I'll walk through the page in slow motion, because I'm loath to pass up an opportunity to be didactic. It'll be slow enough, it'll drag through two posts.
It's not much of a premiss. Sarantakos spelled Acharnians in a post as Αχαρνείς, and the first comment off the cab rank asked, simply: Εν τέλει είναι “Αχαρνείς” ή “Αχαρνής”;
To the classicists whose teeth have been set on edge by the monotonic, I'll render that as: "Should it be spelled Ἀχαρνῆς or Ἀχαρνεῖς?" I'm going to spend most of this post and the next working through the historical morphology of -ευς nouns. In this post, from Homeric backwards, and in the next post, post-Homeric (including why Attic had the plural Ἀχαρνῆς and then Ἀχαρνεῖς.) In the second post, I'll add a codicil on why a Modern Greek speaker would hesitate over the spelling. Yes, once again, Puristic is to blame—this time, because it wasn't as pure as it claimed it was.
Ancient Greek had lots of dialects, and dialects had different ways of declining the same nouns. In historical linguistics, we reconstruct proto-forms, not because we particularly care to know what some tribe's language was five thousand years ago, but because we can use those proto-forms to explain the diversity of later forms. There's several dialects to explain; we want to explain the Attic of Aristophanes ultimately, but we're going to start with the variant of Greek that everyone starts with, Homeric Greek.
Now, there are three declensions of Ancient Greek nouns, and -ευς nouns are third declension nouns. That means that, underlyingly, it shares the following inflections with all other third declension nouns:
That in itself is a reconstruction. In some nouns, the pattern is clear: φύλαξ, φύλακος, φύλακι, pʰýlak-s, pʰýlak-os, pʰýlak-i... For other nouns, the pattern is harder to see: λύσις, λύσεως, λύσει, lýsi-s, lýse-ɔːs, lýse-i. With historical linguistics, comparison with other dialects, and some imagination, we can work out that they all belong to the same underlying pattern.
But what's obvious to us was not obvious at the time. The Roman-era Greek grammarians were pretty good at reconstructing forms to explain their grammar, but the three underlying declensions eluded them: they were happy to have fifty-odd declensions on the books. When in the Renaissance Greek grammarians discovered that Latin grammar had managed to get Latin declensions down to five, they realised they needed to take another look at their own grammar. In the first attempt, they managed to get it down to ten.
Homeric Greek is a few centuries older than Attic, which means there are a few centuries less phonetic change in Homeric Greek to need to disentangle. In Homeric Greek, -ευς nouns *almost* follow the third declension pattern. Here's the Greek characters:
And here's the IPA:
This almost follows the pattern of pʰýlak-s, pʰýlak-os, pʰýlak-i...; but there is a problem. The thematic vowel—the ending of the stem, which goes before the inflection—alternates between /ɛː/, in front of a vowel, and /eu/, in front of a consonant.
That alternation between /ɛː/ and /eu/ isn't right: surely underlyingly there should be just one vowel or diphthong that the stem ends in, throughout. We're going to work out what that vowel or diphthong is, by internal reconstruction: we'll use just the forms in Homeric Greek, and what we know of language change elsewhere in Greek—to arrive at a pre-Homeric Greek form.
Note the distinction: pre-Greek works backwards from just one dialect; proto-Greek takes all the data into account, from other dialects of Greek as well as related languages. The two will look different, but not that different.
To explain the alternation between /ɛː/ and /eu/, we need two sound changes:
- Either the stem adds a /u/ (or a /w/) in front of a consonant, or else the stem drops a /u/ (or a /w/) in front of a vowel.
- Either the stem lengthens its /e/ to /ɛː/, when it drops its /w/, or else it shortens its /ɛː/ to /e/, when it adds its /w/.
Second change first: we know that when languages drop consonants, they lengthen the preceding vowel. This is called compensatory lengthening, and it makes phonetic sense: if you drop one phoneme, and lengthen the preceding phoneme, the result vaguely takes the same length of time to pronounce as before. Here's some examples from other languages—because phonetic change happens along the same lines across languages: it involves facts of articulation, which are universal to humans:
- The spelling cart in English preserves the original pronunciation /kaɹt/, which has survived in American /kɑɹt/. When Commonwealth English dropped its r's, it lengthened the previous vowel: Australian English /kaːt/.
- Modern Greek preserves the old pronunciation of Turkish <ğ> as /ɣ/ in loanwords. In Modern Turkish, /ɣ/ was dropped before consonants, and the previous vowel is lengthened: Greek τσογλάνι /tsoɣlani/ "knave", Turkish iç oğlanı /itʃ oːlanɯ/ "inside boy = palace servant".
