Accent in Ancient compounds

We left off with the tendencies on how to accent words in Ancient and Modern Greek. But our target (or at least, my target) is to work out the rules behind the accent of ξέμαγκας, a compound. Which means we now get to look at the rules for how to accent a compound in Ancient Greek—before we sneak a peek (which may not be linguistically impeccable) at how Modern Greek tried to reinterpret those rules.

Accentuation of simple stems in Ancient Greek, we saw, is mostly recessive; two thirds of the time for nominals. Accentuation of compounds is also mostly recessive. On top of that default, Ancient Greek has a couple of floating islands of delicate semantic distinctions made by accent. And in between the floating islands and the recessive default, there is a sludge of accent confusion.

To navigate the confusion, I'm going to refer to a German, as one always should when looking at Ancient Greek grammar. Todays's (Swiss) German is: Albert Debrunner. 1917. Griechische Wortbildungslehre. Heidelberg: Carl Winters. (Available at archive.org.) On pp. 77–79, Debrunner tries to work out rules for where accents go in compounds, and he almost succeeds:

  • If the second word in the compound is a monosyllable, the compound is accented on the ultima if the monosyllable is long (εὐ-κράς < κρᾱς), and on the penult if the monosyllable is short (εὔ-ζυξ < ζῠξ). It's like the three-mora rule has turned into a two-mora rule for these compounds, and there's an extra mora somewhere that we're not seeing.
    In fact, there is an extra mora, at least for some of these monosyllables: it's the lengthening of syllable through the final cluster. A final short vowel followed by /-ps/ or /-ks/ counts as two moras, and cannot be accented on the antepenult: φύλᾰξ "guard" produces οἰκο-φύλᾰξ "house-guard", not *οἰκό-φυλᾰξ. But working out what's happening here properly would need more Indo-European skills than I command; so I'm going to move on from these.
  • Exocentric compounds with simple nouns are recessive. Exocentric compounds (bahuvrihi) are compounds where X–Y is not a Y, but something else—typically an adjective; so εὔ-θυμος "merry" does not describe a θυμός ("mood"), but a person who is "good-mooded". Notice that the original accent of θυμός has been scratched: whatever the accent on Y used to be, an exocentric X-Y is recessive.
  • Exocentric componds with verbal nominals (nouns and adjectives derived from verbs) are recessive, so long as X is not a preverb: θεό-δμητος "built by a god", ἄ-γνωστος "un-known".
  • Proper names in Ancient Greek are often exocentric compounds, which is why they are usually recessive: Δημό-κριτος "judged by the people = Democritus", Θρασύ-μαχος "bold in battle = Thrasymachus".
  • Endocentric compounds where Y is a simple nominal—are also recessive: σύν "with" + δοῦλος "slave" = σύνδουλος "fellow slave", ἰατρός "doctor" + μάντις "soothsayer" = ἰατρόμαντις "doctor-cum-soothsayer".
  • So what *isn't* recessive? For starters, compounds where Y is an adjective ending in -ης -ες. The rule there seems to be, the ultima is accented if the penult is short (εὐ-γενής "good-breeded" = "noble"), and accent is recessive if the penult is long (εὐ-ώδης "good-smelled" = "fragrant"). This is an exception to the exocentric rule, and it has exceptions of its own.
    • Proper names ending in -ης -ες are recessive no matter what the penult length—because of analogy with all those other exocentric proper names: Σω-κράτης (κρᾰτης) "having saving power = Socrates", in contrast to ἰσο-κρατής "having equal power"
    • On the other hand, there are -ης -ες compounds where the penult is long but which are still stressed on the ultima, and Debrunner has no idea why: ἀ-ψευδής "un-false = truthful", νη-μερτής "un-erring".

