The modal particle in the dialects of archaic and classical Greek has five forms, of which one is a phonologically conditioned allomorph. The dialect division of the forms is as follows:

ἄν

Attic, Ionic, and Arcadian; and Homeric epic

κε

Lesbian, Thessalian (Pelasgiotis, Thessaliotis), Cypriot; and Homeric epic

κεν

‘eastern Aeolic’, i.e. Lesbian and literary Lesbian; and Homeric epic

κᾱ

West Greek dialects, but never in ‘high’ poetry; Boeotian, Thessalian (Histiaeotis)

κ’

optional prevocalic form of κε and (apparently) κᾱ

In spite of the morphological variety, the function of the particle across the dialects seems more or less identical.Footnote ¹ The occasional trivial differences in usage which are attested are no greater than the sort of variation which occurs in modal grammar in other languages, and there are some developments over time. There is more flexibility in the usage of the particles in the language of Homeric epic, which is peculiar in many respects: it is oral formulaic poetry which may have been composed as early as the eighth century BC, and the stock of traditional language which the monumental composer(s) inherited includes features of Greek which reflect periods earlier than the period of composition (in a few rare cases going back as early as the late Bronze Age, the Mycenaean period). In addition to features which reflect different periods and different regions of the Greek world, epic language also contains features (phonological, morphological and lexical) which are ‘artificial’ in the sense that they were innovated by and within the poetic tradition and were never features of the spoken language.Footnote ²

There are a number of peculiarities in the form and distribution of the five forms above. On the whole the form of the particle correlates reasonably neatly with the traditional dialect groups: ἄν in Attic-Ionic, κᾱ in West Greek, and κε in two of the three ‘Aeolic’ dialects. That Boeotian has κᾱ is not very surprising, though it is traditionally grouped with Lesbian and Thessalian in the Aeolic group: this dialect is heavily overlaid with West Greek features, to the extent that even the classical dialectologists of the early twentieth century, who relied more heavily on a Stammbaum model than modern linguists, regarded it as a ‘mixed’ dialect.

The three dialect groups are (more or less) inherited from the ancient Greek world, where they were correlated with ethnic subdivisions among the Greeks. A fourth group, Arcado-Cypriot, is a modern (twentieth-century) linguistic grouping; if the Greeks had views on the dialectal affiliations of Arcadian and Cypriot, they have not survived. Of the attested dialects they seem to have the closest affinity to Mycenaean Greek, a form of the language written in the Linear B script on clay tablets from the late Bronze Age, approximately 1400–1200 BC. The putative Bronze Age ancestors of Arcadian and Cypriot are usually grouped with Mycenaean in a dialect grouping (sometimes labelled ‘Achaean’) assigned to the late Bronze Age Peloponnese, before the collapse of Mycenaean palatial civilisation around the end of the thirteenth century. There are few sentences in the surviving Linear B tablets (which are lists of inventory and economic records from the bureaucracies of the Mycenaean palatial centres), and almost no complex sentences: as a result there are no clearly attested instances of the particle in the tablets.Footnote ³ Arcadian, Cypriot and Mycenaean are East Greek dialects: they share the assibilation of -ti to -si in verbal endings (δίδωτι > δίδωσι) and certain other common features which mark them off from West Greek. Arcadian shares ἄν with Attic-Ionic, the other East Greek dialect grouping, while Cypriot shares κε with the Aeolic dialects Lesbian and Thessalian.

