- If the second half of a compound is a single syllable, stress never recedes back further than the last syllable of the first half: γριά /ɣrja/ "old woman" > παλιόγρια /pa.ˈljo.ɣrja/ "damned old woman", not *πάλιογρια /ˈpa.ljo.ɣrja/. We saw the same rule in Ancient Greek, and there it presumably has phonological reasons behind οἰκοφύλαξ rather than *οἰκόφυλαξ. For Modern Greek, there are no such phonological reasons: compare *πάλιογρια /ˈpa.ljo.ɣrja/ with πάπια /papja/ "duck" > αγριόπαπια /a.ˈɣrjo.pa.pja/ "wild duck".
Rather, this is a matter of analogy. Ancient Greek has not bequeathed any compounds to Modern Greek in which the first half was accented two syllables back—like *οἰκόφυλαξ. Modern Greek kept the restriction on how far back the accent could go back on the first half.
- Notice the accent on αγριόπαπια, btw. It's impossible in Ancient Greek—*/a.ɡri.ó.pa.pi.a/. For this accentuation to be possible, you need /ia/ to be reduced to /ja/ in vernacular Greek. Greek speakers do manage to keep vernacular and learnèd phonologies distinct. In particular, Ancient nouns in -ía went to -ˈja in the vernacular: παραγελλία "order, bid" in learnèd Greek, παραγγελιά in vernacular Greek. But that means that how recessive accent is realised differs by register.
- As we saw with υπολοχαγός and παραστρατός, modern learnèd compounds will leave accent alone, out of a timidity that gives rise to artificiality. Vernacular preposition + noun compounds, like ancient compounds, are recessive, even if vernacular compounds tend more towards being endocentric (which I guessed might encourage accent-retention). So cf. learnèd στρατός > παραστρατός "paramilitary" with colloquial παιδί "child" > παραπαίδι "stepchild, apprentice".
- Proper names retain their accent—as you might expect from the general principle I guessed, that if the second half of the compound is unchanged in meaning, it is likelier not to change in accent. (Proper names are *supposed* to have a fairly strict denotation.) So "mad Nick" is τρελο-Νίκος, not τρελό-Νικος. (Athens U Math forum: Επίσης σκέφτομαι ότι γράφω αυτό το ποστ για να δω αν δουλέυει το αβαταρ που με τόσο κόπο έβαλε ο τρελονίκος, "I'm also thinking that I'm writing this post to see whether the avatar Mad Nick set up with so much effort is working.")
The slang.gr site, that jewel of informal lexicography, has what looks like a counterexample in ντελήσαββας, "Mad Sabbas, a silly person". But the contributor notes that he heard this from a Pontian in Central Macedonia, and Pontic has different accentuation rules to Standard Greek: it's quite comfortable with stresses more than three syllables back. So this lemma doesn't count; indeed, the contributor notes the (non-Pontic?) Central Macedonian variant ντελημπάσχος, from Πάσχος "Pascal", with the stress unchanged. The only way to change the stress in Standard Greek for names in compounds is to switch inflection; and indeed, the slang.gr editors note Γιάννης > Στραβό-γιανν-ος "Blind John (novice sailor)", Αστραπό-γιανν-ος "Lightning John (dullard)".
- We know that Ancient nouns ending in -η /ɛː/ could only have their accent go up to the penult, because of the mora rule. Ancient nouns ending in -ις /is/ could have their accent go up to the antepenult, because /is/ was short. But in Modern Greek, these have fallen together as /i/. Modern Greek respects the Ancient restriction on -η: θήκη "case" > vernacular αβγοθήκη "egg-holder"—recapitulating learnèd ᾠοθήκη "ovary". But when it comes to nouns that formerly ended in -ις, Modern Greek seems to have remembered that they are recessive, even when the nouns are only two syllables long: Ancient βράσις, Modern βράση "boiling" > κουφόβραση "muggy weather", Ancient βρύσις "bubbling up", Modern βρύση "fountain" > κεφαλόβρυση "main fountain".
