Three
Comparative Prosody

Prologus

Any study of traditional oral phraseology and narrative structure must very early come to terms with the prosody that exists in symbiosis with them. Milman Parry recognized this need when he stipulated (1928a, in 1971, 272) that the Homeric formula must occur "under the same metrical conditions," and Albert Lord (1960, 20) extended the metrical requirement to the formula in Serbo-Croatian epic. Francis P. Magoun, Robert P. Creed, and others were also careful to prescribe prosodic rules for the recurrence and morphology of formulaic phraseology in Old English poems.^[1] In more recent scholarship, however, prosodic strictures on the formula have been relaxed, or at least de-emphasized.^[2] Although the declared motive behind this drift away from philological rigor—the putative search for "originality" or "creativity" in reaction against the supposedly mechanistic model of Parry and Lord^[3] —may at first sight seem laudatory, in fact we cannot hope to gain an informed perspective by abdicating precision and thoroughness. Aesthetics does not emerge by default from the short shrift of incomplete analysis. We need to know as accurately and intimately as possible all dimensions of formulaic structure before we can responsibly begin to criticize a traditional text. And, as the linear and recurrent formant of all phraseology, prosody deserves highest priority as the most basic stylistic feature of formulaic structure.

[1] For a history of scholarship on the formulaic phrase in Old English, see Foley 1981b, 52-79, 103-16; also Olsen 1986, 1988.

[2] E.g., Nagler 1974, 1-63.

[3] This type of reaction began with articles such as Calhoun 1933 and 1935, and has manifested itself more recently in works such as Norman Austin's "The Homeric Formula" (1975, 11-80) and Vivante 1982.

With the fundamental nature of prosody established, the next step is to ask how the prosodic rules of the three poetic traditions compare. Since each language fosters an individual, idiosyncratic meter, we should expect the Homeric hexameter, Old English alliterative line, and Serbo-Croatian epic decasyllable to differ in distinct and recognizable ways. Even if it proves possible to trace each line back to a common Indo-European ancestor, the singular character and development of the three language traditions will have impressed a particular stamp on each prosody. Thus, if we mean to enforce the spirit of philological rigor and the letter of tradition-dependence championed in the preceding chapters—that is, if our comparison is to aspire to a sensitive, articulate account of differences as well as similarities—we cannot hold on to the illogical and demonstrably faulty assumption of either an "archetypal prosody" or an "archetypal formula." Because languages differ, the prosodies formed within them differ; thus the formulaic structures interactive with those prosodies must likewise differ. The universal theorem of formulaic style must be tempered by the corollary of tradition-dependence; if our later discussions of phraseology and narrative structure are to have the meaning that only a truly comparative philology can give them, then we must begin the comparison at the fundamental level of prosody.

Method

An interpretive essay on comparative prosody in these three epic poetries would, if exhaustively done, constitute a sizable project in itself. A great deal has been written on the hexameter, alliterative line, and decasyllable, and simple review of that scholarship would demand an expenditure of space unwarranted in these studies as a whole. As a methodological premise, then, I concede from the start that my goal is certainly not a full and exhaustive treatment of the subject per se, but rather an examination of those aspects of the subject most crucial to an understanding of oral traditional structure. In other words, I will be concerned primarily with those prosodic factors which determine the shape and texture of formulaic diction and therefore of the more extensive narrative units that also rely, in the last analysis, on structure at the level of the individual line or even line-part. This is not to say that I intend to ignore or intentionally shortchange any particular areas; I wish merely to indicate from the outset that the emphasis is on the comparison of prosodies, and especially on their role in oral traditional structure .

I advocate three principles of evaluation as an attempt at a single, unified methodology. First, what follows in this chapter will straddle the gap between an exacting description of each epic line in and of itself and an approach that permits productive comparisons in the directions indicated. Second, uniform and consistent recourse to prevailing scholarly opinion on individual prosodies will give way to consideration of minority views only when no clear consensus

on a given point can be identified or when we pass beyond the present state of knowledge. Third, I offer the entire exposition as a useful explicative tool, without proclaiming its necessary "rectitude" and without pronouncing it ex cathedra .^[4] In this manner the available data on the three individual epic lines can be gathered, collated, and presented in a suitable format. In the subsequent process of comparison for the purpose of understanding the traditional verbal structures based on these lines, I hope to return to each prosody as much new comparative knowledge as was taken from each for the preliminary descriptions.

The Indo-European Question

Any proposal to compare three prosodies from related Indo-European language families—Hellenic, Balto-Slavic, and Germanic—must take account of the extent to which similarities may be due to emergence from a common prototype or prototypes in Indo-European and of the significance that such genetic similarities might have for the present study. To a degree, then, we should address two interlocking problems. First, what does comparative historical metrics have to tell us about the meters involved, and how convincing are the conclusions? Second, does the historical perspective bear importantly on the particular kind of investigation underway in this and subsequent chapters?

The methods of reconstruction originally developed for recovering elements in the Indo-European lexicon were first applied to the study of meter by Antoine Meillet. In his Les Origines indo-européennes des mètres grecs (1923, 76), the great French philologist characterized a prehistoric parent meter deduced from ancient Greek and Vedic Sanskrit evidence:

Le vers indo-européen, fait pour une langue dont le rhythme était purement quantitatif, était caractérisé par des cadences définies au point de rue de la durée.

[4] I do not expect in all or most cases to have chosen the "right" path, or for that matter the "wrong" path, since I do not believe that a descriptive prosody is only either fight or wrong, elect or damned. Varying perspectives often camouflage deeper correlations; to put it another way, the nature of any approach must assist more than is customarily assumed in determining the nature of the results. As I suggested in "Formula and Theme in Old English Poetry" (1976a, 208), "a method of scansion is only as successful as it is both thorough and exclusive in accounting for what exists in the Old English poetic corpus. A metrical system which can be thorough and exclusive without devolving into a catalog is extremely useful. No scop thought in terms of lambs or measures, but if we can develop a way to talk about what a singer intuitively felt and, at the same time, if we do not seriously distort the continuity of his art into our own critical perspective, then we may proceed without doing violence to that art. We may produce a useful idiom." Compare Martin L. West's dosing statement to "A New Approach to Greek Prosody" (1970, 194): "In conclusion, I should like to point out that what is sketched in the foregoing pages is not a 'theory' to be accepted or rejected, but a formula for classifying empirical data; it is no more conjectural than the Dewey system of classifying books, it is neither true nor false. It must be judged on its ability to match the complexity of the facts."

Les vers longs, à partir de 11 ou 12 syllabes, avaient de plus une coupe à place légèrement variable, dans la première partie du vers. La partie initiale du vers —dans levers à coupe, presque tout ce qui précédait la coupe—ne comportait pas une répartition fixe de longues et de brèves.^[5]

Testimony from other language families has since been added to Meillet's work: by Roman Jakobson (1952) for Slavic; by Calvert Watkins (1963) for Celtic; by Gregory Nagy (1974, 1975) for ancient Greek and Sanskrit; by Berkley Peabody (1975) for Indic, Iranian, and ancient Greek; by John Vigorita (1976, 1977a,b) for Slavic and Greek; and by M. L. West (1973a,b) for all of these plus Italic, Germanic, and in smaller compass other groups.^[6] The imaginative efforts of these scholars have uncovered much that is valuable and extremely suggestive.

At the same time, it is only fair to report that labors in this incipient field have not been universally accepted without reservation. Jerzy Kurylowicz (1970, 421-22, 429), for example, argues that "the conclusion that the appearance in two Indo-European languages of, e.g., decasyllables or hendecasyllables, with a partially iambic rhythm, catalectic or acatalectic, enlarged or shortened, etc., points to a common origin of such a metrical pattern, is no more justified than the assertion of the common origins of the IE and Semitic plural, genitive, or subjunctive" and further that "as for the question of a common IE origin of Greek and Indic verse, it must remain open." Likewise, in a briefexposition published in 1976, Henry Hoenigswald reminds us (p.275) of methodological problems associated with the extension of comparative reconstruction from the hard linguistic data of lexical roots to meter. The comparative method as originally conceived, he maintains, was "uniformitarian":

In denying that some traits are inherently and characteristically innovative, the comparatists believe that their unreconstructed ancestor language is just another language, different from its descendants to be sure, but, typologically speaking, no more so than descendants may be from each other; whatever structural damage changes in time may cause is sure to be repaired by other changes. It must be admitted that the tenets of historical and prehistorical metrics are non-uniformitarian.^[7]

[5] Note also his statement (p. 61) that "l'épopée homérique est toute faite de formules que se transmettaient les poètes."

[6] See also O'Nolan 1969.

[7] He concludes: "It seems to be difficult to get away from the notion that in certain metrical histories (say, in that of the hexameter from Homer to Nonnos) the nature , and not just the accidental properties, of the verse changes (say, from 'loose' to 'rigid,' although one could in particular cases argue for something different, or even contradictory; or merely from 'simple' to 'complex'). Actually, Meillet's basis is consensus among the comparanda rather than triangulation from them, though his qualitative judgments were admirable" (Hoenigswald 1976, 275). See also the reviews of G. Nagy (1974) by Hoenigswald (1977) and by Brough (1977).

On balance, however, the pursuit of Indo-European meters has received more support than criticism, and it seems prudent to agree with M. West (1973^b , 161) that "the assumption that Indo-European prototypes underlie the metrical forms at least of Indian, Iranian, Greek, Slavic, and Celtic poetry is now respectable."

Roman Jakobson's work on the deseterac , the Serbo-Croatian epic decasyllable, to take the next chronological step from Meillet's Origines , is thorough and wide-ranging, although his textual sample could be more extensive.^[8] He covers all of the Slavic meters cognate to the deseterac and, after carefully considering the idiosyncrasies of each separate language tradition, derives a Common Slavic prototype antecedent to and generative of the various descendant types. Although the Serbo-Croatian epic line itself is more properly the subject of the latter part of this chapter, we may note in capsule form the congruity that Jakobson demonstrates between Meillet's Indo-European line and the Common Slavic reconstruction. Both have what may be termed regular syllabicity , a consistent number of syllables; anceps , indifferent quantity in the final syllable; right justification , a relative freedom from any pattern of quantities in the earlier part of the line and a correspondingly more regular succession of quantities near the end of the line; and caesura , an obligatory word-break at a constant or slightly variable position within the line. Through discovery and delineation of these regularities, Jakobson (1952, 66) relates the Yugoslav epic line and meters from other Slavic traditions to an ancient precursor, complementing the analysis of Meillet: "The testimony of the third witness to the foundations of the Indo-European verse may now join those of Greek and Vedic."^[9]

A number of possible historical derivations for the Homeric hexameter have been offered in the years following Jakobson's (1952, 64-65) championing of Bergk's ([1854] 1886) hypothesis of its development from an Aeolic paroemiac. Prominent among these is Gregory Nagy's account (1974), in which he describes ancient Greek and Indic correlations at some length, taking as a point of departure the cognate Homeric and Vedic formulas and srava^[*] (s) áksitam ("imperishable fame"). In speaking of his two subject meters and their Indo-European parent, he remarks (pp. 30, 36): "The verse is divided into an opening and a closing, which are marked by flexible and rigid rhythms respectively"; accordingly, "the comparative approach, in short, suggests that freedom in the rhythm of the opening is a feature inherited from

[8] Jakobson's body of texts included 783 lines performed by the Montenegrin guslar T. Vucic^[*] and certain unspecified samples from published texts and the Parry Collection. My comments on the heroic decasyllable later on in this chapter are based on computer analysis of more than five thousand lines from the local tradition of Stolac.

[9] Vigorita (1976, 209) concurs with Jakobson's analysis of the structure of the deseterac , concluding that of all the South Slavic meters "the epic decasyllable ... has perhaps the strongest claim to antiquity. It is related to the Greek paroemiac and the rare Vedic decasyllable."

the archetypal Indo-European poetic language." He sees the progressive restriction of freedom as a right-to-left process—"the lineal direction is from line-final to line-initial," the same directional movement posited by Jakobson and Meillet which I called "right justification."^[10] Nagy derives the hexameter from an Aeolic pherecratic with dactylic expansion; the pattern that he cites (p. 49) from Alkaios and a generalized schema of the hexameter are as follow:

pher3d		(16 syllables)
hexameter		(17 syllables)[11]

To accomplish the transformation he stipulates (1) the optional replacement of the dactyl by the spondee, that is, of ; and (2) the replacement of , which in turn is optionally replaced by the dactyl,. One particularly attractive aspect of Nagy's theory is how thoroughly it explains the caesuras and diaeresis, the major word-breaks in the hexameter.^[12]

Berkley Peabody also sifts the evidence from ancient traditions in the Indo-European family, specifically Vedic Sanskrit and Iranian (the Avesta ), and comes up with an alternate derivation for the Homeric line (1975, 47-48):

The hexameter seems to be a hybrid primary combination that resulted from the fusion of dimeter and trimeter verse forms. The fusion of lines into one integrated verse form correlates with the tendencies noted in both the Avestan and Vedic traditions.... The two-against-three structure of the Greek fusion seems to have produced such a tight formal continuum that the verse tended in time to become functionally transformed into a single diploid line form (a single line—not verse—with twice the usual number of parts).

Although the entire demonstration is much too complex to be conveniently summarized here, we may note in passing that, as with Nagy's hypothesis,

[10] Cf. O'Nolan 1969, 17: "The hexameter cannot have sprung fully fledged into existence but is likely to have developed (as Watkins shows for the paroemiac) from a prototype which has a fixed tail-end and a free fore-part. One may imagine a sort of creeping paralysis of versification starting at line-end." Compare also Austerlitz (1985, 46) on a similar tendency in the Finnish (and non-Indo-European) Kalevala .

[12] As we shall see later on, Nagy argues that phraseology is diachronically precedent to meter, and so he understands (1974, 61) the word-break pattern in the hexameter as the product of an interaction between formula and meter: "the pher pattern of the archetypal epic hexameter had accommodated the formulas of shorter verses, and it is these formulas that eventually led to the attested caesura- and diaeresis-system of the larger verse."

Peabody's theory accounts well for the major word-breaks in the line.^[13] What is more, both formulations manifest a deep concern with diachronic roots and with establishing the hexameter as a metrical shape continuously in the process of reshaping itself.

Martin West's synthetic articles (1973a,b) serve as an overview of the subject of Indo-European meter. Following the original tenets of comparative reconstruction, which require evidence from at least three language families for valid triangulation back to Indo-European Ur-forms, he adduces data from the Vedas, the Gathas of the Avesta , and the Aeolic meters of ancient Greek. His simple series of prototypes, generalized and compounded later on, are two:

Acatalectic	Catalectic

where x is a syllable of variable or indifferent quantity; 5 is a protasis of five syllables; and (|) is a word-break after the protasis. He conceives of the hexameter as the possible product of "a pherecratean and expanded reizianum, , welded together and regularized in rhythm throughout" (1973b, 169, n. 10), or, in a manner closer to that of Nagy, as an "expanded pherecratean" (1973a, 186).

Besides the Aeolic meters and ultimately the hexameter, West also tests his prototypes against meters in Slavic (Serbo-Croatian, Czech, Polish, and Russian), Celtic (Old Irish), Italic (Latin), Germanic (Old English and Old Norse), Hittite, Lydian, and Tocharian, with various results. In reference to those meters under consideration in the present study, I find his comments on the deseterac , parallel to those of Jakobson, especially suggestive. He explains the decasyllable as a four-syllable protasis of variable quantities followed by the same version of the Indo-European prototype that also generates the Greek reizianum and Vedic colon of the tristubh line. In addition, he sees the quantitative regulation of the second portion of the line only, a phenomenon I have termed "right justification," as an argument for the extreme archaism of the Serbo-Croatian decasyllable. West's comments on the Germanic "standard line" are necessarily sketchier and more hypothetical, given the modulation from a quantitative Indo-European line to the stress-

[13] "It appears, however, that if the central caesura is seen as the relict of a joint between two earlier line forms, each of which already possessed its caesura, not only are all three caesuras accounted for, but the fact that the central caesura moves in a way different from that of the caesuras on either side is also explained" (Peabody 1975, 51; see further pp. 30-65).

based Germanic vernacular meters that were its eventual descendants. Nonetheless, he offers a plausible explanation of the very developments that obscure the genetic picture.^[14]

To sum up, most Indo-European metrists agree on an Ur-meter that was (1) quantitative, (2) of consistent syllabic definition, (3) relatively free in the distribution of quantities early in the line while relatively rigid in the pattern of quantities later in the line (that is, "right-justified"), and (4) marked by a regularly placed caesura, or word-break, within the line. Through combination and recombination of prototypical metrical forms, through linguistic adaptations resulting from the development of language families and of individual languages within those families, and through the myriad interactions between phraseology and meter, singular verse forms became differentiated and gained identities of their own. It is essential to keep in mind, especially as we continue the discussion by consulting each language tradition separately, that available evidence indicates that the various meters which have emerged and now stand as entities in the extant texts both stem ultimately from one or a series of Indo-European prototypes of unitary and extremely ancient provenance and also exist as integral and dynamic instruments of individual poetries. In linguistic terminology, then, these meters thus have both a diachronic history and a synchronic identity or definition.