The Aeolic dialects do not fit easily into the East/West classification, since the status of the descriptor ‘Aeolic’ is unclear and disputed. The three dialects share a rather small number of overlapping features, which do not allow clear conclusions about the historical affiliations of a supposed ‘proto-Aeolic’ dialect group in the Bronze Age. Some have made this an eastern group, and others a western group;Footnote ⁴ and since the 1970s a number of scholars have questioned the retrojection of Aeolic to the Bronze Age, arguing that the dialects are (either largely or entirely) a post-Mycenaean development which acquired their distinctive features through contact and areal development.Footnote ⁵

The difference between the Arcadian and Cypriot forms of the particle is not particularly surprising. Even after the decipherment of Linear B, we have a very limited insight into the varieties of the spoken language in late Bronze Age Greece. If we suppose that the Greek language was taken to Cyprus by Mycenaean mercantile activity and migration at the end of the Bronze Age,Footnote ⁶ which would explain the similarities of the Cypriot dialect with Mycenaean and Arcadian, there is nevertheless no reason to assume (and it would be remarkable coincidence) that the variety which prevailed in Cyprus was exactly similar to the one which gave rise to Arcadian. Linear B texts have been discovered across the Peloponnese, and elsewhere. The Peloponnese alone is a large and geographically challenging area; although the language of the Linear B tablets lacks regional variation (its uniformity suggests a chancellery language), it would be surprising if the Mycenaean Greek of Messenia in (say) the mid-thirteenth century was identical to that spoken in the Argolid (Mycenae and Tiryns).

More striking is the confinement of κεν to eastern Aeolic, and the conditions under which it alternated with κε, which, in both Thessalian and Cypriot, could be reduced to κ’ before a vowel (which makes it clear that κεν is not a simple pre-vocalic form of κε). The particle κεν is found in the poetry of Sappho and Alkaios; it is also found in Homeric epic, where it is regarded as an ‘Aeolicism’.Footnote ⁷ However, even in Sappho and Alkaios κεν looks like a literary form. It co-exists with, and is outnumbered by (in a ratio approximately 2:3), κε and κ’. It occurs before vowels, where it breaks hiatus and adds a short syllable to the line, and before consonants, where it adds a long syllable to the line.Footnote ⁸

Non-poetic inscriptions are generally recognised as the best indicators of whether a form or usage is vernacular in a particular dialect. The data for κεν from the Lesbian-speaking area (Lesbos and the adjoining region of Asia Minor) do not support the assumption that κεν was a feature of the spoken language. There are just five instances of κεν in surviving inscriptions, as against over one hundred instances of κε. Of the five instances of κεν, only two occur in inscriptions before 300 BC:

1. (1) Hodot myt 05 (ZPE 63 (1986) 109; Mytilene), line 2

   ὤς κεν οἰ πόλιται, ‘in order that the citizens …’ (but αἰ δέ κε ἄλλον τινά ibid. 16)

2. (2) Hodot nas 01 (IG xii/2.645, Buck 27; Nesos), line 51

   τῶγ κεν εὐεργέτη τὰμ πόλιν, ‘of the services he renders to the city’ (but ὄτα κε ἀ πόλις ibid. 33)

It is not obvious why κεν was used in these two cases, since both inscriptions also contain numerous instances of κε, including before a vowel. Both the syntactic and phonological contexts can be paralleled in sequences which have κε. If κεν reflects literary Lesbian, its use in a public inscription may have been a marker of local pride and local identity. Both of these texts reflect a crisis in the region provoked by Macedonian expansion in the 330s and 320s. It could be argued that in (1) the particle, which occurs in the preamble giving the reason for the decree (attempt at national reconciliation after a crisis), is in a far more salient clause than the four instances of κε in the text: ‘… in order that the citizens inhabit the city under a democratic government perpetually, being as well-disposed to one another as possible …’ (tr. Heisserer and Hodot (Reference Heisserer and Hodot1986)). In (2) κεν appears in the very final clause of the inscription and may throw weight on to the critical last words ‘of the services he renders to the city’.

