||This article includes a list of references, but its sources remain unclear because it has insufficient inline citations. (January 2015) (Learn how and when to remove this template message)|
The Proto-Greek language (also known as Proto-Hellenic) is the assumed last common ancestor of all known varieties of Greek, including Mycenaean Greek, the subsequent ancient Greek dialects (i.e., Attic, Ionic, Aeolic, Doric, Ancient Macedonian and Arcadocypriot) and, ultimately, Koine, Byzantine and Modern Greek. The unity of Proto-Greek would have ended as Hellenic migrants, who spoke the predecessor of the Mycenaean language, entered the Greek peninsula sometime in the Neolithic or the Bronze Age.
The evolution of Proto-Greek could be considered within the context of an early Paleo-Balkan sprachbund that makes it difficult to delineate exact boundaries between individual languages. The characteristically Greek representation of word-initial laryngeals by prothetic vowels is shared, for one, by the Armenian language, which also seems to share some other phonological and morphological peculiarities of Greek; this has led some linguists to propose a hypothetically closer relationship between Greek and Armenian, although evidence remains scant.
Proto-Greek is mostly placed in the Early Helladic period (3rd millennium BC), i.e. towards the end of Neolithic Europe. Russell Gray and Quentin Atkinson in a 2003 paper using computational methods on Swadesh lists have arrived at a somewhat earlier estimate, around 5000 BC for Greco-Armenian split and the emergence of Greek as a separate linguistic lineage around 4000 BC.
The primary sound changes separating Proto-Greek from the Proto-Indo-European language include:
- Shortening of long vowels before a sonorant in the same syllable (Osthoff's law): *dyēws "skyling, sky god" > Zeús /sdeús/
- Debuccalization of /s/ to /h/ in inter-vocalic and pre-vocalic positions (between two vowels, or if word-initial and followed by a vowel).
- Devoicing of voiced aspirates.
- Strengthening of word-initial y- (not Hy-) to dy- (later z-).
- Palatalization of consonants followed by -y-, producing various affricate consonants (still represented as a separate sound in Mycenaean) and palatal consonants; they later simplified, mostly losing their palatal character.
- Dissimilation of aspirates (Grassmann's law), possibly post-Mycenaean.
- Vocalization of laryngeals between consonants and initially before consonants to /e/, /a/, /o/ from *h₁, *h₂, *h₃ respectively (unlike all other Indo-European languages).
- Other unique changes involving laryngeals; see below.
- Loss of final stop consonants.
- Final /m/ > /n/.
- Raising of /o/ to /u/ between a resonant and a labial (Cowgill's law).
- Merging of sequences of velar + *w into the labiovelars, with compensatory lengthening of the consonant in some cases. For example, PIE *h₁éḱwos > PG *íkkʷos > Mycenaean i-qo /íkkʷos/, Attic híppos, Aeolic íkkos.
Loss of prevocalic *s was not completed entirely, evidenced by sȳ́s (also hȳ́s, pig, from PIE *suh₁-), dasýs (dense) and dásos (dense growth, forest); syn (with) is another example, contaminated with PIE *ḱom (Latin cum, Proto-Greek *kon) to Homeric / Old Attic ksyn. sélas (light in the sky, as in the aurora) and selḗnē/selā́nā (Moon) may be more examples of the same if it derived from PIE *swel- (to burn) (possibly related to hḗlios "Sun", Ionic hēélios < *sāwélios).
Dissimilation of aspirates (so-called Grassmann's law) caused an initial aspirated sound to lose its aspiration when a following aspirated consonant occurred in the same word. It was a relatively late change in Proto-Greek history and must have occurred independently of the similar dissimilation of aspirates (also known as Grassmann's law) in Indo-Iranian, although it may represent a common areal feature:
- It postdates the Greek-specific de-voicing of voiced aspirates.
- It also postdates the change of /s/ > /h/, as it affects /h/ as well: ékhō "I have" < *hekh- < PIE *seǵʰ-oh₂, but future heksō "I will have" < *heks- < Post-PIE *seǵʰ-s-oh₂.
- It postdates even the loss of aspiration before /j/ that accompanied second-stage palatalization (see below), which postdates both of the previous changes (as well as first-stage palatalization).
- On the other hand, it predates the development of the first aorist passive marker -thē- since the aspirate in that marker has no effect on preceding aspirates.