It is time now to consider the place and importance of the Indo-European question in these studies as a whole; in doing so it will be convenient to examine each meter separately and then conclude with some more general remarks. First, the hexameter's relation to Indo-European prototypes is far from absolutely clear. Scholars seem to be able to agree on the rough outlines of the Homeric line's prehistory, so that we can avoid having to posit with Meillet (1923, 62-63) an external "Aegean" origin for the hexameter, but there is no consensus on the particulars of that history. More immediately, one can in the light of competing hypotheses about its evolution easily lose sight of the fact that the hexameter in its present form has to have been a

relatively late development. Peabody (1975, 65) makes this case, buttressed with both historical and synchronic arguments:

While the hexameter is a pattern universal in the epos, it seems not to have been a static, rigid form. Its shape was in slow but constant redefinition. At the same time that it became crystallized forever (apparently through writing and the analytic establishment of canonic norms), it seems to have been emerging from a period of relatively active change.^[15]

As far as the Homeric line as a structure in itself is concerned, then, we must temper the virtual certainty of an Indo-European ancestry with an admission of uncertainty about the particulars of the lineage and with a general conception of the line's evolution as an ongoing process that culminated rather late in the line of Homer and Hesiod. If genetic survivals are to enter the picture, they must enter subject to these conditions.^[16]

The Serbo-Croatian epic decasyllable, according to all commentators, is surprisingly archaic in structure given that it is still the verse form of the heroic songs sung by guslari in this century. Jakobson (1952, 64-66) posits a "gnomic-epic decasyllable" as the Indo-European prototype, finding parallels to the Slavic in Greek, Vedic, Iranian, and Lithuanian (see note 9 above). M. West (1973^b , 171-72), arguing that the Slavic situation is "parallel with the Vedic rather than the Greek situation,"^[17] proposes another derivation: he compares the opening four-syllable colon of the deseterac to the Indo-European protasis of variable quantities, and the second colon of six syllables to the prototype that also resulted in the Greek Aeolic reizianum:

	1	2	3	4	5	6	7	8	9	10
General Structure	s	s	s	s	s	s	s	s	s	s
Quantitative Series	x	x	x	x	x	x			__	x

[15] Thus M. West (1973 , 187-88), for example, notes: "We have seen that dactylic verse was a South Mycenaean development dating probably from the second half of the [second] millennium, while the stereotyped stichic hexameter represents a further development in the Ionian branch of the tradition, perhaps late Mycenaean, perhaps post-migration"; as for epic poetry, "by 1100 it may have existed in south Greece in something like hexameters, though I imagine that they might be rather looser in technique than what we are used to." Cf. Hoenigswald 1977, 82-83.

[16] I postpone detailed discussion of the interrelationship between formula and meter until later in this chapter and in the individual chapters (4-6) on phraseology in each tradition. It is enough to say at this point that the possible diachronic priority of phraseology over meter, as posited by Nagy, does not confute but only confirms the intimate dependence of metrical shape and structure on natural-language characteristics. Thus the principle of tradition-dependence, introduced in chapters 1 and 2 at the level of general methodology and textual documents, is confirmed at the level of meter and phraseology.

[17] The Common Slavic form entailing "quantitative regulation only in the last four syllables" (M. West 1973 , 173) leads him to this conclusion.

	1	2	3	4	5	6	7	8	9	10
Proto-Slavic (catalectic form)	x	x	x	x	x	x			__	__
IE Prototype (before expansion)	x	x	x	x	__		__	__

What presents itself again, in short, is uncertainty in the particulars of derivation, although the existence of some relationship between the decasyllable and an Indo-European precursor is almost beyond question. And while the apparent archaism of the meter seems to relieve us of the problem of late structural fixation that obscures the evolution of the ancient Greek hexameter, we have still to consider the varying assignment of pitch and stress in the various Slavic vernaculars. In sum, the situation as a whole is similar to that of the hexameter: uncertainty about particulars demands caution in composing the prehistory of the deseterac (cf. Petrovic^[*] 1969, 1974). Nevertheless, it is a verse form that can be well studied analytically in the extant texts, particularly since the textual record is so much more extensive than those of the dead-language traditions.

The situation in Old English is at once more complicated conceptually and simpler from a methodological point of view. Since the emergence of stress as the primary metrical determinant has obscured the quantitative roots of Germanic cola, the possibility of confident theorizing about the genetic prehistory of the alliterative line does not present itself. In this case our problem is not an embarrassment of riches but rather an impoverishment of clear, plausible hypotheses. If we suppose the Old English half-line to have been the earlier verse-unit, one that at some later time was doubled and then knit together by alliteration, and if we further understand the stress-accent typical of the Germanic vernaculars to have replaced the long-short quantitative sequence as the distinctive metrical feature of the verse, we can see with West and Lehmann how the relative number of short unstressed syllables in the line could increase. With the modulation from quantity to stress prominence would come a corresponding modulation from dependence on to relative independence of syllabic count. This tendency toward a larger proportion of unstressed syllables must, however, have been offset to a degree in Old English by the maintenance of traditional patterns of phraseology from earlier Germanic tradition.^[18] At any rate, by the time of the first recorded Old English texts, the modulation to a stress-based meter was complete, and the genetic prehistory of the line—not to mention the relative chronology of fixation of the line in its present form—was largely lost in the transition. For this reason,

[18] "Tradition, therefore, played a greater role in the maintenance of the alliterative line in England than in any of the West Germanic dialects. In early Old English, the strict Germanic line was maintained through retention of an old poetic vocabulary and syntax" (Lehmann [1956] 1971, 102).

and also because our surviving sample of texts from the period is so small,^[19] I advocate special attention to a descriptive account of the meter of Beowulf .

To these specific reasons for concentrating on a descriptive comparison of the three meters may be added some more general considerations. First, the idea of tradition-dependence, developed in theory in the first two chapters, would seem to demand attention at the level both of prosody and of the formula and more complex levels of tradition. As languages and meters evolve away from their common Indo-European prototypes and continuous, variable interaction takes place between phraseology and meter, tradition-dependent systems come more and more to deserve characterization as entities in themselves. Geography, chronology, and the myriad other factors that contribute to their separate developments also generate increasing numbers of peculiarities, such that by the time the ancient and medieval traditions are committed to manuscript, the group of originally similar members has, as we have seen, grown quite dissimilar. The traditions that preserve and evolve phraseology and meter are most certainly of great age and persistence, as Lord, Georges Dumézil, and others have shown,^[20] but this prehistory should not obscure the fact that at a later date different traditions take very different forms, evolve in various ways, preserve and discard elements at all levels, and in general behave idiosyncratically. And we should be aware that the fundamental level of prosody—which must be more immediately reactive to linguistic changes than must narrative units such as typical scenes and story-patterns^[21] —is perhaps the most tradition-dependent of all structures.

A second observation on the Indo-European question and its pertinence to this comparative inquiry has to do with the real nature of a descriptive account of the three meters. In focusing primarily on the verse structure of the texts as we find them, are we condemned to the thinnest synchronic slice of the tradition? Will we, in other words, ignore important diachronic roots and settle for a superficial view of the subject? Part of the problem, I believe, lies in the very form of such questions, for the history of the hexameter is in a real sense the hexameter itself: if we pay careful attention to its synchronic structure, noting its rules and tendencies and tracking its tangible variants in pursuit of useful generalizations, we are at the same time studying at least

[19] The entire Old English poetic corpus contains only slightly more than thirty thousand lines, about 10 percent of which are epic. As indicated in chapter 1, the comparisons made in these studies must for methodological reasons be limited to a single genre, that of epic. The Anglo-Saxon sample is thus only about one-tenth the size of the ancient Greek epic corpus, and both canons are dwarfed by the available Serbo-Croatian material in the Parry Collection and elsewhere.

[20] See, e.g., Lord 1960, 158-97; Dumézil 1973.

[21] In his Foreword to Peabody 1975 (p. xii), Lord advocates a hybrid approach: "The study of oral and traditional literature must of necessity be both diachronic and synchronic." I would add that any study is and must be to some degree both, that the kind of study depends on an emphasis or focus rather than a total exclusion of one or the other approach.

part of the story of how it came to be. By combining what we know about Indo-European metrics with what is ascertainable from a synchronic study, we can read the history and present reality of a metrical form—for the structures confronted in the frozen moment of the text are nothing less than the products of a diachronic process, a process that has left telltale vestiges of earlier stages in verse evolution. In this way our descriptive accounts of the hexameter, deseterac , and alliterative line can mesh with what has been said about Indo-European prosody and reach beyond the synchronic surface of the texts to their diachronic roots.

Prosody and Prosodies

In borrowing W. Sidney Allen's phrase (1966) for the heading of this section, I mean to indicate an emphasis more general than his. As Allen and others have shown, the notion of what is properly designated by the term prosody has varied considerably from ancient times to the present. In this chapter and throughout these studies, I use the term in its widest possible application to refer to all of the "elements and structures involved in the rhythmic or dynamic aspect of speech, and the study of these elements and structures as they occur... in the compositions of the literary arts"; this so-called literary prosody thus "studies the rhythmic structure of prose and verse, not as exemplifying linguistic norms but as functioning ... as an aspect of poetic form" (La Drière 1974, 669). We may therefore list among our possible interests not only meter per se, but also alliteration, rhyme, assonance and consonance, stress, tone, hiatus, juncture and elision, and, in general, sound patterns of any sort. There will also be reason to touch on allied linguistic phenomena, such as syntax and morphology, as they relate to prosodic features, although most of our discussion of these last two areas belongs in the chapters on formulaic structure.

Within this large field of prosodic possibilities the chief emphasis will continue to be on those particular properties of the verse that are most important to the structure and deployment of phraseology and therefore to traditional narrative design. The first narrowing of focus, then, is from prosody as the set of all possible metrical and euphonic factors to traditional prosody , the more restricted set of factors involved in traditional form and dynamics. Additionally, this second and more refined set of characteristics will vary considerably from one poetry to another; in fact, there is evidence within the three traditions treated in these studies that prosody is genre-dependent as well as tradition-dependent.^[22] In Old English, for instance, alliteration

[22] Besides the more obvious differences among the various meters of different genres within a single given tradition (see, e.g., Jakobson's discussion of Serbo-Croatian laments, or tuzbalice [1952, 33-35], or Vigorita's comments on the Indo-European origins of "short lines" in lament and epic [1976, 208-9]), I would point to the example of prosodic structure in oral charms, or bajanje , as contrasted to epic within the Serbo-Croatian tradition. Rhyme, both end-line and in-line, and consonance are on balance much more pervasive in the largely octosyllabic charms. Cf. Lord 1960, 54-58; Lord 1956, 1981; Kerewsky Halpern and Foley 1978; Foley 1980a.

between paired half-lines is not a desideratum but a requirement: without agreement of these initial sounds bridging the gap, one does not have a viable metrical line. (For example, "B eowulf mapelode, / b earn Ecgpeowes" ["Beowulf spoke, son of Ecgtheow"; Beowulf 529 et seq.: alliteration in b ]; and "G rendel g ongan, / G odes yrre bær" ["Grendel going, he bore God's wrath"; Beowulf 7 11: alliteration in g ].)

But while absence of this prosodic feature in Beowulf calls for emendation of the manuscript text, its absence in a Serbo-Croatian or ancient Greek epic line is no violation at all, not even (necessarily) of euphonic propriety. In the latter two traditions, alliteration may constitute a sound-pattern,^[23] but it is not in any way required. In contrast, rhyme of the end of the first with the end of the second colon, or leonine rhyme, is a fairly common phraseological feature in Yugoslav epic; perhaps more importantly, lines that manifest this end-colon (as opposed to end-line) rhyme only rarely show variation from one occurrence to the next.^[24] (Examples of leonine rhyme in the deseterac include, for instance, "I danica da pomoli lice " ["And the morning star shows its face"] and "U becara^[*] nema hizmecara^[*] " ["A bachelor has no helpmate"].) These facts allow us to assign a compositional status to this euphonic feature in the deseterac ; although obviously not a requirement, leonine rhyme is for our purposes a significant prosodic element because it plays a part in formula-shaping and, apparently, in maintaining formulaic shape over time. Prosodic features other than meter, or at least those usually classed as euphonic, seem less important in Greek, but, as Parry (1928a, in 1971, 68-74) showed long ago and Nagler (1974, 1-63) has illustrated more recently in some detail, agreement in sound provides a basis for analogy in Homer. Nagler gives many examples of what he calls "phonemic corresponsions," among which we may cite the following six, all of which occur in the adonean clausula (or final two metra) of the hexameter (1974, 8):

[23] For examples in Serbo-Croatian, see the articles cited in note 22 above; further illustration is available in chapter 5. For ancient Greek, see Packard 1974; Stanford 1969.

[24] As Jakobson emphasizes (1952, 31) "It is noteworthy that as soon as rhymes begin to appear in the normally unrhymed Serbo-Croatian epic decasyllable (in literary productions of the eighteenth century), the quantitative close weakens considerably." See also Stankiewicz 1973.

Both of the example lines that follow occur too often and too widely in Serbo-Croatian epic tradition to be referenced at any single spot. I have drawn them in this instance from the Zenidba Becirbega^[*]Mustajbegova (The Wedding of Mustajbey's Son Becirbeg^[*] , Parry text no. 6699 ) of Halil Bajgoric^[*] , a guslar recorded by Parry and Lord in the Stolac area in 1933-35 and 1950-51. The first example illustrates the partial rhyme often induced by declension of a formula-part or, as here, produced by leonine near-rhyme of nouns from two different declensions (danica = nominative singular diminutive, feminine; lice = accusative singular, neuter).

These acoustically related elements—pioni dêmôi ("[amid] the flourishing populace"), together with the others—argue, Nagler feels, for phonemic relationships that operate outside of the usual formulaic context. Again we encounter a prosaic feature that contributes to traditional structure.

The instances cited so far arc few, and we shall need to consider many more at length below. But perhaps enough has been said to indicate that when speaking of "prosody" and "prosodies," two avenues of differentiation must be kept in mind. The fine is the acknowledged bias of these studies toward analysis of those factors which affect traditional structures. The second has to do with the varying set of pertinent factors for a given individual tradition—with, in short, the tradition-dependence of prosody.

Prosody and Formulaic Structure: Their Interrelationship

Implicit in my opening discussion of methodology was an assumed relationship between prosody and formulaic phraseology and, by extension, a secondary relationship between prosody and larger narrative units, which, for all their apparent structure as action-patterns or typical scenes, depend finally on phraseology for their expression. In the early stages of development of the oral-formulaic theory, such an assumption would have encountered no resistance whatever, since prosody was a fundamental part of the concept of formula: thus Parry's original definition, as "an expression regularly used, under the same tactical conditions , to express an essential idea" (1928a, in 1971, 13; my emphasis).^[25] Without those "same tactical conditions," the formula could not exist; prosody was a crucial and limiting factor in the process of definition. This procedure amounts to claiming that formulaic diction is, to use a favorite philological term, metri causa , that it arises from the constraints of meter and, the argument would continue, is retained because it fits the meter. In fact, this was precisely the direction Parry took in illustrating the thrift of Homer's diction. Observing that almost always only one noun-epithet formula was available to name a given god or hero in a given metrical segment of the line, he maintained that this characteristic was a sign of a traditional diction, a sign that the phraseology was itself a dynamic poetic entity epitomized by generations of individual singers.^[26]

[25] Later, in "Studies I" (1930, in 1971, 272), Parry defined the formula in rightly different terms as "a group of words which is regularly employed under the same metrical conditions to express a given essential idea" (italics deleted).

[26] In discussing the "formulaic system," which he defines (1930, in 1971, 275) as "a group of phrases which have the same metrical value and which are enough alike in thought and words to leave no doubt that the poet who used them knew them not only as single formulas, but also as formulas of a certain type," Parry makes the following observations: "The length of a system consists very obviously in the number of formulas which make it up. The thrift of a system lies in the degree in which it is free of phrases which, having the same metrical value and expressing the same idea, could replace one another" (p. 276; emphasis added). Length may well be a reasonable measure of a formulaic system in traditions other than ancient Greek, but the case is not nearly so clear for thrift; see Fry 1968c and Foley 1981d. See further Parry 1932, in 1971, 325-64.

This fundamental aperçu was, like so many of Parry's theses, a uniquely rigorous and characteristically creative extension of the preliminary work of others, in the present case of Ellendt, Düntzer, and other classical linguists of the late nineteenth century.^[27] Parry was proceeding, in other words, from a belief in the shaping and determining function of meter, and he would maintain this view throughout his published and unpublished writings on Homer and Serbo-Croatian epic. Some scholars who have followed Parry, upset with the supposed mechanistic operation and suppression of aesthetic choice that they see in metri causa , have tended to "soften" the prosodic requirements originally a part of the formula, or at least to redefine or investigate the flexibility of those requirements.^[28] But for all except the most subjective of critics,^[29] meter has remained an integral part of the formula's definition and of its very identity.

