer 1959; and Buck 1955. [193] Exemplary of individual studies are those of Jacobsthal 1907 and Lejeune 1940. Besides steady progress since the nineteenth century in the collection and analysis of the evidence, there has also emerged an important contemporary trend in attempts at a broader perspective on the bewildering variety of attested dialects. The foremost proponent of this trend has been Risch (1949; cf. also 1955), who “used principles of dialect geography to show that when isoglosses do not agree with tribal boundaries (e.g. the treatment of *-ns), the most likely inference is that the innovations in question are relatively late, rather than that tribal mixture and overlaying has occurred.” [194] For a cautious attempt at correlating attested dialectal geography with reconstructed dialectal prehistory, I cite Coleman (1963). [195] {70|71}


[ back ] 1. In the case of intervocalic --, the decipherment of Linear B forces a reassessment of what is said by Meillet 1935/1965:22–24.
[ back ] 2. See the remarks of Blanc 2008:79n98, who disagrees with the approach of Wyatt 1969.
[ back ] 3. Kuryłowicz 1956:264–272, 276–285.
[ back ] 4. For a possible parallel, see Meister 1921:14.
[ back ] 5. Kuryłowicz 1956:284.
[ back ] 6. Parry 1934:132.
[ back ] 7. Meillet 1935/1965:160.
[ back ] 8. Parry 1934:131.
[ back ] 9. Parry 1934:136–137.
[ back ] 10. Parry 1934:138. Relevant in Lejeune’s Traité (1955) is a valuable index analytique (345–362), with such important rubrics as chronologie absolue (350) and chronologie relative (350–351).
[ back ] 11. Lejeune 1955:59 and Dow 1967. Though I concede that an immediate succession of two aspirated stops is not typologically impossible (Allen 1968a:24–26), the κ in compromise-spellings like <δεδοκχθαι> suggests that aspiration is not the assimilating feature here (Dow 1967:220–221).
[ back ] 12. Buck 1955:190.
[ back ] 13. Buck 1955:28–30.
[ back ] 14. Gelb 1963:181–183; for a more specific study using similar methodological principles, see Einarson 1967.
[ back ] 15. For typological commentary on the limitations of a syllabary like Linear В in representing Greek, see Householder 1964 and Cowgill 1966:83–84n21. For an illuminating inquiry into the linguistic background of the Linear В script, see Lejeune 1966.
[ back ] 16. I have already discussed stress in Part I, with specific reference to Allen 1966.
[ back ] 17. See especially Kuryłowicz 1958:106–113.
[ back ] 18. See Allen 1967:46–47, raising the question of syntactic leveling.
[ back ] 19. Allen 1968b:306.
[ back ] 20. Lejeune 1955:79–81.
[ back ] 21. Lejeune 1955:81. The relevance of such morphological restructuring to the language of Linear Β is discussed by Lejeune in the sub-article “Restauration analogique de la sifflante intervocalique” in “Notes de morphologie mycénienne” (1965:1–7).
[ back ] 22. Schwyzer 1939:405–406.
[ back ] 23. Schwyzer 1939:534–535.
[ back ] 24. Kuryłowicz 1945/1949:169.
[ back ] 25. Watkins 1962c:16–17.
[ back ] 26. On this function see Watkins 1962a:113–15, 1962c:13–16; Kiparsky 1968:45.
[ back ] 27. On Greek heteroclitics specifically, see Egli 1954.
[ back ] 28. Kastner 1967:131–132.
[ back ] 29. Cf. Pagliaro 1961:114n16 and Schmitt 1967:111n678.
[ back ] 32. Kiparsky 1968:44.
[ back ] 33. For the concept of deep structure, see Chomsky 1965:16–18.
[ back ] 34. Kiparsky 1968:48.
[ back ] 35. Smyth 1963:429.
[ back ] 36. Smyth 1963:431.
[ back ] 37. Kiparsky 1968:54–55.
[ back ] 38. Wackernagel 1926:7.
[ back ] 39. Benveniste 1946/1966; cf. also 1956a/1966, 1958/1966.
[ back ] 40. Watkins 1962a:90–96.
[ back ] 41. Cf. Thumb 1910/1964:170–171.
[ back ] 42. Löfstedt 1956a and 1956b (especially 1956a:xiii-xxv).
[ back ] 43. Wackernagel 1926:113.
[ back ] 44. Schwyzer/Debrunner 1950:6n4.
[ back ] 45. Schwyzer/Debrunner 1950:9–10.
[ back ] 46. Schwyzer/Debrunner 1950:10.
[ back ] 47. Schwyzer/Debrunner 1950:12.
[ back ] 48. Schwyzer/Debrunner 1950:12.