Greek is unique in reflecting the three different laryngeals with distinct vowels. Most Indo-European languages can be traced back to a dialectal variety of late Proto-Indo-European (PIE) in which all three laryngeals had merged (after colouring adjacent short /e/ vowels), but Greek clearly cannot. For that reason, Greek is extremely important in reconstructing PIE forms.
Greek shows distinct reflexes of the laryngeals in various positions:
- Most famously, between consonants, where original vocalic *h₁, *h₂, *h₃ are reflected as /e/, /a/, /o/ respectively (the so-called triple reflex). All other Indo-European languages reflect the same vowel from all three laryngeals (usually /a/, but /i/ or other vowels in Indo-Iranian):
|dʰh̥₁s- "sacred, religious"||θέσφατος / thésphatos "decreed by God"||धिष्ण्य / dhíṣṇya- "devout"||fānum "temple" < *fasnom < *dʰh̥₁s-no-|
|sth̥₂-to- "standing, being made to stand"||στατός / statós||स्थित / sthíta-||status|
|dh̥₃-ti- "gift"||δόσις / dósis||दिति / díti-||datiō|
- An initial laryngeal before a consonant (a *HC- sequence) leads to the same triple reflex, but most IE languages lost such laryngeals and a few reflect them initially before consonants. Greek vocalized them (leading to what are misleadingly termed prothetic vowels): Greek érebos "darkness" < PIE *h₁regʷos vs. Gothic riqiz- "darkness"; Greek áent- "wind" < *awent- < PIE *h₂wéh₁n̥t- vs. English wind, Latin ventum "wind".
- The sequence *CRHC (C = consonant, R = resonant, H = laryngeal) becomes CRēC, CRāC, CRōC from H = *h₁, *h₂, *h₃ respectively. (Other Indo-European languages again have the same reflex for all three laryngeals: *CuRC in Proto-Germanic, *CíRC/CúRC with acute accent in Proto-Balto-Slavic, *CīRC/CūRC in Proto-Indo-Iranian, *CRāC in Proto-Italic and Proto-Celtic.) Sometimes, CeReC, CaRaC, CoRoC are found instead: Greek thánatos "death" vs. Doric Greek thnātós "mortal", both apparently reflecting *dʰn̥h₂-tos. It is sometimes suggested that the position of the accent was a factor in determining the outcome.
- The sequence *CiHC tends to become *CyēC, *CyāC, *CyōC from H = *h₁, *h₂, *h₃ respectively, with later palatalization (see below). Sometimes, the outcome CīC is found, as in most other Indo-European languages, or the outcome CiaC in the case of *Cih₂C.
All of the cases may stem from an early insertion of /e/ next to a laryngeal not adjacent to a vowel in the Indo-European dialect ancestral to Greek (subsequently coloured to /e/, /a/, /o/ by the particular laryngeal in question) prior to the general merger of laryngeals:
- *CHC > *CHeC > Italic textCeC/CaC/CoC.
- *HC- > *HeC- > eC-/aC-/oC-.
- *CRHC > *CReHC > CRēC/CRāC/CRōC; or, *CRHC > *CeRHeC > *CeReC/CeRaC/CeRoC > CeReC/CaRaC/CoRoC by assimilation.
- *CiHC > *CyeHC > CyēC/CyāC/CyōC; or, *Cih₂C > *Cih₂eC > *CiHaC > *CiyaC > CiaC; or, *CiHC remains without vowel insertion > CīC.
A laryngeal adjacent to a vowel develops along the same lines as other Indo-European languages:
- The sequence *CRHV (C = consonant, R = resonant, H = laryngeal, V = vowel) passes through *CR̥HV, becoming CaRV.
- The sequence *CeHC becomes CēC/CāC/CōC.
- The sequence *CoHC becomes CōC.
- In the sequence *CHV (including CHR̥C, with a vocalized resonant), the laryngeal colors a following short /e/, as expected, but it otherwise disappears entirely (as in most other Indo-European languages but not Indo-Iranian whose laryngeal aspirates a previous stop and prevents the operation of Brugmann's law).
