DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

The Greek Phonological Phrase 
A. M. Devine and Laurence D. Stephens 

RECONSTRUCTING THE PROSODIC' domains' 1 of ancient Greek 
continues to play an important role in classical philology. 
The syntactic components of the appositive group,2 as 

well as its accentual and rhythmic properties, have been 
analyzed in fine detail, primarily because of the value of metrical 
bridges in textual criticism. The major phrase-that is, the 
phonological counterpart of the syntactic clause or simple 
sentence-has also been the object of study, especially for its 
importance in defining the domain of the avoidance of hiatus 
and in the stylistic evolution of enjambement. But little or no 
progress has been made in reconstructing the accent and 
rhythm of the minor phonological phrase (hereafter simply the 
'phonological phrase'), nor in analyzing its syntactic com­
position. Almost nothing has been discovered about how the 
words of even the simplest sentences were joined together into 

1 Intuitively, a prosodic domain is the stretch or chunk of language, e.g. 
syllable, word, word plus enclitic, phrase, on up to paragraph, within which 
or to which a prosodic process such as lengthening, rhythmic grouping, 
lowering of pitch, etc., applies. For example in Greek, crasis applies in a 
domain different from and smaller than the domain in which elision applies. 

2 The term appositive group refers to a prosodic structure consisting of a 
host word, usually lexical, and one or more non-lexicals, and not phonologi­
cally identical either to the word or to the minor phrase. The class of non­
lexical words comprises not only the clitics (the cover term for enclitics and 
[the traditionally understood] proclitics), but also the Article, Conjunctions, 
particles, negatives, pronominals, Prepositions, and other 'little' words; the 
class shades off into semilexicals such as the lower cardinal numerals, the 
Adverb tU (e.g. Aesch. Chae. 693), the modal Ott (e.g. Eur. Or. 1035), seman­
tically bleached imperatives such as <PEpt (e.g. Ar. Eq. 145), ayE (e.g. Ar. Eq. 
155), etc. Many non-Iexicals tend to lose the status of autonomous words and 
to be treated, to varying extents, as part of a contiguous word. Those non­
lexicals losing their phonological autonomy are termed 'appositives'. Pre­
positives are those appositives that cohere with what follows, postpositives 
those cohering with what precedes. See A. M. Devine and L. D. Stephens, 
"The Greek Appositives: Toward a Linguistically Adequate Definition of Cae­
sura and Bridge," CP 73 (1978) 314-28, and "Semantics Syntax, and Phono­
logical Organization in Greek: Aspects of the Theory of Metrical Bridges," CP 
78 (1983) 1-25. 

421 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

422 THE GREEK PHONOLOGICAL PHRASE 

prosodic subunits in ordinary Greek speech. How can we 
demonstrate that there actually were prosodic domains 
intermediate between the word or appositive group and the 
utterance? And if we can, how do we know which sequences 
of words can make up these phonological phrases and which 
cannot? Under what circumstances, if any, can a subject be 
phrased with its Verb? Can a direct object modified by an 
Adjective be phrased with its Verb? What happens if there is 
also an indirect object? Is a Prepositional Phrase always a 
phonological phrase by itself or can it be joined with an adjacent 
constituent? In addition to discovering its syntactic constituen­
cy, we should also like to know what the Greek phonological 
phrase sounded like; specifically whether, in addition to the 
temporal demarcation (final lengthening) normally found at the 
end of phonological phrases, the Greek phonological phrase 
was also characterized by tonal properties such as the scaling of 
accentual peaks wi thin the phonological phrase. 

It is not that anyone doubts that Greek must have had 
phonological phrases, just that there seemed to be no really 
tangible metrical evidence for their definition and analysis. A 
separate study 3 has shown far more metrical evidence than 
previously understood, and this clearly confirms the impor­
tance of syntactic constituency4 in the phonological phrasing of 

3 A. M. Devine and L. D. Stephens, "The Syntax and Phonology of 
Hyperbaton" (forthcoming). 

4 Syntactic constituents are groups of words that go together to make 
grammatical and semantic units of some kind within a clause, e.g. Prepo­
sitional Phrase, Noun Phrase, etc. Syntactic constituency is hierarchical; e.g. a 
Prepositional Phrase may consist of a Preposition plus a Noun Phrase that in 
turn may consist of an Article plus an Adjective Phrase plus a Noun, and the 
Adjective Phrase may consist of an Adverb Phrase plus an Adjective, and so 
on. Not just any string of contiguous words forms a constitutent: e.g. in a 
Prepositional Phrase, Preposition plus Article does not form a constituent, 
since the Article does not belong with the Preposition to the exclusion of the 
Noun. Conversely, in a language with free word order like Greek, a syntactic 
constituent, in the sense used here, need not be continuous, i.e., consisting of 
only contiguous words, but may be discontinuous, interrupted by some other 
constituent or word belonging to some other constituent. In fact interrupted 
structures such as Adjective plus Verb plus Noun, where the Adjective and the 
Noun form a Noun phrase, are common in Greek, e.g. where all three belong 
to the Verb Phrase, as in K01VQV ayytiA£1~ £1tO~ (Eur. Tro. 55): [Adj V N]vP. 
For the depth and rich conceptual content of the notion of constituency, see 
A. Zwicky, "Arguing for Constituents," Proceedings, 14th Regional Meetings 
of the Chicago Linguistics Society (Chicago 1978) 505-12; A. Radford, 
Transformational Grammar (Cambridge 1988) 50ff. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 423 

Greek. But the metrical evidence for phrasing remains a veiled 
and rather imprecise basis for an explicit reconstruction: it 
establishes principles and tendencies without necessarily leading 
to an exhaustive phrasing of any particular stretch of text. 
Fortunately, this lacuna is filled by a substantial body of 
epigraphical evidence pertaining precisely to the temporal 
chunking of speech at a level between the appositive group and 
~he major ~hrase. !his evide~ce .cons~st~ in the placement of 
mterpuncts In certaIn punctuatIng InscnptIons. 

Whereas one class of these inscriptions, including the well­
known Teiae Dirae (SEG XXXI 985), punctuates exclusively or 
almost exclusively appositive groups, another class punctuates 
phrases, often with a sufficient degree of consistency to justify 
detailed study. This latter class includes the following inscrip­
tions chosen as particularly suitable for analysis: 

Acrop. = IC 13 45 
Elis = Meiggs/Lewis 17 (LSAC2 42.6) 
Ephesus = I.Ephesos 1 (LSAG2 pI. 66.53) 
Miletus = LSAM 41 (LSAG2 64.33) 
Naup. = IG IX.P 718 (ICBM IV p.119) 
Oeanth. = IG IX 2717 (LSAG2 15.4) 
Priene = O. Masson and J. Yoyotte, "Une inscription ionienne 

mentionnant Psammetique Ier," EpigAnat 11 (1988) 171. 

Our object is not the epigraphical-philological one of account­
ing for all instances of punctuation in all inscriptions of a par­
ticular corpus but the linguistic-theoretical one of establishing 
that phonological phrases existed in the Greek language and 
analyzing their syntactic composition. Punctuation in metrical 
inscriptions is a less reliable guide to phrasing, partly because it 
can also reflect metrical structure. Although inscriptions in 
which there is a formalized use of punctuation to demarcate the 
items in a list are excluded from this corpus of texts, Miletus has 
been included since it is phrase-punctuated throughout and 
shows how list punctuation originates from consistent phrase 
punctuation: items in a list, even when they comprise only a 
single word, are separated from each other by phrase boundary 
and in many languages are also intonationally demarcated. 

