🗣️ Script and sounds

This chapter describes how Attic Greek of the fifth and fourth centuries BC is written and how it sounded: the twenty-four letters, the vowels and diphthongs, the breathings, the consonants, and the sound changes (vowel gradation, contraction, elision, crasis, consonant euphony) that every later chapter assumes. Accents have their own chapter (2).

How to read the examples. Each example gives a reference, the Greek as our corpus holds it, a transliteration where it helps, and a literal gloss. Classical texts are the Perseus and First1KGreek editions (CC BY-SA), cited by their catalogue number and section; they are quoted briefly and never copied in bulk. The ids tlg0032.tlg006 and the like are the catalogue numbers of the author and the work (here: Xenophon, Anabasis).

Contents

1. The alphabet
2. Vowels and diphthongs
3. Breathings, diaeresis, punctuation
4. Consonants
5. How Attic sounded
6. Vowel gradation
7. Lengthening and shortening
8. Contraction
9. Hiatus: elision, crasis, synizesis
10. Consonant euphony
11. Movable and final consonants
12. Syllables and quantity
13. Unicode and the digital text
14. How Mova handles it
15. Sources

1. The alphabet

The Athenian state adopted the Ionic alphabet in the archonship of Eucleides (403/2 BC); before that, Attic inscriptions used the local alphabet, which had no separate signs for η and ω and wrote Ε for ε, η and the spurious diphthong ει alike, and Ο for ο, ω and ου. Every literary text we read was copied into the Ionic alphabet later, so the spelling of the manuscripts and of our editions is the Ionic one (smyth-1920 §1; kuhner-1890).

The twenty-four letters, with their names, the transliteration used in this manual, and an approximate Attic value of about 400 BC:

Capital and small letters differ only in modern editions: ancient inscriptions are in capitals without word division or accents, and the minuscule belongs to Byzantine copyists. The sigma has two small forms, σ inside a word and ς at the end of a word (and in the middle of a compound only when the editor chooses).

Three letters were dropped from the alphabet but live on in the numerals and in the dialects: digamma ϝ [w] (value 6), koppa ϙ (90), sampi ϡ (900). Digamma is not written in Attic, but its trace is the hiatus it leaves in Homer, the long vowels of some compensatory lengthenings, and the Attic forms ἔτος, οἶκος beside Cretan ϝέτος, ϝοῖκος (see chapter 16). The letter yod [j] is also lost in Attic and is detectable only in the change it leaves in the preceding consonant (φυλάσσω from *φυλακ-jω).

Letters served as numerals from the third century BC: α´ 1 … θ´ 9, ι´ 10 … ϙ´ 90, ρ´ 100 … ϡ´ 900, with a lower stroke ͵α for thousands.

2. Vowels and diphthongs

Greek has seven vowel letters for five qualities with two lengths: α ι υ are short or long and the writing does not tell which (the length is learned with the word, and marked here by a macron only when it is under discussion: ᾱ, ῑ, ῡ); ε and ο are always short; η and ω are always long.

Diphthongs. The proper diphthongs are αι, ει, οι, υι, αυ, ευ, ου, ηυ, ωυ. Of these ει and ου have two origins. Those that come from an old diphthong (εἰμί, θεοῦ in the older stage) were pronounced [ei] and [ou] and had become long monophthongs by the fifth century; those that result from a contraction of ε+ε or ο+ο, or from compensatory lengthening (ἔμεινα from *ἔμενσα), are called spurious diphthongs and were written Ε and Ο in the old Attic alphabet because they were the long close vowels [eː] and [oː]. The two types merged in sound in the fifth century, and by the fourth the sounds were rising towards [iː] and [uː] (section 5). The long diphthongs ᾳ ῃ ῳ carry an iota which in classical Attic had already ceased to be sounded: it is written below the letter (iota subscript) in lower case and beside the letter (iota adscript) in capitals: τῇ Ἑλλάδι, ΤΗΙ ΕΛΛΑΔΙ. The iota is required by the grammar: it marks the dative singular (τῇ, τῷ, ἡμέρᾳ), the subjunctive (λύῃ) and many verb forms.

Quantity of diphthongs. αι and οι count as short for accent purposes when final (except in the optative and in the adverb οἴκοι); all other diphthongs are long.

