Pronouncing Ancient Greek: Food for Classics Students' Thought

How did pre-modern Greeks speak?

More importantly, what is the modern learner of pre-modern Greek to do about it?

And which version of Greek, while we're at it? Greek was, as indeed it is, like any other language in having regional and chronological variation. The the development of one sound in different versions of Greek will illustrate this. What we normally represent with the Greek letter θ (known these days by its Byzantine name of "theta") was pronounced as a voiceless aspirate [tʰ]  (like the "t" of English "top") in the Attic dialect of Athens during the fifth century B.C. In the Doric dialect of Sparta in the same period, however, it was a fricative [θ] (the "th" of English "thug") as in Modern Greek, having changed from [tʰ] some time in the preceding century. It was the [tʰ] version that made it into the Koine, the roughly standard version of the language that developed all over the eastern mediterranean at the end of the 4th century B.C.. By the 4th century A.D, though, that intrepid  sound  had become a fricative [θ], just like its Spartan counterpart had a thousand years earlier. In the meantime that selfsame Spartan dialect, which survived in what is today Tsakonia, had transformed its own erstwhile [θ] into a plain old [s], which it still is in modern Tsakonian words of original Doric vintage.  

Similar issues obtain for other sounds. Limiting ourselves diachronically to the middle of the 5th century, a cross-dialect synchronic comparison of the evidence shows that Ionic, Cretan and Lesbian dialects no longer had their /h/ whereas Attic and some other dialects still preserved it. Doric had an additional /h/ in positions where everybody else has /s/ (rather like many modern Spanish dialects today.) Attic lost its /w/ sound when most other dialects still had it and developed a high front rounded /y/ (i.e. the u of French "lune" [lyn] or the ü of German "über" ['yːbɐ]) where most dialects had /u/. Attic's own /u/ sound, on the other hand, corresponds to the /oː/ of most other dialects (the "ô" of French "drôle" [dʁoːl]) Arcadian, unlike the rest, had what was probably a /tʃ/ (like the "tch" of English "match" [mætʃ]) complete with a special letter to represent it. This by no means exhausts the differences, which go beyond articulatory minutia to issues of morphology, lexicon and even prosody. Thessalian, for example, was probably stress-accented and stress-timed in contrast to the rest of the Hellenosphere which at the time had moraic timing and a pitch-accent. 

Given that the surviving texts written in something we call "Ancient Greek" come from a great many regions and were produced over the course of about a millennium, one cannot countenance an attempt at complete historical accuracy as a practical matter. Otherwise one will have to adopt different pronunciations and even wildly different prosodic systems for different periods, and even different authors of the same period. Even leaving aside the uncertainties in the relative chronology of sound changes and prosodic developments, it hardly makes sense for learners to be forced to use one pronunciation for Homer, another for Sappho and Alcaeus, another for Aeschylus, another for Demosthenes, another for Menander and so on.


The Classical Attic Option

For reading classical texts, many often recommend using the pronunciation of late 5th century B.C. Athens, since it is the best attested "classical" pronunciation and since Attic became the literary standard. The idea is that, though one may not be able to read Homer and Sappho like an 8th century Ionian or a 7th century Lesbian, one can at least read them with the pronunciation that, say, Plato would have given them. This is a sound solution for those who wish to limit themselves to pre-hellenistic Greek and who, as we'll see, don't mind implicitly making a few assumptions about Athenian sociolinguistics.

Below are some audio-examples (courtesy of yours truly) in conservative formal 5th century Athenian (with pitch accent, durational prosody, the works) as it is envisioned by modern historical linguists (the phonological reconstruction mostly follows Geoff Horrocks and W.S. Allen give or take a vowel. For data used to reconstruct the realization of pitch-accent and durational rhythm see Allen's "Accent and Rhythm" and A.M. Devine and L. Stephens "Prosody of Greek Speech")

A dialogue  (Click for MP3)

A: χαῖρε (Hi)
B: καὶ σύ χᾶιρε (Hi yourself)
A: τί πράττεις? (How're you)
B: καλῶς πράττω, καὶ σύ τί πράττεις? (I'm alright, and how about you?)
A: κακῶς πράττω! (Pretty horrific!)

A sentence from Plato: (Click for MP3)

Δοκῶ μοι περὶ ὧν πυνθάνεσθε οὐκ ἀμελέτητος εἶναι.