Phonological phrasing in any language tends to be quite 
variable, depending on rate of speech and discourse factors as 
well as semantic, syntactic, and phonological properties. To cite 
just one instance, in Tokyo Japanese minor phrases may consist 
of from one to three words, and the same sentence can have 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

424 THE GREEK PHONOLOGICAL PHRASE 

different pronunciations: the syntagm takonado tabeta ("ate 
octopus, etc.") can be pronounced as two major phrases in 
unnaturally slow speech, as a single major phrase containing two 
minor phrases, or, in more rapid speech, as a single minor 
phrase. 5 Greek should not be an exception. Although much 
apparent variation in inscriptional punctuation is motivated by 
syntactic constituency or by focus, some instances appear to be 
random: 

EXOEVUl 90uv (IG 13 4b.7) 
EXOEVUl : 90uv (IC P 4b.12) 

This variability naturally inherent in phonological phrasing is 
compounded by epigraphical and editorial 'noise': the eviden­
tiary value of any inscriptional instance of punctuation rests on 
the assumption that it is neither an inscriber's error nor a false 
transcription. The latter category of error frequently occurs, 
partly because interpuncts can be difficult to recognize on 
damaged stone surfaces, and partly because the importance of 
punctuation for an understanding of Greek phrasing has been 
generally underestimated. Editorial error can involve omission 
(LSAC2 3.20,4.31,16.4) or addition (SEC XXX 1283); but the 
assumption of widespread inscriber's error is methodologically 
dangerous. There are apparently absurd word internal punctua­
tions of the type 

: Ot OlKUSOV:'tE<; (LSAM 30.B.5) 
TUAEtOE<; 1t:OlE<; (JHS 52 [1932] 171, pI. VII.2), 

but not even all these are outright errors. Problems with a nail 

5 T. J. Vance, An Introduction to Japanese Phonology (Albany 1987); E. 
SELKIRK and K. TATEISHI, "Constraints on Minor Phrase Formation in 
Japanese," Papers from the 24th Annual Regional Meeting of the Chicago 
Linguistics Society I (1988) 316. Other aspects of the syntax and phonology of 
the Japanese minor phrase are investigated in E. SELKIRK, "On the Nature of 
Prosodic Constituency," in J. Kingston and M. E. Beckman, edd., Papers in 
Laboratory Phonology I (Cambridge 1990) 179; E. SELKIRK and K. TATEISHI, 
"Syntax and Downstep in Japanese," Interdisciplinary Approaches to 
Language. Essays in Honor of S.-Y. Kuroda, edd. C. Georgopoulos and R. 
Ishihara (Dordrecht 1991) 519-44. The results presented in these three papers 
are referred to hereafter as 'Selkirk'. There is little cross linguistic uniformity in 
the terminology used for the phrasal hierarchy: in some, but not all, respects, 
what is called the phonological phrase in English and what we call the minor 
phonological phrase in Greek correspond more closely to what is called the 
major phrase in Japanese. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 425 

hole explain the oddity of : 'ta: e; :: A8avatae;. 6 Two other in­
stances seem to involve vacillation between punctuating word 
end and syllable end in the presence of elision: KataA£l1tOv:ta 
EV tat touat (Naup. 7), where other factors have been sug­
gested7 (the expected location for the punctuation is after 
tCTtUll), and: Jl' a:vE6EKE : (LSA G 2 55.3). More troublesome is 
the apparent tendency for some inscriptions to vary between 
minor phrase punctuation and major phrase punctuation. 
Oeanth. B seems to begin with word punctuation, proceed to 
minor phrase punctuation, and end with major phrase punctua­
tion; the length of the punctuated segments also tends to in­
crease towards the end in Ephesus; other inscriptions just lose 
interest in punctuation as the text proceeds (LSA G2 42.8). 
Punctuation is often but not always omitted at the end of the 
physical line; this feature of inscriptional orthographic practice 
can be a source of additional indeterminacy. 

It is intuitively reasonable to assume that the placement of in­
terpuncts in inscriptions is intended to demarcate phonological 
domains and not syntactic constituents, an intuition that is 
confirmed by instances in word punctuating inscriptions where 
punctuation is placed after a phonological composite structure 
consisting of two prepositives: 8 

: Kat 'toe; : hu:po1toLOe; : (IG 13 244.C.3) 
: EV 'tOt : 1tOAqWt : (IG 12 929.2) 

E1tt tEe; : tpt'tEe; (IG 13 261.1); 

This makes good phonological sense-such composites can also 
appear before the caesura in the trimeter (Soph. Aj. 1228, OC 
280), where single prepositives are strongly avoided-but no 
syntactic sense at all, since in such cases a Determiner (article or 
demonstrative) is typically detached from its Noun and at­
tached to a preceding Preposition or Conjunction. Since word 
punctuation systematically reflects prosodic structure, it is un­
likely that phrase punctuation should simply be random, 
haphazard and linguistically unmotivated. The phrase-punctuat­
ing inscriptions form a substantial body of text, so that it is 

6 M. H. Jameson, "A Treasury of Athena in the Argolid (lG IV.554)," in D. 
W. Bradeen and M. F. McGregor, edd., cPOPOE: Tribute to Benjamin Dean 
Meritt (Locust Valley 1974) 67. 

7 R. G. Kent, The Textual Criticism of Inscriptions (=Language Mono­
graphs 2 [Philadelphia 1926]) 19. 

8 See supra n.2. An appositive group need not contain a lexical host. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

426 THE GREEK PHONOLOGICAL PHRASE 

possible, with the above qualifications, to reconstruct at least 
the outline of an algorithm for the formation of the phono­
logical minor phrase in Greek on the basis of the system of 
punctuation. It should be noted that in some inscriptions punc­
tuation may be omitted at the end of a major phrase (Naup. 12, 
Oeanth. 3, LSAG2 71.43/4.19); in others, minor phrases are 
punctuated by a single row of interpuncts and major phrases by 
multiple rows (Ephesus; IG P 4). Apart from this latter device, 
the binary nature of inscriptional punctuation precludes repre­
sentation of any higher level prosodic structure. That the class 
of phrase-punctuating inscriptions is significantly different from 
the class of word-punctuating inscriptions is well illustrated by 
differences in the punctuation of precompiled phraseology: 

: 11 t€XV11 t : 11 Jl11Xavllt : (Teiae Dirae A.8) 
: t€xvat Kat Jlaxavat : (Naup. 38) 

: Kat autov : Kat y€vo<; : (Teiae Dirae B.38) 
: autov Kat to y€vo<; : (Naup. 4). 

On one interpretation, word punctuation is simply phrase punc­
tuation at a rate of speech at which every appositive group is a 
separate minor phrase, as it might be in slow dictation. 

The first step in the phrasing algorithm is to form phrases 
from phonologically binarily branching constituents,9 that is, 
from contiguous appositive groups making a syntactic constit­
uent: 

: taAaVtOV K' apyupo : (Elis 5) 
: tupov ayvov : (Miletus 4) 
: oupacna Jl€AtXJlata : (Miletus 2) 
: €<; ~aCnA€W<; <>tOotat : (M iletus 3) 
: €K 7tOA£(J)<; llv€lXt91lcrav : (Ephesus 2) 
: €V tpta90vt' aJlapat<; : (Naup. 42) 
: to 'v NaU1taK'tov FOtK£OV'tOC; : (Naup. 29) 
: 'tOY ~€VOV Jl€ hay€v : (Oeanth. 1) 
: at 'V€UO€a 7tpO~€V€Ot : (Oeanth. 8). 