3. Breathings, diaeresis, punctuation

Every word beginning with a vowel or diphthong carries a breathing: the rough breathing ῾ marks the sound [h]; the smooth ᾿ marks its absence. In a diphthong the mark goes on the second vowel (οἶκος, εὖ); with an initial capital the mark stands before the letter (Αἰγύπτιος, Οἶκος, Ἑλλάς). The initial υ and ρ are always rough: ὑπό, ῥήτωρ. A double ρ inside a word after an augment or in a compound is often written ῤῥ in older editions (ἔρρηξα, *ἐρρηξ-), now almost always ρρ. The rough breathing appears as h in transliteration. Attic had the sound [h]; the East Ionic of Asia Minor, Aeolic and Elean had lost it (psilosis), which matters for the dialects (chapter 16).

The diaeresis ¨ marks that two vowels are to be read apart (προΐστημι, Ἀϊδης) and is regular in editions of Homer. The coronis ᾽ marks crasis (τοὔνομα) and the elision apostrophe has the same shape, which is the first problem for a program (section 13). Modern editions punctuate with the comma, the full stop, the raised dot · (our colon and semicolon) and the Greek question mark ;. Ancient texts had none of these.

4. Consonants

The voiceless aspirates were stops with a puff of breath, not fricatives, until the Hellenistic period: φ θ χ were [pʰ tʰ kʰ]. This is why an aspirate and a plain stop of the same place cannot be adjacent in the same root: the first one loses its aspiration (Grassmann's law): τίθημι, not *θίθημι; κέχηνα, not *χέχηνα.

The letter γ before γ κ ξ χ is the velar nasal [ŋ]: ἄγγελος, ἄγκυρα, σφίγξ, ἐλέγχω. The same sound written ν is found in some inscriptions.

5. How Attic sounded

What we know about Attic pronunciation rests on the grammarians, on spellings in inscriptions, on the way later languages (Latin, Coptic, Armenian, Gothic) borrowed Greek words, and on metre. The classical picture (smyth-1920 §23):

• Vowel length was phonemic: ὁ / ὦ, ε / η, ο / ω, and also long and short α, ι, υ. Metre counts it; the accent depends on it (chapter 2).
• υ was a front rounded vowel [y], as in French lune, not [u]. The spelling ου stood for [oː] in the fifth century and [uː] from the fourth (the sound a Latin ū was written with), and ει for [eː] and then [iː]; the older diphthongs [ei], [ou] existed only before.
• ζ stood for a cluster; inscriptions fluctuate between σδ and δσ.
• The accent was a pitch accent, not a stress accent: the acute raises the pitch of a short or long syllable, the circumflex marks a rise and fall on one long syllable.
• Aspirated stops φ θ χ were [pʰ tʰ kʰ], so that θεός was close to [tʰeos].
• σ assimilated to the voicing of the following consonant: ἄσβεστος [azbestos].

By the first century AD, in the Koine, φ θ χ had become fricatives, η υ ι ει οι υι converged on [i], ω and ο converged, and αι became [ɛ]; the reasons why late manuscripts confuse these letters are in the manual for Koine Greek (grc-x-koine, chapter 2).

6. Vowel gradation

Greek, like every Indo-European language, has a system of vowel alternations (ablaut) inherited from the proto-language. A verb or noun root appears in three grades: the e-grade (normal), the o-grade and the zero grade (no vowel), with lengthened grades beside them. The practical consequence is that the vowel of a root changes from form to form, and the learner who does not know the pattern cannot recognise λοιπ-, λιπ-, λειπ- as one word.

The lengthened grade follows a pattern: ε→η, ο→ω, α→ᾱ/η. Hence the root of τίθημι (θε-/θη-), of ἵστημι (στα-/στη-), of δίδωμι (δο-/δω-) – chapter 9. In the third declension the same alternation gives the stem change of πατήρ / πατρός / πατέρα, ῥήτωρ / ῥήτορος (chapter 3).

7. Lengthening and shortening

• Compensatory lengthening: when a consonant (σ or ν with σ, or a liquid) drops out between a vowel and another consonant, the vowel lengthens to keep the weight of the syllable: *παντ-σι → πᾶσι; *ἔμενσα → ἔμεινα; *ἔφανσα → ἔφηνα; λύοντσι → λύουσι (smyth-1920 §37). The vowels become: α→ᾱ, ε→ει, ο→ου, ι→ῑ, υ→ῡ.
• Attic ᾱ→η. Proto-Greek long ᾱ became η in Ionic everywhere and in Attic except after ε, ι, ρ: Attic χώρᾱ, ἡμέρᾱ, σοφίᾱ, but τιμή, νίκη, μοῦσα (the last from *μοντ-jα, a different source). Ionic has χώρη, ἡμέρη, σοφίη. This is the quickest sign of Attic against Ionic.
• Quantitative metathesis: two quantities exchange in Attic: Ionic and Homeric βασιλῆος (genitive) becomes βασιλέως, and νηός becomes νεώς, λαός becomes λεώς. This is the origin of the "Attic second declension" (ὁ νεώς, ὁ λεώς; chapter 3).
• Shortening before a vowel (the correption of epic verse): a long vowel or diphthong at the end of a word may be shortened before a following vowel in Homer (chapter 16); Attic prose does not do this.