Did the coiners of κουφόβραση and κεφαλόβρυση know Ancient Greek grammar? Highly unlikely. Did they work out that βράσις and βρύσις are verbal nominals, and associate that kind of recessive accent? True, Ancient βράσσω survives as βράζω—in use already in Koine; and the noun κουφόβραση can in fact be plausibly derived from the compound verb κουφο-βράζω "to boil in a deaf (= silent) manner, to simmer". But the verb βρύω did not survive into Modern Greek—and even if it did, the modern water tap of βρύση does not call "bubbling up" to mind.
Greek speakers aren't born knowing Ancient morphology, but they don't need to: whereas the -σις suffix was a highly productive verbal nominal suffix in Greek, there are exactly 7 nouns in LSJ ending in -ση: two are nominalised feminine adjectives, and don't count (μέση, ὑπερμέση), and the others did not survive into Demotic (ἀποφράση, ἄση, ἕρση, κόρση, Τεμέση).
So Greek speakers would have been able to work out that most nouns ending in /i/ could not be accented on the antepenult, but nouns ending in /si/ could. They could work that out without any recourse to derivational morphology. And they could carry that insight across to compounds.
We can then speculate that, if another Ancient noun ending in -ῐς survived into Greek, and didn't look like a verbal nominal, its Modern Greek compounds will not be recessive—unless there are surviving compounds from Ancient Greek, which also have recessive accent, and serve as a precedent. So μύτις "snout", ῥάχις "back", μάντις "augur" and κόνις "dust" have survived into Modern Greek as μύτη, ράχη, σκόνη, μάντης; in compounds, they should accent like any another noun ending in -η, on the penult. And that is demonstrated by αετοράχη "eagle-back = sheer mountain peak", χειρομάντης "palm reader". (Yes, there's nothing vernacular about χειρομάντης—except for its ending, which was remodelled from the actual Ancient χειρόμαντις. In its μάντης guise, it's still a productive pattern, as in slang.gr's predictably scatological κοπρομάντης, "practitioner of copromancy". You may not want to google that...)
I'm *not* vindicated by ασημόσκονη, καρβουνόσκονη, χρυσόσκονη, "silver dust, coal dust, gold dust", which are accented as if someone knew about Ancient recessive κόνις after all. But χρυσόκονις does turn up in Timothy of Gaza (§32). It's entirely possible that this word survived into the vernacular, and set up an analogy for other "dust" compounds. At least, I hope that's what's happened.
- Compounds based on nouns ending with consonants, as modern loans, are outside the bounds of Greek morphology; so they will leave the accent alone. βλαχομπαρόκ "hillbilly baroque, kitsch" is literally "Vlach Baroque", with the timehonoured Greek denigration of the Vlachs as mountain folk. There was no way μπαρόκ would be reaccented as *βλαχόμπαροκ. (The Vlachs translated their transhumant pastoralism into transnational financial acumen, and Vlachs financed many of the major cultural institutions of Athens. But stereotypes don't pay attention to such niceties.)
Ditto the delightful coinage of Kaliarda (the Greek gay cant), ζουζουνοσάιλοκ, "Shylock bug = ant", because ants hoard food. OK, it's not delightful because of the easy antisemitic trope; but Kaliarda is full of such far-fetched (and overeducated) cleverness.
One more speculation—although this is being vindicated from what I'm seeing of compounds. The inflection mattered in Ancient Greek for where the accent would end up in compounds—depending on whether the inflection was a long or short vowel. Modern Greek does not have long and short vowels, and it has boiled down the ancient declensions into what, for the nominative singular at least, is a simple pattern: masculines end in -Vs (where V can be any vowel), feminines and neuters end in -V. (Other cases and plurals are a lot messier, but that messiness is not added to with compounding.)
So how should the modern declensions behave with recessive accent?
- Masculines and neuters ending in -os, -o are the simplest case, since this ending has changed the least from Ancient Greek second declension. -o nouns were recessive in Ancient Greek, and this has remained the case. There are lots of exceptions to recessiveness for -o, and they are all verbal nominals or psychopomp-compounds; that too has remained the case.