The fact of such a relationship, whatever its exact nature may be, is important. For whether we choose a formula-meter model as reductive as the "Lego-set" or one as complex as the newer generative or formalist-traditional theories (e.g., Devine and Stephens 1984), we are dealing most basically with language indissolubly allied with prosody. Laying aside for the present the many fascinating questions that could be asked about meters other than the Homeric hexameter and concentrating on the ancient Greek texts out of which formulaic theory was born, we must be struck by the overwhelming consensus (not to say observable fact) that Homer's traditional words are metrically defined. That is, rather than being merely lexical, phonological, morphemic, and syntactic entities, they are metrical or prosodic entities as well, and that prosodic character emanates not from lexical features but from verse structure.^[30] Moreover, we would do well to remember that this metrical dimension also proclaims unambiguously the identity of these words as traditional sound, as opposed to the printed transcriptions we have trained ourselves to interpret back toward their original form. Perhaps this is why Yugoslav guslari , when

[27] See A. Parry 1971, xix-xx; Foley 1988: chap. 1.

[28] I borrow the "hard" and "soft" designations from Rosenmeyer 1965.

[29] Nagler (1974, 18) would posit "a preverbal Gestalt generating a family of allomorphs" as a model for Homeric formulaic diction. Cf. Ingalls 1972, who illustrates the colonic form of Nagler's allomorphs.

[30] Thus Eugene O'Neill, Jr.'s, concerns about "The Localization of Metrical Word-Types in the Greek Hexameter" (1942). See further the section on the "inner metric" of the hexameter below, and also chapter 4.

asked what a "word" (rec ) in an epic song is, respond with a couplet or a single ten-syllable poetic line rather than with what we might expect—the dictionary denotation of word (see chapter 2). For them too, it would seem, a word is no word unless it is a prosodic word .

This line of inquiry holds out the promise of productivity, and I shall return to it at the appropriate time. For the moment, however, let us consider a relatively recent development which cannot help but cast some doubt on the doctrine of metri causa in the generation of the formula. In his 1974 monograph Comparative Studies in Greek and Indic Meter , Nagy proposes three bold new hypotheses, all of them interrelated. First, he traces the Homeric hexameter to a pherecratic^3d precursor, as illustrated earlier. Second, he dismisses the usual chronology of Greek lyric poetry growing out of epic and argues for independent Indo-European roots of lyric. But most important for the present discussion is Nagy's third and overarching hypothesis, which unites the first two. By taking a diachronic or evolutionary view of Greek meter and collecting comparative evidence from Vedic and Homeric material, he formulates a history of development which seems to reverse the accepted relationship between formula and meter (p. 145):

At first, the reasoning goes, traditional phraseology simply contains built-in rhythms. Later, the factor of tradition leads to the preference of phrases with some rhythms over phrases with other rhythms. Still later, the preferred rhythms have their own dynamics and become regulators of any incoming non-traditional phraseology. Recent metrical developments may even obliterate aspects of the selfsame traditional phraseology that had engendered them, if these aspects no longer match the meter.

Far from adopting the consensus correlation, then, Nagy posits that "traditional phraseology had generated meter rather than vice versa." Although this is at best a telegraphic restatement of his position, the essentials are clear enough: the new theory threatens to overturn the mechanical-generation theory of Homeric formulaic diction by placing phraseology in the diachronically determinate position.^[31]

Without questioning either Nagy's methods or his evidence,^[32] which provide imaginative and far-reaching insights into nagging problems in a number of fields, I would make one simple point about his results, for his

[31] In "Formula and Meter" (1976, 251), Nagy describes the process in this way: "Predictable patterns of rhythm emerge from favorite traditional phrases with favorite rhythms; the eventual regulation of these patterns, combined with regulation of the syllable-count in the traditional phrases, constitutes the essentials of what we know as meter . Granted, meter can develop a synchronic system of its own, regulating any incoming phraseology; nevertheless, its origins are from traditional phraseology."

[32] But see Jaan Puhvel's response (1976) to Nagy's "Formula and Meter," and the citations in note 7 above.

proposal that diachronically formula generates meter^[33] may well seem to qualify the present approach to comparative prosody and the tradition-dependence of the formula. As time goes on, what was originally a phraseology-based interaction between formula and meter becomes a meter-based interaction. Another way of explaining the same development is to observe that, over time, the diachronic generation of meter gives way to the synchronic generation of formula. Because my major focus in these studies must remain on the texts that have survived and are the products of the development, I shall be viewing the process from the chronologically later end of the shift and shall thus look to meter as the prosodic "partner" of phraseology in the Homeric and other oral epic poems.^[34]

To this distinction may be added an observation that will indicate further the indissoluble link between formula and meter and the necessity for a comparative study of prosody to begin by recognizing that link. If Nagy's description of the earliest stage of formula-meter interaction is correct, then we have in the generation of meter from phraseology perhaps the most direct proof possible of the influence of natural-language characteristics on meter. What is more, we have evidence of a tradition-dependent meter based on the singularity of a given language from a very early time, since a meter must, we can suppose, be as singular as the language from which it arose. Even when, in the latter stages of the process, prosody came to have a life of its own and was therefore able to accept or reject combinations of elements from its parent language, it was originally that very language which gave it birth and to which it still owed its tradition-dependent identity. With this sort of direct link between a formulaic phraseology and an incipient prosody, there can be no question of persisting in philological reductionism: meters must be as different and idiosyncratic as the languages that spawned them.

In the remainder of this chapter I provide short accounts of each prosody assembled according to the principles explained above.

The Homeric Hexameter

Outer Metric

Let us turn first to a consideration of what O'Neill (1942) has conveniently designated the "outer metric," that is, the foot-based or podic structure of

[33] The emphasis is Nagy's own—see 1974, 140-49; 1976, 251-57.

[34] A second perspective on the diachronic process should also be briefly mentioned. If Nagy's formulation is correct, and if we someday have enough hard proof of the earliest stage in the process to deem a significant part of Homeric formulaic diction pre- or proto-metrical, then we will still have to deal with the metrical dimension of formulas. For in this case the metrical aspect will become even more important to an understanding of traditional structure than it is now: the foregrounding of certain patterns as proto-metrical will indicate, at least crudely, the relative age of various elements in the diction and the calibration of the hexameter over time. Compare the linguistic archaeology practiced by Hoekstra (1964) and Janko (1982).

the hexameter.^[35] After this standard description, we shall address the more complex questions involved with the "inner metric," the internal structure of the line.

The hexameter may be schematized as follows:

	(Od. 9.287)
	(Od. 1.107)

According to conventional notation, the line is here shown to be composed of six "feet," or metra, numbered 1 through 6. Each of the first five metra may be eider dactylic () or spondaic (), although a spondee in position 5 is rare. The sixth metron is always disyllabic, its second element being regarded as long by position if not by nature (brevis in logo ). The last two metra, the fifth almost always dactylic and the sixth disyllabic and spondaic by necessity, thus sound the closing cadence to the line, a cadence that, as explained above, some metrists have found echoed in other Indo-European meters. From this perspective the hexameter is a so-called quantitative meter, depending for its shape on the distribution of relatively long and short syllables over six podic units. In this seine it is not primarily stress-emphasis but rather phonologically determined quantity that yields the metron structure of the hexameter.^[36]

[35] For a survey of modern metricians to 1905, see Porter 1951, 3-8; on colometry per se, see Barnes 1987.

Quantities are distributed in a number of ways, the variability represented by the thirty-two possible types arising (from the point of view of outer metric only) from dactylic-spondaic substitution in the five substitutable metra:^[37]

Theoretically, then, the number of syllables in a Homeric line can vary from twelve (wholly spondaic) to seventeen (five dactyls plus the brevis in longo or "final anceps"). Table 3 documents the syllabic distribution in the nearly twenty-eight thousand lines of the Iliad and Odyssey .^[38] As these figures make evident, while all six possible syllabic categories are occupied,^[39] the great majority of lines are either fifteen or sixteen syllables in length (over 72 percent for both poems)—that is, almost three-quarters of the Iliad and Odyssey is composed of lines consisting of either three or four dactyls and, correspondingly, either three or two spondees, counting metron 6 as spondaic. This favored con-

[37] As noted above, the sixth and final metron is invariably disyllabic, with the last syllable understood as long by convention.

[38] I derive statistics on outer metric from LaRoche 1898 and from Jones and Gray 1972.

[39] Examples of all six possible patterns exist, but the only statistically important types are those with two to five dactyls, or fourteen to seventeen syllables.

figuration is quantitatively the longum-breve rhythm of the hexameter, an illustration of what I call the "syllabicity" of the line. From this point of view, the hexameter, while not absolutely syllabic (as are for example many Latin and Romance meters), demonstrates a relatively focused syllabicity, taking a fifteen-or sixteen-syllable shape 72 percent of the time. If we add in the wholly dactylic seventeen-syllable possibility, the total comes to over 91 percent and the syllabic focus becomes even more apparent.

This statistical profile may be more finely articulated by determining how the dactyls are distributed. In other words, in those hexameter combinations of dactyls and spondees which can take more than a single form, we may profitably inquire about the absolute position of each metron-type. Within the fourteen-, fifteen-, and sixteen-syllable lines, the placement of metra is as shown in table 4.^[40] Avoidance of the fifth-metron spondee—very probably the tradition's effort to preserve the Indo-European closing rhythm and, synchronically, the line-ending cadence for the hexameter—may be seen at the root of many of these statistical phenomena. In the sixteen-syllable line, for example, the spondee is conspicuously unusual in metron 5, and this pattern continues through the fifteen- and fourteen-syllable lines. The chart of metron configurations in the hexameter above also shows the spondee to be relatively rare in metron 3, the predominant locus of the mid-line break; again the pattern extends through the shorter lines. Exactly what the significance might be of the third and fifth metrons' strong tendency toward dactylic shape will be developed in the section below on inner metric. For the moment, let us simply notice that this tendency occurs at points of structural closure, marking half- and whole-line segmentation.

[40] On the comparison of the Iliad and the Odyssey , Jones and Gray (1972, 208) remark: "As far, then, as the outer metric is concerned the similarities far outweigh the differences." The Iliadic Catalog of Ships, filled with proper names and other spondaic forms, is apparently an exception to this general rule; see Rudberg 1972, 20-21.

Another perspective on the dactylic-spondaic texture of the hexameter may be gained by noting the frequency of occurrence of various metron configurations. Table 5 documents the ten most common patterns. It is worthy of mention that these configurations, the ten most frequently occurring of the

thirty-two possibilities, account for a full 87 to 88 percent of the Iliad and Odyssey . In fact, the first three alone make up almost half of all lines in the two epics.

These observations on the podic rhythm of the metra provide an initial characterization of structure in the hexameter. It is clear even from these few remarks that the line is not merely a six-part symmetrical unit that repeats over and over again; it should be equally evident that conceiving of a simplistic verse form cannot help but obscure any conceptions of formulaic structure based on it. The possibilities inherent in the hexameter are many, and all are at one time or another realized; from a statistical standpoint, however, relatively few of these multiple possibilities account for most of the actual lines in the Homeric corpus. Speaking generally and from the perspective of outer metric only, we can make the following points about the hexameter: (1) over 90 percent of all lines arc composed of between fifteen and seventeen syllables, over 70 percent of either fifteen or sixteen syllables; (2) spondaic substitution is common in metra 1, 2, and (somewhat less frequently) 4, but much rarer in metron 3 and positively avoided in metron 5; (3) the line ends typically with a "quantitative close," consisting of a dactylic fifth metron and the spondaic anceps. A preliminary view of the hexameter must, in short, emphasize its focused syllabicity and tendency toward certain patterns of dactylic-spondaic alternation. These patterns, with the exception of metron 6, favor spondaic substitution in the opening rather than the closure of half-lines or hemistichs. This tendency is an aspect of right justification, in this case of dactylic feet within half-lines. As we move on to a discussion of inner metric, we would do well to bear in mind these features of the hexameter's outer metric: focused syllabicity, favored patterns of dactylic-spondaic substitution, and right justification.

Inner Metric

In this section I shall treat the internal structure of the line, for it is the rhythm of the inner versus the outer metric that gives the hexameter its characteristic texture. Indeed, as Milman Parry discovered, the internal structure is that dimension of the line which bounds or encapsulates the formula and which constitutes the "same metrical conditions" that Parry cited as a necessary condition for formulaic diction.

The first step in this description is to identify the three principal caesuras,^[41] or prosodic breaks, in the hexameter. The diagram below, showing the inner metric of the hexameter, indicates these breaks in terms of multiples of three basic positions, A (1-4), B (1 and 2, the masculine [or penthemimeral] and feminine [or trochaic] caesuras), and C (1 and 2, the hepthemimeral caesura and bucolic diaeresis).

[41] The term caesura is employed here in a strictly metrical sense without any implication of a pause between elements. On ancient theories of the metrical caesura, see esp. Bassett 1919.

I take this diagram and the theory it summarizes from Hermann Fränkel's "Der homerische und der kallimachische Hexameter,"^[42] in which the author goes on to posit cola , those sections of the line delimited by the caesura system, as the basic constituents of the hexameter. He concentrates not on word-breaks per se, that is, but rather on the material that they enclose—the colon-words in the line.^[43]

As one might expect, total agreement does not exist among all scholars about the location or meaning of caesura boundaries. Howard Porter (1951) would limit the A break to Fränkel's A3 and A4 and move the C1 to a position after the opening longum of the fifth metron:

While this formulation reduces the number of caesuras from eight to a seemingly more workable six, it sacrifices: a full coverage of colon-types, especially in respect to the A and C boundaries; in the case of Porter's reassignment of C1, concurrence with actual observed data and comparative evidence;^[44] and a more general descriptive and analytical adequacy, as will be seen below. But if we cannot entirely harmonize Porter's system with observed fact, some of his generalizations about the nature of cola are apposite. For example, he comments (p. 17):

Positively the colon is an exacted sequence of syllables produced by a brief rhythmic impulse. Four, rarely three, such sequences of syllables constitute the complex unit of the line. They vary in length from 4 to 8 morae. Each colon is usually marked off by word-ends. Any word-end can sere this function. In the hexameter a colon is frequently a short unit of meaning but need not be.

[42] In H. Fränkel 1955, 100-156. Fränkel's original exposition of this material was in "Der kallimachische und der homerische Hexameter" (1926). For more recent remarks, including comments on G. S. Kirk's criticisms of his original proposals, see Fränkel 1968, 6-19.

[43] As a more recent investigator, Berkley Peabody, has put it (1975, 68), "cola properly should be labeled, not caesuras; for, to take a Parmenidean position, caesuras are without substance or meaning in themselves."

[44] As Peabody (1975, 348 n .4) points out; see also Ingalls 1970, 5.

These remarks help to clarify the definition of the colon, with the only qualification being adjustment of the morae, or counts, per colon, depending on the placement and number of caesura-positions.

Another scholar who disagrees with Fränkel's system of breaks, and even with the four-colon structure that they segment, is Geoffrey S. Kirk (1966).^[45] To begin with, he seconds Porter's exclusion of Fränkel's A1 and A2 positions but not his relocation of C1. In fact, Kirk (p. 82) rejects the underlying principle behind binary positions: "Alternatively, of course, we can say that there is no 'alternate' to the bucolic caesura, and that in the nearly 40% of Homeric verses which do not have that caesura the latter part of the verse does not in practice normally fall into two word-groups." He then proceeds to consideration of Porter's data, attempting to demonstrate that the four-colon theory is statistically inadequate, and later to his own analysis of a sample passage from the Iliad for the same purpose. Kirk argues further that the apparent A and C caesuras are the result of "word-length availability" in the Greek language rather than metrical or sense grouping. To summarize, he would explain the hexameter as

a complex of causes, some obvious and others less so: the B caesura is a structural division of the verse primarily designed to integrate it and prevent it from falling into two equal parts; the C caesura tends to introduce a distinct verse-end sequence; the tendency to caesura around the middle of the first 'half' of the verse is due primarily to the average lengths of Greek words available in the poetical vocabulary.... The inhibitions on word-end at 31/2 and 71/2  and  [Meyer's and Hermann's Bridges, respectively] arc caused by the desire to avoid any strong possibility of three successive trochaic cuts, that on 4 [between the second and third metra] being due to the desire to avoid a monosyllabic ending, especially after a heavy word, to a major part of the verse. (p. 103)

In addition to the problems of invoking modern criteria and tautologically assuming that Homeric words—which are, after all, embedded in Homeric meter—should serve as unambiguous evidence for available word-length (both of which problems Kirk admits, pp. 103-4), there arc other logical flaws in his formulation. The first and most basic is his tacit contention that the half-line break, or B caesura, should be the standard against which we measure the "inadequacies" of the A and C breaks. But if all breaks were of the same frequency and occurred without variation in the same positions, the hexameter would be a much less fluid and subtle instrument than it most obviously is. To deny a four-colon structure on the basis of Kirk's disclaimer is to require a meter to become a pattern without flexibility, a hardened and stylized set of stringencies which could never accommodate the "mighty line" of Homeric epic.