The other three instances of κεν occur in two inscriptions which are clearly written in an artificial revival Lesbian. Both are from Kyme:

1. (3) Hodot kym 014 (IvK v.13), c. 130 BC, line 18

   καθότι κεν Ἀρχίππα πρoαιρῆται, ‘in accordance with the preferences of Archippe’Footnote ⁹

2. (4) Hodot kym 016 (IvK v.19, Buck 28), c. first decade of the Common Era, lines 30 and 51

   ἐν πάντεσσι τoῖς ἀγώνεσσιν, oἵς κεν ἁ πόλις συντελέη (line 30), ‘in all the festivals which the city organises’

   καὶ ἐντάφην, ἐν ὧ κεν ἂν εὔθετoν ἔμμεναι φαίνηται τόπω, ‘and be buried there, in whichever place seems most appropriate’ (line 51)Footnote ¹⁰

If κεν is a feature of poetic language, there are two possibilities: either it is an archaism lost from the spoken language, or it is a creation of the epic tradition. If it is an archaism, loss of the final -n in the vernacular form κε needs to be explained. It might be argued that the -n was reanalysed as a moveable n (nu ephelkystikon) and omitted before a following consonant: the form κε was subsequently generalised. There are two objections to this. First, moveable n was a feature of Ionic, but not of vernacular Lesbian, though it was used as a metrical device by Sappho and Alkaios (as by poets in the Doric tradition, including Alkman and Stesichoros). Secondly, it is hard to see why the -n would be generalised away before a following vowel, if the reason for extending it from verbs and nouns in the first place was precisely to avoid hiatus.

As a working hypothesis, therefore, I shall assume that κεν was a creation of a poetic tradition: this was presumably the eastern Aegean epic tradition, which we know as Homer, and which in some form or another must have been in interaction with poetic traditions (epic or monodic) of Lesbos and eastern Aeolis.Footnote ¹¹

If κεν is a Greek poetic creation, then the interesting and influential theory of Forbes (Reference Forbes1958) cannot be exactly correct in its original form. Forbes sought to give a phonological and morphological history of the dialect forms of the particle, which took them all back to an original form *kn̥/*ken (where *kn̥ is the normal zero-grade form of the morpheme, and *ken the e-grade form).Footnote ¹² Zero-grade *kn̥ would give Greek κα and possibly (by the Sievers-Edgerton ‘law’) καν before a vowel.Footnote ¹³ Early Greek would then, by normal sound-change, have κεν, and κα/καν; Forbes suggested that the analogy of κα/καν produced κε (out of κεν). It is hard to see how this analogical process could happen outside of poetic diction, however. κα is a feature of West Greek, and there are no grounds for supposing the type of extended, early contact between κα dialects and κε(ν) dialects which might have led to the creation of κε in Thessalian, Lesbian and Cypriot. Moreover, *kn̥ would give a short vowel in κα: the attested form κᾱ would have to be accounted for separately, and is a weak analogical model for κε.Footnote ¹⁴

Forbes suggested that Attic, Ionic and Arcadian ἄν was the result of a resegmentation of the sequence οὐ καν as οὐκ ἄν. This, in itself, is perfectly plausible. The form *καν is not generally recognised as a form of the potential particle, but Forbes drew attention to a peculiar sequence of particles in three incriptions from Tegea in Arcadia:

1. (5) IG v/2.3 (Buck 18, Dubois (Reference Dubois1986) ii.20–34), line 16

   ΕΙΚΑΝ διελαυνόμενα τύχε̄, ‘if they are being driven through’

2. (6) IG v/2.6 (Buck 19, Dubois (Reference Dubois1986) ii.39–61), line 2

   ΕΙΚΑΝ τι γίνητοι, ‘if anything occurs’

3. (7) Buck 22 (Dubois (Reference Dubois1986) ii.61–77), line 34

   ΕΙΚΑΝ τι αὐτοῖς ἐ[π]απύλογον ἦι, ‘if there is a defence case against them’

There are eight instances in total in these three inscriptions, which date to the late fifth and fourth centuries BC. Forbes (followed by Dubois and others) saw the potential particle καν in this sequence, which she divided εἴ καν. This seems prima facie quite unlikely, given that the same inscriptions also present sequences εἰκ ἐπί (one instance) and εἰ δ’ ἄν (twenty instances), which point to a division εἰκ ἄν rather than εἴ καν for (5)–(7):

