I'm going to start by putting up those Homeric endings again; we'll be treating Homer, rather than proto-Greek as a starting point—because the dialects we're working with will allow that. (Epic Greek is a mishmash of a lot of dialects, but its core is Ionic; and Attic is a deviant sibling of Ionic.)
|Nom||εύς eús||ῆε ɛ̂ːɛ||ῆες ɛ̂ːes|
|Gen||ῆος ɛ̂ːos||ήοιν ɛ́ːoin||ήων ɛ́ːɔːn|
|Dat||ῆι ɛ̂ːi||ήοιν ɛ́ːoin||εῦσι eûsi|
|Acc||ῆα ɛ̂ːa||ῆε ɛ̂ːɛ||ήᾱς ɛ́ːaːs|
|Voc||εῦ eû||ῆε ɛ̂ːɛ||ῆες ɛ̂ːes|
- For one thing, the Homeric table has a whole lot of vowels next to other vowels. That's hiatus: Greek is normally not comfortable with hiatus, and proto-Greek vowel pairs usually merge ("contract") to a new vowel or diphthong. In fact, hiatus in Classical Greek is usually a sign that there used to be a digamma there that has dropped off; while the /w/ was there, it prevented contraction. That's exactly what has happened with these endings: *-ɛːwos > -ɛːos. But Greek kept contracting vowels after Homer, and we will see some of that on the way to Attic.
- The other thing that may strike you as odd is that it's an eta preceding another vowel. Some vowel pairs do make it into Attic: contraction is not universal. But pairs like ηε, ηο, ηα don't show up in words of Attic: they scream out Homer. That suggests that something has happened to eta in pairs in particular, since that eta has not survived into Attic.
I'm going to start with the second observation. There is a rule that long vowels are shortened before another vowel (if they survive contraction). This doesn't happen in all dialects, and it doesn't happen to all vowels; but it does happen in Attic and Ionic with η; for example, Homeric ἠώς /ɛːɔ́ːs/ corresponds to Attic ἔως /éɔːs/.
We can use that rule, on its own, to explain what has happened in Ionic. It's not the only possible explanation, but it will serve for this exposition:
We can explain this table almost entirely by shorting the Homeric /ɛː/ systematically, whenever it occurs in front of another vowel (which, for this declension, is always). There are a couple of peculiarities:
- The table has a long e, /êːs/, where we would expect two short e's, /ees/. That's a new contraction, and I don't think anyone should be surprised at that contraction. The suffix has a circumflex on the final syllable, which is usually a giveaway that contraction has happened.
- The table also has a long e in the dative singular, /êː/, where we would expect /éi/. That's also a contraction (as the circumflex tells you), and the contraction shouldn't be a surprise either. In fact, the shortened /e/ occasionally turns up in this declension in Homer already, which as a result does feature the expected /éi/ -έϊ.
The contraction of /ei/ to /eː/ was early: early enough, that <ei> is actually how you write /eː/ in Ancient Greek: ει. Because /ei/ turned into /eː/, you need historical linguistics, or pre-Classical texts, to tell whether an instance of ει reflects an earlier /ei/ ("genuine diphthong"), or started out as /eː/ ("spurious diphthong").
So we've been able to explain Ionic reasonably easily as well. There's some features that the textbooks are sweeping under the carpet to simplify things, and make the dialects look purer than they are. For instance, the grammars don't highlight the fact that you'll occasionally see short epsilon in Homer; they're describing internally consistent dialect tables, that are abstracted from the messier dialect mixes of texts. But this is a clean table.
The table for Attic is not as clean, which is why you have to go via the cleaner Homeric and Ionic tables to make sense of it:
It looks like the Ionic table, but it's not quite the same. In fact, while the Ionic table is explained by a systematic shortening of /ɛː/, the Attic table shows that the shorting was not systematic.