The Opening of the Odyssey: (Click for MP3)

Ἄνδρα μοι ἔννεπε, Μοῦσα, πολύτροπον, ὃς μάλα πολλὰ
πλάγχθη, ἐπεὶ Τροίης ἱερὸν πτολίεθρον ἔπερσε·
πολλῶν δ’ ἀνθρώπων ἴδεν ἄστεα καὶ νόον ἔγνω,
πολλὰ δ’ ὅ γ’ ἐν πόντῳ πάθεν ἄλγεα ὃν κατὰ θυμόν,
ἀρνύμενος ἥν τε ψυχὴν καὶ νόστον ἑταίρων.
ἀλλ' οὐδ' ὧς ἑτάρους ἐρρύσατο, ἱέμενός περ·
αὐτῶν γὰρ σφετέρῃσιν ἀτασθαλίῃσιν ὄλοντο,
νήπιοι, οἳ κατὰ βοῦς Ὑπερίονος Ἠελίοιο
ἤσθιον· αὐτὰρ ὁ τοῖσιν ἀφείλετο νόστιμον ἦμαρ.
τῶν ἁμόθεν γε, θεά, θύγατερ Διός, εἰπὲ καὶ ἡμῖν.

The Problem With Antiqued Attic Antics 

Exciting, and perhaps convincing, though it may be to be able to read things out loud with this "accent", there are a few problems with adopting it as a general way of reading Ancient Greek beyond its classical period, to say nothing of teaching it for any period to non-specialists.

First, there's the problem of anachronism. Quite a few important texts were produced centuries after the Athenian Golden Age, and reading e.g. a dialogue in Menander with the pronunciation of 200 years previous hardly makes any sense.

There's also the fact that the phonology and especially the prosody of this system are, while definitely the most probable layout for a given period, somewhat typologically rare and therefore both unstable and difficult to learn. The vowels fully contrast for height, backness, length and roundedness in some rather funky ways. For example there's a four-way contrast on the front axis between ε, η, ει and ῃ / e ~ εː ~ eː ~ ε(ː)i/. The symbol /ː/ means that the sound is lengthened. /εː/ is like the "aî" of French "maître" and /eː/ resembles the last vowel of French "parlé". This corresponds to a  two-way contrast on the back axis /o ~ ɔː/. Moreover, since one does not know precisely when the vowel of ω was raised, one is just as justified in pronouncing it as either /ɔː/ or /oː/ (the former is like the vowel of French "fort".) The obvious instability of such a system means that it was probably only around for a short while before subsequent chain-shifts rejiggered the vowel system once again, thereby making it an exceedingly narrow window into the past.

Likewise the combination of pitch-accent with durational prominence -a feature which there's no getting around- will take some getting used to. It is sometimes said, even by some who should know better, that Attic Greek was like modern Norwegian in that it had an accent system of stress+pitch with a lexical component in the alignment of accentual peaks. This seems quite unlikely in the face of the actual evidence (whose details need not detain us now) without some qualification. The likely reality is somewhat more "exotic" (from the perspective of a modern Euroglot, anyway.) Though accoustically reminiscent of French in some ways, the actual prosody of 5th century Athenian speech (and that of much of the rest of the ancient Hellenosphere) was more akin to the rhythmic and pitch patterns of Tokyo Japanese and, even more so, to certain lesser-known languages such as Nubi, Isthmus Zapotec and some southern Bantu languages where lexical tone and durational prominence coexist, as they did in Greek, in some fairly interesting ways. No European language works in this latter fashion, nor does any language a classics student is likely to have previously studied.

For me, personally, it was of some difficulty to acquire and put into practice (and that only once I had finally extracted the information in learnable form from the mountains of data and papers and specialized books.)

Less obviously, but perhaps more crucially, the issue is also one of synchronic variation within a single dialect.  Note the following from Plato's Cratylus (translation mine)

Σωκράτης: οἶσθα ὅτι οἱ παλαιοὶ οἱ ἡμέτεροι τῷ ἰῶτα καὶ τῷ δέλτα εὖ μάλα ἐχρῶντο, καὶ οὐχ ἥκιστα αἱ γυναῖκες, αἵπερ μάλιστα τὴν ἀρχαίαν φωνὴν σῴζουσι. νῦν δὲ ἀντὶ μὲν τοῦ ἰῶτα ἢ εἶ ἢ ἦτα μεταστρέφουσιν, ἀντὶ δὲ τοῦ δέλτα ζῆτα, ὡς δὴ μεγαλοπρεπέστερα ὄντα.
Ἑρμογένης: πῶς δή?
Σωκράτης: οἷον οἱ μὲν ἀρχαιότατοι ‘ἱμέραν’ τὴν ἡμέραν ἐκάλουν, οἱ δὲ ‘ἑμέραν,’ οἱ δὲ νῦν ‘ἡμέραν.’
........
Σωκράτης: καὶ τό γε ‘ζυγὸν’ οἶσθα ὅτι ‘δυογὸν’ οἱ παλαιοὶ ἐκάλουν.
Ἑρμογένης: πάνυ γε.