- In a *VHV sequence (a laryngeal between vowels, including a vocalic resonant R̥), the laryngeal again colours any adjacent short /e/ but otherwise vanishes early on. That change appears to be uniform across the Indo-European languages and was probably the first environment in which laryngeals were lost. If the first V was *i, *u or a vocalic resonant, a consonantal copy was apparently inserted in place of the laryngeal: *CiHV > *CiyV, *CuHV > *CuwV, *CR̥HV possibly > *CR̥RV, with R̥ always remaining as vocalic until the dissolution of vocalic resonants in the various daughter languages. Otherwise, a hiatus resulted, which was resolved in various ways in the daughter languages, typically by converting i, u and vocalic resonants, when it directly followed a vowel, back into a consonant and merging adjacent non-high vowels into a single long vowel.
Proto-Greek underwent palatalization of consonants before *y. This occurred in two separate stages. The first stage affected only dental consonants, while the second stage affected all consonants.
The first palatalization turned dentals + *y into alveolar affricates:
Alongside these changes, the inherited clusters *ts, *ds and *tʰs all merged into *ts.
In the second palatalization, all consonants were affected. It took place following the resolution of syllabic laryngeals and sonorants. The following table, based on American linguist Andrew Sihler, shows the developments.
|*sy > *hy||*yy|
|*wy||*ɥɥ > *yy|
In post-Proto-Greek times, the resulting palatal consonants and clusters were resolved in varying ways. Most notably, *ň and *ř were resolved into plain sonorants plus a palatal on-glide, which eventually turned the preceding vowel into a diphthong.
|Proto-Greek||Attic||Homeric||West Ionic||Other Ionic||Boeotian||Arcado-
|*ňň||in (but *uňň > ūn)|
|*řř||ir (but *uřř > ūr)|
In the time between the first and second palatalizations, new clusters *tsy and *dzy were formed by restoring a lost *y after the newly formed *ts and *dz. This occurred only in morphologically transparent formations, by analogy with similar formations where *y was preceded by other consonants. In formations that were morphologically opaque and not understood as such by speakers of the time, this restoration did not take place and *ts and *dz remained. Hence, depending on the type of formation, the Pre-Greek sequences *ty, *tʰy and *dy have different outcomes in the later languages. In particular, medial *ty becomes Attic s in opaque formations, but tt in transparent formations.
The outcome of PG medial *ts in Homeric Greek is s after a long vowel, and vacillation between s and ss after a short vowel: tátēsi dat. pl. "rug" < tátēt-, possí(n)/posí(n) dat. pl. "foot" < pod-. This was useful for the composer of the Iliad and Odyssey, since possí with double ss scans as long-short, while posí with single s scans as short-short. Thus the writer could use each form in different positions in a line.
Examples of initial *ts:
- PIE *tyegʷ- "avoid" > PG *tsegʷ- > Greek sébomai "worship, be respectful" (Ved. tyaj- "flee")
- PIE *dʰyeh₂- "notice" > PG *tsā- > Dor. sā́ma, Att. sêma "sign" (Ved. dhyā́- "thought, contemplation")
Examples of medial *ts (morphologically opaque forms, first palatalization only):
- PreG *tótyos "as much" > PG *tótsos > Att. tósos, Hom. tósos/tóssos (cf. Ved. táti, Lat. tot "so much/many")
- PIE *médʰyos "middle" > PG *métsos > Att. mésos, Hom. mésos/méssos, Boeot. méttos, other dial. mésos (cf. Ved. mádhya-, Lat. medius)
Examples of medial *ťť (morphologically transparent forms, first and second palatalization):
- PIE *h₁erh₁-t-yoh₂ "I row" > PG *eréťťō > Attic eréttō, usual non-Attic eréssō (cf. erétēs "oarsman")
- PIE *krét-yōs > PreG *krétyōn "better" > PG *kréťťōn > Attic kreíttōn, usual non-Attic kréssōn (cf. kratús "strong" < PIE *kr̥tús)
Other Post-Proto-Greek changes
Sound changes between Proto-Greek and all early dialects, including Mycenaean Greek, include:
- Remaining syllabic resonants *m̥ *n̥ *l̥ and *r̥ are resolved to vowels or combinations of a vowel and consonantal resonant. It appears that the process still occurred within Proto-Greek, and resulted in an epenthetic vowel of undetermined quality (denoted here as *ə). This vowel then usually developed into a but also o in some cases. Thus:
- *m̥, *n̥ > *ə, but > *əm, *ən before a sonorant. *ə appears as o in Mycenaean after a labial: pe-mo (spérmo) "seed" vs. usual spérma < *spérmn̥. Similarly, o often appears in Arcadian after a velar, e.g. déko "ten", hekotón "one hundred" vs. usual déka, hekatón < *déḱm̥, *sem-ḱm̥tóm.
