How Greek accentuation works

In a previous post, I accented ΞΕΜΑΓΚΑΣ "the un-mangas, the ex-mangas" as ξεμάγκας. Nikos Sarantakos pointed out the correct accent is ξέμαγκας. I see why that is the correct accent, though it still looks wrong to me.

To explain why, I'm going to spend the next few posts building up to this explanation of what has happened:

The accentuation of ξέμαγκας follows the normal recessive default, which has survived into Modern Greek for compounds. Though μάγκας as a first-declension noun would not have allowed antepenult accent in Ancient Greek, the Modern system ignores that restriction for -ας, where there is no analogical form imposing the Classical penult stress (unlike -ης nouns). My notion was that the accent was preserved because ξε- was used quotatively, as with μα και ξε-μά.

If that paragraph made sense to you, what I am going to write about won't have any real surprises. If not, stay tuned. What I'm going to build up, post by post, is an explanation of how accents work in compounds in Modern Greek. To get there, I will be going through

• how accentuation works in Ancient Greek;
• how it has changed into Modern Greek;
• how compounds accent in Ancient Greek;
• how Modern Greek has tried to realign the Ancient compound accent rules;
• and finally, why a different construction made me get the accent wrong.

Towards the end, I'm going to do something silly; I'm going to try and work out the Modern accent rules for compounds on my own, rather than look them up. (Unless I find a good statement of the rules, next time I'm going through my photocopies in the garage.) Phonology was never my thing, so I haven't researched the topic explicitly; and I don't have the library access I used to. I'm reasonably sure someone *has* already worked these rules out. Me needlessly duplicating that effort promises to be fun for you, and embarrassing for me. Or vice versa.

So. How does Ancient Greek accentuation work? This is standard stuff for Ancient Greek grammar, but let's run through it, because Ancient Greek phonology is not always taught properly—particularly in Greece. You can get the standard presentation from Smyth's Grammar.

(And I rejoice that, by linking to the CHLT site copy of Smyth, I no longer have to link to Perseus' buggy, clunky, crashy, crappy interface, even if where I link to is still the Perseus encoded Smyth.)

If you're having to learn how to accent in the polytonic, particularly in Greece's ossified paedagogical system, you are pummelled with the various possible permutations of accent on words:

• oxytone: last syllable (ultima) has an acute: ἀνήρ
  
• paroxytone: second last syllable (penult) has an acute: γέρων
  
• proparoxytone: third last syllable (antepenult) has an acute: εὔλογος
  
• perispomenon: last syllable has a circumflex: Ἀθηνᾶ
  
• properispomenon: second last syllable has a circumflex: κοῖτος

If you actually try to work out how Ancient Greek accentuation works, though, things are simpler than that. There is one major distinction between kinds of accent in Greek. The rest are either subclasses, or minor exceptions.

The major distinction is, is the word accented as far back from the end as possible, or not? If it is, its accent is recessive. The Roman-era grammarians of Greek were rather more intelligent than the schoolmasters of Modern Greece, and they worked out that this distinction was more important; so they gave it a separate Greek name: barytone. (I know that Smyth says barytone is anything accented before the ultima, but barytone is usually used to mean recessive.)

With the five types we saw, barytone cuts across them:

	One syllable long	Two syllables long	Three syllables long
Oxytone	Barytone	Not	Not
Paroxytone	—	Barytone	Barytone if last syllable long
Proparoxytone	—	—	Barytone
Perispomenon	Barytone	Depends on proto-Greek vowels	Depends on proto-Greek vowels
Properispomenon	—	Barytone	Not

Let me explain. Accentuation in Ancient Greek is restricted to occuring in the last three (or four) moras of a word. A mora is like a syllable, only a long vowel or diphthong has two of them. So a word has recessive accent, if it occurs on the third last mora: that's as far back as the accent can go. Except, the penult is treated as having only one mora, whether it's long or short. (That's the "or four".)

• If the ultima is short (that's one mora), and we treat the penult as a single mora (that's two moras), then the accent can be as far back as the antepenult (that's three moras). e.g. ἄνθρωπος (short–long–short, moras: 1.1.1)
  
• If the ultima is long (that's two moras), then the accent can be as far back as the penult (that's three moras). e.g. προσῳδίᾱ (short–long–short–long, moras: 1.2.1.2)
  
• If the word is a properispomenon (circumflex on second last syllable), its ultima must be short: that's how circumflexes work. So if that word has three syllables, it can't be recessive: the recessive accent should be three syllables back. e.g. κυπρῖνος (short–long–short, moras: 1.2.1)
  
• If the word is perispomenon, that always means that in proto-Greek, the final syllable was initially two vowels that have been contracted. For example, the citation verb ending -ῶ is proto-Greek *άω, *έω, or *όω. Homeric Greek in fact did still preserve *άω and *έω as verb endings, which is how Roman-era grammarians were able to work out the proto-Greek system.
  