Let's start with the genitive singular, which is /éɔːs/ -έως, where Epic had /ɛ̂ːos/ -ῆος and Ionic had /éos/ -έος. That -εως suffix turns up as a genitive elsewhere in Attic, with both penult and antepenult stress; so Attic λύσις, λύσεως, corresponding to Ionic λύσις, λύσιος. But there's something very wrong about the accent of λύσεως /lýseɔːs/. Greek is not supposed to allow stress on the antepenult if the ultima is long (the three-mora rule): ἄνθρωπος /án.tʰrɔː.pos/, ἀνθρώπου /an.tʰrɔ́.poː/. λύσεως ends in a long syllable; but it is accented on the antepenult.
So λύσεως is accented as if it ends in a short /o/, and it is a case which in proto-Greek (and other dialects) ends in a short /o/. The conclusion is that genitive /eɔːs/ comes from the Old Ionic /ɛːos/, and (in those other declensions where it is stressed on the antepenult) is still accented like the Old Ionic /ɛːos/.
We just saw that Ionic shortened /ɛːos/ to /eos/. Attic is also shortening /ɛː/ in the suffix—but it seems to be lengthening the following /o/, in compensation. This is called quantitative metathesis: the longness of the vowel ("quantity") is being swapped ("metathesis") from the first vowel to the second; and Attic does it routinely for /ɛː/ before /o/. In fact, it does so regularly enough, that it gave rise to a new flavour of the second declension:
- Proto-Greek *λᾱός */laːós/, Doric λᾱός /laːós/, Epic λᾱός /laːós/, Ionic ληός /lɛːós/, Attic λεώς /leɔ́ːs/, Koine λᾱός /laːós/, Modern λαός /laˈos/
- Proto-Greek *νᾱός */naːós/, Doric νᾱός /naːós/, Epic νηός /nɛːós/, Ionic νηός /nɛːós/, Attic νεώς /neɔ́ːs/, Koine νᾱός /naːós/, Modern ναός /naˈos/
So second-declension nouns, which in Greek end in a short /o/, could end in Attic in a long /ɔː/—provided there was compensatory lengthening quantitative metathesis from a preceding /ɛː/. This was an oddity specific to Attic; and although Koine normally went along with Attic, this was an oddity that Koine smoothed over.
Compensatory lengtheningQuantitative metathesis means that Attic didn't just shorten all /ɛː/, like Ionic did. It lengthened the following syllables as well, where it could. If we do quantitative lengthening metathesis to the Homeric table, we end up with the following:
That's close to the actual Attic table. If we do some contractions, it becomes even closer. Attic liked to contract; and even if there used to be a digamma there, Attic would contract two e's together if it caught them. Remember that the aorist *ἔϝεργον /éwerɡon/ became ἔεργον /éerɡon/ in Epic, but εἶργον /êːrɡon/ in Attic—which could not stand two /e/ in a row. Attic also contracted /eɛː/ to /ɛː/; for example, the subjunctives λύ-ῃς, λύ-ητε /lý-ɛːis, lý-ɛːte/ correspond to the subjunctives *ποιέ-ῃς, *ποιέ-ητε /poié-ɛːis, poié-ɛːte/, which in Attic contract to ποιῇς, ποιῆτε /poiɛ̂ːs, poiɛ̂ːte/. (And remember: a circumflex on the final syllable points to contraction.)
If we contract /eɛː/ to /ɛː/, our mapping to Attic is pretty much done:
Chantraine in his historical morphology reports that inscriptions do contain -έης /éɛːs/ plurals at around 400 BC; but the default plural was contracted -ῆς /ɛ̂ːs/.
There are two exceptions to deal with in the table; Chantraine takes it upon himself to address both. The dative is not uncontracted */éiː/ (έῑ), but contracted /êː/ (εῖ), just as in Ionic; its protoform (as with Ionic) is */ei/ not */ɛːi/. Chantraine says that the short /e/ of */ei/ was an analogy from other cases (eús, éɔːs, éaː); maybe, but I can see /éiː/ contracting to /eː/ anyway, and there were plenty of /eː/ datives in other third declension nouns, to serve as an analogy.