- Ancient Greek digamma, ϝ /w/, disappeared from most dialects of Greek. The first instances of /w/ to go were /w/ after another consonant. When it did, Ionic lengthened the preceding vowel—though Attic did not. So we explain pairs like Ionic ξεῖνος Attic ξένος, Ionic κούρη Attic κόρη, Ionic οὖρος Attic ὄρος, by positing an original /w/ (which does in fact turn up in very early inscriptions): Proto-Greek /ksenwos, korwaː, worwos/, Ionic /kseːnos, koːrɛː, oːros/, Attic /ksenos, korɛː, oros/.
So we have good precedent for /w/ being dropped, and /e/ lengthening when it does. We could then reconstruct the stem as ending in /ew/: *-ews, *-ewos, *-ewi > -ews, -ɛːos, -ɛːi.
Neat, but I'm afraid, wrong. The compensatory lengthening happens when /w/ follows a consonant; but our deleted /w/ in *-ewos would be following a vowel. And when Homeric /w/ is deleted, the preceding vowel keeps its length.
We have evidence of that from augment. Normally, the aorist in Greek is formed by adding an /e/ at the start of a verb, if the verb starts with a consonant; but if the verb starts with a vowel, the vowel is lengthened instead. So δράμω, ἔδραμον drámɔː, é-dramon, but ἐγείρω, ἤγειρα egéːrɔ, ɛ́ːɡeːra. (You may have noticed that sometimes ε /e/ is lengthened to η /ɛː/, sometimes to ει /eː/. /ɛː/ is the old lengthening, which is in Homer; /eː/ is the new lengthening, which is post-Homeric.)
But some verbs in Homer do not lengthen the initial vowel: they add an /e/, as if the verb started with a consonant: ἔργω, ἔεργον /érɡɔː, éerɡon/. We would rather not concede an arbitrary exception: it is simpler to claim that there used to be a consonant there, which was dropped after the augment was added: ϝέργω, ἔϝεργον /wérɡɔː, éwerɡon/. We can see the /w/ in old dialect inscriptions. We can also see the /w/ in English: ϝέργω has the same stem as work.
But notice that /éwerɡon/ went to /éerɡon/ in Homeric Greek, without any compensatory lengthening. Attic Greek does have a long vowel there, as εἶργον /êːrɡon/; but the /eː/ is merely how Attic concatenates two /ee/.
So a pre-Greek -ews, -ewos, -ewi would end up in Homeric Greek as -ews, -eos, -ei. That's not what Homeric has. We know that the /w/ must have dropped out between vowels, because it did so all the time: the alternative, that /w/ popped up between /e/ and /s/ in -eus, doesn't make sense. So we're on the right track; but the original vowel must have been a long /ɛː/: if the Homeric genitive was /ɛːos/, the pre-Greek must have been /ɛːwos/. And that means the pre-Greek must have been */-ɛːws, -ɛːwos, -ɛːwi/. What we're now looking for is a rule to explain why */ɛːws/ and */ɛːwsi/ turned into short /ews/ and /ewsi/.
That rule exists, and is called Osthoff's Law. It does more than explain /ɛːws/ going to /ews/: it says that in general, if a long vowel if followed by a resonant and then another consonant (VːRC, where R is one of /m n l r j w/), the vowel is shortened. We know that there used to be long vowels there, because Osthoff's Law did not apply to Indo-Iranian.
So the sky god, the textbooks say was dyā́us (/djáːws/) in Vedic Sanskrit; but /djáːws/ fits the VːRC pattern. The equivalent word in Greek, then, has a short vowel through Osthoff's Law. Can you guess the Greek equivalent?
/dj/ corresponds to Greek /zd/: yes, the Greek for /djáːws/ is /zdéws/, Ζεύς, and we reconstruct the Proto–Indo-European (PIE) Sky God as *dyēws /djeːws/. (Sanskrit defaulted its vowels to /a/; Greek is considered to have preserved PIE vowels better than Sanskrit, because it has a lot more /e/ and /o/.)
With Osthoff and digamma-dropping, we have explained almost all of the Homeric declension of -ευς:
- Pre-Greek: -ɛːws, -ɛːwos, -ɛːwi, -ɛːwa, -ɛːw-, -ɛːwe, -ɛːwojn, -ɛːwes, -ɛːwɔːn, -ɛːwsi, -ɛːwas, -ɛːwes
- Osthoff: -ɛws, -ɛːwos, -ɛːwi, -ɛːwa, -ɛːw-, -ɛːwe, -ɛːwojn, -ɛːwes, -ɛːwɔːn, -ɛwsi, -ɛːwas, -ɛːwes
- Drop digamma between vowels: -ɛws, -ɛːos, -ɛːi, -ɛːa, -ɛːw-, -ɛːe, -ɛːojn, -ɛːes, -ɛːɔːn, -ɛwsi, -ɛːas, -ɛːes
There is one form that this does not explain: Osthoff's Law does not apply to the vocative singular, /-ɛːw-/, but this has ended up as short /-eû/ regardless.