• If you look at what the proto-Greek vowels are behind a perispomenon, the accent could still be recessive. Proto-Greek *όω counts as three moras (short–long); that means *δηλόω is recessively accented, and so is its attested counterpart δηλῶ.

The mora restriction is confirmed by the behaviour of enclitics—words that are suffixed to a phonological word, without their own accent, so they end up adding to the mora count.

• If a proparoxytone has an enclitic suffixed, its accent ends up not three, but four or five moras from the end of the new phonological word. To fix this, a secondary accent is added to the ultima. So ἄνθρωπος τις is phonologically /án.tʰrɔː.pos.tis/, with the accent now four moras back; to fix this, it is accented as ἄνθρωπός τις.
  
• If a properispomenon has an enclitic suffixed, its accent ends up four or five moras from the end as well. Again, to fix this a secondary accent is added to the ultima. So κῆπος τις is phonologically /kɛ̂ː.pos.tis/, and the circumflex is now four moras back (kɛ̂ː is two moras, being long). This is fixed by accenting it as κῆπός τις.

Now with those rules, accent is distributed like this:

• Verbs are in almost all cases recessive. There are a few inflections that aren't recessive (see Smyth)—certain imperatives and optatives; and participles and infinitives, which behave like nominals. There's also some interplay of the augment and accent, and there is real confusion about prefixed subjunctives and optatives; but our endpoint does not involve verbs, so I won't spent any time on them.
  
• Nominals—adjectives and nouns—are by default recessive.

That's only a default, and it's not a strong default. The following counts are taken from around 120,000 noun and adjective stems in LSJ and LSJ Supplement, for which the TLG lemmatiser has a definite analysis:

• Recessive: 79,953
  
• Ultima stress (syllables > 1): 30,401
  
• Penult stress (syllables > 2, short ultima): 9096

So verbs in most inflections, and two thirds of all nominals, are recessive. That includes two-syllable–long words accented in their initial syllable, and one-syllable–long words. Most remaining nouns, and a few inflections, are stressed on the ultima. Penult stress—as distinct from recessive stress—is much less common: only a tenth of all nominals, and a few (though common) verb inflections—notably perfect passive participles, and most infinitives.

That's Ancient Greek accentuation; how about Modern?

Modern Greek no longer has long and short syllables. The three-somethings-back restriction on accent location survives in most dialects of Greek, but now the "somethings" are just syllables (since any syllable just has one of them). Enclitics still generate secondary stress: άνθρωπος μου /ˈan.θro.pos.mu/ is accented as άνθρωπός μου (though nowadays people sometimes forget to write the second accent). κήπος μου on the other hand doesn't need a second accent, since the accent in /ˈki.pos.mu/ is just three syllables back: there are no double moras for the η, because there is no longer a long vowel there.

Accent in Greek is now stress based rather than pitch based—which is why monotonic accentuation can dispense with the circumflex; and accent is still by default recessive. Recessive accent in Ancient Greek is defined by the accent going as far back as the Ancient moras allowed. Recessive accent in Modern Greek is defined by the accent going as far back as the Ancient moras allow.

... Seriously. What's actually happened is, the location of stress is unchanged for any words that survived from Ancient into Modern Greek. So the Ancient rules for where the accent goes still apply to those words in Modern Greek. The only time where accent location is reevaluated, is when Greek-speakers realised there was something odd with where the accent now was; and the only reason for them to realise it is, they were comparing it to something else. Which means analogy.

Analogy has a long history of rearranging accentuation of feminine nominals; let's go through some snapshots.

• Diphthongs are long syllables, so a word ending in -οι should not be accented on the antepenult: the -οι already counts as two moras. However, most instances of -οι as an inflection, including the nominative masculine plural, are treated as short: ἄνθρωπος :: ἄνθρωποι, πρῶτος :: πρῶτοι.
  
• The feminine nominative plural ending -αι is also treated as short: θάλασσα : θάλασσαι , πρώτη :: πρῶται
  
• But does not mean -αι is recessively accented. If the feminine noun ends in short -α, it is accented on the antepenult in the nominative singular (θάλασσα), and the same happens in the plural, with its short -αι (θάλασσαι). Most feminine nouns, however, end in long -ᾱ (or -η, which developed from Proto-Greek -ᾱ). When those nouns are recessively accented, their accent stays on the penult in the singular, following the mora rule. But in the plural, their accent stays on the penult, even though -αι is supposed to be short: ἡμέρᾱ :: ἡμέραι, not *ἥμεραι. The accent may be recessive, but it's more important for the nominative plural to have the same accent as the nominative singular.
  