8. Contraction

Where two vowels meet inside a word because a consonant has been lost, Attic contracts them into a long vowel or a diphthong. This is the origin of the contract verbs (chapter 8), of the Attic declension of ὀστέον, νοῦς, and of many nouns of the third declension (τριήρης).

The rules (smyth-1920 §48; goodwin-1892):

The practical memory aids are: (a) like vowels give a long vowel (αα→ᾱ, εε→ει, οο→ου); (b) o beats a and e (αο→ω, εο→ου, οε→ου); (c) a beats e (αε→ᾱ, and εα→η, but εα→ᾱ after ε, ι, ρ: ἀργυρέα → ἀργυρᾶ); (d) when the second element is a diphthong, the contraction includes its iota as subscript: αει→ᾳ, οει→οι.

Not every pair contracts. Verbs with a stem in υ and a monosyllabic stem in ε (πλέω, πνέω, ῥέω, θέω, νέω, δέω "lack") contract only εε→ει and εει→ει, and leave ε+ο, ε+ω open: πλέομεν, πλέω. The datives and genitives of nouns in -ις, -υς, -ευς keep their vowels open or only partly contracted (πόλεως, βασιλέως). Ionic and Homeric leave nearly all of these open (chapter 16). The contraction of the stem vowel with the thematic vowel is the whole of the contract conjugation (chapter 8).

Example. ἐπεμελεῖτο (Xen. An. 1.1.5, tlg0032.tlg006) "he took care", from ἐπεμελέ-ετο with εε → ει; ἠσθένει (1.1.1) "he was ill" from ἠσθένεε.

9. Hiatus: elision, crasis, synizesis

The Attic of prose avoids the hiatus of a vowel ending followed by a vowel beginning, in particular within a phrase; in poetry elision is regular. There are four devices (smyth-1920 §46, §62, §70).

Elision. A short final vowel (α, ε, ι, ο; never υ, and rarely ι in datives and in the particles) drops before a word beginning with a vowel; the apostrophe marks it, and a preceding unaspirated stop turns aspirate before a rough breathing (ἀφ᾽ οὗ from ἀπὸ οὗ, ἐφ᾽ ἡμῖν from ἐπὶ ἡμῖν). In Xenophon the following are all in one place:

An. 1.1.4–5 · tlg0032.tlg006 ὁ δʼ ὡς ἀπῆλθε κινδυνεύσας … ἀντʼ ἐκείνου … τῶν παρʼ ἑαυτῷ δὲ βαρβάρων … ὅστις δʼ ἀφικνεῖτο "and he, when he had gone away after danger … in his place … of the barbarians in his own land … whoever arrived"

Here δέ → δʼ, ἀντί → ἀντʼ, παρά → παρʼ; the elided forms keep their accent only in the manner of the proclitics (chapter 2).

Crasis. A word ending in a vowel merges with the following word that begins with a vowel and writes the result as one word, with a coronis over the contracted vowel (identical in form to the elision apostrophe): τὸ ὄνομα → τοὔνομα, καὶ ἐγώ → κἀγώ, τὰ ἀληθῆ → τἀληθῆ, ὁ ἀνήρ → ἁνήρ. Crasis is common with the article, with καί, with πρό and with ἐγώ.

Pl. Ap. 22a · tlg0059.tlg002 δεῖ γὰρ πρὸς ὑμᾶς τἀληθῆ λέγειν "for I must tell you the truth" (τἀληθῆ = τὰ ἀληθῆ, "the true things")

Synizesis. Two vowels that are written apart are pronounced as one syllable (in Attic verse θεοί, θεῶν, ἐπεὶ οὐ can be one syllable). It is a feature of verse and has no spelling.

Aphaeresis (inverse elision) is the loss of an initial short ε or α after a long vowel or diphthong: μὴ ἐγώ → μὴ ʼγώ, ποῦ ἐστιν → ποῦ ʼστιν. Common in Attic comedy.

Hiatus avoided in other ways. The movable ν (section 11), the prefixed form of some words (οὐκ, οὐχ), and the absence of hiatus in the style of Isocrates, who rarely lets a vowel end meet a vowel start.