When in doubt, we appeal to analogy: in third declension nouns based on stems ending in /i/ and /u/, the vocative singular is the same as the nominative, except for dropping the /s/: ὀφρύς, ὀφρύ /opʰrýs, opʰrý/; πόλις, πόλι /pólis, póli/. Following that pattern, the vocative *-ɛːw was also made to look like the nominative, except for dropping the /s/: βασιλεύς, βασιλεῦ /basileús, basileû/. If any of the cases was going to fall under another's sway, it would be the vocative, which is a minor case—and which is conflated with the nominative in everything but the masculine singular.
So for a good Homeric word like βασιλεύς "king", then, we have just reconstructed the pre-Greek nominative as *βασιληύς, /basilɛ́ːws/.
The proto-Greek, as opposed to the pre-Greek, is based on more evidence than just Homeric: it seeks to explain *all* forms that have appeared in Greek, not just one dialect's. And the proto-Greek for βασιλεύς is not /basilɛ́ːws/. It's /ɡʷatilews/, with a genitive of /ɡʷatileːwos/.
That's nuts, I know. But:
- /b/ in Greek corresponds to Indic /g/, Latin /b, w/, and Germanic /k/ (English cow, Sanskrit gau, Greek βοῦς). This means we're looking for a single Proto–Indo-European consonant to explain all of these, which has something in common with both the labial /b/ and the velar /g/. The convention in Indo-European is to call that something a labio-velar, /ɡʷ/. In fact, there has been no need to assume PIE had a /b/ at all.
- PIE /ti/ ended up in Greek as /si/, through assibilation: affricating a palatalised consonant. It's the same process through which Latin natio ended up pronounced in French as /nasjɔ̃/; and (applied to /k/ rather than /t/) it's what Modern Greek linguists call tsitacism.
The proto-form has PIE /ɡʷ/, not /b/, and PIE /ti/, not /si/. On the other hand, it has Osthoff's Law already operating, unlike PIE (ɡʷatilews, not ɡʷatileːws). The reason for all that is, there are forms of Greek with /ti/ instead of /si/ (Doric); and the point of a proto-Greek form is to explain the forms in all variants of Greek.
Likewise, the proto-form already has Osthoff's Law, because Osthoff's Law applies to all variants of Greek, so it doesn't need to account for any forms in which Osthoff's Law is absent. The proto-Greek form explains the Greek variants of the word, not how the Greek variants differ from the non-Osthoff Sanskrit.
The proto-form also has /ɡʷ/, not /b/, which means we're claiming there is a form of Greek which still had the PIE labiovelars. There's no Greek letter for /ɡʷ/, so no form written in the letters introduced by the Phoenecians recorded such a /ɡʷ/.
There is, however, a Greek *syllable* for /ɡʷa/. βασιλεύς turns up in Linear B, as qa-si-re-u: 𐀣𐀯𐀩𐀄. We also have a genitive of -εύς turn up in Linear B, complete with its digamma: ἱερῆϝος /hierɛ̂ːwos/ "of the priest" (Attic ἱερέως) turns up in Linear B as i-je-re-wo, 𐀂𐀋𐀩𐀺.
- The astute reader may have noticed that proto-Greek has a /ti/ in ɡʷatilews, but Linear B has a /si/ in qa-si-re-u. /si/ is supposed to be a later development than /ti/. But that just proves that Linear B is not proto-Greek: it is the earliest form of just one dialect branch of Greek—which did not retain /ti/ like Doric did. So in some features, Mycenaean is more innovative than Doric was a millenium later. (But only some.)
- We know now that 𐀣 qa corresponds to PIE /kʷa, ɡʷa/. (Linear B also conflates /k/ and /g/.) But the labio-velar interpretation wasn't immediately obvious to Ventris: in the first edition of the decipherment, the sign was transcribed as pa2.
- No, there wasn't really an overwhelming reason for me to cite Linear B in the original. Actually, I'm curious to know: how many of you can see Linear B in the paragraph above?
So far, we've explained Homeric Greek. The plural of Acharnians in Homeric Greek would be Ἀχαρνῆες—which is neither Ἀχαρνῆς not Ἀχαρνεῖς. That involves a few more changes within the dialects of Greek, and it will need to wait for the next post.