[ back ] 49. Schwyzer/Debrunner 1950:17.
[ back ] 50. Schwyzer/Debrunner 1950:20–21.
[ back ] 51. Schwyzer/Debrunner 1950:21.
[ back ] 52. Schwyzer/Debrunner 1950:23.
[ back ] 53. Schwyzer/Debrunner 1950:23n1.
[ back ] 54. Schwyzer/Debrunner 1950:21n8.
[ back ] 55. Schwyzer/Debrunner 1950:21n11.
[ back ] 56. Schwyzer/Debrunner 1950:24.
[ back ] 57. Schwyzer/Debrunner 1950:24.
[ back ] 58. Schwyzer/Debrunner 1950:24.
[ back ] 59. Schwyzer/Debrunner 1950:56.
[ back ] 60. Schwyzer/Debrunner 1950:61.
[ back ] 61. Schwyzer/Debrunner 1950:61n2.
[ back ] 62. Schwyzer/Debrunner 1950:59.
[ back ] 63. Schwyzer/Debrunner 1950:67.
[ back ] 64. Schwyzer/Debrunner 1950:67.
[ back ] 65. Schwyzer/Debrunner 1950:70.
[ back ] 66. Schwyzer/Debrunner 1950:71.
[ back ] 67. Schwyzer/Debrunner 1950:81.
[ back ] 68. Chomsky 1965:136, 198–199.
[ back ] 69. Schwyzer/Debrunner 1950:90.
[ back ] 70. Schwyzer/Debrunner 1950:90.
[ back ] 71. Schwyzer/Debrunner 1950:100.
[ back ] 72. Schwyzer/Debrunner 1950:119.
[ back ] 73. Schwyzer/Debrunner 1950:138.
[ back ] 74. Schwyzer/Debrunner 1950:139.
[ back ] 75. Benveniste 1954/1966:289.
[ back ] 76. Benveniste 1954/1966:296.
[ back ] 77. Benveniste 1954/1966:297–298.
[ back ] 78. Frisk GEW II 579.
[ back ] 79. Benveniste 1954/1966:298.
[ back ] 80. Cf. Palmer 1950, 1956; also Chantraine 1968:284.
[ back ] 81. Cf. Frisk GEW I 659.
[ back ] 82. See the previous note.
[ back ] 83. Schulze 1966/1885:50.
[ back ] 84. Ruijgh 1957:37–38.
[ back ] 85. Ruijgh 1957:38.
[ back ] 86. Nilsson 1921:13–14.
[ back ] 87. Nilsson 1921:16.
[ back ] 88. The terms “theme I” and “theme II” stem from Benveniste 1935:155.
[ back ] 89. Benveniste 1935:18.
[ back ] 90. Thieme 1952a:5–6.
[ back ] 91. Thieme 1952a:15.
[ back ] 92. Schmitt 1961:88.
[ back ] 93. Schmitt 1967:190n1131.
[ back ] 94. Palmer 1963a:90–91.
[ back ] 95. Forbes 1958:180.
[ back ] 96. Forbes 1958:180.
[ back ] 97. Forbes 1958:182.
[ back ] 98. Forbes 1958:182.
[ back ] 99. Forbes 1958:180.
[ back ] 100. Chantraine 1968:vii.
[ back ] 101. Chantraine 1968:viii.
[ back ] 102. Chantraine 1968:ix.
[ back ] 103. Chantraine 1968:ix.
[ back ] 104. Chantraine 1968:vii-viii.
[ back ] 105. Benveniste 1951/1966:316–317.
[ back ] 106. Havers 1946:51 and Benveniste 1949/1966:311.
[ back ] 107. Benveniste 1962:126–131.
[ back ] 108. van Brock 1959:119.
[ back ] 109. van Brock 1959:125ff.
[ back ] 110. Janko 1992:339.
[ back ] 111. Whitman 1958:136.
[ back ] 112. Whitman 1958:137.
[ back ] 113. Whitman 1958:345n50.
[ back ] 114. Whitman 1958:200.
[ back ] 115. Cf. Frisk GEW I 124–5.
[ back ] 116. Whitman 1958:201–202.
[ back ] 117. Keilschrifturkunden aus Boghazköi XIX 5 Vs 8; brief discussion in Page 1959:24n4.
[ back ] 118. For a collection of arguments for the existence of Hittite/Greek contacts, supplemented with bibliography presenting the opposing arguments as well, see Harmatta 1968.
[ back ] 119. Benveniste 1964:34.
[ back ] 120. On the ordinals, see Benveniste 1948:114–168; for more on the syntax of the ordinal, with significant Greek examples included, see Watkins 1965:287–297.
[ back ] 121. See especially Latte 1924:143.