To carry this idea a bit further, the greater stability of the C (as compared

[45] See also Kirk's later remarks (1985, 18-24).

to the A) caesura need not be only a function of the line-ending closure identified by Kirk and described earlier, though it most certainly is in part due to that closing cadence. Instead, we may understand the 60 percent occurrence of the bucolic diaeresis as in part the opening boundary of the fourth colon in the form of the typical adonean rhythm, which is also the familiar final cadence in the line. Indeed, the two concepts seem to be complementary rather than mutually exclusive. Finally, Kirk's notion of "word-length availability" as determining colon extent does not prove out when applied to the line as a whole. If all caesuras are to be held up to the standard of the B break and therefore taken as equal, the assortment of word-lengths that open the line should multiply variabilities at mid-line and further multiply possibilities at the C caesura. At the very least, the A and C breaks should occur in approximately corresponding variability, with the two-position B boundary making the C perhaps more flexible than the A, which takes its departure from the one-position line-beginning. But although Kirk's model would predict these phenomena, they do not occur. Apparently the A and C caesuras must be explained in another way.

The most recent extensive study of the colometric structure of the hexameter is that of Berkley Peabody (1975), who would assign line-breaks in the following positions:

His schema thus differs from Fränkel's in the deletion of the first two of the A positions. Peabody's placement of the three breaks covers most of the lines in his object text, Hesiod's Works and Days (A1 or A2, 90 percent; B1 or B2, 99 percent; and C1 or C2, 90 percent), and he founds his theory of cola on comparative diachronic studies of Greek, Indic, and Iranian meters. From this caesura-system he derives twelve principal colonic forms that populate the four-part structure (p. 68; some examples with spondaic substitution):

1	2	3	4

I find this recension of colometric theory far the most satisfying of the modifications of Fränkel's original proposals, both because of the comparative diachronic evidence on which it stands and also because it best acknowledges the complexity and flexibility of the hexameter. While one cannot expect absolute congruity of every Homeric or Hesiodic line with any abstraction (for as one moves toward that pattern-example congruity, one also moves toward a schema at the expense of a meter), a theory that accounts for variability and subtlety is inherently more useful and appealing than one that does not. Before going any further with Peabody's modifications, however, let us return to Fränkel's original ideas to clarify some issues.

Fränkel's exposition of cola, based ultimately on an earlier study by Eduard Fränkel (1932), was the first to posit the four-part structure of the Homeric lines. Contrary to Porter's and Kirk's later claims, he does not demand an absolute and firm sense-break (Sinneseinschnitt ) at every point of caesura, but admits throughout that some caesuras are stronger demarcations than others.^[46] Rather, he places the colon boundaries at the positions most commonly marked by editors' punctuation. To be fair, this method is not unambiguous, since it tends to isolate an editor's idea of syntactic boundaries over a sample of lines and then to apply that information back to metrical structure.^[47] Still, his results do partition the Homeric line effectively and give an excellent first approximation of the dynamics of the inner metric.

Fränkel explains the exceptions to his rules in terms of an occasional "heavy word" (schweres Wort ), which, because of its length of six morae or more, bridges a colon boundary and makes the given colon overlong.^[48] As one of many examples, he offers Iliad 9.145, which contains a bridged B caesura (verschobene B-Zäsur ):

Without a mid-line break, the result of the eight-morae schweres Wortbild,^[49] the line seems to be divided into three rather than four cola. Of course, the bridging of the hemistich boundary occurs, as we know, only

[46] E.g., H. Fränkel 1955, 104: "Zwischen beiden Extremen, Satzgrenze und Wortgrenze, gibt es Einschnitte jeden Grades. Ein Mass für die absolute Stärke eines Einschnittes gibt es nicht, abet die relative Stärke der Sinneseinschnitte in einer Wortfolge ist oft unmittelbar ersichtlich."

[47] Fränkel's statistics are taken from a table compiled by A. Ludwich in Rossbach and Westphal 1867-68.

[48] He remarks (1955, 107): "Dem verspäteten Einschnitt geht ein Wort oder Wortbild...von mindestens 6 Moren voraus . Unter dem Gewicht eines solchen 'Schweren Wortes' (SW), wie wir es nennen wollen, kann ein Einschnitt um eine oder mehr Stellen zurückgedrückt werden."

[49] The idea of Wortbild includes both compounds and words combined with enclitics or proclitics. Cf. the "accentual groupings" in the Serbo-Groatian deseterac as described by Maretic^[*] (below).

about once in a hundred Homeric lines. Somewhat more frequent is the bridging of A and C: caesuras, as in the following examples:

	(Od. 15.425 [No A])
	(Od. 15.433 [No C])

In these two lines the A and C breaks are, respectively, verschobene , and the hexameters again appear to divide into three rather than four cola. However, through Fränkel's combination of multiple possibilities for caesura placement with "heavy word" bridges, he is able to account for all the complexities in these and other particular actualizations—including even those places in the line at which word-break is seldom tolerated, Meyer's and Hermann's Bridges. Both of these zeugmata, the former after a second-metron trochee and the latter after a fourth-metron trochee, are explained as interruptions of the colon system and for that reason are avoided.^[50]

In sum, Fränkel's original system well suits the protean flexibility of the hexameter, a flexibility which fosters a correspondingly supple formulaic diction, as the author himself recognizes (1955, 116): "Den Sängern war es darum möglich die Inhalts- und Kolongliederung ohne harten Zwang so weitgehend zusammenfallen zu lassen, well ihnen das Zäsurensystem eine grosse Zahl von legitimen Varianten zur Auswahl stellte—vier für A, und je zwei für B und C."

In a series of articles dating from 1970, Wayne B. Ingalls has championed Fränkel's system and argued against its critics. Using Porter's own data on the Iliad and Odyssey , he points out (1970, 6) that Porter's displacement of the C caesura is not statistically justified and shows that "even Fräkel's additional alternative A-caesuras at 1 and 11/2 [after the longum in metron 1 and after the first trochee, respectively], rare as they are, are more common than Porter's C2, at 9 [after the longum in metron 5]."^[51] In dealing with

[50] I mint in part agree with Kirk, especially in the case of Hermann's Bridge, that these zeugmata are avoided on the basis of a general prohibition against "trochaic cuts." The trochee will falsely signal line-end by imitating, in combination with a previous dactyl, the "quantitative close" or final adonean, that portion of the hexameter from the bucolic diaeresis on. But again I see no reason why these aspects of Fränkel's and Kirk's explanations must be viewed as mutually exclusive.

[51] O'Neill and Porter refer to various line-positions by means of a schema that counts every two morae as a whole integer and every mora as a half integer:

1, : Fränkel's additional A caesuras (A1, A2); 2, 3: Common A caesuras (A3, A4); 5: B151/2; : B2; 7: C1; 8: C2.

Kirk's objections, Ingalls demonstrates the essential subjectivity of arguments that depend on assertion and are not buttressed at all points with statistics. He then turns to a re-evaluation of Kirk's colonic analysis, which is shown to derive from eliminating two of Fränkel's A caesuras and all C breaks except the bucolic diaeresis. As Ingalls reveals, Kirk's methodology is flawed, for one cannot use one set of rules to determine breaks and another set to define cola: "In so doing [Kirk] naturally precludes the possibility of much concurrence between colometric and semantic units" (p. 11). On the basis of this re-analysis, then, Ingalls is able to dismiss Kirk's objections to the original colon-system and argue that, in the meaning Fränkel had intended, each colon is indeed a sense-unit.

In two later studies (1972, 1976) Ingalls widened his perspective to take in the relationship between metrical cola and formulas. Though this relationship is more properly the subject of later chapters, we may look briefly at some of his more important conclusions which have to do with our present concerns about metrical structure. First, he feels that Parry's original definition of the formula is tied too closely to its syntactic identity as noun plus epithet, and that this first approximation cannot be generalized to other kinds of formulas without becoming misleading. This argument bears directly on our understanding of the link between meter and formula and, if accepted, seems to indicate the need for a finer articulation of the blanket phrase "under the same metrical conditions." Second, he prescribes the colon as a metrical rationalization (Ingalls 1972, 122):

The formulae from Parry's analysis, then, confirm the intimate connection between formular usage and the colometric structure of the hexameter. Just as the formulae are the linguistic building blocks of the verse, so the cola are the metrical blocks. In other words, the metrical shapes of the formulas tend to coincide with those of the cola with which the verse is composed.^[52]

As we shall see in chapter 4, the formulaic process is more complicated than a simple one-to-one relationship between formula and colon, but Ingalls's suggestions, based on Fränkel's colometry, are an important step toward a deeper understanding of that process.

As mentioned above, Peabody's discussion of inner metric in the hexameter is in many ways the most productive approach so far advanced. I shall follow the main descriptive outlines of his presentation both in this chapter and later on, but first let us make clear that the difference between the Fränkel-Ingalls and Peabody theories is basically one of statistical and diagrammatic convenience.^[53] From a statistical point of view, Peabody's elimination of the

[52] Since Parry was of course working with caesura-bound phraseology, this approach h perhaps not as novel as it might seem. What makes it convincing is the impressive colonic analysis of Nagler's traditional phrases (see esp. Ingalls 1972, 115-18).

[53] For figures on relative percentage occurrence of the original Sinneseinschnitte , see Fränkel 1955, 104-5.

A1 and A2 caesuras, the only difference between the two systems of colon-structure, equalizes the variance among A, B, and C breaks at two apiece and affects only 4 percent of the sample. Further, even the 4 percent affected is not lost entirely, but simply reassigned to the "all others" category. The number of possible cola is greatly reduced by this simplification, since the possibilities for the beginning point of colon 2 and the ending point of colon 1 are cut in half, from four to two each. From a statistical and presentational perspective, then, more is gained than lost by Peabody's simplified view of the A caesura.^[54]

Summary of Inner Metric

What is needed at this point is a standard for ranking, a quantitative measure of the importance of each word-break. As Kirk and others have stressed, the B caesura is the most regular, with the B1 and B2 positions providing a mid-line break in 99 percent of Homeric lines. There simply are no other points in the hexameter where word-break is so regularly observed. If Peabody (1975, 45-65) is right about the prehistory of the Homeric line as two shorter verses joined in a single hybrid (cf. G. Nagy 1974, 49-102), then we have a diachronic explanation of why this hemistich boundary should be so prominent in the synchronic sample of Homeric hexameters. But however we conceive of that prehistory, the shape of the hexameter as we know it indicates that the first and primary segmentation of the line is into two parts—two unequal hemistichs or half-lines:

Possibility 1 (B1 caesura):

Possibility 2 (B2 caesura):

Or, in mora-count:

Possibility 1: 10x | 12x + 1
Possibility 2: 11x | 11x + 1

In either case a slightly longer colon follows a shorter one. This is the most consistently observed dimension of the inner metric: a segmentation at the half-line level.

The A and C: boundaries operate on another level of segmentation, as their

[54] While any descriptive model can be more or less useful according to its innate complexity and ability to represent faithfully the empirical facts, any concept of inner metric will remain an abstraction that names and explains rather than is that inner metric. Our task in all areas of this comparative prosody chapter is to choose the abstraction that communicates the most valuable data in the most convenient manner. If by admitting two positions per caesura we can control 96 percent of the sample, then we have, formally speaking, a very workable model.

lower frequency figures make apparent. While this does not mean that they are to be discounted or questioned as caesuras, it does indicate that the cola which they form will have one boundary less consistently marked than the other. The colon system as a whole thus depends on Schnitte of at least three different types: (1) the absolutely regular line-beginning and line-end, always separated by twenty-three counts or morae and ending with the typical closing rhythm; (2) the B caesura, or hemistich boundary, which may take one of two positions and thus form two possible pairs of half-line segments; and (3) the A and C caesuras, whose two positions (four in all) account for a somewhat smaller percentage of all Homeric lines, about 90 percent. To look at the same situation in another way, we could say that Fränkel's "heavy words" bridge breaks according to the following schedule: (1) line-beginning and line-end can never be bridged; (2) the two-position B caesura can be bridged only 1 percent of the time; and (3) the two-position A and C caesuras can be bridged about 10 percent of the time. These are three distinct levels of segmentation in the hexameter, coordinated to be sure, but distinct from one another in frequency and variability of position.

Yet there seems also to be a fourth level of segmentation: Fränkel's A1 and A2 and the two word-breaks commonly found within colon 4. These four structure points occur decidedly less regularly than the main caesuras, but they appear often enough to beg the question of whether any or all of them should be classed as caesuras. Because they seem to function at another level of segmentation, I prefer to distinguish these four positions by labeling them juncture points , thus preserving the integrity of the colon structure and avoiding ambiguity in terminology. I believe, however, that they must be considered along with the more regular aspects of inner metric if we are to obtain a full description of the hexameter. By recognizing these juncture points as structural markers but not as caesuras, we can, first, encode in our model for the Homeric line important structural details beyond the hemistich and colon which may lead to a better understanding of phraseology and, second, preserve a relatively simple assortment of what I shall call, with Peabody, "principal colonic types." We can, in other words, limit the A, B, and C caesuras to two possibilities each and thereby limit the number of cola that they can bound to twelve. The levels of segmentation can be seen in table 6.^[55] The principal colonic types, after Peabody (1975, 68) are as shown in figure 1. If we thus dispense with Fränkel's A1 and A2 caesuras, preferring to interpret these two breaks and the breaks at  and 10 as juncture points within the first and final cola, and adopt with modifications Peabody's statistically more presentable model,

[55] The percentage occurrences cited for juncture points consist of ranges based on the following individual percentages: a (at 11/2), Il = 39 percent and Od = 38 percent; b (at91/2 ), Il = 30 percent and Od = 34 percent; g (at 91/2 ), Il = 44 percent and Od = 51 percent; and d (at 10), Il = 35 percent and Od = 32 percent. Note that these are simply word-end percentages and are not meant to indicate a metrical boundary.

Figure 1. Principal Colonic Types in the Hexameter
1	2	3	4

we shall have at once a more detailed and a more flexible descriptive instrument. The resulting schema for the hexameter is thus as follows:

The six main caesuras (A1, A2, B1, B2, C1, C2) and the juncture points (a , b , g , d ) determine the inner metric of the line. In the next chapter I shall study the relationship of this colonic structure to the verbal data of the Homeric epos, stressing Peabody's insight (1975, 74) that "the remarkable statistical coincidence of the forms of the elements used in the epos with the forms of the principal cola is significant. This coincidence goes far toward proving the essential unity of the metrical and linguistic traditions in the epos. It also shows that the colon, both in origin and function, is a linguistic period, a 'word form.'"^[56]

For the moment, however, let us conclude this description of inner metric in the hexameter with a consideration of "right justification," the metrical phenomenon mentioned above in relation to Indo-European meter and the outer metric of the Greek epic line. In the case of the podic structure of the hexameter, the evidence of the tendency toward right justification was the relative

[56] See Russom 1987 on the coincidence of metrical units and word-forms in the Old English alliterative meter.

frequency in various positions of the dactylic metron, Which occurred more regularly toward the end than toward the beginning of each hemistich and most regularly as part of the closing cadence, or adonean clausula, of the line as a whole. Diagrammatically we may represent this two-level function as follows:

This characteristic texture means that the syllabically more extensive metra, those which have either a greater number of syllables (or, to put it another way, more short syllables) tend toward the right or end of metrical units within the hexameter. Still from the perspective of outer metric, then, syllabic extent and short syllables tend statistically to migrate toward the end of these units. Conversely, the left-hand or beginning portions of line and half-line will lean statistically in the opposite direction, that is, toward a shorter syllabic extent and the long syllables of spondaic substitution. This characteristic distribution is, of course, a tendency rather than a rule, but it will prove significant for our discussion of formulaic diction, since the texture of metrical units will be affected by right justification. In other words, lexical elements will typically arrange themselves, both over time and synchronically, in a right-justified order, a relative placement which is to a discernible degree overseen by the ending dactylic cadence of line and hemistich. Since the quantitative close or final cadence is more frequently observed than the preference for a dactyl in metron 3, the second hemistich and whole line will, in general, show the effect of right justification more regularly than will the first hemistich. The basic inclination, however, affects all parts of the line.

The same tendency toward situating longer metrical elements with more short syllables toward the end of metrical units in the hexameter is also apparent in the inner metric, although here it is of course the measure of mora-count which determines the "length" of increments. The hemistich patterns of 10x | 12x + 1 and 11x | 11x + 1, determined as they are by the B1 and B2 caesuras in the third metron, divide the hexameter into two unequal parts: in both cases a shorter first half is followed by a longer second half. By using Peabody's method for schematizing cola, we can analyze these hemistichs for the same tendency toward right justification at the level of the colon (1975, 69):^[57]

[57] But at this point the tendency ends; right justification seems not to enter into the internal texture of the first and fourth cola, specifically that texture created by the juncture points examined above. In colon 1, where these points occur at 1 and 11/2 , two of the four possible segmentations can yield relatively longer colon-parts in the latter section of the unit, measuring by syllables. Measuring by morae, however, the situation is more balanced, with one possibility being right-justified, a second left-justified, and the remaining two symmetrical. The fourth colon operates in a similar manner; in fact, the segments in this last colon tend, if anything, toward left rather than right justification.

Computing by morae (symbolized as x )—that is, by assigning a count of 2 to each longum and 1 to each breve—one arrives at the table above. We can point out a number of instances of right justification in this diagram. First, cola 2 and 4 are on the average considerably longer than their half-line partners, cola 1 and 3, respectively. Second, the most spacious colon of the four is the last one, which, it will be recalled, involves the closing cadence to the line. At the levels of both whole line and hemistich, then, the hexameter forms itself in larger metrical units as one moves from left to fight, from beginning to end. What is more, this tendency is apparent from the perspective of both inner and outer metric. But the pattern does not extend to segments formed by juncture points; apparently the influence of this Indo-European characteristic ends at the hemistich level.