  • But the real source of compounds that aren't recessively accented, as I hinted, are preverb + verbal nominal compounds. In such compounds, each suffix has its own rule. Some suffixes keep their accent in compounding: βολή "throwing" > ­ὑπο-βολή "throwing under; secretly substituting; reminding". Others are recessive: στάσις "standing, station" > ἀνά-στασις "resurgence; resurrection". (You can tell that the -σις ending is recessive from other verbs, like αἰσθάνομαι "I sense" > αἴσθησις "sensation".)

Still other preverb + verbal nominal compounds allow both accents; those are the two floating islands I referred to earlier.

-τος distinction

The first floating island of a neat distinction is with verbal adjectives ending in -τος—which are equivalent to passive participles. Tucked away in grammars (Smyth 425c) is the observation that if they are recessive, they indicate a permanent state of affairs; if they are oxytone, they do not. So ἐξαιρετός is something that can be picked out at any time—removable; the removal is not permanent. But ἐξαίρετος is something that has been picked out for good—something that is not "selected", but "select": something choice, special.

I get the impression this rule gets more attention from students of New Testament Greek than of Classical Greek, because it explains John 14:16: the Holy Spirit as the Paraclete is not who you call on to your aid as a one-off (παρακλητός), but as a long standing intercessor (παράκλητος).

The distinction is real enough in Ancient Greek; but I have to question how real the rule was as late as the New Testament. Dio Cassius, writing a century later, uses the same word παράκλητος (in the genitive plural παρακλήτων); but LSJ glosses his use as "summoned" (46.20.2: τὴν ἀγορὰν καὶ τὸ Καπιτώλιον ἄλλων τέ τινων καὶ δούλων παρακλήτων πληρώσας "Did you not [...] fill the Forum and the Capitol with slaves, among others, whom you had summoned to help you?") That is a temporary rather than permanent state of affairs; but Dio Cassius is not taking the opportunity to make the fine distinction.

If you want to understand why John the Evangelist use παράκλητος, it's more useful to look at how the word was used further back in time, when the rule was still operative: in Demosthenes, to mean "legal assistant, advocate". That meaning, too, refers to a permanent state of affairs. But the coding of permanence through the accent location is something that happened in Athens, not Judaea. John the Evangelist probably didn't have access to the fine distinction between παρακλητός and παράκλητος any more, if Dio Casssius didn't. (And whaddaya know, while KJV has "Comforter" in John 14:16, and the NASB has "Helper", NIV has "Advocate".)

It's noticeable in fact that the semantic distinction in accent does not get a lot of prominence in general. Debrunner book-length treatise on compounding doesn't mention it at all; and LSJ gives it short shrift as well in its definitions. This was a nice distinction, but not a distinction that endured. As Nikos Sarantakos likes to say, and I'll quote him again: "Nice distinctions burn nicely."

o-grade adjective distinction

The second floating island is noticed by Debrunner, and he spends some time on it. Greek has nouns of the form CοCος, related to verbs of the form C(ε/ο)Cω. These nouns typically indicate an agent doing the verb: so τρέφω :: τροφός "I feed :: feeder", πέμπω :: πομπός "I send :: sender", λέγω :: λόγος "I say :: speech" (but "speaker" in compounds).

There is ablaut going on, which is why we can refer to these nouns as o-grade. As the ablaut in the stem indicates, this alternation is Indo-European–old, and the noun is not considered as derived from the verb, the way πεμπ-τός "send" or λεκ-τέος "to be spoken" would be. But speakers of Ancient Greek were not historical linguists, and the nouns looked like verbal nominals. Which allowed an accent distinction to be set up.

Ancient Greek used these nouns in X–Y compounds in an active sense, to indicate that Y was doing X: ψυχο-πομπος "sender of souls, psychopomp" (Hermes escorting souls to the underworld). But Ancient Greek also used these nouns in X–Y compounds in a passive sense to indicate that Y had X done to it: θεο-πομπος "sent by God, God-sent".