Socrates: You know that our ancestors used the sounds ῑ and δ well- especially women, who are most conservative of older pronunciations. But these days people change ῑ to ει or η , and δ to ζ, hoping to increase the grandeur and sublimity of the sounds.
Hermogenes: What do you mean?
Socrates: Well, for example, in ancient times the word for day was ἱμέρα /hi:méra/, whereas others said ἑιμέρα /he:méra/, but now we say  ἡμέρα /hεːméra/.
.........
Socrates: And you know the ancients pronounced ζυγόν /zdyːgón/ as δυογόν /dyogón/, right?
Hermogenes: That's right, they did.

Plato, obviously, was no great historical linguist. That said, what his ruminating puppets are describing is in fact likely not archaism, but rather innovation among Athenian women which had its origins in non-standard speech and, depending on the analysis adopted, was either later generalized into ubiquity or leveled into oblivion in later developments of the Athenian dialect.

By the latter analysis (the one I prefer), the two sound-changes here described /εː/ -> /eː/  and /zd -> (d)d/ were to be found outside of Athens. The presence of /dd/ (word-initial /d/) where Attic had /zd/ (both descending from an original /dz/ rather than the /d(y?)/ Plato postulates) is known to have been a feature of many of the ancient dialects in the immediate vicinity of Plato's stomping ground. For example, the Gortyn law-code from the mid-5th century B.C. shows Cretan καταδικαδδετο /katadikadːétːoː/ "let him condemn" as against early Attic κατδικαζέτω /katadikazdétɔː/. This was also true of the Beotian dialect of Thebes, spoken not too far from Athens, which also had /e:/ where Attic had /ɛ:/ (e.g. Theban χρειματα /kʰré:mata/ "money, property, capital" as against Athenian χρήματα /kʰrɛ́ːmata/.) In other words, Theban Greek shows possibly two features which Plato seems to perceive in the speech of the (presumably Athenian) women he had contact with. Though the consonant-shift is not a feature of later Attic, and we have no evidence for Beotian influence on standard Attic in this period, neither of these tell us much about non-standard (non-male?) speech of the time and place in question. It should be noted that since early antiquity, diverse circumstances of human and commercial mobility had acclimatized all manner of Greek dialects in Athens, as stated by Solon (24, 31.f) and Pseudo-Xenophon (II 8).

The second way to interpret the evidence is to conclude that the vowel shift is not /ɛ:/ -> /e:/, but rather /ɛ:/ -> /i:/ and, like the consonant-shift, is a purely independent innovation which presages later vowel-shifts known to have taken place in the Athenian vernacular more generally in subsequent centuries. In this analysis, the original /zd/ is not turned into /d/ but has been metathesized in women's speech into /dz/ and is on the way toward becoming /z/ as it did in all of Attic by the mid 3rd century at the latest. 

Either possibility suffices for our purposes. While the depiction of female behavior as a continuation of a tradition rather than (as one would expect) a mere perversion of, and deviation from, the male norm may be interesting for the classical historian, what I want to draw attention to is that Plato has furnished evidence that the speech of women, back there and back then, differed phonologically from that of men enough for attentive speakers like Plato to notice and ponder the difference.

This is not as far-fetched as it may seem to those familiar only with the urban and standard versions of modern western European languages. In many language-communities around the world, men and women have different pronunciation habits no less than people of different socioeconomic status.

For example, working-class male speakers of Jordanian Arabic in some urban regions often have [g] for classical Arabic /q/ where their female counterparts have [ʔ] (a glottal stop, pronounced like the articulatory break represented by the dash in "uh-oh"), such that قلقان qalqān "nervous" is pronounced [gəlgaːn]  or as [ʔǝlʔaːn] depending on gender. The [g] sound is in general construed as "manly" among the working class and [ʔ] the opposite. Thus, for example, [g] is the only socially-acceptable pronunciation in such contexts as army life or among law enforcement personnel, with the male use of [ʔ] somewhat stigmatized in such circles. On the other hand, among more affluent Jordanians some studies have found [ʔ] to be the dominant realization for both  men and women, with [g] demeaned as unrefined, coarse or low-class. In this particular case, we have good evidence showing that the [g] version is a borrowing from neighboring rural and Bedouin dialects which later entered urban speech and entered into competition with the native [ʔ] along gender and social lines.