1. (8) IG v/2.3 (Buck 18, Dubois (Reference Dubois1986) ii.20–34), line 21

   εἰκ ἐπὶ δō̂μα πῦρ ἐποίσε[.], ‘if one brings fire near the temple’

2. (9) IG v/2.3 (Buck 18, Dubois (Reference Dubois1986) ii.20–34), line 2

   εἰ δ’ ἂν καταλλάσσε̄, ‘if one makes any change’

It seems clear that Arcadian has merely innovated εἰ/εἰκ on the analogy of οὐ/οὐκ, where the final -κ prevents hiatus before a following vowel.Footnote ¹⁵

Forbes suggested that the putative Greek form καν was cognate with forms in Hittite and Sanskrit. Specifically, she drew attention to the Homeric collocation νύ κε(ν) in sentences such as Iliad 5.311–12:

1. (10)

   καί νύ κεν ἐνθ’ ἀπόλοιτο ἄναξ ἀνδρῶν Αἰνείας,

   εἰ μὴ ἄρ’ ὀξὺ νόησε Διὸς θυγάτηρ Ἀφροδίτη

   And there would Aeneas, leader of men, have perished,

   had not Aphrodite daughter of Zeus quickly perceived him.

She compared the Sanskrit sequence nú kam and the Hittite nu-kan. The Sanskrit particle kam is a vague asseverative ‘indeed, yes, well’ in Vedic sentences such as the following:

1. (11)

   vidád gávyaṃ sarámā dṝḻhám ūrváṃ yénā nú kam mā́nuṣī bhójate víṭ

   Saramā found the cattle's firm-built prison whereby nú kam the race of man is still supported.

   (Rig Veda 1.72.8)

In Hittite, nu is a clause connector which stands at the head of a sequence of particles; kan stands last in a sequence of particles, so nu and kan will appear next to each other only if there are no other elements in the sequence. The meaning and function of kan are not completely understood, however. It is particularly associated with verbs and adverbs of movement or location; it may also (Hoffner (Reference Hoffner and Carruba2006)) intensify the telic force of a verb. In (12) the verb šamen- ‘withdraw, pass by’ becomes ‘do without, forfeit’ with kan:Footnote ¹⁶

1. (12) Law 30 (OH/NS): Hoffner (Reference Hoffner1997) 39–40

   takku LÚ-š=a DUMU.MUNUS naui dāi | n=an=za mimmai | kūšata=ma kuit piddāit | n=aš= kan šamenziFootnote ¹⁷

   But if a man has not yet taken a girl and refuses her, the bride price which he paid he forfeits.

Here, n=an=za is nu- [connective] + an [enclitic pronoun acc. ‘her’] + za [morphosyntactic middle marker], while n=aš=kan is nu- [connective] + aš [enclitic pronoun ‘he’] + kan. In spite of the superficial similarity of the collocation with nu, a historical connection between Greek κεν and the Sanskrit and Hittite words seems extremely unlikely. A serious linguistic objection to the Hittite parallel is that the marking of an apodosis with nu seems to be a feature of later Hittite texts (i.e. it is a development within the history of Hittite).Footnote ¹⁸

On this scenario, Greek inherited *nu-km̥ and *nu-kem; *nu-kem, ancestor of the Homeric form, became νύ κε by analogy while κεν speakers were in contact with καν/κα speakers. This process must presumably have been complete by the start of the late Bronze Age (c. 1500 BC), to allow for the contact-induced innovation κε not only to spread but to become the dominant variant. But it cannot have been too early, because the sound change *km̥ > κα must have already happened for the analogy καν : κα :: κεν : x (x → κε) to work.Footnote ¹⁹ It might be argued that the innovation κεν → κε was influenced instead by the ancestor of Attic-Ionic, since the scenario supposes that Attic and Ionic (like West Greek) derive from a *km̥ dialect. However, early Attic-Ionic speakers would need to be isolated relatively early after the (supposed) change *km̥ > κα, καν, or the innovated form ἄν from οὐκαν would be expected in other *km̥ dialects (i.e. in West Greek dialects); and furthermore, the development of ἄν presupposes a dialect which had in fact generalised καν and discarded κα.