The passive sense looks like any number of exocentric compounds: X–Y is not a Y (a sender), but something else (a sendee), just as εὔ-θυμος is not a mood, but a mood-ee. The passive sense also looks like passive compounds with verbal nominals, like θεό-δμητος "built by God". "Built by God" has a recessive accent, and so does "sent by God": θεό-πομπος.

The active sense, on the other hand, is not exocentric, and it did not look like the normal endocentric compounds. What it looked like was a verbal nominal compound: remember, Greek speakers couldn't really differentiate between πομπός "sender" and πεμπτός "sent"—or for that matter πεμπτήρ "sender". Verbal nominals tended to keep their accent in compounds. So ψυχο-πομπος was also made to keep its accent: ψυχο-πομπός. And this has turned into a general distinction in accentuation between active and passive senses: βούστροφος "ox-ploughed" :: βουστρόφος "ox-guiding", πολύτροφος "well-fed" :: πολυτρόφος "nutritious", φυτοσκάφος "digging around plants" :: φυτόσκαφος "dug in preparation for plants".

This distinction too breaks down eventually; while Diodorus Siculus and Lucian correctly use εὔτροφος to mean "well-nourished", Theophrastus uses it to mean both "well-nourished", and instead of εὐτρόφος to mean "nourishing".
Ἐνιαχοῦ δὲ οὐ γίνονται τὸ ὅλον ἶπες ὅταν εὔπνους τε καὶ μὴ ἔνυγρος μηδ’ εὔτροφος ὁ τόπος ὑπάρχῃ. "In some places woodworms do not appear when the place is well-ventilated, and neither damp nor nourishing. (De Causis Plantarum 3.22.6)
οὐ δύναται δ’ ἅμα ταῦτα διὰ τὴν πρότερον λεχθεῖσαν αἰτίαν, ὥσπερ τὰ εὔχυλα καὶ εὔτροφα "And these cannot take place at the same time [both harvesting and begetting fruit] for the aforementioned reason, like juicy and well-nourished [trees]." (De Causis Plantarum 1.20.3)
Yes, Theophrastus' is the most corrupt text of the Classical Canon; but corruption happens where confusion is easy. Again, nice distinctions burn nicely.

The oddity about the distinction made between a recessive passive and an accent-preserving active is, it seems to misapply the rule for verbal nominal accents. Verbal nominals preserve their accent in compounding, it's true; but as Debrunner put is, that only happens with preverbs—prepositions and particles. "God" is not a preposition, and following a noun like "God", the compound is meant to be recessive.

As it turns out, the verbal suffixes for agents all end in long syllables: -τωρ, -τήρ, -της, -τής. And they would have looked like they retained their accent, notwithstanding the rule:
  • The agent endings in -τωρ and -της would retain their accent anyway, because of the mora rule: κυν-ηγέτης "dog-leader", θεο-δέκτωρ "god-receiver".
  • The accented -τής ending could have gone recessive in compounds; but accenting -τής is a classical innovation (by analogy with-τήρ). So there would have been no instances of -τής at the time, to serve as a counterexample. And by Plato, the rule has broken down anyway: δημ-εραστής "friend of the people", not *δημ-εράστης.
  • I don't see any recessive noun-agentive compounds ending in -τήρ in LSJ.
  • μηλο-βοτήρ "sheep-grazer" in Homer is not accented recessively; if it is a compound of μῆλον and βοτήρ, it is violating Debrunner's rule. We could argue instead that it is formed straight from the compound verb μηλοβοτέω "to sheep-graze", which would explain why the -τήρ suffix remains accented. Except that μηλοβοτέω is from Hesychius, and could well have been made up by Hesychius to explain μηλοβοτήρ. Likewise, we can't derive δημ-εραστής from δημεραστέω "to love the people", because Olympiodorus made up the verb δημεραστέω to explain δημ-εραστής—at around the same time as Hesychius.

Greek-speakers wouldn't have seen those subtleties; they would have seen a bunch of agentive suffixes keeping their accents in compounds. So, with the misunderstanding (logical abduction) that underlies linguistic change, they treated o-grade nominals like -τρόφος as yet another agentive suffix preserving its suffix.