It is in fact not uncommon for a language's pronunciation to be sensitive to gender, religious affiliation, age-group, social class, political inclination, tribal identity, professional vs. domestic settings or any combination thereof. In the Arabic dialect of Nazareth, Muslim women are more likely than Christian women to use [k] instead of [ʔ] for /q/. Among Tunisian Muslims the pronunciations of young men and young women resemble one another more than either of them do the speech of women over 40 who in turn display marked differences from men over 40, but Tunisian Jews display different patterns from this entirely. In some communities use of features typically associated with one's opposite gender can also be sensitive to whether one is talking about themselves as the main focus or another person or a particular event, whether they feel positively or negatively about what they're describing etc.

I mention this because it is a handy example of how sociological phenomena and linguistic habits can interact in some fairly complicated, circumstance-specific ways. You'll notice that above I described the recordings not as a demonstration of the pronunciation of 5th century Attic, but as a "conservative, formal" register thereof. It seems clear, at the very least, that conservative and progressive pronunciations coexisted even within the same city in the classical period (as Sven-Tage Teodorsson showed in his detailed study of Attic and Koine phonology- even if the precise nature of those pronunciations is off by a century or three.) If the trends associated with high vs. low speech registers in modern languages are any indication, pronunciation would also have varied by context within a single speaker's idiolect. By another analogy with colloquial Arabic, it may have made a difference in phonology, probably a somewhat significant one, whether an educated 5th century Athenian was speaking on stage or in a brothel. Moreover, if Plato's evidence means what it appears to, modern reconstructions of 5th century Attic may only represent the speech of privileged males, and then perhaps only in the most formal of contexts!

One of the most pervasive problems of historical linguistics, and historical dialectology in particular, is the researchers' neglect of the fact that the languages they study were once spoken by real frickin' people, and no group of real people is linguistically homogenous.

Taking the authenticity argument to its logical extreme, even if we had better information about how male and female speech differed in 5th century Attic, would anyone seriously then contemplate having female classics students be taught a different pronunciation from males, all for authenticity's sake?

Of course, authenticity isn't always worth a soldier's damn for the modern student. But even if it were, the version of Greek speech you hear in my recordings, then, the one typically imagined to be  that of 5th century Athens, is at best just that: a version. Heavily reliant as it is upon privileged male grammarians who had both a vested interest in presenting affluence and maleness as normative and femaleness/poverty a regrettable deviance,  it likely only tells one side of a very complex story.

In a sense, then, using a reconstruction of acrolectic 5th century Athenian speech is counterproductive. By putting the learner in the odd position of having to pretend to be an ancient Athenian, with hopelessly incomplete sociolinguistic knowledge, it merely reminds us of the futility of pretending to be what you're not. Learners of Greek, I feel, need something that it's possible to mispronounce and appropriate while maintaining access to a voice of the ancient world when reading ancient texts out loud, a version of Greek hospitable to foreign adoption, which would not be hopelessly "put on" in historical context and which retrospectively encompasses the great Hellenistic poets such as Callimachus and the unjustly neglected Apollonius of Rhodes. (I'm assuming for the purposes of this writing that historical pronunciations matter when it comes to appreciating ancient literature. I have some reservations about that myself but this isn't the place to air them. I'll do that later.)

The best candidate I can see is a streamlined version of middle Hellenistic Koine, by which I mean the rough supraregional norm of the late 3rd century B.C. See below for more details

2. Hellenism

Here's how that Hellenistic system looks:

Vowels (Click for MP3)
High front unrounded /e eː i iː/
High front rounded /y yː ø øː ~ øɥ/
High back rounded /o oː uː/
Low central /a aː/
Rising diphthongs /ai oi yi/
Falling diphthongs /iw ew aw/

Consonants (Click for MP3)
Labial:  /p pʰ b m f w/
Dentals:  /t tʰ d n/
Alveolars: /r l s z/
Palatal: /j/
Velars: /k kʰ g ŋ/
Glottal: /h/

Below is a description of how these sounds correspond to the graphemes used today to represent ancient Greek, along with example words and a recording thereof. The examples are taken from Athenian and Hellenistic comedy, classical epigrams, ancient graffiti and the writings of a few bloviating moralizers. To show a different side of Ancient Greek, I have taken care to include lexical items and semantic ranges which the beginning or intermediate student is unlikely to learn from a Classics teacher.