9 Constituency, whether phonological or syntactic, is represented by graphs 
known as trees (see Radford [supra nA]). Each point, or node, on the tree 
represents a constituent. If a constituent is complex, i.e., has internal structure, 
then its node branches, each branch connecting it with lower nodes for its 
subconstituents. Thus a binary branching constituent is one that contains two 
subconstituents at the next hierarchical level down. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 427 

There is no reason to suspect that this premium on binary 
structures is a mechanical or artificial epigraphical feature: some 
speakers of Japanese apparently have just such a preference for 
binary accentual structures, particularly in slower speech (Sel­
kirk). A minor phrase can be formed whether the modifier 
precedes the head10 or follows it: 

: £ NaU1taKtO avxop£ovta: (Naup. 19 [PP V]) 
: Kapu~al EV 'tayopal : (Naup. 20 [V PPJ), 

but there may be a preference for phrasing prehead modifiers 
with the head when both a prehead and a posthead modifier are 
present: in : at Ka hU1t' avaVKa<; a1t£AaOVtat : £ N aU1taKtO : 
(Naup. 8) the prehead adjunct is phrased with the head and the 
posthead complement is orphaned. Proper names having the 
structure head plus modifier tend to constitute fixed phrases in 
phrase-punctuating inscriptions: 

: tOt ~t OAUV1ttot : (Elis 6) 
: tOt Zt tOAUV1ttOt (I.Olymp. 3.4) 

A8£vatllt Da'tpotllt : (LSeG 113 [Guarducci, EpigrGr IV 
13 fig. 5J) 

: A1tOAAWVO<; ~EAq)tvto (Miletus 12) 
A1tOAAWvt ~EAq)tVtWt: (LSAG2 65.50); 

sometimes also in word-punctuating inscriptions: : HEPJltt 
EvayovtOt : (IG P 5.3). Unarticulated forms probably have a 
stronger tendency to lack internal punctuation. 

A nonbranching subject or the head of an interrupted 
branching subject phrase may be phrased with a verb, at least 
when the verb is intransitive or when, for whatever reason, its 
complement is not phrased with the verb: 

: Opa1t£tE<; JlE £CHtt : (Acrop. 4) 
: ha OE ~oAa 7t01:EAa'to : (IG IV 554.6) 
: Kat xpEJlata 1taJlato<paytt(Jtat : (N aup. 40, 44) 
: £OptT) KT)pu(J(JEtat : (Miletus 12). 

to Intuitively, the head of a constituent is the part of speech that gives it its 
basic syntactic and semantic properties. Thus a noun is the head of a Noun 
Phrase, a verb is the head of a Verb Phrase. In Greek concord and rection are 
determined by heads. For a recent review of the notion of head, see R. A. 
Hudson, "Zwicky on Heads," Journal of Linguistics 23 (1987) 109-32. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

428 THE GREEK PHONOLOGICAL PHRASE 

These data suggest that the verb is phrased with the subject 
only by default, at least in the absence of interfering phonologi­
cal factors. In one case where a subject seems to be phrased 
with the indirect object in a dedication, the inscriber left space 
for the missing punctuation (Raubitschek, DAA 220). In 
Korean sentences with fronted objects, the rule of obstruent 
devoicing indicates that the subject can be phrased with the 
Verb, but not if the Verb is followed by an Adverb: for instance 
"[balls] [Cuni catches]," but "[balls] [Cuni] [catches quickly]," 
where the brackets indicate how the Korean words correspon­
ding to the glosses would be phrased. t t 

In Cyprian,12 apart from a few instances with phonologically 
heavy non-Iexicals (see supra n.l) 

I FPE'tac; 'tacroE I (/d. 29) 
I a(v}n 'to Jllcr90v I (ld.5) 

and a few fixed phrases with proper names 

I 'ta(v) 1t'tOAtV EoaALOV I (/ d. 1) 
I 'tOY Ovacruru1tpOV 'tOY ya'tEpavl (l d. 2) 
10 I1a<po ~acrlAEuc; I (ICS 6, cf 15, contrast 7), 

most of the clear cases of the divider sign punctuating phrases 
involve a Verb phrased with its subject or complement: 

I Ka'tEFopyov Maool I (ld. 1) 
I avoyov OvacrtAov I (Id. 2) 
I1tElcrEl OvacrtA-oll (Id. 12,25) 
I 0 APJlEVEUC; EXE I (Id. 21) 
I avocrl ya FOl YEVOl'tU I (I d. 29) 
I OlKOV vaov I (lCS 306.4). 

The preponderance of verbs in Cyprian punctuated phrases 
may indicate that in this dialect particularly Verbs lacked 
prosodic salience. The nonverbal element in the phrase tends to 
be free of appositives. The high incidence of proper names 

11 Y. Y. Cho, "Syntax and Phrasing in Korean," in S. Inkelas and D. Zec, 
edd., The Phonology-Syntax Connection (Chicago 1990) 47-62. 

12 O. Masson, Les inscriptions chypriotes syllabiques 2 (Paris 1983: hereafter 
• ICS'). '1 d.' indicates the main Idalion inscription (ICS 217). 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 429 

could be a focus 13 related phenomenon or it could be a reflex of 
the constraint on appositives. In the latter case, it follows that 
for the easiest and most restrictive type of phrasing, the simple 
lexical word phrases more readily than the group consisting of 
lexical word plus appositive(s). 

In general. however, it is clear that the minor phrase is based 
on phonologically branching structure, not on a syntactically 
branching structure. The phrasing algorithm is largely blind to 
appositives. Simple two-word constituents can become a 
phonological phrase: 

: 1tpoo'ta'tav Ka'tao'taoat: (Naup. 34) 
: OtOJlooat hop90v : (Naup. 45), 

but so can syntactically much more complex structures, such as 
branching Prepositional Phrase with articulated Noun and 
articulat.ed m.odifier plus any additional appositive material such 
as a conJunctlOn: 

: EK 'tEe; <pUAEe; 'tEe; ltpU'tavEUOOEe; : (Acrop. 16) 
: at ne; hU1to 'tOY VOJltOV 'tOY £lttFot90v : (Naup. 27). 

The importance of syntactic constituency in the phrasing 
algorithm is illustrated by cases in which a nonbranching 
constituent is flanked by two branching constituents, for 
example NP V NP 

: 'taAav'tov K' apyupo : a1tonvotav : 'tOt ~t OAUV1ttOt : (Elis 5) 
: 'tOte; au'tov VO~ltOte; : xp£o'tat : Ka'ta 1tOAtv FEKao'tOUe; : 

(Naup.28). 