The other exception is the dual genitive/dative, which is not *έῳν /éɔːin/, but έοιν /éoin/—again, as it would have been in Ionic, with no quantitative metathesis. The dual was already dying out in Attic, and was dead in Ionic; that's why I didn't give any Ionic duals above. So Attic can't have borrowed έοιν from Ionic. Chantraine's take on it is cheeky, with the kind of cheekiness that neogrammarians can display: if the manuscripts have a dual that doesn't match the predictions of historical linguistics, then the manuscripts must be wrong.
In particular, Chantraine latches on to a genitive plural βασιλέων in Aescyhlus (Suppliants 327), which refers to two kings: this might have been a correction of the original dual βασιλέῳν. And the 1972 Page edition does indeed present βασιλέωιν in this text: πῶς οὖν τελευτᾶι βασιλέωιν νείκη τάδε; "So how did this strife between kings end up?"
Again, maybe; but again, έοιν duals turn up elsewhere in Attic, and could have served as a pattern of analogy to get rid of the idiosyncratic (and rare) dual. The same ending shows up in the dual for "ships": proto-Greek *νᾱοῖν /naːoîn/, Homeric *νηοῖν /nɛːoîn/ should have generated Attic *νεῷν /neɔ̂ːin/, but instead Thucydides has νεοῖν /neoîn/—bringing it back in line with all the other duals ending in /oin/. And if speakers of Attic didn't do the analogy, then subsequent scribes would: "A dual in -έῳν? That doesn't look right."
(Analogy did not get rid of -εως, because it was just as idiosyncratic, but it was also really really common: speakers were just too used to it. In fact, even when Koine did back away from -εως where it could—λύσις λύσιος instead of λύσις λύσεως, ναός instead of νεώς; but it did not back away from βασιλέως. Soon enough, βασιλέως was pronounced identically to the more normal Ionic βασιλέος, anyway.)
So it looks like Ionic shortened the -η- consistently, while Attic shortened it with quantitative metathesis, lengthening the following vowel. This resulted in several strange endings, and some of them stuck (notably -έως), while for others, the lengthened final vowel went away, and went back to looking like Ionic (-εῖ, έοιν). In a few cases, the metathesis wouldn't make a difference: ῆες and έης both contract to ῆς.
Now, all that explains why Acharnians would be written Ἀχαρνῆς: *Ἀχαρνῆϝες > Ἀχαρνῆες > Ἀχαρνέης > Ἀχαρνῆς, *akʰarnɛ̂ːwes > akʰarnɛ̂ːes > *akʰarnéɛːs > akʰarnɛ̂ːs. But why would Sarantakos have used the Ionic Ἀχαρνεῖς /akʰarnêːs/ instead?
There are a couple of causes to go through. The immediate cause is that Athenians also started using the Ionic Ἀχαρνεῖς, but not at the time of Aristophanes. The Acharnians were written in 425 BC. With the 404 BC spelling reform, Attic acquired the letter Η from Ionic, and started differentiating /eː, e, ɛː/, as the Ionic alphabet did. At that time, the plural was written as -ης. But between 350 BC and 325 BC, the plural changed to being written as -εις, /êːs/ instead. Which happens to be the Ionic plural, derived from */ées/, not /ɛːes/ > /eɛːs/. Later on, this -εις nominative was copied to the accusative as well—by analogy with other third declension nouns (nom. sg. πόλις, ἡδύς, nom. pl. and acc. pl. πόλεις, ἡδεῖς).
Why would the switch have happened? Sihler thinks that the first -e- in */ées/ came in by analogy with the -e- in the other plural cases, /éɔːn, éaːs/. But that means Sihler thinks /ɛ̂ːes/ changed to */ées/; and we don't think it did: we think it had already changed to /éɛːs/ by quantitative metathesis. So that explanation doesn't make sense.