In fact, now that I look at μηλοβοτήρ, I'm not convinced Debrunner's rule does apply here. Debrunner's examples of recessive noun+verbal nominal compounds all involve -τος adjectives, which are usually semantically passive. If Homer accented the noun as μηλοβοτήρ, that shows that agentive suffixes, which are semantically active, kept their accent from the beginning, whether preceded by a preposition or a noun. And in that case, making an agentive -τρόφος or -λόγος preserve its accent was not particularly wrong.

Our second nice distinction, then, seems to hinge on a rule that Greek-speakers may have worked out on the spot—or may have got more right than Debrunner did, to begin with.

That second distinction came close to being stillborn in Proto-Greek: nice distinctions burn nicely, because they are so fragile to phonological change. The suffix -πομπός corresponds to the noun πομπός, but the suffix -τρόφος corresponds to the noun τροφός. The suffix is -τρόφος, not *-τροφός, because of an old phonological rule Debrunner reports, that the accent went to the penult in words with dactylic metre (short–short–long). (In -οτροφος, the final /s/ makes the final syllable metrically long, so the rule applies; in -οπομπος, the middle /mp/ makes the middle syllable metrically long, so the rule does not apply.)

If that rule had made the accent go to the antepenult instead of the penult for -τροφος, we would have lost our fine distinction between active and passive compounds. Or rather, we would have lost it a millennium before Theophrastus forgot about it.


The system Debrunner describes is early Greek, and compounding gets more complicated the later you get. There is an innovating class of compounds where there is reason not to have recessive accent, but to have the accent preserved. These are X–Y compounds where X is originally a nominal, but has turned into a prefix, with a very simple meaning. Many of those prefixes are familiar to English: philo-, neo-, pseudo-, miso-, archi-, homo-.

Old Greek had prefixes that were prefixes: a-, dys-, eu-, hemi-. Being old, those prefixes had recessive accentuation. But when a new batch of such prefixes was created, Greek started making a new accent distinction. If the first half of the compound was a full-fledged word, accentuation was recessive—because the X–Y compound was a different thing than the Y itself; so it was accented as a new word. But if the first half of the compound was a prefix like "philo-", Y could keep its accent, because X was semantically subsidiary to Y. So Diogenes Laertes could accent φιλο-γενναιος "loving what is noble", not as φιλογένναιος, but φιλογενναῖος.

The accents get preserved in Y only in late Greek; a lover of the grape in Phanocritus (preserved in Athenaeus) is still φιλόβοτρυς, not φιλοβότρυς. We can follow the development of pseudo- with Debrunner (p. 57).
  • To begin with, Homer has ψευδ-άγγελος "messenger of falsehood". This is a normal endocentric compound, and it's recessively accented—although ἄγγελος itself is recessive, so you could not tell if the compound were not recessive. Semantically, this is an agentive compound, like μηλοβοτήρ "sheep-grazer"; so it follows that established pattern.
  • Aristophanes has ψευδολόγος "falsehood-talker". This is an o-grade compound like πολυτρόφος "nourishing" or ψυχοπομπός "soul-sender"; and again, it is an active, agentive compound.
  • Sophocles has ψευδοκῆρυξ "herald of falsehood", which is still recessive—remember that final /-ks/ counts as an extra mora.
  • ψευδό-μαντις "false prophet" in Herodotus is unambiguously recessive.
  • Debrunner confidently brings up ψευδόμαρτυς "false witness" as another recessive instance. The problem is, we can only tell between a recessive ψευδόμαρτυς and an accent-preserving ψευδομάρτυς in the nominative singular; the plural ψευδομάρτυρες could belong to either. But we have no good ancient instances of either ψευδόμαρτυς or ψευδομάρτυς.
    We have an instance reported for Critias by Pollux, but Pollux is a ii AD grammarian, and it's not clear whether Critias uses the actual form. (Pollux 6.152: "[plural] ψευδομάρτυρες is used in Critias, and ψευδομάρτυς—I don't know where. And the same author says ψευδομαρτυρεῖν 'to bear false witness' somewhere, and Demosthenes says καταψευδομαρτυροῦμαι 'to have false witness borne against one'.")