Examples (Click for MP3) 
ᾱ, - /a:/ as in στυγάνωρ /stygáːnoːr/ "man-hater, Amazon, disobedient wife"
ᾳ - /a:/ as in  ᾅδης /hâːdeːs/ "hell"
ᾰ - /a/ as in τέτᾰνος /tétanos/ "tension, hard-on"
αι - /ai/ as in ἡταιρηκώς /heːtaireːkóːs/ "gay prostitute, rent-boy"
αυ- /aw/ as in προσαυλῶ /prosawlôː/ "to give a blowjob, to fellate"

β - /b/ as in κασαλβάς /kasalbás/ "hooker,"

γ - /g/ as in ἀγροῖκος /agrø̂ːkos/ "slut, easy lay"
γμ - /ŋm/ as in πρᾶγμα /prâːŋma/ "act, deed, thingy, dick"
γγ - /ŋg/ as in συγγίγνομαι /syŋgígnomai/ "to have sex"

δ - /d/ as in δέφω /dépʰoː/ "to jerk off"

ε - /e/ as in πέος /péos/ "cock, schlong"
ευ  - /ew/, /ewː/ before a vowel as in ἐκμοχλεύω /ekmokʰléwːoː/ "to fuck the shit out of" (literally "to wedge open with a lever")
ει  - /eː ~ eːj/ before a vowel as in κιναιδεία /kinaidéːa/ "lust for young boys"; /i:/ in all other environments as in λείχοντες /líːkʰontes/ "lickers, men who eat pussy, muff-munchers"

ζ - /z/ at the beginning of a word; doubled as /zː/ in the middle of a word as in κασαλβάζω /kasalbázːoː/ "to whore out"

η- /eː/ as in σποδησιλαύρα /spodeːsiláwra/ "street-walker"
ῃ- /iː/as in χαμαιτύπῃ /kʰampaitýpiː/ "unto a whore"
ηυ- /iw/ as in ηὐρέθην /iwrétʰeːn/, and as /iwː/ before a vowel

θ - /tʰ/ as in ἄρθρα [ártʰr̊a] "joints, private parts"

ῑ - /iː/ as in φοινῑκίζω /pʰøːniːkízːoː/ "to eat pussy, to go down on"
ῐ- /i/ as in ἐγκολπίζω /eŋkolpízːoː/ "to take into one's bosom", "to insert into one's vagina, to fuck oneself with" (with female subject), "to take it up the ass" (with male subject)
ι - /j/ in borrowings as in Ἰανουάριος  /jaːnuáːrios/ "January" from lat. iānuārius

κ - /k/as in πορνοβοσκέω /pornoboskéoː/ "to waste money on hookers" 

λ - /l/ as in κόλπος /kólpos/ "lap, gulf, crotch"

μ - /m/ as in μείγνυμι  /míːgnymi/ "to mingle, to have intercourse"

ν- /n/ as in νεῦρον /nêwron/ "tendon, penis".

ξ - /ks/ as in ξύλον /ksýlon/ "wood" (in both the literal and figurative sense)

ο - /o/ as in σκατός /skatós/ "shit"
ου -/uː/ normally as in ἀναισχυντογράφος /anaiskʰyntográpʰos/ "one who writes obscene words"; also /w/ in foreign borrowings as in οὐαλεριος /walérios/ from lat. "Valerius"
οι-  /øː ~ øɥ/ as in ὀιφῶ /øːpʰôː/ "to fuck" ; /yi/ in borrowings when used to render the Latin "qui-"  as in κοίντος /kyíntos/ "Quintus."


π - /p/ as in κατάπυγον /katápyːgon/ "the middle finger" (used as an obscene gesture. Related to  καταπύγων /katapýːgoːn/ "buttboy") 

ρ - /r/ as in κασωρίς /kasoːrís/ "call-girl"; pronounced as a voiceless [r̊] at the beginnings of words ῥ as the second element of a doubled -ρρ- [rr̊], and when preceded by θ, χ or φ as in ἀφροδισιάζω /apʰr̊odiːsiázːoː/ "to have sex with"

σ -  /s/ normally as in σαύρα /sáwra/ "lizard, a long penis"; voiced [z] before μ, ν, β and δ as in the etymologically ironic λεσβιάζω /lezbiázːoː/ "to suck dick."