Chantraine takes a broader view of what would have happened: inscriptions around 400 BC had -εης /eɛːs/, Attic poets used -έες /ées/—"possibly an Ionicism"; and then /ées/ contracted into /êːs/. So while Chantraine isn't outright claiming it, he is allowing that /êːs/ is indeed a borrowing from Ionic.
This means that Aristophanes would have written Acharnians as Ἀχαρνῆς, but Athenians a century later were pronouncing and writing it as Ἀχαρνεῖς. I have no idea what the earliest manuscripts of Aristophanes have it as; I can easily see people, a century on, respelling the Acharnians they way they would now pronounce it, because they did not know yet that every vowel of Aristophanes was sacred. Even if the entire manuscript tradition of Aristophanes had the Ἀχαρνεῖς spelling, modern editors are doing their job in trying to work out how Aristophanes would originally have spelled it, and would see it as in their rights to restore the spelling as Ἀχαρνῆς.
At any rate, the Hellenistic grammarians certainly knew about -ῆς as a spelling specific to the Athenians; so people two or three centuries on were certainly aware what would have been the original spelling. Herodian in ii AD spells out our reconstruction implicitly:
Nouns ending in -εις are recessively accented, if their singular is recessively accented: Δημοσθένης, Δημοσθένεις. But if the singular has an acute or circumflex on the ultima, they take the circumflex: εὐγενής εὐγενεῖς, Ἡρακλῆς Ἡρακλεῖς. But Athenians contract these with an η, and they similarly put a circumflex on that: ἱππῆς from ἱππέες, βασιλέες becomes βασιλῆς. (De prosodia catholica p. 424 Lentz)
In ix AD, George Choeroboscus spells it out explicitly, if mistakenly:
Rarely, in the dual, εε are contracted into the diphthong ει, as in ταρίχεε ταρίχει, πόλεε πόλει [...] and in the plural they are contracted into η, as in ἱππέες ἱππῆς and βασιλέες βασιλῆς. These are Attic, and the Attic-speakers do this only with ευς nouns, I mean contracting εε into η in the nominative plural. (Prolegomena et scholia in Theodosii Alexandrini canones isagogicos de flexione nominum, p. 182)
That is, Choeroboscus doesn't know about *έης, and thinks the Attic ending -ῆς is an exception to the normal contraction of /ee/ to ει /eː/. (He repeats elsewhere that outside of Attic, the plural is of ἱππεύς βασιλεύς is ἱππεῖς βασιλεῖς.) He goes on to offer two theories why this exception has happened; another's, that it's because of the following /s/; and his own, that first and second declension nominative plurals also end in ι (φίλοι, κοχλίαι).
So that's why there could be a question about whether to write Acharnians as Ἀχαρνῆς or Ἀχαρνεῖς. The -ῆς ending was identified as peculiarly Old Attic, and it did not make it into Koine: the plural there is the Ionic and New Attic -εῖς.
Classicists have no compunction spelling it the Old Attic way. Puristic was supposed to revive the glories of Ancient Greek (which predate New Attic); you'd have thought they would have tried to get all Greekdom speaking of βασιλῆς. Yet you'll never see βασιλῆς written in Puristic, only βασιλεῖς. In fact, Modern Greeks will only be familiar with the βασιλῆς spelling if they paid attention in Ancient Greek class (and sometimes, not even then: I have been asked whether one instance was a spelling mistake).
The reason for that is, the mission of Puristic was not defined positively, as restoring Attic, but negatively, as cleansing Modern Greek of foreign elements and corruption. The cleansing of foreign elements in vocabulary was pretty successful; but the cleansing of corruption could mean anything, depending on how far back you consider corruption to have set in. The result was that Puristic morphology became pretty eclectic and inconsistent, because different writers decided on different levels of corruption to deign to put up with; but outside a few decades of insanity in the end of the 19th century, Puristic gave up on trying to revive Ancient Greek outright. What it did seek to do consistently was use anything but what the contemporary vernacular was using.