  • Then ψευδο- starts being used as a real prefix, without Y being an agent. At the start, the accent is still recessive; ψευδόδειπνον < δεῖπνον "fake dinner" (Aeschylus), ψευδοπάρθενος < παρθένος "fake virgin" (Herodotus), ψευδηρακλῆς < Ἡρακλέης "fake Heracles" (Pherecrates)—remember that final circumflexes are recessive in proto-Greek.
  • And then the recessive accent stops. Debrunner quotes Lucian as using ψευδάττικος < Ἀττικός "fake Atticist"; but our edition of Lucian has Ψευδαττικόν, and Phrynichus castigates ψευδαττικοί for their poor usage. (Yet another instance of the fish-in-a-barrel sport of catching a pedant erring against the language standard he fetishises.) Diogenes Laertius cites a comic using ψευδαλαζών "lying braggart"; Josephus has ψευδιερεύς "false priest"; Dio Cassius has ψευδαντωνῖνος "pseudo-Antoninus", Ignatius of Antioch has ψευδοϊουδαῖος "fake Jew", Epiphanius has Ψευδοβραχμᾶνες "fake Brahmins".

It could be that these examples reflect a general breakdown in accent placement, and aren't specific to prefixes like pseudo- and philo-. I hope they don't, because I plan to use them in my argument in a couple of posts.

Conclusion for now

So. It's an analogical mess, as language synchrony is once you look at the diachronic fine print; but we do have an overall picture:
  • Compound stress is recessive by default, except for:
  • X–Y compounds where Y is a single syllable. (They didn't survive into Modern Greek)
  • X-Y compounds where Y is an adjective ending in -ης -ες after a short syllable—and X–Y is not a name. (They didn't survive into Modern Greek either.)
  • Lots of X–Y compounds where Y is a verbal nominal, and X is a preverb.
  • It looks like some X–Y compounds where Y is a verbal nominal, and X *isn't* a preverb—notably, Y as an agentive.
  • The verbal nominals are consistent for each distinct suffix—except for a fine distinction of permanent vs temporary -τος compounds (which didn't survive), and a less fine distinction of active vs passive o-grade nominals (like -πομπός and -τρόφος).

What does this mean for Modern Greek? With all the distinctions that didn't survive, a safe default of recessive accent, and lot of exceptions involving verbal nominals.

With no clear sense of what was and what wasn't a verbal nominal, because the verbs those nominals came from often had not survived. Which throws the Modern Greek system into disarray, and that is the topic of the next post.


  1. As far as I can tell, not all of the 'active' ψυχο-πομπός type compounds have o-grade in the verbal element: there's δημιοεργός 'craftsman', ἀργός 'idle' (<ἀ-εργος). (The accent of the second of these puzzled me until I read this post, since alpha-privative compounds are generally recessive, and I'd been thinking of it as a bahuvrihi, 'one whose work is not'; but it must be an active verbal compound just like ψυχο-πομπός, except with a non-nominal first element.)

    Then again, I can't think of other such compounds with e-grade in the second element, and maybe there's some special rule of PIE ablaut that accounts for these.

    (Fascinating blog, by the way! Here through Language Hat.)

  2. Tom, welcome!

    I think you're right about ἀϝεργός > ἀεργός > ἀργός following the ψυχοπομπός pattern; I certainly don't see another explanation for the accent being where it is.

    If ἀϝεργός has the active sense, "not working", we expect that there would also be a passive sense ἄϝεργος, "not worked on". As it happens, such a passive adjective was coined later on, as ἄνεργος "not done"; it was coined in Classical times (Euripides), when the initial digamma of ἔργον was long forgotten.