If in these instances the phrasing had been completely random, 
constituency relations would not have been reflected by punc­
tuation at a greater than chance rate. If the phrasing had simply 
paired appositive groups from left to right, the Verb would 
have been phrased with the first following appositive group of 
the right-hand Noun Phrase; if the phrasing had simply paired 
appositive groups from right to left, the Verb would have been 

13 Focus is the new information that the sentence is intended to convey. 
Focus can be seen as typically foreground information, including contrastive 
and counterassertive information, while (unfocused) topic is typically 
background information, similar to the old notion of "psychological subject." 
Both focus and topic may be marked in nonneutral utterances. This may be 
done by word order, particles, or prosodic prominence. Topicalization is the 
name for the special marking of the utterance topic. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

430 THE GREEK PHONOLOGICAL PHRASE 

paired with the second element of the left-hand Noun Phrase, 
again violating constituency. The Verbs are left unpaired be­
cause the algorithm, on its first pass, paired appositive groups 
according to constituency (UUtOV is not appositive: c(. Teiae 
Dirae B.38, Naup. 4). Other instances of phonologicafly non­
branching constituents phrased separately from flanking branch­
ing constituents are the second Prepositional Phrase in : Ul 1(U 
hUTt' UVUV1(U<; UTtEAUOVtUl: E NUUTtU1(to: Ao9pOl tOl HUTt01(VU­
J.llOlOl : (Naup. 8) and the phrasing of the modifiers in : Ol<; 
AEU1(11 : E'Y1(UUP : AEU1(Wl UVU~E~UJ,lEV11 : (Miletus 6), where 
E'Y1(UUP modifies the Noun Phrase Ol<; AEUK1l to form a con­
stituent that is itself modified by the following Participial 
Phrase. Contrast : ()UO yuAAOl E(J'tE9J,lEVOl : (M iletus 2) with 
prepositive ()UO and nonbranching Participial Phrase. 

It should be noted that some of the usual approaches 14 to 
phrasing are too directional in character to handle the Greek 
data. An approach in which phrases are built from the head and 
material on the nonrecursive side of the head 15 privileges the 
nonrecursive side to a degree that appears unwarranted in 
Greek. An approach that phrases right-branching structures 
with a left-hand node, or vice versa, but allows a new phrase to 
start only at the end, or respectively at the beginning, of a 
constituent, cannot account for data that allow inclusion of non­
branching nodes from both the left and the right; for instance, a 
Verb can be joined to a Prepositional Phrase from either the left 
or the right. 

The limitation of first-pass phrasing to two appositive groups 
per phrase means that (phonologically) branching constituents 
can be phrased differently from their nonbranching counter­
parts: 

14 M. Nespor and I. Vogel, Prosodic Phonology (Dordrecht 1986); E. O. 
Selkirk, «On derived domains in sentence phonology," Phonology Yearbook 
3 (1986) 371. For a recent collection of articles on this and related topics, see 
Inkelas and Zec (supra n.11). The book on this topic by A. M. Zwicky and 
G. K. Pullum had not appeared at the time of writing. 

15 The recursive side of the head is the side on which complements or 
adjuncts may freely be added, e.g. the right side of the Noun Phrase in French 
and Italian; the nonrecursive side is the side on which only a limited class of 
items may occur, e.g. the left side of the Noun Phrase in French and Italian. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 431 

: SUEtV Kat AaVxaVEtv : (Naup. 2) 
: ocna AaVxaVEtv : Kat SUEtV : (Naup. 3) 

: 'to 't£<; Nq.l£o£o<; apyupw : (IG 13 248.3) 
: 'to 8£ aAM apyupw : 'to 't£<; N£jl£ow<; : (IG 13 248.9) 

: A09pov 'tov HU1tOKYallt8wv : (Naup. 26, 34) 
: £1ttFOtKOU<; A09pov : 'tov HU1toKYajlt8wv : (Naup. 5) 
: £ NaU1taK'to aVXop£ov'ta : (Naup. 19) 

: £v NaU1taK'tOt : KapU~at £v 'tcXyopat : (Naup. 20) 
: at n<; hu1to 'tov vOjlwv 'tov £1ttFOtKOV : avxop£Et: (N aup. 

27). 

Since the first-pass pairs only contiguous items forming a 
consti tuen t, some sequences of apposi tive groups will remain 
unpaired after the initial pass. These may optionally be paired in 
a second pass through the unpaired material: 

'tov hop90v E~EtjlEV : (Naup. 12) 
: 'tov A09pov 'tc>1ttFot90t : (Naup. 34) 
: at Ka jl£ y£vo<; £v 'tat tonat : (Naup. 16) 
: F£Kao'tov vOJlo<; £on : (Naup. 30). 

This phonological pairing in the absence of syntactic motivation 
is also found in the phrasing of unaccented words in Japanese, 
where optionally indirect object and direct object may be 
paired without intervening initial lowering (Selkirk). But there is 
apparently some resistance to joining either a direct object or an 
indirect object to its Verb to the exclusion of the other 
complement: 

: 'tc>VKaA£tjlEVOt : 'tav 8tKaV : 80jlEV 'tov apxov : (N aup. 41) 
: at Ka jl£ 8t8ot : 'tOt £VKUAEtjl£VOt : 'tav 8tKUV : (Naup. 43). 

The next step is to phrase single appositive groups orphaned by 
the preceding steps. This may be accomplished by allowing the 
orphans to stand as autonomous phrases: 

: 'taAav'tov K' apyupo : u1tonvotav : (Elis 5) 
: 'tOt 6t OAUV1ttOt : 'tOt KaOaA£jl£VOt : (Elis 6) 
: £v u8ptav : 'tav 'Vacpt~LV Etjl£V : (Naup. 45) 

'tEKYat Kat jlaxavat : jlE8Ejltat : FE90v'ta<; : (Naup. 12) 
: 'tOVKUAEtjlEVOt: 'tuv 8tKaV : (Naup. 41, cf 43) 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

432 THE GREEK PHONOLOGICAL PHRASE 

: EV NUU1tUK'tOl : KUpU~Ul EV 'tUYOPUl : (Naup. 20) 
: J.1.E cpupElV : EV Ao9pol<; 'tOl<; HU1tOKVUJ.1.l~hol<; : (Naup. 5). 

Alternatively, the orphan appositive group may be adjoined to a 
preceding or following phonological phrase according to its 
syntactic constituency relations: this solution may have been 
favored when one of the three appositive groups was phono­
logically light or lacked appositives: 

: E1tUYEV J.1.E'tU 'tPlUKOV'tU FE'tEU : (Naup. 13) 
: Evop90v 'tOl<; E1tlFOlKOl<; EV NUU1tUK'tOV : (Naup. 11) 
: h01to<; Upl<J'tU KUl EU'tEAE<J'tU'tU <JKEUU<JUl : (Acrop. 8) 
: Etv 'tOOl 1tpoo'tool XPU<Jool TlVElx't6Tl<JUV : (Ephesus 3) 

'tu ~EVlKU E 6UAU<JU<; h UyEV : (Oeanth. 3) 
: EPOE'tUl no'tEpoov E'tro<; n:ATl : (SLCC 113) 

In the phrasing of Japanese accented words, a ternary branching 
subject phrase may optionally be treated as a single domain; and 
in the phrasing of Japanese unaccented words an object Noun 
following a branching subject phrase is optionally joined to the 
latter without initial lowering (Selkirk). 

In an early fourth-century Boeotian inventory with list punc­
tuation (SEC XXIV 361), simplex numerals are adjoined to a 
preceding branching phrase: 

: AUV1t'tEPooXOl <JlOUPlOl 'tpl<; : (20) 
: XUAKlU nAu'tEu 1tEV'tE (7) 

but a complex numeral is phrased separately : FO l v oxo lU 
XUAKlU : 1tEV'tEKUlOEKU : (13). In particular, a syntactically inter­
rupted structure consisting of three appositive groups can be 
treated as a single phonological phrase: 

TITloroJ.1. J.1.' uVE6TlKEV ooJ.1.cplVVEoo: (Priene 1) 
: «ool ~U<JlA£U<; £0009' rolyt>1t'tlO<; : (Priene 3) 
: 'to 1tUpU 'tOl<Jl 'tU<; OlUKO<JlU<; OPUXJ.1.U<; OCPEAO<Jl (IC I J 248.5). 