- *l̥, *r̥ > *lə, *rə, but *əl, *ər before sonorants and analogously. *ə appears as o in Mycenaean Greek, Aeolic Greek and Cypro-Arcadian. Example: PIE *str̥-tos > usual stratós, Aeolic strótos "army"; post-PIE *ḱr̥di-eh₂ "heart" > Attic kardíā, Homeric kradíē, Pamphylian korzdia.
- Loss of s in consonant clusters, with compensatory lengthening of the preceding vowel (Attic, Ionic, Doric) or of the consonant (Aeolic): *ésmi "I am" > ḗmi, eîmi or émmi.
- Creation of secondary s from clusters, *nty > ns (it was, in turn, followed by a change similar to the one described above, loss of the n with compensatory lengthening: *apónt-ya > apónsa > apoûsa, "absent", feminine).
- Conversion of labiovelars to velars next to /u/, the "boukólos rule".
- In southern dialects (including Mycenaean, but not Doric), -ti- > -si- (assibilation).
The following changes are apparently post-Mycenaean:[why?]
- Loss of /h/ (from original /s/), except initially, e.g. Doric níkaas "having conquered" < *níkahas < *níkasas.
- Loss of /j/, e.g. treîs "three" < *tréyes.
- Loss of /w/ in many dialects (later than loss of /h/ and /j/). Example: étos "year" from *wétos.
- Loss of labiovelars, which were converted (mostly) into labials, sometimes into dentals (or velars next to /u/, as a result of an earlier sound change). See below for details. It had not yet happened in Mycenaean, as is shown by the fact that a separate letter ⟨q⟩ is used for such sounds.
- Contraction of adjacent vowels resulting from loss of /h/ and /j/ (and, to a lesser extent, from loss of /w/); more in Attic Greek than elsewhere.
- Rise of a distinctive circumflex accent, resulting from contraction and certain other changes.
- Limitation of the accent to the last three syllables, with various further restrictions.
- Loss of /n/ before /s/ (incompletely in Cretan Greek), with compensatory lengthening of the preceding vowel.
- Raising of ā to ē /ɛː/ in Attic and Ionic dialects (but not Doric). In Ionic, the change was general, but in Attic it did not occur after /i/, /e/ or /r/. (Note Attic kórē "girl" < *kórwā; loss of /w/ after /r/ had not occurred at that point in Attic.)
Note that /w/ and /j/, when following a vowel and not preceding a vowel, combined early on with the vowel to form a diphthong and so were not lost.
Loss of /h/ and /w/ after a consonant was often accompanied by compensatory lengthening of a preceding vowel.
The development of labiovelars varies from dialect to dialect:
- Due to the PIE boukólos rule, labiovelars next to /u/ had already been converted to plain velars: boukólos "herdsman" < *gʷou-kʷólos (cf. boûs "cow" < *gʷou-) vs. aipólos "goatherd" < *ai(g)-kʷólos (cf. aíks, gen. aigós "goat"); elakhús "small" < *h₁ln̥gʷʰ-ús vs. elaphrós "light" < *h₁ln̥gʷʰ-rós.
- In Attic and some other dialects (but not, for example, Lesbian), labiovelars before some front vowels became dentals. In Attic, kʷ and kʷʰ became t and th, respectively, before /e/ and /i/, while gʷ became d before /e/ (but not /i/). Cf. theínō "I strike, kill" < *gʷʰen-yō vs. phónos "slaughter" < *gʷʰón-os; delphús "womb" < *gʷelbʰ- (Sanskrit garbha-) vs. bíos "life" < *gʷih₃wos (Gothic qius "alive"), tís "who?" < *kʷis (Latin quis).
- All remaining labiovelars became labials, original kʷ kʷʰ gʷ becoming p ph b respectively. That happened to all labiovelars in some dialects like Lesbian; in other dialects, like Attic, it occurred to all labiovelars not converted into dentals. Many occurrences of dentals were later converted into labials by analogy with other forms: bélos "missile", bélemnon "spear, dart" (dialectal délemnon) by analogy with bállō "I throw (a missile, etc.)", bolḗ "a blow with a missile".