To summarize these remarks on fight justification, then, I would stress its prominent motive force in the formation of both inner and outer metric in the hexameter. To take the latter first, dactyls migrate toward the ends of both lines and half-lines, making these terminal sections more expansive by both syllable- and mora-count and more densely populated by short syllables. For its part, the inner metric manifests fight justification in a longer second hemistich and relatively more expansive second and fourth as compared to first and third cola. Although the principle does not extend to the inner texture of cola, it does figure in all units of the line that recur with regularity. In general, the Homeric hexameter locates the more extensive elements to the right, or toward the end, of a given metrical unit.

We recall that the features ascribed by most scholars to the reconstructed Indo-European ancestor of the hexameter and other poetic lines are four: (1) a quantitative basis, (2) consistent syllabic extent, (3) a regularly placed caesura within the line, and (4) right justification. Because all these features are to varying degrees reflected in both the inner and the outer metric of the Homeric line, we may envision a diachronic history underlying the synchronic patterns of the extant texts. Homeric phraseology takes its shape from these prosodic patterns, ancient Greek reflexes of Indo-European compositional habits culminating over time in the hexameter diction we find in the Iliad and Odyssey . For the moment, however, the important point is that, notwithstanding a seminal ancestry with many interrelated progeny, Homeric prosody is itself a singular prosody, with rules and tendencies very much its own. To put the matter succinctly—however much it may genetically owe to earlier forms, the Homeric hexameter is tradition-dependent.

The Junacki Deseterac (Heroic Decasyllable)

Serbo-Croatian oral epic tradition takes two primary metrical forms, the fifteen-syllabic bugarštica , also called the "long line," and the ten-syllable junacki deseterac or "heroic decasyllable."^[58] While both have venerable histories in the poetic tradition—the bugarštica , for example, serving as the medium for a song in Petar Hektorovic's^[*] Ribanje , the earliest extant recorded folk poetry in the language (published in 1568)—we shall be concerned in these studies strictly with the deseterac , the meter of the Stolac Return Songs, our Serbo-Croatian comparand for the ancient Greek Odyssey and Old English Beowulf . As an organizational procedure, we shall approach the decasyllable as we did the hexameter, from the perspective first of outer metric and then of inner metric, seeking in each case to establish both the lineage and the idiosyncratic form of the deseterac . Only when our grasp of the prosody is sure and unambiguous can we productively proceed to study of the phraseology with which it exists in symbiosis.^[59]

Outer Metric

The outer or podic structure of the heroic decasyllable seems at first sight more regular than that of the hexameter. While the Homeric line tolerates spondaic-dactylic substitution freely in the first four metra and grudgingly in the usually dactylic fifth metron, the Serbo-Croatian line seems always to conform to a pentameter scheme involving five two-syllable feet:^[60]

In addition to this five-part regularity, we observe a complementary consistency in a syllable count of ten, as the following examples illustrate:^[61]

Posle toga dva cifta pušaka
Behind him two paired rifles

[58] For a convenient comparison of the two meters, see Stolz 1969.

[59] On the importance of prosody in the study of formulaic structure in Serbo-Croatian epic, consider Lord's remark (1960, 31) that "the formula is the offspring of the marriage of thought and sung verse. Whereas thought, in theory at least, may be free, sung verse imposes restrictions, varying in degree of rigidity from culture to culture, that shape the form of thought. Any study of formula must therefore properly begin with a consideration of metrics and music." Cf. Petrovic^[*] 1969, 178.

[60] With, of course, the exception of the relatively rare hyper- and hyposyllabic verses, which are treated below and in chapter 5. I will refer to examples and discussion from Tomislav Maretic's^[*] seminal studies of the deseterac throughout the discussion of Serbo-Croatian prosody, citing passages according to the following abbreviations: Maretic^[*] A = 1907a; B = 1907b; C = 1935; D = 1936. On shortening and lengthening lines to answer the ten-syllable constraint, for example, see Maretic^[*] A:80-112, B:76-122, C:218-42, and D:1-27, where he explains with some care the various methods involved.

[61] Unless otherwise noted, all examples arc taken from the Stolac texts in the Parry Collection.

Poce pisat' knjigu šarovitu
[He] began to write a multicolored letter

In these respects, the deseterac exhibits the quantitative basis and syllabicity characteristic of both the hexameter and its Indo-European precursor.

When we inquire about the disposition of these quantities in each of the five metra, however, the situation rapidly becomes more complicated and passes from the certainty of dependable rules to the uncertainty of irregularly observed tendencies. Jakobson (1952, 26) has argued that "the verse inclines toward a trochaic pentameter pattern," noting that the great majority of word-accents fall on the odd syllabic positions in the decasyllable. The model toward which he views the deseterac as tending is thus

Of course, absolute coincidence of lexical accent and verse ictus is not often observed in all five positions in the line. The guslar may bend individual lexical patterns to the recurrent and generalized influence of his rhythmic vocal and instrumental melody,^[62] and as Lord (1960, 37) observes, "there is a tension between the normal accent and the meter." Because we can actually listen to the Stolac songs (as we cannot do with dead-language texts), we know of other performance variables, such as the particular prominence of the ninth or penultimate syllable, but even without these firsthand observations it is plain that the putative trochaic pentameter pattern must remain a tendency, not a rule.

Other scholars who have considered the possibility of regular trochaic ictus include Svetozar Petrovic^[*] (1969) and Pavle Batinic^[*] (1975). Petrovic^[*] (1969, 183) examines selections from four of the most famous songs collected and published by the Serbian linguist-ethnographer Vuk Karadzic^[*] in the nineteenth century, setting his findings alongside Jakobson's figures for coincidence of word-accent and position (table 7).^[63] Throughout the five samples, the strongest correlation between lexical accent and position is at syllables 1 and 5, a fact that is in part explained by noting the prevalence of accented monosyllables in those positions^[64] and the regular onset of the two cola (1-4 and 5-10). After syllables 1 and 5 only positions 3 and 9 coincide with accent at all regularly, and these two are not much more prominent than some surrounding even-numbered positions. What syllables 3 and 9 do share, however—and this contributes to their relative importance from the point of view

[62] On the melody of Serbo-Croatian oral epic performance, see Herzog 1951; the transcription by Béla Bartók from song no. 4, SCHS 1:435-67; Lord 1960, 37-41; and Bynum, "The Singing," SCHS 1 4: 14-43.

[63] Note that the use of the Vuk songs violates the principle of genre-dependence, for many of them are quite short and more lyric than epic. On this point, see Foley 1983b.

[64] See Maretic^[*] A:41 and C:186-96, as well as the description of right justification below.

of inner metric—is their penultimate spot in each of the cola. To put the matter another way, each precedes a closing syllable that shows a zero correlation with word-accent. As Petrovic^[*] indicates, position 7 has the weakest correlation of all odd—and supposedly ictus-bearing—syllables in the line, with text B actually suggesting a dactylic shape for syllables 5-7 and text C presenting a sequence (5:69, 6:40, 7:27) that seems unmetrical.

Also working strictly from the perspective of outer metric and on a sample of some 1,274 lines of mixed material,^[65] Batinic^[*] attempts to solve the rhythm of the deseterac by advocating attention not merely to stress-accent but also to two other properties of Serbo-Croatian lexical units: tone and quantity. Instead of assuming an exclusive correlation between accent and metrical position, he widens the search to include a survey of the three possible tones (rising, falling, and unmarked) and two possible quantities (long and short) in his characterization of syllables. The result, displayed in table 8 (from Batinic^[*] 1975, 102), is a description of various syllable-types in terms of their suitability for ictus, that is, of their likelihood of coinciding with an odd-numbered metrical position in the decasyllable. In Batinic's^[*] view, any syllable with two or more of these expressive features will almost certainly bear metrical ictus, those with either stress (word-accent) or quantitative length prove ambivalent and can occur in either odd- or even-numbered positions, and those lacking all three linguistic features will occupy the off-beats of the verse.

While this graded schedule of suitability for metrical ictus does provide us with a more exacting description of syllable distribution than is elsewhere available, it suffers both from the tautological assumption of binary feet as the sole and necessary metrical foundation of the deseterac and from the "catalog" mode of explication. To take the second objection first, note that more refined

[65] Of the entire sample, 815 lines are taken from the Karadzic^[*] songs, 305 from the Matica Hrvatska collection of Luka Marjanovic^[*] , and 154 are left unidentified. The relatively small size and internal imbalance of the material call Batinic's^[*] results into question.

TABLE 8. Batinic's[*] Syllabic Theory of the Deseterac
	Stress	Length	Fall	Character	Classes
1	-	-	0	Depression	Depression	1
2 3	- +	+ -	0 -		Ambivalent	2
4 5 6	+ + +	+ - +	- + +		Ictus	3

description may better characterize whatever synchronic design one finds in a text or group of texts, but unless that description is not only pertinent to the observed data but also revelatory of the structures underlying those data, it cannot explain them. Trochaic pentameter models, however elaborated, do not completely explain the deseterac lines of Serbo-Croatian epic because they do not penetrate to the fundamental structures underlying the verse. A pentameter of binary feet may serve as a helpful first approximation of the decasyllable, but the lack of a uniform fit between syllable (whether characterized by one or three expressive features) and metrical position should warn against accepting that model out of hand. Like the hexameter, the deseterac also exhibits an inner metric, and, again like the hexameter, it is on this inner metric that traditional phraseology rests.

Before turning to the description of inner metric, it is well to establish the important outer metrical constraints on the deseterac as evidenced in the Stolac material used in these studies. In particular, I am interested in the consistency of syllabic count in these Parry Collection songs, and in the reasons behind any variance from the ten-syllable norm. To begin, let us separate out two distinct categories: oral-dictated and sung texts. The former group will be represented by three poems comprising a total of 4,077 lines: Texts A, Mujo Kukuruzovic's^[*]Ropstvo Ograscic^[*] Alije (The Captivity of Ograscic^[*] Alija ; no. 1287a, 1,288 lines); B, Kukuruzovic's^[*] Ropstvo Alagic^[*] Alije i izbavinje Turaka (The Captivity of Alagic^[*] Alija and the Rescue of the Turks ; no. 1868, 2, 152 lines); and D, Halil Bajgoric's^[*]Halil izbavi Bojicic^[*] Aliju (Halil Rescues Bojicic^[*] Alija ; no. 6703, 637 lines). Sung texts are represented by text (3, Bajgoric's^[*]Zenidba Becirbega^[*] Mustajbegova (The Wedding of Mustajbeg's Son Becirbeg^[*] ; no. 6699 , 1,030 lines). The raw figures and references for hypersyllabic and hyposyllabic verses in these four songs are as shown in table 9. In all, the oral-dictated texts (A, B, D) show only five hypermetric and seventeen hypometric lines out of 4,077 (0. 12 percent and 0.42 percent, respectively), while the sung text (C) contains twelve eleven- or twelve-syllable lines and six of less than ten syllables out of 1,030 lines (1.2 percent and 0.58 percent, respectively).

Clearly, then, analysis of our sample of over five thousand verses yields no substantial evidence for syllabic variation in the decasyllable, and we shall

discover that even this minuscule percentage of hyper- and hyposyllabic lines consists principally of verses that generally follow the syllabic rule. To put it positively, the junacki deseterac proves itself a highly consistent verse form from the point of view of syllabicity, with over 98 percent of our extensive sample unambiguously adhering to this first principle of outer metric. Even if we choose to leave the uncertain approximation of podic structure and trochaic rhythm aside, preferring to explain coincidence of phonological features and ictus through principles of inner metric, we can be sure of this much: in the songs that are in part the subject of these studies, the fundamental metrical measure of ten syllables is a precise and limiting feature of the line.

With this information in hand, we may profitably make two inquiries about the 1.8 percent of the sample that seems not to conform to the metrical rule of syllabicity. First, why do twelve of seventeen hypermetrics occur in the sung as opposed to the oral-dictated texts? Second, and more generally, to what kinds of "errors" do we owe the forty variations reported above? To begin with the sung text C, note that six of the twelve eleven-syllable lines owe their incongruity to initial extra-metrical syllables, in the form either of interjections:

or what I call "performatives":

The lines involving interjections arc typical at the beginning of songs, at the

resumption of a song following a rest break, and at moments of dramatic intensity. Because they have a rhetorical force in performance, interjections are more appropriate to sung than to dictated texts, and the Parry-Lord amanuensis Nikola Vujnovic^[*] was more likely to transcribe them than other kinds of variations from the decasyllabic norm. Indeed, it is well to remember that Nikola was himself a guslar and a member of the tradition he helped to record,^[66] so it is only prudent to be aware of the filter he provided between what was performed and what he either took down from dictation or transcribed from acoustic recordings in later years.

The "performatives" offer an avenue for inquiry into the nature of metrical "flaws." Although we may argue that the extra-metrical I (or hI , with the aspirate [h] customarily acting as a hiatus bridge between the tenth syllable of one line and the onset of the next line) represents the very common conjunction i , or "and," not every case will tolerate that interpretation syntactically (cf. Maretic^[*] 1935-36, 255: 10-16). In listening to the recording of Bajgoric's^[*] performance, it becomes apparent that the sound-image itself can vary from a fully pronounced i , at times arguably the conjunction, to the simple glide [j], ostensibly a hiatus bridge or continuant between lines. Its function seems on the whole to be phonological rather than syntactical; this is the reason that I have not felt it correct to "translate" the sound in the examples above. All in all, the concept of a performative—that is, a sound that promotes the phonological continuity of the performance without interceding in the syntax or meaning of the contiguous lines—seems closest to the true function of this sound. And we may assume that Nikola tacitly recognized these sounds for what they were, since he regularly left them untranscribed.

Of the remaining six hypermetrics in text C, two involve the insertion of an excrescent vowel, as in this example:

Nikola does not transcribe the aberration, resorting instead to the standard form, kalpak . At first sight the addition of a might seem idiolectal, as if some peculiarity of Bajgoric's^[*] performance style generated an occasional excrescent vowel. This supposition gains some support from the unusual assortment of hiatus bridges in Bajgoric's^[*] repertoire: [v], [j], [h], and even [m] and [n] can prevent glottal stops and the momentary interruption of vocal continuity both between and within the words.^[67] As a counter-example, we may compare

[66] On the collection process, see Lord, "General Introduction " to SCHS 1:1-20; Lord 1951b.

[67] This kind of phonological bridging h consistent with the general tendency of the poetic medium to make colonic "words" out of lexical items.

Ibro Basic's^[*] habit of doubling a stem vowel to eke out the ten-syllable norm of his verse.^[68] Whatever the case, the excrescent vowels are, like the extrametricals, phonological variants which do not affect the lexical or syntactic structure of the decasyllable.

Two or more eleven-syllable lines result from Bajgoric's^[*] using the longer of two dialectal forms when the shorter one would have suited the metrical environment:

In choosing the diminutive gunjinu instead of the simplex noun gunju , the singer exceeds the ten-syllabic norm; likewise, by selecting the four-syllable ijekavian dialect-form dijeliti instead of the three-syllable jekavian bi-form djeliti , he makes a hypermetric verse. While this choice amounts to a metrical error, it is only fitting to point out that it is also further evidence of the polymorphism of oral tradition, a characteristic found at all levels of structure. The doublet gunju /gunjinu —or, more generally, the simplex/diminutive substitution that can involve many nouns—is evidence of a compositional flexibility: the poet in performance can select either bi-form on the basis of syllabic count in the rest of the line. Of course, many of his "choices" have been made for him by the tradition that he inherits and in the formulas that he employs, and in any case he does not ponder the choice in performance but rather trusts his "ear" to produce a metrical verse. But the very fact that this kind of hyper-syllabicity can and does occur in the synchronic moment of performance proves a degree of flexibility at some points. The dijeliti /djeliti doublet reinforces the argument: perhaps under the influence of cognate formulas,^[69] Bajgoric^[*] uses the metrically "wrong" bi-form in combination with mejdane and produces an eleven-syllabic line. In both "errors" we detect an important compositional principle: although this sort of flexibility can occasionally lead to metrical infelicities, like other kinds of traditional polymorphism it serves the needs of the composing oral poet and makes possible over time the evolution of a traditional diction. The price one pays for that synchronic flexibility and

[68] E.g., line 421 from a sung version of his Alagic^[*] Alija and Velagic^[*] Selim (text no. 6597 ), second colon: "banii ce mlade" instead of the expected (and hyposyllabic) "banice mlade."

[69] Compare the following, all examples drawn from Stolac texts: "Da mi, bane, mejdan dijelimo " (6699 .814), "Ve' dvojica mejdan dijeliše " (6699 .857), "Dijeliše , pa se rastadoše" (6699 .858), "Treba nema pravo dijeliti " (1868.2062), and "A danak se s noci^[*] dijeljaše " (1287a.976).

diachronic development is the relatively rare lapse exemplified by gunjinu and dijeliti .