The special status of such YXY syntactic structures is con­
firmed by metrical evidence: in the iambic trimeter of tragedy 
positional lengthening (-V#CC-) between X and Y or Y and X 
is permitted to a significantly greater degree in YXY structures 
than in their flanking counterparts XYY or YYX.16 

16 Devine and Stephens (supra n.3). 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 433 

Scope17 affects phrasing, since scope affects constituency and 
constituency affects phrasing: 

: oota AaVxaVEtv : Kat SUEtV (Naup. 2); 

if the punctuation were after o(na rather than AavxaV~tv, the 
phonology would have indicated that oota was the complement 
of both verbs. When a phonologically heavy preposition has 
scope over two conjuncts, there may be some preference for 
not linking the first conjunct with the preposition to the 
exclusion of the second conjunct: 

: EXSOC; 7tpO~EVO Kat FtOtO ~EVO: (Oeanth. 11: Kat begins a 
new line) 

: aVEuc; : ~oAav : Kat SaJlOv 7tAaSuov'ta (1.0Iymp. 3.8). 

Although articulated modifiers can follow the basic rules of 
phrase formation as in various examples already cited, in certain 
contexts they can also show some degree of prosodic auton­
omy. A probably branching articulated modifier is phrased 
separately from its Noun in an example involving a head-inter­
rupted Verb Phrase: 

EV 'tE7ttapOt K' EVEXOt'tO : 'tOt 'v'tau't' EypaJlEVOt (Elis 10).18 

Two cases have an articulated modifier located between two 
elements of a Noun Phrase; the articulated modifier is not 
phrased with either element of the Noun Phrase to the 
exclusion of the other: 

: XaAEtEOtC; : 'tOtC; GUV Av'ttcpa'tat : FOtKE'tatC; (Naup. 47) 
Kat NaF7taK'ttoV : 'tOY E7ttFOt90V : 7tA£Sat : (Naup. 40). 

The separate phrasing of articulated modifiers can result 
prima facie violation of constituency: 

: E7ttFOtKOUC; A09pov : 'tOvHU7tOKVaJltOtoV : (Naup. 5) 
: otoJlooat hop90v : 'tOY VOJltoV : (Naup. 45). 

In a 

17 Scope is the part of a sentence or stretch of words over which the 
meaning of a word applies. Varying possibilities for scope are the source of 
well known ambiguities of the type 'old men and women', in which if the 
scope of 'old' is wide only old people are included, but if it is narrow, i.e., over 
'men' only, then young women would also be possible referents. 

18 A. Johnson reports that the punctuation is partly worn but clearly visible. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

434 THE GREEK PHONOLOGICAL PHRASE 

The most likely explanation for the separate phrasing of 
articulated modifiers is that the Noun is treated as predictable 
information and the following articulated modifier is highlighted 
in its function of specifying a subset, which results in its 
prosodic autonomy. This approach is supported by the parallel 
phrasing of Noun Phrases followed by resumptive deictics: 

: al bE np 'ta 'YpacpEa : 'tal KabaAEOl'tO : (Elis 7) 
l'tov a(v)bplya(v)'tav I 'to(V)bE KatE<lta<lE I (/CS 220) 

'tOY a(v)bpla(v)tav [space] tov(v)u I EbooKEV [end line] 
Ka<l IOVE9EKEV [space] (ICS 215; phrase end is denoted 
by space, word end by the divider). 

A significant body of prose evidence relating to the word order 
of postpositive particles and vocatives indicates that topicalized 
and contrastive material in general is phonologically demar­
cated:19 

to\><; ~EV AaKEbaq.lOvloU<;, ib livbpE<; ~u~~axol, OUK av En 
aitla<lai~E9a (Thuc. 1.120) 

KA..EOCProV'tO<; toivuv, C1 livbpE<; OlKa<ltal, EtEpa ~EV liv n<; 
EXOl Katll'Yopll<lal (Lys. 30.12) 

The likelihood that focused constituents will form a separate 
minor phrase may explain certain apparent anomalies, but this 
approach is circular to the extent that the discourse structure 
assumed cannot be independently motivated. Constituents that 
are contrastive or focused, particularly at the beginning of a 
paragraph, tend to appear as phrases consisting of one ap­
positive group only: 

: Ka 'tOVOE : ha '7tlFolKla : (Naup. 1) 
:A09pov tOY : HU7tOKVa~lOLOv : (Naup. 1; elsewhere mostly 

unpunctuated) . 
: IIEp909aplav : Kal MU<laXEOV : (Naup. 22; unpunctuated in 

line 27) 
: KEV A09pOl<; : tOl(<;) HU7tOKVa~loLOl<; : (Naup. 21; contrastive 

with EV NaU7taKtOl in line 20). 

19 E. Fraenkel, "Kolon und Satz," Kleine Beitrage (Rome 1964) I 73-92 (= 
NGG [1932] 197-213), ·Kolon und Satz, II," 193-130 (=NGG [1933] 319-54), 
·Nachtrage zu 'Kolon und Satz'," I 131-39, and ·Noch einmal Kolon und 
Satz," SBMiinch (1965) 2. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 435 

It is well known that in addition to syntactic constituency, 
phrasing is also sensitive to the need for phonologically 
balanced prosodic constituents. 20 This may explain why in a few 
cases constituency is disregarded in favor of phonologically 
balanced phrasing: 

: apyupat 1tEV'tE : Kat Et?Ocr(t) JlVEat : (Ephesus 2) 
: xpEJla'ta 'tOLe; HU1tOKVaJll8LOle; : VOJlLOle; xpEcr'tat : (Naup. 24), 

or when a numeral is phrased with a Preposition and separated 
from its Noun: 

EV'tOe; hEXcrEKov'ta : EJlEpOV : (Acrop. 12) 
JlEXpt 'tptoV : O~EAOV (IG 13 4.B12). 

This restructuring is another indication that inscriptional punc­
tuation is ~n orthographic representation of phonological and 
not syntactIc structure. 

Conjuncts and disjuncts are regularly phrased together: 

: at'tE FE1tOe; at'tE Fapyov : (Elis 3) 
: 'ta 't' aA(a> Kat 1tap 1tOAEJlO : (Elis 4) 
: KE 8aJlo K£ 90tvavov : (Naup. 4) 
: 'tExval Kal Jlaxaval : (Naup. 38, cf 12). 

Pairing of conjuncts and disjuncts proceeds left to right, so that 
a third item is phrased separately: 

: al'tE F£'tae; anE 'tEAEo'ta : al'tE 8aJloe; : (Elis 8). 

A recurrent feature of the phrasing of conjuncts and disjuncts is 
disregard for constituency: 

: 'tPE'tO Kal 8aJlEuEOo8o : EVe; A8Evalav : (IG IV 554.5) 
: E 'tae; XaAEl8oe; : 'tOY OtaV8Ea JlE8E 'tOY XaAElEa : E 'tae; 

Olav8l8oe; : (Oeanth. 2) 
: H01tOV'tLOV : 'tE XlALOV : 1tAE8at Kal NaF1taK'tLOv : 'tOY 

E1tlFOt90v : 1tAE8at : (Naup. 39). 