τ - /t/ as in τρυπάω /tryːpáo:/ (also τρυπῶ) "to bore, to pierce, to fuck"

ῡ- /y:/ as in πῡγίζω /pyːgízːoː/ "to sodomize"
ῠ- /y/ as in κύσθος /kýstʰos/ "cunt"
υι- /yi/ as in ἄγυιος /áːgyios/ "weak-limbed, unable to get it up"

φ - /pʰ/ normally as in φλέβιππος [pʰlébippos] "horse-hung"; used to represent the fricative [f] in borrowings as in φοσσᾶτον /fosːâːton/ "ditch, trench, twat" from lat. fossātum

χ - /kʰ/ as in αἰσχρουργέω [aiskʰr̊uːrgéoː] "to masturbate."

ψ - /ps/ as in ψωλοκοπῶ /psoːlokopôː/ with male object, "to turn someone turn on, to give someone a hardon." note also the middle voice, ψωλοκοποῦμαι /psoːlokopûːmai/ "to get a boner." (This verb is often found in brothel graffiti. I cannot possibly imagine why.)

ω - /oː/ as in λεωφόρος /leoːpʰóros/ "a woman who enjoys sex" (cause it was weird for women to like it too much)
ῳ - /oi/ as in ζῳον /zôion/ "animal, creature, penis" 

Spiritus asper - /h/ as in ἵμερος /híːmeros/ "desire, passion," also "sex-drive, ability to get it up"

If the choice of examples above appears unacceptably obscene to my more decorous readers (assuming my blog hasn't already driven them far, far away), I would advise them to consider what even Saint Paul, closeted heterosexual though he was, once said of the squeamish in another context:

οἶδα καὶ πέπεισμαι ἐν κυρίῳ Ἰησοῦ ὅτι οὐδὲν κοινὸν δι’ ἑαυτοῦ, εἰ μὴ τῷ λογιζομένῳ τι κοινὸν εἶναι ἐκείνῳ κοινόν.
I know, and am convinced by our Lord Jesus, that nothing is unclean in and of itself. Rather, if someone believes it is unclean, then for that person it is unclean. 
-Romans [14:14]

The phonological system I have briefly sketched above offers, in my opinion, the best shot at authenticity inasmuch as this is possible for a dead (or perhaps zombie) language.

First of all, the phonology is (as the term "Hellenistic" suggest) implicitly post-classical, and therefore allows one to read classical and most hellenistic texts without fear of anachronism. Furthermore, the vowel contrasts are somewhat less idiosyncratic in their complexity than those of its 5th century Athenian progenitor, and are probably easier to naturalize for a larger proportion of Greek-learners.

More importantly, and unlike mid-5th century Attic, the Hellenistic option does not necessarily give us the insoluble problem of needing to sound like a native. It reflects an age when Greek was being widely learned by non-natives and bilingualism was unusually common, (and where class and gender distinctions in speech would have been nowhere near as clear-cut as they could be in a stable, heavily stratified, monoglot society such as that of classical Athens.) In the 3rd century BC, the streets of Alexandria, Kos, Palmyra and Sepphoris were indubitably rife with all manner of bilinguals, non-native Hellenophones and Greek-learners, much as Anglophone bilinguals, non-native Anglophones and English-learners fill the streets of New York, London, Amsterdam or Manila today.

The system I offer is, as far as I can determine, a reasonable articulatory facsimile of the kind of Greek a well-to-do non-native in, say, 225 B.C. might have aspired to reproduce. That said, the student need not worry too much (if they're obsessive enough to worry at all in the first place) about whatever traces of foreignness may remain in his/her "Barbarian" accent, for in such a context of widespread adult language acquisition in the target timeframe there is little reason to consider this "inauthentic" in any meaningful sense.

Koine pronunciation was "common" in the most literal meaning of the word in that it seems to have had a somewhat stable, supraregional phonology distinct from the divergent sound-changes operating in various regional versions of Greek -both native and non-native- in the same period. An extremely rough parallel might be drawn to the modern colloquial ad hoc Koine of educated Arabs.. (One might compare also the tendency of Anglophone Arabic-learners to approximate this version of Arabic, at both the phonological and morphological level, in conversing with Arabs.)