The last two examples of hypersyllabicity in text C entail five-syllable first hemistichs:

More than the other examples cited, these two lines seem to be true hyper-metrics. Their extra length results from neither supernumerary elements nor an unfortunate choice, and all of the words involved have syntactic roles to play; in fact, in assigning the asterisk in each case to the third position, I have made a somewhat arbitrary decision about which is the "offending" syllable. On the other hand, all would have been well had Bajgoric^[*] elided sve vodaje to sv'odaje and te vuzdrz'oto t'uzdrz'o . As the lines stand, however, the hiatus bridge [v] precludes elision and memorializes the hypersyllabic construction. Without a thorough search of all of Bajgoric's^[*] sung repertoire and comparison with the Stolac community of singers,^[70] we cannot begin to describe how this formation came to be (that is, whether the aberration was frozen or caused by the hiatus bridge), and in the end the question may not be as important as noticing the deformation as it occurs in the text.

The twelve hypermetric verses that occur in the sung text C may thus be divided into four groups: (1) extra-metricals (interjections and performatives), (2) words lengthened one syllable by excrescent vowels, (3) faulty choices between metrical bi-forms, and (4) "true" hypermetrics caused by lack of elision. The first two categories, accounting for eight of the lines, may confidently be designated as purely phonological variations affecting neither the syntax nor the sense of the line. They are typical of sung performances and were not transcribed by the Parry-Lord amanuensis Nikola. The two instances in the third category derive from the same traditional polymorphism that makes possible multiforms at all levels of traditional epic; these infrequent errors are mere synchronic blemishes that fade into the background of the continually re-created diachronic text. The last pair of hypermetrics are thus the only lines that qualify as outright violations of the syllabic constraint in the deseterac .

Of the six eight- and nine-syllable hypometric lines in the same sung text,

[70] There is no evidence in the computerized concordance of Stolac songs of any corresponding phraseology.

four are attributable to missing connectives and temporal conjunctions that Bajgoric^[*] probably swallowed in the moment of performance. Nikola, demonstrating his singer's ear, fills the short lines out to the usual increment, fashioning common formulas that match the syntax (added elements are italicized in the following examples):

The other two hyposyllabic lines result from, in the case of line 854, an unfortunate elision: *Pa odigra cacina^[*] goluba to

	(C.854)
Then he danced his father's horse out

—another example of multiformity gone awry; and in line 886, the transformation of the connective or performative i to a glide rather than a full morpheme: * I jovako momak progovara to

	(C.886)
The young man spoke in this way

Nikola silently corrects these last two lines as well, causing them to conform to the usual deseterac syllabicity.^[72] The message of the hypometric lines is thus the same as that of the hypermetrics: aberrations are in the main phonological mishaps that stem from the exigencies of composition in performance; the syllabic constraint on the deseterac proves once again a very strict and important one.

The few lines that vary from the standard ten syllables in the oral-dictated texts, a total of only twenty-two out of 4,077 lines (0.5 percent), stem from the same kinds of causes as underlie those in the sung text C, with the expectable exception that we find no hypermetrics or performatives in the dictated texts. Excrescent vowels, faulty selection of metrical bi-forms, and

[71] Here the singer also fills the first two syllables with an especially prominent instrumental passage on the gusle .

[72] Here and elsewhere it may be easy to overestimate Nikola's editing of the received text. To put the matter in proper perspective, it is well to recall that his "emendations" only reaffirm the overwhelmingly consistent syllabicity of the deseterac , representing what the singers would commonly (and in fact do commonly) do, and affect only 15 of 1,030 lines (0.5 percent).

problems with elision all contribute to long and short lines,^[73] but once again the overwhelming impression is of a highly syllabic verse form—a line that varies from its ten-syllable shape only in the moment of performance, and then only very infrequently. With over 98 percent of our sample answering the syllabic constraint, and, further, with virtually all those rare departures from the norm explained as momentary ornaments or errors, it is safe to pronounce syllabicity a constant and rigorous rule in the deseterac . To pursue other regularities in the line, we must turn to its inner metric.

Inner Metric

Caesura

All commentators on the deseterac add to the decasyllabic constraint a regularly recurring caesura between the fourth and fifth syllables, yielding two cola of four and six syllables each:

	(B.1)
What was the shouting in Zadar?

As this example illustrates, no matter what the syntactic form of the line, the word-break will come at precisely the same place in each decasyllable.^[74] In fact, even when a line is hypermetric or hypometric, one or both of the two cola will be preserved, as in the following instances discussed above:

	(C.1; both cola intact)
	(C. 103; colon 2 intact)

From a practical standpoint, the caesura is never bridged in the deseterac . Caesura placement and colon formation are constant throughout the sample of more than five thousand lines and show even greater stability, as these two lines help to prove, than syllabicity.

Before moving on to describe the nature of each colon or hemistich, let us make two related theoretical points. First, it will be recalled that the Homeric hexameter exhibited three caesuras, a highly regular mid-line break at two possible positions (B1 and B2, 99 percent) and two somewhat less frequent breaks within the half-line, each of which could also occur at either of two

[73] We may suppose that some metrical infelicities are due to lapsus linguae or lapsus calami , but without an aural recording of the dictated texts it is difficult to be sure. See further Maretic^[*] A:76-112; B:76-123; C:218-42; and D:1-28.

[74] Maretic^[*] affirms the same principle for his study of the Karadzic^[*] (A:62) and Moslem texts (C:183).

positions (A1 and A2, 90 percent; C1 and C2, 90 percent). Only by allowing two slots for each caesura do we attain such high percentages, but this controlled variation is to be expected in a verse form in which dactylic-spondaic substitution affects metrical shape so strongly. The point is that the hexameter has three movable caesuras and a correspondingly complex assortment of colon-types; the deseterac , in contrast, with its more focused syllabicity and the absolutely regular placement of its single caesura, has only two possible colon-types: the four- and six-syllable increments that together constitute a whole stich. Given this idiosyncratic situation (both traditions exhibit colon formation but the repertoire of types in Serbo-Croatian epic is far less elaborate), it behooves us to consider the patterns within the four- and six-syllable increments, for within the remarkable regularity of syllable count, caesura, and colon formation, the deseterac allows and even promotes a complicated display of traditional word-craft.

At the foundation of the singer's art, from both an evolutionary and a performance-oriented standpoint, lies the Indo-European principle of right justification. Much as in the hexameter, this increasing metrical (and therefore phraseological) conservatism as the line progresses from the beginning of a unit toward the end governs the shape of prosody and diction. But just as the rule took a tradition-dependent form in the hexameter—one resulting, for example, in varying hemistich and colon lengths—so it takes another series of forms in the deseterac . Indeed, right justification emerges as the principle behind the idiosyncratic texture of both cola, each with its own appreciable collection of individual features. We can trace the synchronic designs created by this diachronic pattern in the textual record of the Stolac songs.

Colon 2 and The Shape of The verse

Perhaps the most obvious evidence of right justification is the asymmetry of the decasyllable. As in the hexameter, the second hemistich is longer than the first, leading to a correspondingly greater metrical and phraseological stability in the latter part of the line. As a first approximation, we may recall Jakobson's (1952, 25) demonstration of the "quantitative close" over the last four syllables of the deseterac : "An accented short is avoided in the penult (ninth) syllable, and an accented long practically never occurs in the two antepenults (eighth and seventh syllables)." Schematically, then, the close follows this sequence:

	(C.504)
Then he attacked the hollowed-out ravine

The quantitative close thus governs the placement of words to some extent,

especially by inviting a long syllable in position 9, far the heaviest stress in the deseterac . So strong is this penultimate ictus that the poet composing in performance will often stress a lexically short syllable in the ninth position, in which case metrical ictus momentarily outweighs the rules of accent.^[75] The singer's rhythm and vocal melody emphasize the quantitative close, and the demands of performance override uncontextualized lexical values. The quantitative close and its focus on the ninth-syllable ictus are important features of the deseterac , and they offer one example of the operation of right justification in the prosody and its constraints on verse-making in the decasyllable.

Just as the longer second colon becomes firmer in its distribution of quantities toward the end of the unit, so the shorter first colon also reveals a looser-to-firmer progression of quantities from beginning to end. Although the effect of right justification is, logically enough, less pronounced in the shorter hemistich, we still observe the heaviest performance stress on the third syllable,^[76] followed by a complete lack of ictus on syllable 4. All in all, the close of the first hemistich is to an extent a mirror of line closure, with syllables 3-4 and 9-10 serving as boundaries:

	(C.216)
As he entered, the aga gave him a selam

We may now correlate these observations with Petrovic's^[*] figures on coincidence of lexical accent and metrical position, and note that the stress on syllables in positions 3 and 9 stems not necessarily from a trochaic pentameter tendency but from the Indo-European rule of right justification as imaged in the tradition-dependent Serbo-Croatian deseterac . Absolutely consistently in the rhythm and melody of sung performance (Lord 1960, 37-38), and with reasonable regularity in the verbal component regulated by the prosody (see Petrovic^[*] figures above, p. 87), the third and ninth syllables bear the strongest performance stress of any positions in the line, while stress is forbidden in the fourth and tenth syllables. We may thus consider each colon, as well as the entire line, to have a recurrent closing cadence of long-short and stressed-unstressed.^[77]

Corresponding to these dosing cadences, and in accord with the demands of

[75] The most common instance of this phenomenon is the guslar's stressing the penultimate syllable of verb forms (chiefly past participles, infinitives, or aorists) when they occur in final position; e.g., no. 6699 .224: "Tri, cetiri, cejifugrabijo^[*] " ("Three, four, the feeling seized him").

[76] The third-syllable ictus is not nearly as heavy as that involving the ninth syllable, but the two together are performatively much the most prominent in the line.

[77] As will be illustrated in more detail below, this 3-4 and 9-10 configuration of s^[*] s^[*] and s^[*] s^[*] also accounts for Jakobson's empirical description (1952, 25) of a bridge, or zeugma, at the end of each colon.

right justification, we find a looser distribution of quantities at the opening of each colon. These beginnings of units provide for rhythmic (and phraseological) variation, and for this as well as additional reasons to be discussed later in this chapter, the first and fifth syllables are thus the primary sites for stressed monosyllables. We may recall, for instance, that Petrovic's^[*] figures reveal the highest coincidence of word-accent and position at syllables t and 5. For the moment, it is enough to remark that the third and ninth syllables do not usually harbor these monosyllables because to do so would violate the bridges at 3-4 and 9-10.^[78] Schematically, then, we expect the configuration

	(C.278; stressed first syllable)
Then the tsar's hero began to shout
Dobro gledaj šta ti knjiga piše	(C.350; stressed fifth syllable)
See well what the letter tells you

Likewise, any initially accented word beginning either of the two cola will bear metrical ictus in positions 1 and 5. When we add to these lexical considerations the tendency of the guslar's metrical stress to fall at the onset of units in the deseterac , the following overall pattern emerges:

That is, there exist four primary sites for ictus, both in the prosody and in the phraseology that it helps to determine: positions 1, 3, 5, and 9. Taken as a group, all four sites are important to deseterac prosody and, as we shall see, affect phraseology in significant ways; all four also derive from the fundamental principle of right justification.

Before moving on to examine some specific instances of these general rules and to promote further articulation of the rough sketch that they assist in providing, we should consider what remains after these six positions, four stressed and two unstressed, are described. It may come as no surprise that the second syllable, for instance, shows no strong tendency toward a particular prosodic value. Syllables 6 through 8 reveal a similar ambiguity, as Petrovic^[*] reminded us and as Jakobson in defining his quantitative close in part illustrated. These four positions are not specifically defined in the deseterac ; to put the same matter another way, they allow for variation much more readily than do the six syllabic positions at either end of the two cola.

Perhaps we can now see how the hypothesis of a trochaic pentameter came to be, and also how oblique such a concept is to the true nature of the verse

[78] Other than an accented monosyllable, only an enclitic could occupy the tenth position alone, and then the formation at 9-10 would amount grammatically and prosodically to a disyllable with initial stress.

form. With stress and length tending to fall on syllables 1, 3, 5, and 9, and with off-beats and shortness coinciding absolutely regularly with 4 and 10, metrists were almost able to fit the deseterac into a Greco-Latin quantitative mold. If positions 2, 6, 8, and particularly 7 did not agree with the classical model, the "aberrations" at these points could be explained away by terming the rhythmic pattern a tendency instead of a rule. In fact, the notion of trochaic pentameter obscures the quantitative shape of the decasyllable, designating as it does a line of five stressed-unstressed doublets. The deseterac consists of ten syllables, to be sure, but they are divided asymmetrically into two cola of four and six syllables each, with each colon characterized by stress placement at particular positions. Right justification has provided each hemistich with a closing cadence, and an initial stress has likewise emerged in the relatively loose sequence of quantities that begins each unit. Greco-Roman models aside, the deseterac takes its own tradition-dependent form.

In addition to the closing cadence and initial stress in each hemistich, right justification provides the second colon with two complementary distribution rules. The first entails the sequence of words according to their syllabic length and prescribes that longer words follow shorter ones, the more extensive words characteristically appearing at the ends of cola and lines.^[79] The second and interdependent rule calls for initially accented disyllables, of the shape s^[*] s^[*] , to seek the ninth and tenth positions;^[80] this sorting, which answers both the penultimate ictus and zeugrna requirements, can, but does not necessarily, override the first distribution rule. To illustrate, consider these examples of the shorter-before-longer constraint in colon 2:

	(C.550; dòbro)
When they were well arisen
	(C.704; trošak)
The cost has been great for me to manage

As long as the disyllable has an initial short vowel, the first rule is the arbiter of placement in the longer hemistich. When a disyllable with a long stem

[79] Cf. Maretic^[*] B:127-29 and D:32-46.

[80] By "initially accented" I mean to indicate those initial syllables with lexically defined natural length , that is, those which are long rising ('), long falling (^), or simply long by nature (with macron or unmarked). Cf. Maretic^[*] A:58 and C:198-200. In the process of description, Maretic^[*] presents a table comparing the preference for line-final disyllables with a long initial vowel over those with a short initial vowel in oral versus written material. This table, derived from his figures (C:198-200), shows how strongly the oral singer favors the line-final disyllable with a long initial vowel and, in the process, offers a possible way to distinguish between oral and "imitation oral" texts. On this last point, cf. Haymes 1980.

vowel is involved, however, the shorter-longer order may be reversed under the influence of the ninth syllable and closing cadence.

	(C. 705; gâdu)
In Kanidza the white city
	(C.713; klêta)
The cursed cloak remained with the ban

It is precisely the firmness of the colon- and line-ending cadence that attracts the initially accented disyllable and causes the 4 + 2 distribution. Both the "normal" shorter-longer order and the reversal as a result of the attraction of ninth-syllable ictus are tradition-dependent realizations of the principle of right justification.

To restate this pair of distribution rules in properly ordered sequence, we can say that in the second hemistich a syllabically longer word will always follow a syllabically shorter word unless the shorter word is an initially accented disyllable, in which case the order may be (but is not always) reversed. Instances of non-reversal do occur:

	(C.768; bâne)
As the ban wrote to him

—so it will prove most accurate to conceive of the ninth-syllable exclusion as a possible rather than certain reversal of the customary order. A corollary to these two rules will further illustrate their interdependence and solve the problem of sequence with syllabically equal words. If two three-syllable words constitute the second colon, leaving nothing to choose between them on the basis of extent alone, then the one with an accented medial syllable will be favored in final position:

The penultimate ictus in the deseterac attracts the medially accented trisyllable,^[81] just as it does the initially stressed disyllable:

1	2	3	4	5	6	7	8
								klêt-	a
							mo-	má-	ka

[81] The trlsyllabic genitive plural (especially partitive), which switches accent from the initial position in the disyllabic nominative singular, becomes a syntactically and prosodically appropriate, and therefore frequent, formation at the end of colon 2.

This combination of distribution rules, both deriving from right justification in the inner metric of the deseterac , yields for our 5,107-line sample the frequencies of principal metrical types in the second colon shown in table 10.^[82] As a generalization, we can say that shorter-longer sequence is preferred in the second hemistich of the decasyllable, with approximately an equal number of ninth-syllable reversals and symmetrical (3/3) cola. These distributional rules override normal prose word order when the two come into conflict, although in many cases there is no conflict and customary word order is maintained.^[83] The major point is that the inner metric of colon 2 reveals a texture ultimately attributable to right justification but also amounts to a tradition-dependent set of constraints.

This characteristic texture becomes ever more apparent as we examine the distribution of elements in colon 2 more finely. Proclitics, such as prepositions, the conjunctions i (and), a (and, but), and ni (neither, nor), and the negative particle ne , are unaccented and cannot be treated metrically or grammatically as single words. From a prosodic viewpoint, a proclitic joins with the word that follows to form an accented unit within the colon and line. Thus the following examples of two-element second hemistichs:

	(C. 19)
The young man jumped to his nimble feet
	(C.23)
He took his horse from the stable

In the syllabically more spacious second colon this proclitic binding , as I shall call it, participates most often in noun doublets, noun-adjective pairs, longer (often prefixed) verbs, and prepositional phrases. Since the two-word units formed by this process of amalgamation are metrically equivalent to one-

[82] For reasons discussed below, I have included in this table those 2/4 and 4/2 cola that form the longer increment by combination with proclitics, as well as those trisyllabic pairs (3/3) of the shape proclitic-disyllable-trisyllable.