It is not clear whether this violation of constituency is due to 
restructuring or simply to the mechanical application on the 
part of inscribers of a rule requiring conjuncts to be phrased 
together. 

20 J. P. Gee and F. Grosjean, "Performance Structures: A Psycholinguistic 
and Linguistic Appraisal," Cognitive Psychology 15 (1983) 411. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

436 THE GREEK PHONOLOGICAL PHRASE 

Of the three questions we would like to answer about the 
phonological phrase in Greek-what are its syntactic com­
ponents? what are its temporal and rhythmical properties? 
what, if any, are its accentual properties?-the above analysis 
has provided a fairly complete answer to the first. The answer 
to the second question follows automatically, at least in broad 
outline. The mere fact of inscriptional phrase punctuation 
implies some form of temporal demarcation, and this in turn 
implies a more cohesive domain for rhythmic organization as 
well as for the application of sandhi processes. 21 Between 
interpuncts, for instance, Priene consistently applies the rules of 
connected speech-elision, crasis, assimilation of final nasals, 
movable nu. It is also reasonable to look for accentual proper­
ties specific to the Greek phonological phrase. Recent work on 
the Japanese pitch accent22 has revealed that long major phrases 
can be broken up into more manageable subunits by means of 
an intermediate raising of the pitch range at the beginning of 
(often, but not exclusively, syntactically defined) subunits. The 
result can be interpreted as a hierarchical structuring of phrasal 
downtrends. Musical records can be used to discover whether 
something of a similar nature occurred in Greek. 

It is necessary to begin with a review of the principal 
phonological process affecting accentual pitch excursions in the 
domain of the major phonological phrase, namely catathesis, the 
name given to the lowering and compression of the intervals of 
rise to and fall from pitch peaks in alternating High-Tone/Low­
Tone structures; catathesis is distinguished from declination, a 
smaller progressive lowering of pitch from the beginning to the 
end of the domain. 23 The effects of catathesis are dramatic 
whether the High-Low structures are tonal, as in Hausa,24 or 

21 Sandhi (from Sanskrit sam + dhii, 'put together') is a term covering the 
different phonological modifications that words undergo when they come 
into contact. Such modifications in Greek are traditionally classified as, e.g., eli­
sion, aphaeresis, crasis. 

22 H. Kubozono, The Organization of Japanese Prosody (diss.Edinburgh 
1987); cf Selkirk. 

23 Full documentation and discussion in A. M. Devine and L. D. Stephens, 
"Dionysius of Halicarnassus, de compositione Verborum XI: Reconstructing 
the Phonetics of the Greek Accent," TAPA 121 (1991, forthcoming). 

24 S. Inkelas and W. R. Leben, "Where Phonology and Phonetics Intersect," 
in Kingston and Beckman (supra n.6) 17. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 437 

accentual, as in Japanese.25 In Greek catathesis applies to 
sequences of fully accented words, that is words bearing the 
acute or circumflex, or appositive groups. Catathesis is very 
strongly respected in the musical settings of the Delphic 
hymns. 26 Although sometimes the accentual peak of the second 
of two successive, fully-accented words is set higher than the 
peak of the first, both the frequency and the magnitude of pitch 
falling from the first to the second peak are so great that an 
overall mean fall from one accentual peak to . the next of a 
semitone occurs. Catathesis is blocked by the grave accent. 
Since the grave does not have the pitch fall after the accentual 
peak that triggers catathesis, a word bearing the grave may form 
a single accentual trajectory with a following, fully-accented 
word. The result is that in strings of full + grave + full accent the 
peak of the second full accent may be higher than than that of 
the first. Allowing for the effect of the grave and for local pitch 
rises due to emphasis, it is typical for a clause or major syntactic 
phrase to have its highest note on the accentual peak of its first 
fully accented lexical word, after which the pitches of the peaks 
and valleys fall all the way to the end. At the beginning of the 
next clause or major phrase the pitch level is reset upward, but 
within periodic sentences generally not quite so high as in the 
preceding clause or major phrase. 

Both the relative magnitude of the lowering of accentual peaks 
and the locations in which the infrequent exceptions to it may 
occur are regulated by the articulation of the catathesis domain 
into constituent minor phonological phrases. To avoid circu­
larity in the definition of minor phonological phrases, the 
analysis will proceed initially in terms of syntactic constituency 
relations. It will be shown (1) that syntactic constituency 
imposes very strict constraints on the relative magnitudes of 
successive peak-to-peak intervals in the musical setting of the 
Delphic hymns; (2) that the syntactic structures involved in 
those constraints correspond to minor phonological phrases; 
and (3) that minor phonological phrase structure not only 
explains the exception to the syntactically formulated con­
straints, but also accounts for the variability of the interval 

25 W. J. Poser, The Phonetics and Phonology of Tone in Japanese (diss. 
Massachusetts Institute of Technology 1985). 

26 E. Pohlmann, Denkmaler altgriechischer Musik (Niirnberg 1970: 
hereafter 'DAM') nos. 19f. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

438 THE GREEK PHONOLOGICAL PHRASE 

relations in syntactic configurations not included in the initial 
syntactic formulation of the rules. 

The unit of analysis is a string of three fully-accented lexical 
words not interrupted by grave accents. Such a string is the 
minimal sequence in which two successive peak-to-peak inter­
vals can be observed. Proper names constitute a special class 
and will be treated separately.27 Description of the interval re­
lations is simplified if falling intervals are defined as positive and 
rising intervals as negative; in other words the interval from D 
to C is + 1 tone, the interval from C to D -1 tone. 28 

The basic principle is that greater peak-to-peak intervals occur 
within smaller, more cohesive syntactic constituents, and, con­
versely, smaller (including rising) intervals occur across such 
constituents, i.e., within larger, less cohesive stretches of the 
sentence. This principle can be rendered sufficiently explicit for 
empirical testing as follows. In any string of three lexical words, 
Ll L2 L), not interrupted by grave accents or non-clitic non­
lexicals, (1.a) if Ll does not belong to the smallest constituent 
containing L2 L), then the peak-to-peak interval between L2 and 
L3 is not less than the interval between Ll and L 2• Conversely, 
(2.a) if L) does not belong to the smallest constituent containing 
Ll L 2, then the interval between L 1 and L2 is greater than the 
interval between L2 and L). If the smallest constituent 
containing Ll L2 is identical to the smallest constituent 
containing L2 L), then the interval relations are determined by 
asymmetry in subconstituency between Ll and L). (1.b) If Ll is 
not a sister of L2, but belongs to a branching subconstituent (i.e. , 
has a sister to its left) and L3 does not, then the interval between 
L2 and L) is not less than the interval between Ll and L 2. 
Conversely, (2.b) if L) belongs to a branching subconstituent 

27 The frequency of interruption by grave accents and non-clitic non-Iexicals 
and damage to the inscriptions reduce the number of such lexical triples, but 
the settings of enough are preserved to permit a definitive statistical test as 
given below in the text. 