    The sound of heads exploding you hear is of Modern Greek-speaking readers, for whom ἄνεργος means not "not worked on", but "not working, unemployed". In later Greek, the meaning distinction between active and passive accented compounds of this kind was erased. In fact the meaning "inactive" of ἄνεργος is as early as Josephus (at least as a variant reading), according to LSJ.

  3. ... And of course, recoining ἄϝεργος as ἄνεργος means that by the time of Josephus (or whoever), noone could connect ἀργός to ἀ + ἔργον. Which is no surprise.

    ἀϝεργός "not working" contrasts with ἐνεργός "active", which confirms that the accent placement is no accident.

    Having ἀ- privative without recessive accent is still decidedly odd...

  4. Hmm, I hadn't thought of ἐνεργός. But it just deepens the mystery, doesn't it? It certainly can't be a ψυχοπομπός-type verbal compound, and prepositional compounds are generally recessive just like alpha-privative ones: ἔνοπλος, ἔνθεος, ἄποικος, etc.

  5. Whenever a rule that should have been followed doesn't get followed, our first thought is analogy.

    I feel dirty inside for suggesting this, because it's a cheap thing to do, and ἀϝεργός looks too old for it to work. But Greek is full of psychopomp-style -εργός, active noun–noun compounds. In Homer, I count ὀβριμοεργός, ταλαεργός, κλυτοεργός, δημιοεργός, ἐντεσιεργός. Analogy from this ending, as particularly productive, back to ἀϝεργός?

  6. I would have thought the analogy of general recessive accentuation in alpha-privative and prepositional compounds would be much stronger... but I can't think of a better explanation.

    Is there a way to search the TLG word index for, e.g., words beginning in a- and ending in accented -os? (I've figured out how to search for #a...os# without the accent, but of course that brings up countless recessively accented compounds.)

  7. Tom: I know what's supposed to work—

    * Advanced search
    * Wildcard search
    * Left sidebar: "With diacritics": select "accents only"
    * Search for: ^A.*O/S$

    —but it's not working; I've reported the bug. I've done a search, and there are some 200 adjectives matching the ἀ.*ός pattern and up to 10 Beta-code characters long; most of them of course are nothing to do with alpha privative...

  8. Waiting with bated breath for the next installment :)

  9. That's the instalment I'm afraid of, where I try to work out the rules (or rather, trends) of Modern Greek. I'm going to do some counts to confirm my hypotheses, so it won't be up for another day or so...

    You already know how it will end though (in the post after that): I thought it was ξεμάγκας by analogy with ξε-μά (or ξε-Τίτο, as in Josip Tito, which I found online). The point of the posts is, Modern Greek is so confused by the Ancient accent system, that it's *all* analogy. But stay tuned for detail.

  10. I am waiting, but to preempt you a bit, it seems to me that the case of ξε-Τίτο is totally different. Consider the following two phrases:

    Μην τον ακούς αυτόν τον ξέμαγκα, τώρα παντρεύτηκε και νοικοκυρεύτηκε.
    (he no longer is a μάγκας)

    Μάγκας-ξεμάγκας, όταν σου μιλάω θα με ακούς

  11. Ah, I don't dispute that you're right. I'm saying that I've never learned the former pattern in my (defective) idiolect, and it was easy for the latter, productive pattern to make more of an impression on me.

  12. The ἄϝεργος > ἀργός and ἄνεργος alternation reminds of English apt and inept. Now Old Latin had initial stress and vowel reduction, which accounts for the e in the latter. The pre-Classical change to the stress rules made sure that it would arrive as inépt. But it's semantic change in English that has pried this pair apart, causing the appearance of unapt 'unsuitable', and jocularly ept 'proficient, not inept'.


The Other Place (opɯcɯluklɑr)

Powered by Blogger Widgets