[83] Cf. Maretic^[*] B:123-64 and D:28-64.

element units of the same syllabic extent, we may include the proclitic (p ) groups with their single-word counterparts in the order shown in table II.^[84] It is worth noting that the proclitic pattern plays a particularly significant role in both the 4 + 2 and the 3 + 3 configurations, accounting in each case for up to half of the examples located, and further that these three sequences taken together make up approximately 80 percent of all second hemistichs in the 5,107-line sample.^[85]

After these major colon-patterns are recognized, the remainder of our sample breaks down into sparsely populated and for the most part statistically unimportant categories. The only category worth tabulating here is that involving three two-syllable words (2/2/2), a sequence that follows the expectable rule of initially accented disyllables in final position and in which the pattern distribution is reasonably consistent.^[86]

Likewise, enclitic binding , by which an unaccented word "inclines" on a preceding accented word to form a grammatical and prosodic group, proves statistically insignificant, not sufficiently affecting any one pattern to merit categorical analysis. Of descriptive importance, however, arc the facts that these accentless elements (1) usually follow the word order of the spoken language, (2) almost never occupy the fifth position (at the beginning of the second hemistich), since this placement would amount to a bridged caesura, and (3) are much rarer in the second than in the first colon.^[87] Whereas the

[84] I include herein only the statistically prominent proclitic patterns.

[85] When the word following a proclitic has a falling accent on the first syllabic, the grammatical stress shifts to the proclitic, whether or not it is expressed in the medium of the sung line. Since that shift would place grammatical stress on syllables 5 (patterns 4/2 and 3/3) and 7 (a much lower percentage), we may see proclitic binding as acting in concert with right-justification rules.

[86] Text A = 3.5 percent, B = 3.6 percent, C = 4.5 percent, and D = 6.0 percent.

[87] For further remarks on enclitics in the spoken language, see Browne 1975. On the rarity of an enclitic bridging a caesura, see Maretic^[*] A:63 and C:210; similarly, I find only three examples of a fifth-syllabic enclitic in 5,107 lines of the Stolac metrical sample. As for relative frequency of occurrence of enclitics, 85.1 percent are found in the first colon.

TABLE 12. Principal Metrical Types, Colon 1 (in Percent)
Colon 1 Pattern	A	B	C	D
2/2	25.2	25.1	22.6	33.6
p /3	14.6	15.8	19.1	15.4
3/	8.4	7.3	3.1	5.5
TOTALS	48.2	48.2	44.8	54.5

process of proclitic binding, the joining of an unaccented word to what follows it, plays a relatively important role in pattern distribution throughout the line, that of enclitic binding is primarily confined to the first hemistich.

Colon 1

In general, the role played by proclitic and enclitic binding reflects the relative lack of prosodic and phraseological fixity of the first as opposed to the second colon. Partly because its brevity precludes setting up regular sequences of nouns and noun-adjective combinations, the opening hemistich reveals a much higher degree of pliability, both in its accommodation of unaccented words and in the large number of lightly populated categories or sequences of elements. This flexibility is a typical manifestation of fight justification in the line as a whole, the first hemistich being more loosely organized than the second. At the level of the colon, we may expect the first four-syllable unit to manifest some evidence of greater firmness or regularity toward its end, much as the six-syllable segment showed above. And in fact, the prominence of the first- and third-syllable ictus is, as mentioned earlier, a sign of right justification.

As an initial approximation of the most important patterns, consider the distribution of the three most frequently occurring (table 12).^[88]

The most common pattern is a balanced one of two disyllables which follows the penultimate ictus rule in its internal arrangement (svlaci ), while the second and third both consist of what Maretic^[*] calls neprave cetvorosloznice , "facsimile tetrasyllables," formed by two words, one of which is unaccented. Neither of these accentual groups owes its formation to right justification, either by the shorter-before-longer or the penultimate ictus constraints that proved so important in the second hemistich. On the basis of these most heavily represented categories, then, we derive very little evidence of right justification apart from the tendency toward penultimate ictus and the very variability of inner metric patterns.

It is important to notice from the start that this flexibility promotes two different but complementary trends in colon formation. The more obvious one will amount to a synchronic freedom for the composing poet, affording him a section of the line that will remain open, for example, to syntactic adjustments.^[89] This is not to say that formulaic structure will weaken or lapse at line-beginning, but, in terms widely used in current scholarship, that verbal repetition is likely to be more systemic than verbatim. The less obvious result of first-colon variability, a corollary to the former trend, will be greater differences in prosodic structures from one text to another and from one singer to another. If this initial section of the line is more open to individual or idiolectal habits of composition, in opposition to the greater regularity of the second-colon inner metric as imaged in its dependence on a few well-represented patterns, then the first hemistich must also prove the site for a degree of prosodic chauvinism. We see some evidence of this latter trend toward individual habits of composition in the figure for the 2/2 balanced pattern over the four texts.^[90] With the first colon revealing a greater degree of flexibility, we expect and indeed find a correspondingly greater degree of individuality in prosodic structures. As an example of this individuality, consider table 13, which displays word-types that begin the first hemistich. Some of the percentages most divergent from the averages, notably occurring only in texts C and D, are italicized.

Summary: The Deseterac

In gathering together our findings and observations on the decasyllable, it is well to recognize the tenuous nature of outer metric as a true, definitive aspect

[89] We may contrast the less flexible second colon, for example, by noting the tendency for second-colon noun-epithet or proper name combinations to be inflected only in the final word. The principle is similar to that of the idiom's characteristic formation of colonic "words" from individual lexical items, a process that, as we shall see in chapter 4, is also typical of Homeric phraseology.

[90] While the relative frequency of the disyllabic sequence over the A, B, and C texts remains approximately the same (22-25 percent), the corresponding percentage for D is considerably elevated, representing an increase of between 8 and II percent or between 51 and 70 lines in a text of 637 lines.

of the prosody. At best, the putative podic structure of the deseterac remains a descriptive approximation, much as the six-metron model formally characterized the hexameter. But whereas the Homeric line responded in a limited fashion to this approach by yielding a profile of dactylic-spondaic substitution, the decasyllable reveals no such clearly defined podic units. Scholars have labored to impose on the Serbo-Croatian line a regularity at the structural level of the Greco-Latin foot, but the verse resists all such attempts, forcing a retreat to unconvincing portraits of trochaic pentameter tendencies and the like. What remains after a fair examination of outer metric in the deseterac is a firm and unambiguous constraint of syllabic count, a constraint that makes the line absolutely regular syllabically, with all but a very few hyper-metric and hypometric lines demonstrably the product of phonological ornamentation and traditional "error." And, as we have seen, these apparent exceptions actually prove the ten-syllable rule rather than bringing its consistency into question.

The same exceptions, when added to the overwhelming regularity throughout the 5, 107-line sample, assist also in proving the systematic recurrence of the caesura or word-break between syllables 4 and 5. This key to the inner metric of the deseterac thus combines with the equally stringent syllabicity rule to mirror the two corresponding principles first found in the reconstructed Indo-European line and also reflected faithfully in the Homeric hexameter. The caesura divides the decasyllable into two hemistichs or cola, the first of four and the second of six syllables, and their regularity as units derives directly from the demonstrated consistencies in syllabic extent and word-break placement. As already mentioned, the deseterac thus involves an externally simpler system of cola than does the hexameter, since the Homeric line divides first into two hemistichs and then into four cola by means of multiple A, B, and C caesuras, although the cola of the decasyllable are, as we have seen, correspondingly complex in their inner make-up. We should note in passing that this examination of the Greek and Serbo-Croatian colon formation typifies the kind of comparison-contrast advocated throughout these studies: the two verses are alike in prescribing regularly recurring intralinear

units that give each prosody a particular texture, and each is distinctive in prescribing certain tradition-dependent characteristics of that texture.

As for the third Indo-European and ancient Greek principle, that of fight justification, the grossest evidence for a trace of its operation in the deseterac consists of the very 4 + 6, shorter-before-longer make-up of the composite line. Even more important for our view of the verse form as the foundation of a traditional diction is the contribution of right justification to the placement of ictus. As shown in detail above, each colon, and particularly the second, maintains a penultimate ictus (on syllables 3 and 9) and an initial ictus (on syllables 1 and 5) that derive from the smaller-to-larger, looser-to-firmer patterns of the line as a whole and of each hemistich individually. With lotus on all odd syllables except the seventh, the assumption of a trochaic pentameter—a podic, outer metric structure familiar to metrists intent on viewing the decasyllable in the canonical Greco-Latin context—seems outwardly logical, but in fact it proves diachronically and dynamically impertinent. Viewed on its own terms, the seventh-syllable ambiguity is not a troublesome flaw in the otherwise straightforwardly trochaic base of the line, but rather a syllable position left unemphasized by the relative distribution of ictus in neighboring positions. In the Serbo-Croatian line, right justification prescribes two cola in an asymmetrical arrangement, the shorter preceding the longer, and further provides for recurrent prosodic ictus on syllables 1, 3, 5, and 9 while forbidding ictus at positions 4 and 10.

Within the second and more extensive hemistich, the same archaic principle finds expression in two complementary distribution rules: (1) syllabically shorter words before longer and (2) initially accented disyllables at positions 9-10, which comprise the conclusion of the ending cadence and collectively constitute a zeugma or bridge. That is, shorter words will precede longer words unless the shorter element is an initially accented disyllable, in which case the disyllable may (or may not) seek the line-ending zeugma by virtue of the coincidence between its lexical accent and the penultimate ictus, thus reversing the more usual order and leaving the syllabically longer element in first position. Although the second hemistich might seem to present greater opportunities for variety in internal design than the less spacious opening hemistich, in fact the longer unit shows itself more conservative prosodically: fully 80 percent of the Stolac sample falls into one of three major colon-patterns—2 + 4, 3 + 3, and 4 + 2—with proclitic binding tending to work toward one of these.

Colon 1, in contrast, utilizes both proclitic and enclitic binding in its general tendency to make larger prosodic elements out of shorter ones. This part of the line is characterized by its relative variability, just as were the first and third cola of the hexameter, with that diachronic flexibility providing a site for synchronic fashioning of the verse in a particular textual situation. Under such conditions we would not expect major colon-patterns to emerge, and

indeed the assortment of observed sequences is made up largely of a great many statistically insignificant categories. But right justification takes shape in more than the shorter-before-longer, highly variable nature of the unit as a whole; we also observe a marked preference for shorter elements (monosyllables, disyllables, and unaccented monosyllabic proclitics) at the onset of the initial hemistich. Even when a trisyllable opens the first colon, it almost always gives way to an enclitic in position 4 and so forms a prosodic tetra-syllable. Within an overall flexibility, then, we do perceive some manifestations of right justification, although they are by no means as pronounced as in the more conservative second hemistich.

Overall, the deseterac takes its prosodic cue from the Indo-European (and Homeric) features of syllabicity, caesura (implying colon formation), and fight justification. As a general observation, then, we might logically expect the two epic prosodies—having at least this much in common as well as a respectable number of disparities—to support comparable phraseologies, that is, poetic dictions that will reveal significant similarities alongside inevitable, tradition-dependent differences. In the history of the evolution of Oral Theory, moreover, this is precisely the Hellenic-Slavic linguistic combination out of which Parry and Lord first forged both the methodology and the interpretive theory that were to serve so many other traditions. Whether the ancient Greek-South Slavic synthesis can be called on to bear such a comparative burden without the aid of other comparanda is a vexed question, as is that concerning the role to be played by other poetic traditions. Let us begin an informed and meticulous answer to both questions by asking how the prosodic descriptions so far developed relate to the prosody of the Old English Beowulf .

The Old English Alliterative Line

As with the hexameter and deseterac , the search for prosodic structure in the Old English line is primarily an investigation into what, in Parry's famous definition, the "same metrical conditions" might mean in the diachronic development and synchronic morphology of the verbal formula. Toward that end I shah once again concentrate on those compositional parameters most influential in the shape of phraseology, seeking where appropriate to distinguish this prosody from each of the others and to ascertain its tradition-dependent nature.

The Beginnings: Sievers and Some Basic Principles

Since before the time of Eduard Sievers's Altgermanische Metrik (1893)^[91] almost one hundred years ago, the alliterative line of Beowulf has occasioned

[91] See also his earlier article, "Zur Rhythmik des germanischen Alliterationsverses I" (1885). A convenient translation of an article containing the major premises of his system is his "Old Germanic Metrics and Old English Metrics" (1968).

a large number of metrical theories, with almost no consensus among their proponents. On the basis of comparative Germanic evidence and versificational features of stress and alliteration, Sievers sought to rationalize the enormously variable unit of the alliterative line into five archetypal patterns. Each of these patterns was to represent and account for certain of the half-lines (or verses ) in the poetry, and collectively they were to comprise the entire metrical foundation for the poetic canon. To say that Sievers's prescriptions were at first accepted is misleading: in fact they were accorded the status of law, and textual emendations, for example, were founded on a supposedly necessary agreement between the metrical abstraction and the received manuscript text. Later years have seen the certainty about these canonical rules fade somewhat, and yet Sievers's basic conceptions are still deeply ingrained in some much more recent influential work on Old English metrics (e.g., Cable 1974). No scholar who proposes to treat Old English prosody can avoid coming to terms with his theory.

Sievers's "Five Types" consist of verse- (or half-line-)length patterns divided into two "feet." The first three have what he calls equal feet:^[92]

and the other two are composed of unequal feet:

Sievers's basic assumptions in assigning prosodic values have become virtually universal among metrists. First, the most fundamental unit of prosody is stress , indicated by an acute accent (s^[*] ) for primary or strongest stress-emphasis and a grave accent (s) for secondary but still major stress; s^[*] marks a syllabic bearing minimal or no ictus. This is the initial point, and it will prove a crucial one: the atom or "prosodeme" of Old English meter is not the syllable or mora of such quantitative meters as the hexameter and deseterac , but rather the stress. Second, Sievers and others reached the hypothesis of the half-line or verse as the most basic metrical unit by observing the other

[92] Line numbers with the notations a and b following the numeral refer, respectively, to the first and second verses in a given line.

[93] Klaeber underdots the o in the second syllable of morporbed to indicate that he considers the syllable syncopated.

indispensable feature of the meter—alliteration. As mentioned in the first part of this chapter, alliteration is not a desideratum but a requirement in the prosody of Beowulf and other Anglo-Saxon poems: unless there exists an agreement of initial stressed sounds between verses, an Old English line simply is not metrical. Alliteration and syntactic units, in fact, furnish the criteria for editing the run-on prose of the Cotton Vitellius A. xv. and other Anglo-Saxon manuscripts into poetic lines and half-lines, yielding a passage such as the following (Bwf 2401-2405):

Gewat pa t welfa sum t orne gebolgen
d ryhten Geata d racan sceawian;
hæfde pa gef runen, hwanan sio f æhð aras,
b ealonið b iorna; him to b earme cwom
m aðpumfæt m ære purh ðæs m eldan hond.

Then a certain one of twelve went, bitterly angered,
Lord of the Geats, to examine the dragon;
He had heard whence the feud arose,
The people's pernicious enmity; to his bosom came
An illustrious treasure-vessel through the informer's hand.

With the alliterating elements (or staves) underlined and space marking half-lines, we can see that the first (a) and second (b) verses alliterate in every case. The a-verse can, optionally, have two staves instead of one, but this property does not extend to the b-verse.

Sievers assigns his stresses systematically to the alliterating elements and other grammatically significant words in the line, such as other nouns, adjectives, adverbs, and verbs.^[94] The phonological criteria are straightforward: if a syllable is long by nature (with a long vowel as its core) or by position (its short vowel followed by two or more consonants), it is stressable. Thus, for example, in the passage above we observe a number of stressable words with initial syllables long by nature: Geata, sceawian, , cwom, , maðpumfæt ; and others in the same category but with prefixes (ge - and a -): Gewat, gefrunen, aras . We also notice words long by position: twelfa, torne, dryhten, hæfde, biorna, bearme, meldan, bond , and a prefixed counterpart, gebolgen , all of which are equally eligible phonologically and grammatically to bear stress on their root syllables. Such are the lexical items of primary importance in the Old English poetic line.

Of course, not all of the half-lines in Beowulf or any other Anglo-Saxon poem maintain a one-to-one correspondence between metrical position and syllable. In the unemended text of Beowulf a verse may consist of from two to ten syllables and a whole line of from seven to sixteen. Thus it is that Sievers and all metrists after him have had to admit to their prosodic descriptions twin.

[94] Rarely a word of lesser grammatical importance, such as a possessive pronoun, will serve as a stave.

rules which we may label resolution and ramification . Resolution entails the distribution of stress over two syllables if the first one is short by nature and position and therefore cannot itself bear ictus. For example, dracan in line 2402b of the passage quoted above cannot answer the metrical description of a trochee, or s^[*] s^[*] , because its first syllable is short and cannot by itself bear stress. Since, however, the word occupies the stave position in the b-verse, alliterating with dryhten in the opening verse, it must as a significant prosodic item in the line somehow shoulder a major stress. Resolution allows the word to take the prosodic shape , with the stress distributed over both syllables rather than localized over the first one, as in dryhten^[*] , for instance. The second and complementary rule of ramification accounts for the proliferation of short or unstressed syllables in the various minimally stressed positions among the Five Types. The infinitive sceawian^[*] , with its two unstressed syllables, provides an example of how ramification—like its fellow principle, resolution—can extend a single abstract pattern to a group of related line-occurrences; the syllable count may vary, but the basic type prevails.