28 The definition has no substantive consequences and in no way prejudices 
the test results; it merely allows the substitution of 'greater' for the periphrasis 
'which falls more / rises less'. If the reverse definition is chosen, then the 
inequalities in rules 1 and 2 are reversed. Defining falling intervals as positive, 
however, has the intuitive appeal of according with the progressive lowering 
of pitch imposed by catathesis. Furthermore, since intervals measured in tones 
are merely the logarithms of fundamental frequencies (I ij=log[f/fj]=log fi - log 
~), when intervals are calculated in the direction that the notes are sung, if the 
hrst is higher than the second the difference log f i-log f j is positive, and, 
conversely, if the second is higher than the first, the difference is negative. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 439 

(i.e., has a sister to its right) and L 1 does not, then the interval 
between Ll and L2 is greater than the interval between L2 and 
L3. 29 If L h L2, and L3 are sisters, or if Ll and L3 are both 
immediate constituents of respective branching subconstitu­
ents, no prediction is made about the relative magnitudes of the 
intervals. Rules 1.a-b and 2.a-b can be more readily com­
prehensible with the help of some shorthand notation. Let 112 
and 123 stand for the magnitudes of the intervals between the 
peaks of the accents of L 1 and L2, and, respectively, between 
the peaks of L2 and L 3. Let square brackets indicate the 
constituency relations among LJ, L 2, and L3. Then 

2.a. [Ll L2] L 3 } 

2.b. [Ll L2 [L3X]] 

Rule 1.a applies to a lexical triple such as 

~ ~bl'~ ~ r I a !p PO I 0 ~ Jl 
3 ~pE-\.lWV a - EI - OA - OI-OIS ~[E] - AE - OW 

DAM 19.14 

since the Adjective and the Noun belong to a Noun Phrase that 
does not include the Participle on the left. The prediction of the 
rule is correct, since 112 = -0.5 but 123 = +0.5 tones. Rule l.b 
applies to a lexical triple formed by the last three words of a 
Noun Phrase consisting of a branching Genitive Phrase + Noun 
+ Adjective, e.g. 

29 Constituents are sisters if they are at the same hierarchical level; thus a 
Genitive Phrase consisting of an Adjective and Noun may be the sister of a 
single Noun that is the head of the Noun Phrase comprising all three words, 
but the Noun in the Genitive is not itself a sister of the Noun that is the head 
of the Noun Phrase. The rather involved definition of rules l.b and 2.b is 
required because we do not wish to beg the questions of formal syntactic 
analysis posed by discontinuous constituents in Greek. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

440 THE GREEK PHONOLOGICAL PHRASE 

t(pl] - 13pO-l.Iov-ov av-ya-Tpes ev - W - A(eVOI,] 

DAM 19.2 

since there are no obvious grounds for grouping 9{ryatpE<; with 
either the Genitive Phrase to the exclusion of the Adjective, or 
vice versa, but the Genitive Phrase branches, so that £pl~p6JloU 
has a sister to its left. The prediction of the rule is correct, since 
112 = +0.5 but 123 = + 1 tone. Rule 2.a applies to a lexical triple 
such as 

I ~a ~ ~ ~ IDr 0 r I br DAM 20.3 
8 at VI - cpo-136- AOVS T!~-TPCXS val - ee' 

since the Adjective and the Noun belong to a Noun Phrase that 
does not include the Verb. The prediction of the rule is correct, 
since 112 = +2 but 123 = 0 tones. 

Excluded from these rules are head interrupted configura­
tions in which neither of the flanking modifier or complement 
element branches, for example 

JlaVtElOV £CPE7tWV mxyov (DAM 19.8) 
VEWV Jlllpa tauprov (DAM 19.12) 

and also configurations of two modifiers preceding (or follow­
ing) their head, for example, otv(J)7ta oacpva<; (DAM 20.23f). 
These configurations will be treated separately. 

We now provide examples of how rules 1 and 2 regulate the 
interval relations in different syntactic structures. Rule 1 pre­
dicts that in a triple in which the first lexical belongs to a 
Prepositional Phrase, the second to a Direct Object Phrase, and 
the third is the Verb, if there is a falling interval and a rising inter­
val, the falling interval will occur between the word of the 
Direct Object Phrase and the Verb, the rising interval between 
the word belonging to the Preposi tional Phrase and word 
belonging to the Object Phrase. 

~ pbi• r 0001 rzorJJCJtI 
t:J.eA- cpov a-vex [lTpJw-w- va \,la-av-



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 441 

ct 0 0 0 I j ]J J II DAM 19.7-8 
T£l - E1 - ov t<p- ~- - 1l'CUV iTO:-yov. 

Here the falling interval of 1.5 tones occurs between the peak of 
1ii and cpt, but the rising interval of 1.5 tones between the peak 
of 1t pro and the peak of n:i. Rule 1 permits equal intervals in 
structures such as verb + branching Noun Phrase 

a 01- ei - &(1) ve-cuv 
I r 0 La I ~F DAM 19.12 
~f1 - po TO-OU- pwv' 

with equal intervals of a semitone, and in structures such as the 
last word of a Noun Phrase + a branching Adjective Phrase 

@~ a r I ~. 0 r 6 r p r IDAM 20.20-1 J 
a T£X-Vl - 'TW-WV lv - 01- KO - OS iT6 - AEI 

with two zero intervals. 
Rule 2 means that in a structure of, for example, branching 

Genitive Phrase + verbal form, if there are two unequal falling 
intervals between the accentual peaks, the larger interval will 
occur within the Genitive Phrase and the smaller between the 
last word of the Genitive Phrase and the Verb. 

,~ r IL be ~ r Ir 
9 YAOV - KO - [Ci]S ~ - Acil - as 

DAM 20.6 

Here the larger falling interval of two whole tones occurs be­
tween the peak of ac; and Aat within the Genitive Phrase, but 
the smaller falling interval of only a semitone between Aat and 
the peak of you. Similarly in a structure of branching Noun 
Phrase indirect object + Verb, if there is one falling and one 
rising interval, the falling interval will occur within the Noun 
Phrase, the rising interval between the last word of the Noun 
Phrase and the Verb. 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

442 THE GREEK PHONOLOGICAL PHRASE 

II lTO - e71 eva - -rol-oiS lTpo-<paf - vel - [eiS A0Y'a.] 

DAM 19.20 
Here the falling interval of 1.5 tones occurs between the feaks 
of 1[0. and 'tolC;, but the rising interval between the peak 0 'tOl<; 
and <pai. Again, in a structure of branching Noun Phrase direct 
object + Verb, if there are two unequal rising intervals, the 
smaller rise will occur within the Noun Phrase, the larger rise 
between the final word of the Noun Phrase and the Verb. 

& L!l 
~ ~ lz 
9 [-rp]f-lTo-oa 

r 
~av -

I Gil 0 r 
-rel- ei - ov Ws EI - el - [AES. 

DAM 19.21 
Here the larger rising interval of 1.5 tones occurs from the peak 
of 'tEL to that of d , but the smaller rising interval of a whole tone 
between 'tpi and the peak of ui within the Noun Phrase. 

The table below presents the results of a statistical test of rules 
1 and 2 in the Delphic hymns. 

Ll [L2 L3], etc. 
[LI L2] L3, etc. 

P = .0041 

Interval Relations Between Accentual Peaks 

The distribution observed is highly significant. Even though 
there are only thirteen lexical triples that meet the structural 
descriptions of rules 1 and 2, there is (by the hypergeometric 
distribution) only a chance of about 1 in 244 that the relations of 
the interval magnitudes would accord with the syntactic struc­
ture as prescribed by the rules purely as a result of chance. 