Feminines in -o (-ω) are a fascinating grabbag of nouns: Ancient names in -ώ like Sappho and Calypso; Mediaeval names in -ω like Κρυστάλω "Crystal", modelled after the Slavonic feminine vocative; the Modern slang derivational suffix -ω (e.g. μαλάκω, σπασίκλω); and the erstwhile normalisation of Ancient feminines in -ος (η Κόρθο, η μέθοδο). It's a fascinating grabbag, but it's marginal; and outside the now abandoned forms like μέθοδο, it provides no reason for accentuation on the antepenult.
- The -e(s) and -u(s) declensions are also quite infrequent: -e(s) nouns are Turkish and Romance loans, -us nouns are a couple of Ancient leftovers (παππούς "grandfather", Ιησούς "Jesus"), -u nouns are the convention for borrowing Turkish feminines in -ı/i, or deriving them locally from Turkish masculines (typically the feminine -τζού "professional of X, wife of professional of X", from Turkish -cı/ci, as in καφετζού "café owner's wife, tasseomancer", στριπτιζτζού "stripper"). Again, no reason to assume accent on the antepenult, and not a whole lot of nouns following the pattern to begin with.
- The -i(s) nouns are mostly derived from Ancient -η(ς), and in Ancient Greek they were accented only as far back as the penult, because of the long syllable. This has been preserved in Modern Greek; as we saw in previous posts, it would take overwhelming analogy to dislodge the accent of -η(ς) to the antepenult; and that hasn't happened with compounds. The exception is with Ancient nouns that originally ended in -ῐς instead of -η; and again, as I've argued, it's analogy that has allowed those nouns to remain an exception.
That leaves the -a(s) nouns. There is good reason for there to be confusion about where to accent such compounds. That's convenient for me, because it justifies me getting the accent of ξέμαγκας wrong, and it explains the variability that was pointed out in comments last post, between παλιοπαπάς and παλιόπαπας.
At least some of the confusion is as old as Ancient Greek. Feminines could end in both -ᾰ -and -ᾱ; the former could be accented on the antepenult, the latter could not. As a result, there are compounds of the former, which are recessively accented on the antepenult (λιμνο-θάλασσα "lake–sea = lagoon"), and compounds of the latter (περι-τιάρα "round cap"), which are recessively accented on the penult. To a Modern Greek speaker, the latter would look accent-preserving, because there are no long syllables in Modern Greek to explain why the accent doesn't move.
There are a lot less penult -ᾱ(ς) nouns in the Modern Greek vernacular: -έᾱ(ς), -ίᾱ(ς) /-ˈea(s), -ˈia(s)/ in the vernacular went to -ιά(ς) /-ˈja(s)/, so if the vernacular were left on its own, few nouns would look like offering a counterexample to the recessive accent of θάλασσα. Except that -ία is a very productive verbal nominal suffix, and would not have been reaccented in the verbal nominals inherited from Ancient Greek. Like ανεμοβλογιά /a.ne.mo.vlo.ˈɣja/ "chickenpox" < *ανεμ-ευλογία /anem-eulogía/ "wind-blessing", which did not end up as *ανεμόβλογια /a.ne.ˈmo.vlo.ɣja/.
Masculines in Ancient Greek only ended in -ᾱς (or -ης, as a regular sound change); so they were never accented on the antepenult. But as we also saw a couple of posts ago, Modern Greek is awash with antepenult nouns ending in /as/, from the erstwhile third declension. So again, we have some compounds ending in -ας whose accent is antepenult; and some compounds which cannot possibly be antepenult—because they are preserving the Ancient Greek moras, as first declension nouns. Again, they will look to Modern Greek speakers as arbitrarily accent-preserving. Hence, the potential for confusion in general with -a(s) nouns, as to whether they are accent preserving or not.
There will be maybe a couple of thousand compounds out of slang.gr, that I will go through in a future post, to see if these claims are borne out by the numbers. Yes, the corpus is problematic; I'll talk about that too, and I'll make up some excuse or other for insisting on using such a fun word list.