[ back ] 122. Cf. Hofmann 1950:86 and Risch 1954a:184.
[ back ] 123. On the probative value of shared innovation vs. shared retention, see especially Adrados 1952.
[ back ] 124. Meillet 1935/1965:183.
[ back ] 125. See especially Horrocks 1997:214.
[ back ] 126. See especially Wachter 2000:64n4.
[ back ] 127. Chadwick 1963:9.
[ back ] 128. Chadwick 1963:9. For a list of correspondences linking “Mycenaean” and Arcado-Cypriote, see Vilborg 1960:20–21. For a list of correspondences linking Arcadian and Cypriote, see Vilborg 1960:22–23.
[ back ] 129. Risch 1966:157.
[ back ] 130. Risch 1966:150.
[ back ] 131. Risch 1966:157.
[ back ] 132. Weingarten 1983.
[ back ] 133. Vilborg 1960:21.
[ back ] 134. Vilborg 1960:21–22.
[ back ] 135. Palmer 1963b:60–64.
[ back ] 136. Risch 1955:70–71.
[ back ] 137. Cf. Buck 1955:152; also Thumb and Scherer 1959:18.
[ back ] 138. For the list, see again Vilborg 1960:21–22.
[ back ] 144. Buck 1955:7.
[ back ] 145. Buck 1955:154.
[ back ] 146. Risch 1955:70–71.
[ back ] 147. Cowgill 1968:182.
[ back ] 148. Buck 1955:147.
[ back ] 149. Ruipérez 1955:166–167, in disagreement with Risch 1955:72.
[ back ] 150. Benveniste 1956b:263.
[ back ] 151. Palmer 1963b:61.
[ back ] 152. Kiechle 1960.
[ back ] 153. Kiechle 1962.
[ back ] 154. For references to such attempts, see Risch 1958:91n1; also Cowgill 1966:86.
[ back ] 155. As we see from the scholia for Apollonius of Rhodes 2.1005; cf. Latte 1924:151–152.
[ back ] 156. Cf. Ruijgh 1957:68.
[ back ] 157. Ruijgh 1957.
[ back ] 158. Risch 1958.
[ back ] 159. For the form, see Bechtel 1921:360.
[ back ] 160. For the form, see Masson 1961:284.
[ back ] 161. Risch 1958:92.
[ back ] 162. Chantraine 1958:306.
[ back ] 163. On this tendency, see Meister 1921:10–27.
[ back ] 164. For an analysis of this innovation, see Palmer 1963a:83–84.
[ back ] 165. Cf. Meister 1921:18.
[ back ] 166. Chantraine 1958:488.
[ back ] 167. For an analysis of the whole text, see Ventris and Chadwick 1956:285.
[ back ] 168. Risch 1966:157.
[ back ] 169. Cf. Frisk GEW II 509; also already Hamp 1960:200. See now Janko 1992:303.
[ back ] 170. For commentary on the morphological variants, see Szemerényi 1966:34.
[ back ] 171. Cf. Chantraine 1958:507–508.
[ back ] 172. Cf. Thumb and Scherer 1959:113, 126.
[ back ] 173. Ruijgh 1957:7.
[ back ] 174. Here I disagree with Meillet 1935/1965:173.
[ back ] 175. Further discussion in Nagy 1970:120–122.
[ back ] 176. Meillet 1935/1965:171.
[ back ] 177. Meister 1921:170.
[ back ] 178. Meister 1921:164.
[ back ] 179. Cf. Hoekstra 1957:202.
[ back ] 180. Cf. Meister 1921:168–169.
[ back ] 181. Cf. also Frisk GEW I 721–722.
[ back ] 182. Allen 1968a:84–85.
[ back ] 183. Cf. Buck 1955:28–29.
[ back ] 184. Further discussion in Nagy 1970:150–151.
[ back ] 185. This formulation is different from the one offered by Palmer 1963a:89.
[ back ] 186. Cf. Meister 1921:256.
[ back ] 187. Cf. Coleman 1963:89–90.
[ back ] 188. Cf. Chantraine 1958:173–178.
[ back ] 189. Risch 1937:115.
[ back ] 190. Meillet 1935/1965:172.
[ back ] 191. Cf. e.g. Palmer 1963a:105–106 and 175.
[ back ] 192. Morpurgo-Davies 1964:157n2.
[ back ] 193. For a particularly useful bibliographical summary of source-material, see Thumb and Kieckers 1932:13–47 (“Die Quellen der griechischen Dialekte. Literarische Hilfsmittel”).
[ back ] 194. Cowgill 1966:78. By “tribal” boundaries, Cowgill has in mind the boundaries that separate, say, Dorians from Aeolians from Ionians.
[ back ] 195. Convenient summary by Coleman 1968:169–170.