There is only one exception to a perfect, bidirectional im­
plication: 

~ ,"" r 
• : 

8 Ko- <TTa-Af-oos e - ov- v-oPOV va - 1Ja-r' ~ - 1Tl - vi - O'E--rOI. 

DAM 19.6 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 443 

Here the greater interval of fall occurs between the direct 
object noun and the verb rather than between the direct object 
vclfla't' and genitive Adjective EUUbpou of the branching 
Genitive Phrase that modifies vclfla't'. This case, however, is an 
instructive exception, for it illustrates that pitch relations are 
regulated directly by minor phonological phrases. Pitch rela­
tions can to a high degree of accuracy be expressed in syntactic 
terms, because the formation of minor phonological phrases is 
strongly determined by syntactic constituency, but it is not 
completely so determined. In DAM 19.6 vclfla't' could not 
form a minor phonological phrase with EUUbpOU to the ex­
clusion of the head of the Genitive Phrase, Kacr'taAlboc;. As 
noted above, minor phonological phrases are preferentially 
formed from two, rather than three, lexical words or appositive 
groups, unless they are in a head-interrupted structure. There is 
a further reason why formation of a single minor phonological 
phrase spanning the entire Noun Phrase Kacr'taAlboc; EUUbpou 
VclflCl't' is unlikely. In the Delphic hymns, proper names such as 
Kacr'taAlboc; are treated as emphatic, focused constituents, and 
given a pitch boost. Not only does emphatic material not form 
minor phonological phrases with contiguous unemphatic ma­
terial, but, as we have seen, in inscriptions that punctuate minor 
phonological phrases, the structure proper name + modifier is 
punctuated separately, not inclosed with other words inside 
interpuncts. Thus in the present case vclfla't' was undoubtedly 
orphaned from the minor phonological phrase containing the 
proper name and its epithet and adjoined to its governing Verb 
£7tlvim:'tul. The pitch relations are now predictable from the 
structure of the minor phonological phrases: the greater fall 
occurs within, not across the pair of words belonging to the 
same minor phonological phrase. 

We have seen (supra n.3) that interrupted structures consist­
ing of three appositive groups may be treated as a single minor 
phonological phrase in inscriptions and in meter. Certainly in 
the Delphic hymns head-interrupted configurations behave 
differently from structures that have more strongly differen­
tiated internal constituency. In these head-interrupted con­
figurations there is no constraint on the location of the greater 
and the smaller peak-to-peak interval. For example in 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

444 THE GREEK PHONOLOGICAL PHRASE 

~ ?"h tv I c:r 0 G 0 I J ]I J II DAM 19.8 
8 ~a-av- 'TEl - eI - OV tcp- t- - Tl"WV Tr<l-yOV. 

the greater interval occurs between the second and third words 
(probably because of the steep drop in pitch resulting from 
terminal fa1l 30 at the end of a periodic sentence), whereas in 

just the opposite relation holds, although in each case the inter­
rupted constituent is a Noun Phrase in the order Adjective, 
Noun. Since there is no internal differentiation into separate 
minor phonological phrases, lexical triples in head-interrupted 
configurations trivially conform to the constraint that the 
interval within the minor phonological phrase is greater than the 
interval across minor phonological phrases, whatever the in­
tervals may be. Accordingly, their peak-to-peak pitch relations 
may be determined wholly by other factors, such as terminal 
fall or considerations of emphasis that are independent of or not 
also signalled by the order of the interrupted structure. 

The interval relations in the configuration consisting of 
modifier + modifier + head Noun in 

suggest that otviima. oaq>va.c; at least forms a minor phonological 
phrase, probably but not necessarily to the exclusion of KAa­
oov . Unfortunately there is no evidence that could indicate a 
difference between complements and adjuncts. 

The syntactically formulated rules 1 and 2 may now be re­
stated in terms of minor phonological phrases. Within minor 
phonological phrases the pitch of accentual peaks falls more / 
rises less, than across minor phonological phrases, where pitch 
falls less / rises more. These rules may be formalized as 

30 See Devine and Stephens (supra n.23). 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

A. M. DEVINE AND LAURENCE D. STEPHENS 445 

where q> indicates that the brackets now refer to the boundaries 
of m~nor phonological phrases. 

Rules l' and 2' may be tested against all pairs of uninterrupted 
fully-accented lexical words in the hymns. This prediction is 
that the mean interval between accentual peaks will be greater 
in pairs that are within the same minor phonological phrase, 
smaller in pairs of words belonging to different minor 
phonological phrases. The size of the data set for the test is 
therefore greatly increased beyond that available for the lexical 
triples above. The results are presented in the table below. 

Within 
P1P2]<J> = 1.12 

s = 1.00 
t = 3.68, df = 46 

p<.OOS 

Across 
Pt<J>][<J> P2 = 0.00 

s=1.11 

Mean Interval between Peaks 

The parsing into minor phonological phrases was done accord­
ing to the algorithm deduced from inscription punctuation. The 
mean fall within minor phonological phrases was found to be 
just over a whole tone, whereas the mean fall across minor 
phonological phrases was observed to be zero. The difference 
between the means is highly significant. 31 The effect of rules l' 
and 2' on accentual pitch excursions may be illustrated by a 
schematic diagram of a major phonological phrase articulated 
into three minor rhonological phrases. Each minor phrase is 
taken to consist 0 only two medially-accented lexical polysyl­
lables. Such polysyllables have a Mid-High-Low, rather than a 
Low-High-Low, pitch pattern. 32 The diagram does not attempt 

31 The parsing into minor phonological phrases employed for the test is not 
the only one compatible with the algorithm discussed above, but one that 
minimizes minor phonological phrases of three appositive groups in length. 
Other parsings will yield somewhat different means, but the mean interval 
within minor phonological phrases will always be greater than the mean 
across them. 

32 See Devine and Stephens (supra n.23). 



DEVINE, A. M., The Greek Phonological Phrase , Greek, Roman and Byzantine Studies, 31:4 
(1990:Winter) p.421 

446 THE GREEK PHONOLOGICAL PHRASE 

to represent to scale the compression effect of catathesis, only 
the average lowering of about a tone of successive accentual 
peaks within minor phrases, and the small pitch boost at the be­
ginning of successive minor phrases that blocks the lowering of 
the first peak vis-a-vis the last peak of the preceding minor 
phrase. 

[Minor Phrase1] [Minor Phrase 2] [Minor Phrase 3] 

The interval relations described by rules l' and 2' are the re­
sult of the processing of pitch excursions in minor phonological 
phrases. There is a major upward reset at the beginning of pri­
mary catathesis domains, which are basically coterminous with 
major phonological phrases. There is also a minor pitch boost, 
much smaller than initial reset, at the beginning of minor 
phonological phrases. As a result, the pi tch of the first full ac­
cent in a minor phonological phrase is not as much lower than 
the last peak of the preceding minor phonological phrase as is 
the following pitch peak within a minor phonological phrase. 
The initial pitch peak may in fact sometimes be higher than the 
last peak in the preceding minor phonological phrase, but on 
the average it is about the same, since the mean interval was 
observed to be O. It follows that catathesis works in terms of a 
self-similar hierarchy: (1) the minor phonological phrase that 
has a minor pitch boost on its first lexical full accent and strong 
catathesis lowering of the following full accent; (2) the major 
phonological phrase, or primary catathesis domain, which has a 
major pitch reset at its beginning and general fall over the minor 
phonological phrases within it; (3) the periodic sentence, which 
has an even greater initial pitch level and long-term pitch fall 
over the primary catathesis domains within it. 

STANFORD UNIVERSITY 

UNIVERSITY OF NORTI-I CAROLINA, CHAPEL HILL 

March,1991