10. Consonant euphony

Consonants assimilate to their neighbours; the rules are a consequence of the voicing and aspiration pattern of the stops (smyth-1920 §§77–132).

Stops before stops. A stop assimilates in voicing and aspiration to a following stop: λέγω gives λεκτός, but ἐλέχθην; γράφω gives γραπτός, but ἐγράφθην; τρίβω gives τριπτός, but ἐτρίφθην; πέμπω gives πέπεμπται and ἐπέμφθην. Before τ a labial is π and a velar κ; before θ they are φ and χ. A dental before another dental yields σ: ἐπείσθην from πείθω, ἐψεύσθην from ψεύδω.

Stops before μ. A labial becomes μ (τέτριμμαι from τρίβω, γέγραμμαι from γράφω), a velar becomes γ (πέπλεγμαι from πλέκω), a dental becomes σ (πέπεισμαι from πείθω, ἔψευσμαι from ψεύδω).

Nasals. ν before a labial becomes μ (ἐμβάλλω, συμπίπτω), before a velar it becomes the γ nasal (ἐγκαλῶ, συγγράφω), before a liquid it assimilates (συλλέγω, ἐλλείπω, συρρέω), before σ it drops with compensatory lengthening of the preceding vowel.

Dentals before σ. τ δ θ are lost before σ (the sibilant stays): πείσω, ἐλπίς – ἐλπίσι, *ἐλπιδ-σι; a labial or velar gives ψ or ξ: λείψω, φλέξω.

σ. Intervocalic σ is lost in Greek, the origin of the vowel groups that produce contraction in the inflection of γένος (γένεος → γένους), of the endings of the middle voice (λύεσαι → λύῃ), and of the future of liquid verbs (ἀγγελῶ from *ἀγγελ-έ-σω). Between consonants and after a lost τ the σ remains or the group simplifies by the pattern of the inflection: λελυκότ-σι → λελυκόσι.

Aspiration. If a word with a rough breathing is compounded with a prefix ending in a stop, the stop takes the aspirate: ἐφ᾽ ἡμέρᾳ, ἀφ᾽ οὗ, καθ᾽ ἡμέραν. The prefix ἐκ- becomes ἐξ- before vowels and ἐκ- before consonants.

Voiceless ρ. Initial ρ is voiceless and takes the rough breathing; ρρ and ῤῥ appear when the sound is reached in the middle of a word after a short vowel: ἔρρεον from ῥέω.

Lost consonants. Digamma and yod and the final dentals give results that look odd in the paradigms: ναῦς / νεώς (stem ναϝ-), βοῦς / βοός (βοϝ-), οἶκος for *ϝοῖκος; φέρων from *φέροντ-ς, with the loss of τ and compensatory lengthening, μέλας from *μελαν-ς, σῶμα for *σωματ (chapter 3).

11. Movable and final consonants

Movable ν (ν ἐφελκυστικόν) is added at the end of a number of words that end in a vowel (third person singular verbs in -ε, the dative plural in -σι, ἐστί, εἴκοσι) when the next word begins with a vowel, and also at the end of a sentence or line: ἔδωκεν αὐτῷ, but ἔδωκε τοῦτο. In prose manuscripts it often appears before consonants, and Xenophon and Plato follow the rule very irregularly (smyth-1920 §134).

Movable ς. οὕτως / οὕτω, ἐκ / ἐξ, ἄχρι(ς), μέχρι(ς), ἀμφί(ς). Before a vowel οὕτως and ἐξ stand, before a consonant οὕτω and ἐκ. The form οὐκ appears before a smooth vowel, οὐχ before a rough, οὐ elsewhere.

Final consonants. Greek words end only in a vowel or in ν, ρ, ς (with ξ ψ) – and in the two words ἐκ and οὐκ with a velar stop, which are proclitics and lean on the next word. All other consonants have dropped off: the nominative in -μα from -ματ-, the accusative in -ν.

12. Syllables and quantity

A word has as many syllables as vowels and diphthongs. A single consonant between vowels goes with the following vowel (πο-ρεύ-ω); a group that can begin a word begins the syllable (ἔ-χθρα, ἐ-γρή-γορα); a stop + liquid combination, which cannot always begin a word, may either begin the syllable or divide it (τέ-κνον vs. ἄ-γρα, or in poetry both) (smyth-1920 §§138–148).

A syllable is long by nature if it has a long vowel or diphthong, and long by position if its short vowel is followed by two consonants or a double consonant (ψ, ξ, ζ). Before a stop + liquid, the syllable is common in Attic poetry. This is the base of Greek metre: the dactylic hexameter of Homer, the iambic trimeter of Attic drama, with the quantities as above.