Starting with οὐκ, ἄν and the new possibility of εἰκ, I suggest a split development in the dialects along the same lines as Forbes, except that the inherited particle ἄν was retained (on this scenario) by Attic-Ionic and Arcadian; Lesbian, Thessalian and Cypriot ended up with κε/κ’ as the result of the reanalysis of εἰκ ἄν and οὐκ ἄν in conjunction with a phonetic variant κε of the particle τε < *k^we. Dunkel and others have suggested that the way to explain the first-millennium diversity is to reconstruct multiple forms *ke, *ken (from *kem) and *an.Footnote ²⁰ West Greek κᾱ is explained by analogy (e.g. adverbial forms in -ā). This is perfectly possible, though it ignores the interesting coincidence that all forms of the particle either begin or end with an /n/ or a /k/, i.e. precisely those consonants which were used to prevent hiatus in Greek. I shall argue that ἄν was an inherited form (rather than back-formed within Greek from *kn̥), and conversely that the k-forms are the result of a Greek mixture of reanalysis of word-final -k and the vestiges of an old topicalising/conditional force of the IE particle *k^we (which appears elsewhere as τε).

My starting point is an article by Cowgill in Reference Cowgill1960, which set out to demonstrate that the Greek negative οὐ(κ) has its roots in an IE phrase *ne H₂oyu (k^wid) ‘never’, composed of *ne ‘not’, an adverbial accusative * H₂oyu ‘ever’, and an optional indefinite pronoun *k^wid. The form of the negative in Greek is otherwise a puzzle: the well-attested IE *ne is not found in Greek, and Greek οὐ(κ) has no obvious etymology. A third form οὐκί is found in Homer: Cowgill started from the assumption that the prevocalic οὐκ is elided from οὐκί, and that this form contains the neuter indefinite pronoun *k^wid as a reinforcing particle (the root of *H₂óyu ‘life, age’ appears in αἰών, αἰεί from *αἰϝ-, i.e. *H₂eyu-).Footnote ²¹ In Greek, οὐ and οὐκί were reduced to preconsonantal οὐ and prevocalic οὐκ.Footnote ²² The phonotactic rules of Greek do not generally allow any consonant in word-final position except -ν, -ρ, -ς. For this reason the form οὐκ’/οὐκ is likely to have been marked as a prevocalic variant (originally at least), licensed perhaps by the prevocalic perfect tense variant ἔθηκ’, ἔδωκ’ etc.Footnote ²³

An adverbial particle ἄν in early varieties of Greek may have been a potential or indefinite marker which became increasingly restricted to situations which depended on the prior achievement of some condition, and clauses with a verb in the subjunctive mood, including the indefinite relative.Footnote ²⁴ Greek also inherited a particle *k^we which presents a range of overlapping functions in the IE languages, including conjunction, coordination and subordination. In addition to its semantic force (and here it differs from a marker of potentiality), it also plays a role in the ordering and presentation of discourse.Footnote ²⁵ It is familiar in Greek as the connective and ‘epic’ τε, and (if my argument is right) it was also a formal and functional ingredient in the attested k-forms of the modal particle. In the emergent Greek Sprachbund the dialects in contact formed, with some morphological variation, a distinct subordinating conjunction ‘if’ (εἰ, αἰ and ἠ in the dialects), and a specialised modal particle which developed a similar range of functions as the syntax of conditional and indefinite relative clauses took on a distinctively Greek shape.

It is not possible to reconstruct a word ‘if’ in Indo-European (the word is innovated in IE languages which have it). There are, however, some commonalities in conditional syntax across the languages. Cross-linguistically, conditional grammar can be effected without additional marking by pragmatic or discourse-structuring devices, of which juxtaposition is a straightforward example:

1. (13)

   πρεσβύτερός τις τῆι θύραι

   ὑπακήκο’· εὐθὺς πατέρα καὶ πάππα[ν λέγω].