• That's nouns. In adjectives, the feminine singular nominative also ends in in long -ᾱ (or -η). So if the adjective has recessive accent, the masculine and neuter singular (ending in short vowels) have the accent in a different place from the feminine, because of the feminine's extra mora: δεύτερος :: δευτέρᾱ :: δεύτερον
  
• But whereas the plural of ἡμέρᾱ is ἡμέραι, the plural of δευτέρᾱ is δεύτεραι. For the noun, the powerful analogy was between nominative singular and nominative plural. For the adjective, the analogy is not with the singular, but with the other genders in the plural: δεύτεροι :: δεύτεραι :: δεύτερα.
  In the singular, the extra mora prevented the masculine and feminine from having their accent in the same place. In the plural, with -αι considered short, the extra mora is no longer in the way, and the pressure to match the masculine and neuter accent carries the day. For an adjective, as opposed to a noun, the other genders matter, and can force analogical change. In fact, that analogy is the only difference between feminine adjective and noun declension.
  
• So far, we've seen analogy in Ancient Greek. Ancient Greek was constrained by moras not to clean up the accent discrepancy in the singular δεύτερος :: δευτέρᾱ :: δεύτερον. But Modern Greek has no moras.
  
• Which meant Modern Greek speakers noticed that the accent of δευτέρᾱ was odd, compared to δεύτερος and δεύτερον, and fixed it. In addition, whereas Attic stopped -ᾱ going to -η after /r, i, e/, Modern Greek allowed -η after /r/ as well, which makes adjectives ending in -ρος -ρη -ρον look like other adjecives. (Adjectives ending in /-ios/ remained separate.)
  
• Hence the singular of the adjective is now δεύτερος :: δεύτερη :: δεύτερο. That violates the Ancient mora rule; but there are no moras now to make that matter. Still, Modern Greek only violates the Ancient mora rule here, because there was good analogical reason to. ἡμέρᾱ has still kept its accent, as μέρα, and ἀγάπη has not changed to *άγαπη.

Unlike Ancient Greek, there are now nouns ending in -η which are accented on the antepenult—violating the mora constraint like δεύτερη does. But they too are the result of analogy: they are the third declension feminine nouns which in Ancient Greek ended in -ῐς.

Since -ῐς was short, if those nouns were recessive, they were accented on the antepenult: ἄνοιξις. Modern Greek got rid of its third declension quite comprehensively; when it came to these nouns (nom. -is, gen. -eos, acc. -in), the obvious thing to do was to move them across to the feminine first declension (nom. -i, gen. -is, acc. -i). The endings are pretty much the same in Modern Greek, so there was no good reason to shift the accent location: ἄνοιξις, ἀνοίξεως, ἄνοιξιν became άνοιξη, άνοιξης, άνοιξη.

If you're brought up on Ancient Greek, άνοιξη is nonsensical: an eta has two moras, it can never be accented like that. For that reason, the conservative spelling was άνοιξι, keeping the etymological iota to avoid challenging the mora. But of course Modern Greek didn't particularly care where the first declension endings were accented, when it carried them across to these third declension nouns. It didn't bother changing where the old first declension feminines were stressed; but it had no reason to change where the old third declension feminines were stressed either. In the first case, accent conservatism preserves the old mora rule; in the second case, accent conservatism, plus analogy in inflection, throw out the old mora rule.

We just saw that Ancient Greek had first declension feminines with both a long -ᾱ/η (ἡμέρᾱ, δευτέρᾱ) and a short -ᾰ (θάλασσα). In the masculine, though, the first declension only ever ended in a long -ᾱς (or -ης). So καταρράκτης is recessively accented, and accent conservatism means it has not gone to *κατάρραχτης.

Modern Greek does have recessive nouns ending in -ας, like γέροντας. These have a similar origin as the recessive nouns in -η, such as άνοιξη. Masculine third declension nouns were all remodelled to look like first declension nouns, and the suffix of choice was /-as/, following after the third declension accusative /-a/. So γέρων, γέροντος, γέροντα was remodelled as γέροντας, γέροντα, γέροντα, after first declension ταμίας, ταμία, ταμία(ν): nom. -Ø, gen. -os, acc. -a was remodelled as nom. -as, gen. -a, acc. -a.

In fact the transition was so successful that most vernacular nouns in -ας come from third declension nouns; first declension nouns like ταμίας did not survive in colloquial use. Once the pattern was in place, proper names and other remodelled nouns could follow—like Καβάσιλας, or μάγειρας. Once again, the accent was wrong for Ancient Greek; once again, that did not matter in Modern Greek—but the accent disruption only happened because analogy introduced an innovation into the system.

Because of the innovation of recessive nouns like γέροντας, nouns like μάγειρας were introduced, which were impossible in Ancient Greek. Because of that pattern being established in turn, it because possible for the compound ξε + μάγκας to be accented ξέ-μαγκας—whereas Bacchylides could only accent προ + κόπτᾱς as Προκόπτας.

But that brings us to the recessive accentuation of compounds, which is next post.