- Original PIE labiovelars had still remained as such even before consonants and so became labials also there. In many other centum languages such as Latin and most Germanic languages, the labiovelars lost their labialisation before consonants. (Greek pémptos "fifth" < *pénkʷtos; compare Old Latin quinctus.) This makes Greek of particular importance in reconstructing original labiovelars.
The results of vowel contraction were complex from dialect to dialect. Such contractions occur in the inflection of a number of different noun and verb classes and are among the most difficult aspects of Ancient Greek grammar. They were particularly important in the large class of contracted verbs, denominative verbs formed from nouns and adjectives ending in a vowel. (In fact, the reflex of contracted verbs in Modern Greek, the set of verbs derived from Ancient Greek contracted verbs, represents one of the two main classes of verbs in that language.)
As Mycenaean Greek shows, the PIE dative (suffix -i), instrumental (suffix -phi) and locative (suffix -si) cases are still distinct and are not yet syncretized into other cases.
Nominative plural -oi, -ai replaces late PIE -ōs, -ās.
The superlative in -tatos becomes productive.
The pronouns hoûtos, ekeînos and autós are created. The use of ho, hā, ton as articles is post-Mycenaean.
Proto-Greek inherited the augment, a prefix e-, to verbal forms expressing past tense. That feature is shared only with Indo-Iranian and Phrygian (and to some extent, Armenian), lending some support to a "Graeco-Aryan" or "Inner PIE" proto-dialect. However, the augment down to the time of Homer remained optional and was probably little more than a free sentence particle, meaning "previously" in the proto-language, which may easily have been lost by most other branches.
The first person middle verbal desinences -mai, -mān replace -ai, -a. The third singular phérei is an innovation by analogy, replacing the expected Doric *phéreti, Ionic *phéresi (from PIE *bʰéreti).
The future tense is created, including a future passive as well as an aorist passive.
The suffix -ka- is attached to some perfects and aorists.
Infinitives in -ehen, -enai and -men are created.
- "one": *héms ~ *héns (masculine), *hmía (feminine) (> Myc. e-me /heméi/ (dative); Att./Ion. εἷς (ἑνός), μία, heîs (henos), mía)
- "two": *dúwō (> Myc. du-wo /dúwoː/; Hom. δύω, dúō; Att.-Ion. δύο, dúo)
- "three": *tréyes (> Myc. ti-ri /trins/; Att./Ion. τρεῖς, treîs; Lesb. τρής, trḗs; Cret. τρέες, trées)
- "four": nominative *kʷétwores, genitive *kʷeturṓn (> Myc. qe-to-ro-we /kʷétroːwes/ "four-eared"; Att. τέτταρες, téttares; Ion. τέσσερες, tésseres; Boeot. πέτταρες, péttares; Thess. πίτταρες, píttares; Lesb. πίσυρες, písures; Dor. τέτορες, tétores)
- "five": *pénkʷe (> Att.-Ion. πέντε, pénte; Lesb., Thess. πέμπε, pémpe)
- "six": *hwéks (> Att. ἕξ, héks; Dor. ϝέξ, wéks)
- "seven": *heptə́ (> Att. ἑπτά, heptá)
- "eight": *oktṓ (> Att. ὀκτώ, oktṓ)
- "nine": *ennéwə (> Att. ἐννέα, ennéa; Dor. ἐννῆ, ennê)
- "ten": *dékə (> Att. δέκα, déka)
- Ancient Macedonian language
- Paleo-Balkan languages
- Pre-Greek substrate
- Proto-Indo-European language
- A comprehensive overview in J.T. Hooker's Mycenaean Greece (Hooker 1976, Chapter 2: "Before the Mycenaean Age", pp. 11–33 and passim); for a different hypothesis excluding massive migrations and favoring an autochthonous scenario, see Colin Renfrew's "Problems in the General Correlation of Archaeological and Linguistic Strata in Prehistoric Greece: The Model of Autochthonous Origin" (Renfrew 1973, pp. 263–276, especially p. 267) in Bronze Age Migrations by R.A. Crossland and A. Birchall, eds. (1973).