   γραῦς· μητέρ’.

   An older man opens the door: right away [I say] ‘Father’ and ‘Grandad’. An old woman – Mother’.

   (Men. Dys. 493–5)

2. (14)

   Negat quis, nego; ait, aio

   Somebody denies something, I deny it; they affirm it, I affirm it.

   (Ter. Eun. 252)Footnote ²⁶

Here the first clause sketches the situation or context in which the second clause (main clause or apodosis) will be valid. The similarity between a condition (protasis) and a topic was noted by Haiman during the development of discourse analysis as a specialised field in the 1970s.Footnote ²⁷ The examples above are asyndetic; in English and other languages a frequent connector between the ‘topic’ (condition) and the main clause is a word meaning ‘and’. Greek (like Latin and English) uses and when the protasis has the form of an imperative:Footnote ²⁸

1. (15)

   αἰτεῖτε καὶ δοθήσεται ὑμῖν

   Ask and it shall be given to you.

   (Matthew 7:7)

In this case the ‘left-subordinating and’ has the effect of making the imperative ‘ask’ the structural equivalent of a subordinate clause.Footnote ²⁹ Blakemore (Reference Blakemore1987) argues persuasively that discourse connectives such as and are procedural rather than representational words: they point the hearer to the context and indicate the inference (how to take the representational elements).Footnote ³⁰

In (16) and (17) the connective introduces the subordinate clause in a more straightforward way. In early Latin apsque could link a condition to a principal clause:

1. (16)

   nam apsque te esset, hodie numquam ad solem occasum uiuerem

   if it hadn't been for you, I would never have lived till sunset today

   (Plaut. Men. 1022–3)

Klein and Condon (Reference Klein and Condon1993) have shown that Gothic nih from *ne k^we has a subordinating value on four occasions in the Gospel of John, as in the following example:

1. (17)

   nih wesi [= εἰ μὴ ἦν] sa fram guda, ni mahtedi [= οὐκ ἐδύνατο] taujan ni waiht

   if he were not from God, he would not have been able to do anything

   (John 9:33)

Latin neque/nec does not have this subordinating function, though it adds a negative condition in early Latin:

1. (18)

   Si intestato moritur, cui heres nec escit, adgnatus proximus familiam habeto.

   If a man dies intestate, and if he does not have an heir, the nearest agnate kinsman shall have possession of his household.

   (Twelve Tables, 5)Footnote ³¹

Gonda, in his critical review of Ruijgh's (Reference Ruijgh1971) Autour de TE épique, noted that the Sanskrit particle ca ‘and’ < IE *k^we can in Vedic also mark a conditional clause:Footnote ³²

1. (19)

   imāṃ ca vācaṃ pratiharyathā naro viśved vāmā vo aśnavat

   and if you gladly receive this speech, o noble men [= gods], it will attain all things of yours worth winning

   (Rig Veda 1.40.6; tr. Jamison and Brereton (Reference Jamison and Brereton2014))

He compared this to Iliad 1.218, ‘where the protasis is virtually conditional and te no doubt helps to combine the two parts of the sentence into a whole’:

1. (20)

   ὅς κε θεοῖσ’ ἐπιπείθηται, μάλα τ’ ἔκλυον αὐτοῦ

   Whoever obeys the gods, readily do they listen to him.

Patri has added data on *k^we from Slavic to the discussion, and in a wide-ranging analysis of the function of the particle in Indo-European, argues persuasively that the appearances of the particle across the various languages can be understood by supposing an early unitary value: at the head of a phrase it introduces a condition, or topic, and in other places in the phrase it is a connector (more or less ‘and’).Footnote ³³ A plausible case can (in this context) be made for traces of *k^we in the conditional grammar of Greek.