It must be noted, however, that even Koine pronunciation could hardly have been completely uniform anymore than that of any other lingua franca (which is why I call this compromise pronunciation of mine a "streamlined" version of Koinizing Hellenistic Greek.) I have based my system on the learned Koine of Alexandria (where I actually happen to be as I type this), mainly because of its prominence as an ancient cultural center, but also because the papyrological and other data on Alexandria are (for sundry reasons) much greater than elsewhere. There were regions and regional registers in the late 3rd century where, for example οι was not realized as /øː/ or /øɥ/ but might be anything from /yː/ (in Boeotia) to /øi/ (in Attica) to the original pronunciation of /oi/ among the most educated Romans. Likewise, there must have those who, under the influence of surviving non-Attic speech, pronounced the voiced stops β δ γ, and even aspirated stops χ θ φ, as fricatives (just as in Modern Greek) although their speech must have been regarded as substandard. This all allows a reasonably wide latitude of possible pronunciations for the student today.

Pitch-Accent, Vowel Quantity and Other Stuff

Speaking of possible pronunciations, now is a good time to discuss the issue of accent. 5th century Attic, as we have seen, was -at least on the surface-  pitch-accented (we'll get to the qualification later.) The use of a stress accent of the kind found in English,  German or Russian, was no more than a peculiarity of Thessalian. If the comedies of a certain asshole named Aristophanes be any indication, foreigners who spoke with a stress-accent could expect a rain of almighty piss and ridicule to pour down on them.

There is good evidence, however, that in Hellenistic times both the conservative pitch-accent (of the type heard in Golden Age Athens) and a progressive stress-accent (of the type heard in English or Modern Greek) could be widely encountered.

That different regions had different prosodic habits is likely not only because of the general tendency of speech communities to develop regional differences when a language is spread over so massive an area, but because there is a great deal of evidence that suggests not only the existence of such differences but even some of their distinguishing features. One of the most significant is different regional spelling habits which developed in e.g. inscriptions, non-literary papyri, graffiti and the like. If one compares the spellings found in Athenian vs. Egyptian writing of the same period, one finds a considerable amount of difference.

The most widely-accepted explanation for the loss of vowel-quantity and pitch-accent is that the former conditioned the latter. Geoffrey Horrocks puts it thusly:

The loss of long vowels and diphthongs destroyed the environment for the occurrence of the circumflex accent (rise-fall on a single 'long' syllable) and so neutralized the contrast between the circumflex and acute (rise+fall over two successive syllables). This in turn highlighted the increase in volume that was almost certainly associated secondarily with the rise in pitch, a development aided regionally by substrate effects (e.g. Egyptian/Coptic had a strong stress accent) and one that led to the perception that increased amplitude rather than a rise in pitch was the primary marker of word accent. 

First, Horrocks' presumption that the contrast between circumflex and acute would have been neutralized by a loss of contrastive vowel-length has less truth to it than a pop star's autobiography. In languages which have a stress accent, particularly if they don't have phonological vowel-length, stress often manifests itself as not just greater intensity but also greater duration of the syllable's nucleus. I.e. accented vowels will tend to be not just louder but also longer than their unaccented counterparts (as in e.g. modern Greek.) This can even happen when there is phonological vowel length. (as in standard Serbo-croatian). When this is the case it is quite possible to have an  intermediate system of pitch-differentiated stress, (again as in Serbo-croatian) whereby a stress accent is combined with lexical contrast in the  alignment of the accentual peak. i.e. the pitch can rise early or late in a stressed vowel. This could easily have remained a contrastive feature in Greek for accented syllables that were also historically long and could thus carry either an acute (manifesting itself as a rising pitch) or a circumflex (a fall in pitch). There is thus no reason why a neutralization of phonological vowel-length need prevent either phonetic vowel length as one of the manifestations of stress or contrastive pitch contours on a stressed syllable. Indeed, the evidence available suggests quite the opposite. In other words, the tonal contrast of minimal pairs such as  φῶς "man" vs. φώς "light" could have easily survived both the stress-reenforcement of the accent and the loss of phonological vowel-length. In truth there is scant evidence that what Horrocks describes really happened. What is far more likely is that a stress accent arose before the general loss of quantity distinctions. This resulted in the accented syllable attracting additional accentual exponents, namely increased duration. The quantitative pattern was in conflict with the tendency to lengthen stress-reinforced accented syllables and shorten unaccented ones, and therefore eroded.