So sometimes Puristic went back to Early Modern Greek (futures in θέλω + Infinitive); but often, Puristic would go back to Koine, as opposed to going all the way back to Attic. Puristic displaced the Arabic targumān > δραγουμάνος "interpreter", and replaced it with the Byzantine διερμηνεύς (instead of the Koine διερμηνευτής). And it never used the Attic plural *διερμηνῆς: Puristic used the same plural as the Byzantines would have used, the Koine διερμηνεῖς.
Modern Greek also uses διερμηνεῖς, because of Puristic; but that is not the vernacular development of -ευς nouns (which διερμηνεῖς is not). The vernacular switched all its third declension nouns to first declension in the singular, so βασιλεύς /basileús/ became βασιλέας /vasiˈleas/. Initially, the vernacular left the nominative plural of the third declension alone, so the plural would have remained βασιλεῖς /vasiˈlis/.
But avoidance of hiatus made βασιλέας into βασιλιάς /vasiˈljas/, a first declension noun accented on the final syllable. And the vernacular gave such oxytone first declension nouns a plural in -άδες: βασιλιάς βασιλιάδες /vasiˈljas vasiˈljaðes/, μαθητής μαθητάδες /maθiˈtis maθiˈtaðes/. Modern Greek has rolled that plural back for -ης nouns, under indirect Puristic influence, but βασιλιάδες is still the Modern plural.
Why is -άδες a plural? Well, in Koine, -ᾶς became a widespread agent suffix: ζυτᾶς κασσιτερᾶς κλειδᾶς "brewer, tinker, locksmith". The suffix still exists in Modern Greek: ψωμάς, κομπιουτεράς, Ζητάς "breadmaker, computer geek, Zeta Force guy (police motorcyclist)". This meant that there were suddenly a lot of first declension /-as/ nouns accented in Greek on the ultima; in Attic they had been few and obscure, and even fewer had a plural.
But Greek had a lot of third declension nouns ending in accented /-as/, since that was a widespread feminine suffix: Κρονιάς, κυκλάς, λαμπάς, ναϊάς "Saturnalia, encircling, torch, naiad". And its plural, -άδες, was much less odd-looking than the correct, contracted first declension plural -αῖ would have been. So the plural of torches, λαμπάδες, was carried across to these new agent nouns, κλειδᾶδες, and eventually to the normal first declension nouns, ending in -ής, μαθητάδες.
As linguists have pointed out before me, the clash of Puristic and Vernacular Greek has meant that Modern Greek now has two ways of saying "kings": the kind of mess that was the natural legacy of diglossia. Colloquial Greek has βασιλιάς βασιλιάδες. Puristic had βασιλεύς βασιλεῖς; high register Modern Greek cleans up the singular to be first declension rather than third declension, but it goes back to Middle Greek to do so, with βασιλέας βασιλεῖς.
So if you want to show royalty more deference than is usual in contemporary Greece, you will use the pseudo-Puristic βασιλέας βασιλεῖς, instead of the colloquial βασιλιάς βασιλιάδες.
Markos Vamvakaris managed to show deference to the king with the colloquial βασιλιάς (Καλώς μας ήρθες Βασιληά, "Welcome back, King!"; but 1935 was a very different time.
I'm curious how many Greeks' heads are exploding right now, to hear Vamvakaris wrote a royalist paean in rebetiko (right after recording Κάν' τονε Σταύρο Κάντονε, "Get that bong ready, Stavros" no less, his catchy ditty about group drug intoxication.)
I could go back and report what the historical grammars say about other Ancient dialects' declensions, but I've established what I needed to establish, and there are posts on Modern Greek compound accentuation to write.