- Georgiev 1981, p. 156: "The Proto-Greek region included Epirus, approximately up to Αυλών in the north including Paravaia, Tymphaia, Athamania, Dolopia, Amphilochia, and Acarnania), west and north Thessaly (Hestiaiotis, Perrhaibia, Tripolis, and Pieria), i.e. more or less the territory of contemporary northwestern Greece)."
- Renfrew 2003, p. 35: "Greek The fragmentation of the Balkan Proto-Indo-European Sprachbund of phase II around 3000 BC led gradually in the succeeding centuries to the much clearer definition of the languages of the constituent sub-regions."
- Clackson 1995.
- Vladimir I. Georgiev, for example, placed Proto-Greek in northwestern Greece during the Late Neolithic period. (Georgiev 1981, p. 192: "Late Neolithic Period: in northwestern Greece the Proto-Greek language had already been formed: this is the original home of the Greeks.")
- Gray & Atkinson 2003, pp. 437–438; Atkinson & Gray 2006, p. 102: "Hittite appears to have diverged from the main Proto-Indo-European stock around 8700 years ago, perhaps reflecting the initial migration out of Anatolia. Indeed, this date exactly matches estimates for the age of Europe’s first agricultural settlements in southern Greece. Following the initial split, the language tree shows the formation of separate Tocharian, Greek, and then Armenian lineages, all before 6000 BP, with all of the remaining language families formed by 4000 BP. We note that the received linguistic orthodoxy (Indo-European is only 6000 years old) does approximately fit the divergence dates we obtained for most of the branches of the tree. Only the basal branches leading to Hittite, Tocharian, Greek and Armenian are well beyond this age."
- Sihler 1995.
- Lengthened -ei /eː/ due to Attic analogical lengthening in comparatives.
- Atkinson, Quentin D.; Gray, Russel D. (2006). "Chapter 8: How Old is the Indo-European Language Family? Illumination or More Moths to the Flame?". In Forster, Peter; Renfrew, Colin. Phylogenetic Methods and the Prehistory of Languages. Cambridge: McDonald Institute for Archaeological Research. pp. 91–109. ISBN 978-1-902937-33-5.
- Beekes, Robert Stephen Paul (1995). Comparative Indo-European Linguistics. Amsterdam: John Benjamins. ISBN 90-272-2150-2.
- Buck, Carl Darling (1933). Comparative Grammar of Greek and Latin. Chicago: University of Chicago Press.
- Clackson, James (1995). The Linguistic Relationship Between Armenian and Greek. Oxford: Wiley-Blackwell. ISBN 9780631191971.
- Fortson, Benjamin W., IV (2004). Indo-European Language and Culture. Malden, MA: Blackwell Publishing. ISBN 1-4051-0316-7.
- Georgiev, Vladimir Ivanov (1981). Introduction to the History of the Indo-European Languages. Sofia: Bulgarian Academy of Sciences.
- Gray, Russel D.; Atkinson, Quentin D. (2003). "Language-tree Divergence Times Support the Anatolian Theory of Indo-European Origin". Nature. 426 (6965): 435–439. doi:10.1038/nature02029. PMID 14647380.
- Hooker, J.T. (1976). Mycenaean Greece. London: Routledge & Kegan Paul.
- Renfrew, Colin (1973). "Problems in the General Correlation of Archaeological and Linguistic Strata in Prehistoric Greece: The Model of Autochthonous Origin". In Crossland, R. A.; Birchall, Ann. Bronze Age Migrations in the Aegean; Archaeological and Linguistic Problems in Greek Prehistory: Proceedings of the first International Colloquium on Aegean Prehistory, Sheffield. London: Gerald Duckworth and Company Limited. pp. 263–276. ISBN 0-7156-0580-1.
- Renfrew, Colin (2003). "Time Depth, Convergence Theory, and Innovation in Proto-Indo-European: 'Old Europe' as a PIE Linguistic Area". In Bammesberger, Alfred; Vennemann, Theo. Languages in Prehistoric Europe. Heidelberg: Universitätsverlag Winter GmBH. pp. 17–48. ISBN 978-3-82-531449-1.
- Schwyzer, Eduard (1939). Griechische Grammatik: auf der Grundlage von Karl Grugmanns Griechischer Grammatik (in German). Munich: C.H. Beck.
- Sihler, Andrew L. (1995). New Comparative Grammar of Greek and Latin. New York: Oxford University Press. ISBN 0-19-508345-8.