The widespread idea that vowel-length was lost first seems to be the unfortunate result of research  conducted by classicists and papyrologists based upon flawed assumptions. The method of investigation has usually been to compare misspellings of etymologically long vs. short vowels found in inscriptions and papyri with no regard for accentuation, on the rather questionable assumption of equal accentuation for long and short vowels. But in fact, graffiti, papyri and inscriptions show that even in the first century or so of the Christian era, an etymologically short vowel was far more likely to be confused with an etymologically long vowel when accented and that speakers still had an intuition of length being a distinctive feature in unaccented syllables. This suggests that vowel-length had yet to be completely neutralized and was accessible to speakers who equated accent-lengthened, but etymologically short, vowels with historically long vowels elsewhere. Moreover, even this by itself doesn't even necessarily mean that pitch ceased to be the main accentual feature. All it definitely suggests is that by this time the pitch accent had come to be reenforced by duration. Modern greek, however, shows articulatory intensity as an exponent of accent as well, and also it seems clear that intensity had jumped onto the accentual bandwagon in most areas at least by the end of the 2nd century AD. 

Thus, where the old pitch-accent had become a stress-accent, the quantitative distinctions of vowel-length were not immediately obliterated. Contrary to what some people who faint at the sight of data suggest, it is, in fact, possible for a language to have a contrastive stress-accent and contrastive vowel length at the same time. An example is Hebron Arabic where not only does the initial stress of شفنا  /ˈʃufnaː/ "we saw" contrast with the final stress of شفناه /ʃufˈnaː/ "we saw him" but the long /aː/ of حاضر  /'ħa:ḍar/ "he harangued, he lectured" contrasts with the short /a/ of حضر  /'ħaḍar/ "he showed up".  Other languages that allow contrastive stress to coexist with contrastive vowel-length include Estonian, Komi and some versions of Australian English ( e.g. Australian [bɪd] "bid" vs. [bɪːd] "beard".)
   
W.S. Allen, whom I hate to love, recommends against attempting a reconstruction of the ancient pitch accent in practice, on the following basis:

We probably have sufficient knowledge to achieve a rough approximation to the melodic pattern of isolated Greek words...but...we know virtually nothing about 'melodic syntax' i.e. the way in which such patterns interacted with one another and with clause- and sentence-intonations in continuous speech. To judge from what we find in living tonal and melodically accented languages, these interactions may be extensive and complex. Given the melodic patterns of the word-isolates in such languages, it is of course possible to derive the melodic sentence-pattern from them- but the latter is not usually a simple summation of the former. The author has listened to a number of recordings, recent and less recent, of attempted melodic-accentual recitations of ancient Greek, and, whilst some are less objectionable or ridiculous than others, has found none of them convincing; and, as W.G. Clark commented on such efforts over a century ago, the less gifted exponents of this practice 'may fancy that they reproduce it when they do nothing of the kind'. The carefully considered advice is therefore given, albeit reluctantly, not to strive for a melodic rendering but rather to concentrate one's efforts on fluency and accuracy in other aspects of the language

Allen's advice was sound when he gave it several decades ago. It probably still is, and it has at least some basis in antiquity. In using a stress-accent with hellenistic pronunciation, one is at least producing the kind of Greek which, in addition to being widespread across the social spectrum in many areas, was certainly current among many a non-native of the 3rd century B.C.

For those who wish to use a pitch accent, however, and have some background in phonetics and phonology, Devine and Stephens' book The Prosody of Greek Speech is full of useful data employing linguistic universals, inscriptional evidence, metrical particularities and surviving fragments of ancient musical settings. That said, the book is mercilessly technical and a lot of it makes for pretty dry reading.

But here are some general pointers:

Acute accent: raised (high) pitch on short vowels, rising (low to high) pitch on long vowels. In both cases the pitch of the following syllable is significantly lowered below the onset pitch of the accented syllable.

Grave: a more modestly raised pitch on short vowels.

Circumflex: falling pitch from high to low

Greek intonation exhibits "tone-terracing"- a tendency for pitch-contours to be constrained to smaller and smaller intervals as a phrase progresses. In unmarked utterance (i.e. without especial emphasis applied to any one word or word-like group) the highest pitch of a major syntactic phrase is usually the pitch-bearing syllable of the first lexical word. The pitch-intervals of the remaining tone contours decrease progressively, resetting slightly for the beginnings of minor phrases.


ADDENDUM ON PEDAGOGY:

The practical difficulties of actually teaching this pronunciation, even with stress accent, are now apparent to me. There seems to be no way around the fact that contrasts such as /pʰ kʰ tʰ/ vs /p k t/ are difficult to produce, and even more difficult to perceive reliably, for most Euroglots. There is some pedagogical justification for the common practice of pronouncing φ, χ, θ as /f x θ/.