13. Unicode and the digital text

Greek text in a computer has three layers of encoding, all of which appear in our corpora, and a program that reads the corpus must meet them.

• Greek and Coptic block (U+0370–03FF): the monotonic letters with the single accent tonos, for example ά U+03AC; and the unaccented base letters.
• Greek Extended block (U+1F00–1FFF): the polytonic letters, precomposed for every combination of breathing, accent and iota subscript, for example ἄ U+1F04, ᾠδή. Some of them are singletons canonically equivalent to the Greek-block letters: the oxia letters (U+1F71 ά) decompose to the tonos ones (U+03AC), so that normalisation turns them into the same character, and the ambiguity disappears in NFC.
• Combining marks: U+0301 acute, U+0300 grave, U+0342 perispomeni (circumflex), U+0313 psili, U+0314 dasia, U+0345 ypogegrammeni, U+0308 diaeresis. NFC composes them, NFD decomposes them.

The apostrophe has several candidates, because the same symbol is the elision mark, the coronis, and a punctuation: U+02BC modifier letter apostrophe (Perseus and the eBible texts: δʼ), U+2019 right single quotation mark, U+1FBF Greek psili, which the PTNK files use for the elision mark (551 times in the three files, for example ἀπ᾿, ἀφ᾿, δ᾿), U+1FBD Greek koronis (8 times in the same files) and U+0027. A tokenizer that splits on non-letter characters cuts δʼ in two if it does not treat the modifier apostrophe as a letter.

The final sigma ς is a different code point (U+03C2) from σ (U+03C3); case folding of Σ gives σ, so that a lower-case/upper-case round trip loses the distinction unless the code knows the word boundary. On 2026-10-22 we checked the three PTNK files, Brenton's Genesis (eBible grcbrent), SBLGNT Matthew (eBible grcsbl) and Perseus' Anabasis: all are in NFC, with precomposed letters, no combining marks, and tonos (not oxia) for the acute. Elision is written U+02BC in the Perseus and eBible texts (δʼ), but U+1FBF in PTNK (δ᾿), so a lookup that crosses the two must fold the marks.

14. How Mova handles it

• Normal form. Greek enters Mova in NFC with the precomposed letters; the three sources above need no conversion. A text from outside (a PDF extract, an OCR) must be normalised before use, because the oxia/tonos duplicates and decomposed marks are common there.
• Tokens. The elision mark (U+02BC, U+2019, U+1FBF, U+1FBD, U+0027 after a Greek letter) belongs to the word it ends, and is folded to one character when two sources are compared. Without that rule δʼ is not a word and the tagger sees two tokens.
• Accents and case. The polytonic accent is part of the lexical identity of the word (chapter 2): ἡ article / ἥ relative, ἤ "or" / ἦ "indeed", οὐ "not" / οὗ "where". A model that strips accents merges them. The PTNK treebank and the other UD treebanks keep the accents in FORM, and their LEMMA carries the accent of the citation form.
• Contraction and euphony are the reason for the gap shown in the parser baseline. On the UD Ancient Greek PTNK treebank (the Septuagint in Codex Alexandrinus) Mova's parser, trained on the train part and measured on dev, reaches a labelled attachment score (LAS) of 74.41 with the tags that en xpos-fill makes for itself and 78.29 with the treebank's gold tags (UD 2.18, 2026-10-22). So the tags carry about four points. We have not yet split that gap by feature, and this chapter only gives the reason to expect it to lie in the forms: the tagger sees the letters, not the paradigm, and Greek letters change by contraction, elision, augment and ablaut. A Greek morphology module would apply the rules of this chapter and of chapters 3, 7, 8 and 9. None exists yet. The first pieces that code would need are the normalisation and the elision rule above, which are fully decided and can be tested on the corpora.

15. Sources

• Smyth, H. W. 1920. A Greek Grammar for Colleges (smyth-1920): §1 alphabet, §4 vowels, §9 breathings, §15 consonants, §23 pronunciation, §27 vowel change, §37 compensatory lengthening, §48 contraction, §62 crasis, §70 elision, §77 consonants, §134 movable consonants, §138 quantity. The text of the Perseus edition is CC BY-SA and is paraphrased here.
• Goodwin, W. W. 1892. A Greek Grammar (goodwin-1892), public domain.
• Kühner, R., Blass, F., Gerth, B. 1890–1904. Ausführliche Grammatik der griechischen Sprache (kuhner-1890), public domain.
• Xenophon, Anabasis and Plato, Apology: Perseus canonical-greekLit (CC BY-SA 4.0).

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/grc/01-script-sounds/