The Idiosyncratic Nature of Old English Meter

With only these few broad generalizations about the shape of the meter, it becomes clear that the Old English line is very different from the hexameter and deseterac . For one thing, we observe an extremely large variation in syllable count, with no apparent restriction on when a certain length is to be used; this allows the poet to juxtapose such lines as Beowulf 51-54:

secgan to soðe, selerædend e,
hæleð under heofenum, hwa pæm hlæste onfeng.
        Ða wæs on burgum Beowulf Scyldinga,
leof leodcyning longe prage

To say in truth, hall-counselors,
Heroes under the heavens, who received the burden.
        Then was in the strongholds Beowulf of the Scyldings,
Beloved nation-king for a long time

This sequence includes in succession lines of ten, thirteen, ten, and eight syllables. As for the half-line subdivisions (indicated in the quoted passages by spaces), these verses, like the lines they compose, have little or no syllabic definition. Nor are the half-line units necessarily symmetrical in length or structure; they seem rather to have a semi-independent metrical life of their own.^[95] As indicated above, the prosodeme on which the meter is founded proves to be the stress-position rather than the syllable or mora, and the rather

[95] This half-line model of metrical organization accounts for the attempt of most metrists to formulate a theory of rhythm in terms of the verse unit. It also accounts for most investigators' focus on the verse as the length of the Old English formula. The so-called single verses—many of these combine with whole lines to form "triplets"—are evidence of a hybrid line structure that exhibits both a whole-line and half-line identity. See further Bliss 1979; Foley 1980b.

loose paratactic relationship of the two verses is formalized by required alliteration.

Unlike the hexameter and deseterac , then, the alliterative line obeys no rule of syllabicity; indeed, the expansion rules of resolution and ramification work directly against syllabic consistency. And since the half-line division is correspondingly variable, we can speak neither of a caesura, which would have to occur in a regular spot in the line, nor of the cola demarcated by a caesura. Obviously, under such conditions the notions of anceps and right justification are totally without meaning. Moreover, this synchronic portrait also indicates that, diachronically, the alliterative line has developed much farther away from a possible Indo-European precursor than have the ancient Greek and Serbo-Croatian verse forms. What intervened in this development was the shift of prosodeme from syllable to stress during the Common Germanic period, the various results of which are described by Winfred Lehmann ([1956] 1971, esp. 23-63). The birth of Germanic alliterative verse, he reminds us, was coincident with the shift of stress from a variable position to the initial syllable of a word. Under these circumstances, the Indo-European characteristics imbedded in a syllabically regular line-type would be lost. The general conclusion to be drawn is that the line of Beowulf is a verse form tellingly different from those employed by Homer and the guslar , and that the difference manifests itself both synchronically in the evidence of the text and diachronically in the history of versification.

Where does this catalog of idiosyncrasies leave us as we enter on a description of the alliterative line, which we aim to make sufficient both to allow comparison with the hexameter and deseterac and later to provide a basis for understanding formulaic structure in Beowulf ? The approach from Indo-European features is blocked by the linguistic reality of the stress shift and its foregrounding as the functional kernel of Germanic prosody. Indeed, it cannot be overemphasized that this is a stress rather than a quantitative meter, a prosody that depends on the stressed position rather than a sequence of syllables for its identity. It is obviously illegitimate to impose a Greco-Roman podic model; outer metric has already proven a treacherous because finally external and superficial concept, and it surely has no possible application here. Nor does the distinction of inner metric—the foundation of formulaic structure in the other two traditions—succeed in addressing the metrical issues of the Beowulf line; without consistent syllabicity and a regular caesura- or diaresis-system and its assortment of cola, there can be no inner metric. And yet, if we are to examine Old English traditional phraseology we must confront the prosodic nature of the formulaic structure on its own terms, just as was done above for the Greek and Serbo-Croatian traditions.

The Metrical Foundation

If the history of the alliterative line precludes an approach similar to that employed in studying the hexameter and deseterac , we would do better to shift

focus and attempt to determine what it is about the line that does remain constant from instance to instance. Besides the alliteration mentioned above, most scholars agree that the verse form requires four heaviest or primary stresses per line, and two per half-line:

s s | s s

where s is a syllable or syllables bearing primary stress (s) and the rest of the line consists of a varying number of secondary (s^[*] ) and minimal stresses (s^[*] ). In many cases, this much information allows complete and unambiguous scansion, as in this example from the earlier passage:

(Bwf 2402)

All four primary stresses arc assigned, two to the alliterating elements and the remaining two to the phonologically stressable core syllables of grammatically significant words; resolution and ramification arc active in the second verse, yet the line takes a relatively simple shape with a fundamentally trochaic pattern.

But the issue of line- and verse-types rapidly becomes more complicated. As we pass beyond the recurrence of the four "stress maxima" (or SMs), we must deal with what has proven a bewildering variety of permutations. Not only are the most basic features of the line wholly different from those of the quantitative, colonic Greek and Serbo-Croatian prosodies, but the ways in which these different features—SMs, the variable number of secondary stresses and unstressed syllables, resolution, and ramification—combine and recombine to produce the lines of Beowulf are also idiosyncratic. To begin to appreciate this aspect of tradition-dependence, consider that

(Bwf 2404)

and

(Bwf 1975)

are, from a metrical point of view, equivalent phrases. Both answer to the requirements of Sievers categories A plus B, even though their syllabic distribution and secondarily stressed positions are unequal:

Basic Pattern / x / x | x / x /

It becomes clear that any attempt to catalog line structure, to seek its systematic and recurrent sequence, must resort to a series of generative patterns. Sievers recognized this necessity for an expansible system from the start and made it a feature of the Five Types, and all later metrists have in various ways responded to the same obvious need.

Patterns and Systems

Among these scholars, John C. Pope offered in 1942 a sweeping and significant revision of the Five-Type catalog. Claiming four isochronous

measures to a line (and two to a half-line), Pope discovered that by beating a regular cadence there emerged an unvocalized stress at the head of the metrically acephalic B and C verses.^[96] In other words, he substituted a rest for the "missing" initial stress and thus brought both parts of the line into accord by supplying initial ictus where there was no lexical item to bear it. To use his own example ([1942] 1966, 39) alongside the Sievers reading, consider:

Sievers B: Basic Pattern xx/|x/
Pope B: Basic Pattern (/)xx|/x\

He proceeded to apply the initial stress-rest, marked ('), or stress taken in vocal silence, to all types of B and C verses, using as a leveling device the theory of isochrony among the four measures of a line. In experimenting with various lines in the poem, Pope ([1942] 1966, 247-409) argued that this realignment of stresses expressed the natural rhythm of the alliterative line more faithfully than the Sievers system.

From the hypothesis of initial rests in Types B and C and general isochrony, Pope derived another and more daring theory—that of the use of a harp^[97] to accompany the performance and specifically to mark time. Although instrumental accompaniment had certainly been considered before,^[98] no one had proposed that the lyre actually bore one of the major stresses in the line during a vocal rest. Whether the Old English scop really used an instrument in his performance, and what the musical aspect of his narrative might have been like, are problems probably beyond our ability to solve given the present state of knowledge. And although most metrists now resist the harp hypothesis in an effort to fashion theories with as few ambiguous or uncertain features as possible, it is interesting that Pope was able to suggest the use of the lyre purely on metrical grounds. Once the isochrony and initial rest are accepted, the way is left open to the possible participation of a device to mark rhythm.

It is not difficult to see that Pope's theories consist, from our comparative vantage point, of attempts to supply another kind of regularity to a poetic line ungoverned by the more familiar parameters of syllable count, caesura, and the like. The principle of four isochronous measures provided this consistency and rationalized the apparent near-chaos of syllabic count, stress distribution, resolution, and ramification. In place of six metra or four cola, or of five feet or two hemistichs, one had four recurrent rhythmic units; some may have been different, some may have begun with a stress taken on the

[96] Pope ([1942] 1966) acknowledges his debt to the work of Sievers and Andreas Heusler throughout; see the latter's Deutsche Versgeschichte, mit Einschluss des altenglischen und altnordischen Stabreimverses (1925-29).

[97] Called se hearpa by the Old English poets, the instrument probably more closely resembled a lyre; see Bruce-Mitford and Bruce-Mitford 1974.

[98] E.g., Sievers (1893, 186ff.), who doubted that any accompanied poetry had survived.

harp, but all were rhythmically equivalent in extent. In positing isochrony and in thus changing the internal divisions in the Five Types (for example, Sievers B or x / | x / becomes (/) x | / x \ and Sievers C or x / | / x becomes (/) x | / \ x), Pope did more than modify "feet" to "measures." More importantly, he attempted to formulate not just a description that suited the observable facts, but a metrical theory that would explain them. However we judge the value of his innovations, we must admire his sense of an underlying order and of the importance of reaching beyond the metrist's catalog to the aural reality of the poetry.

Some sixteen years after Pope's work first appeared, A.J. Bliss argued for a return to Sievers's major principles in The Metre of Beowulf (1967). Having undertaken a thorough re-examination of the poem, he explicitly dismisses the hypothesis of isochrony and derives Types B through D from the various displacements of stress in Type A, the core pattern which he considers the "norm of Old English verse" (p. 108). He also groups half-lines into light, normal, and heavy categories, depending on whether they contain one, two, or three stresses. Apart from his philologically indefensible discussion of the so-called caesura, which as we have seen cannot by definition exist in the Old English poetic line, Bliss succeeds in recataloging the lines of Beowulf into the Sieversian scheme; but of course this exercise cannot fail, because the Five Types had already proven themselves an adequate descriptive metaphor, if not an explanatory system.

More successful in evaluating the systematic nature of Old English meter was Robert P. Creed's "A New Approach to the Rhythm of Beowulf " (1966). Agreeing with Pope on the issues of isochrony and initial rests, Creed goes a step further in elevating the measure to a metrical unit. This is an important rationalization: while Pope posited measures of equal temporal duration, he still based his morphology of metrical types on the half-line or verse. Creed, in contrast, sees the measure as the kernel of prosody from a functional as well as descriptive point of view. With a few modifications made since the original article was published,^[99] he proposes seven basic measure-types:

[99] For a more recent report, see Foley 1976a, 1978a, and Creed 1989.

One can see the roots of this system of units in Pope's catalog of half-line types: the a represents one element of Sievers-Pope A; a + is an augmented version of a ; and g and b consist of the second element of Pope's D1 and D2 respectively. It may not be so obvious, but even Creed's is equivalent to the first part of Pope's B or C. The d measure, which includes a vocal rest and minimal harp stroke, allows Creed to scan single-syllable measures and maintain a basically trochaic rhythm, while the h is a logical development from the initial-stress hypothesis intended to handle unstressed syllables at the head of a verse:

In this last measure-type, the two syllables in the prefix ofer- are themselves unstressed but coincide temporally with the stress taken on the lyre. Thus the necessity for anacrusis, the assigning of such unimportant syllables to the foregoing metrical unit (in this case to the end of the second verse of the preceding line), is avoided and the prosodic wholeness of the line maintained.^[100]

One of the more recent thorough treatments of Old English meter is that by Thomas Cable (1974), who presents a strong argument for a rigorous and systematic scansion of Beowulf that harmonizes with and builds on the original work of Sievers.^[101] Cable, however, goes far beyond a mere reworking of the Five Types, even proposing a few changes at the level of categorization; his concern (p. 85) is to establish "that Beowulf can be scanned with four metrical positions to the verse."^[102] If one accepts his revisions, the result, as shown in table 14, is a group of five contours that correspond to Sievers's Five Types, each contour denoting either rising (/) or falling (\) stress between successive positions (p. 88). Cable's most telling point is that the four positions inexorably generate the Five Types or contours, with the meter being fundamentally the

[100] On the subject of anacrusis in B and C verse-types (which until his work was the standard method of dealing with prefixed verbs in double-alliterating lines and similar configurations involving metrically isolated initial syllables), Pope remarks ([1942] 1966, 64-65): "When there is only one such syllable, it may be treated either as anacrusis or as the up-beat of the first measure, after a rest, the choice depending mainly on the degree of intimacy with the preceding verse. Since mere prefixes cannot be made to fill half a measure even after a rest, these should always be read as anacrusis.... Two unimportant syllables may be treated as anacrusis if the connection with the preceding verse is so close that the two verses together form a single long phrase." Note that the h measure solves these problems systematically, without having recourse to non-metrical and non-grammatical criteria.

[101] Geoffrey Russom's generative study of Old English meter (1987) came to hand just as this volume went to press; there was therefore no opportunity to take it into account in this chapter.

[102] For example, he shifts Sievers D2 verses (with the pattern / | / x \) to the E category. By position he means (1974, 85) "not only the main metrical stresses but also the intermediate stresses, if there are any, and the metrically unstressed sequences, of which there are often two and always at least one."

TABLE 14. Cable's Five Contours for the Line of Beowulf
	Contour	Sievers Pattern
A	1 \ 2 / 3 \ 4	/ x | / x
B	1 / 2 \ 3 / 4	x / I x /
C	1 / 2 \ 3 \ 4	x / | / x
Dl	1 \ 2 \ 3 \ 4	/ | / \ x
D2, E	1 \ 2 \ 3 / 4

underlying four-unit pattern. Instead of wrestling with what I earlier called the descriptive level of the Types, he understands the A through E verses as the inevitable product of the four-position system. With the single qualification that the second of two "clashing stresses" cannot be the heavier,^[103] he is able to predict contours and Types from positions: of the eight possible configurations, the only three that are not observed in Beowulf are those three in which the clashing stress rule is broken. Cable's system of explanation is satisfying because self-contained and inherently logical, and it deserves serious consideration by any scholar in search of the regularities of Old English prosody.

Prosody and Composition

From the metrical foundations of the alliterative line, on which all scholars agree, and from these various accounts of the systemic structure of Old English meter, we can derive a workable model for comparison with the hexameter and deseterac . To begin, we have learned that the quantitative, colonic verse form of the ancient Greek and Serbo-Croatian epic poetries contrasts sharply with the stress-based alliterative line, not only in the synchronic evidence of the Beowulf text but also in the diachronic reality of the history and development of Germanic verification. Indo-European features are not to be found in the Old English line because they could not suave the Common Germanic shift of lexical stress to the initial root syllable and consequent generation of a stress-based rather than syllabic meter. Synchronically, this shift of prosodeme from syllable to ictus means a line without syllabic constraint, without a caesura, and so on.

But we have seen that the verse form did evolve its own set of metrical regularities, and if we are to heed the caveat of tradition-dependence, we must follow out three regularities on their own terms. Fortunately, all metrists

[103] Clashing stresses are successive major stresses. See especially chapter 5, where Cable claims (1974, 73) that "the evidence of alliteration and syllabic quantity indicates that we shoed redefine those types with clashing stating explicitly that the first of two consecutive stresses must always be the heavier."

agree on the crucial recurrence of stress and alliteration—the two major characteristics of the Old English line—and also on the consistent half-line dimension of the prosody. Verses, in other words, stand together and they stand alone: they are bound into a whole line by the alliterative constraint, and yet they complementarily maintain—like so many subunits in oral epic tradition—an independent, integral aspect as well. Not unlike the Greek hemistich, the half-line retains a full prosodic and a normative syntactic unity; we need only think of the typical Germanic device of poetic variation to appreciate this unitary character.^[104] Of course, each half-line can vary tremendously in count and texture, and so we cannot summon the comparative notion of colon. What can be said is that a tradition-dependent subunit does exist and recurs consistently. We may add to these standard qualities the four major stresses (or SMs) in each line, occurring as they do according to regularly observed phonological rules of placement and extent.

As noted above, with these regularities—a stress-based meter, alliteration between verses, half-line units, and four stress maxima per line—we reach the limit of general consensus among metrists. From this point on, all seem to go their own way, whether to isochrony, reconstruing of Sievers's rules, the measure, or stress contours. Their catalogs reflect the nature of their individual procedures: Sievers's Five Types descriptively rationalize all lines but offer no real explanation; Pope's innovations add the regularity of isochrony (modifying Sievers's B and C verses), show attention to aural characteristics, and move toward explanation; Bliss refines Sievers's system to deal with the variability of types; Greed employs the rationalizing power of the measure to fashion a generative seven-item scansion system; and Gable shifts Sievers's D2 verses to the E category, posits a sequence rule for clashing stress, and explains the five contours as the inevitable development of four metrical positions per verse. All of these methods also represent responses to the twin principles of resolution—by which a stress may be borne by two syllables if the first one is metrically short—and ramification—by which the number of syllables in an unstressed position can increase markedly, to as many as five or six in some verse-initial configurations and very frequently to two or three in almost any position. With these two avenues of variation so open to syllabic traffic, the only possibility for systemic simplicity is through a generative series of patterns. Each metrist fashions his own, and, bearing the stamp of their makers, the resultant patterns seem mutually contradictory, or at least exclusive.

But even though their premises and explicative power may differ, the metrical theories we have summarized do "translate" from one to the next, and some basic correspondences among the ostensibly dissimilar descriptive

[104] On variation, see esp. Brodeur 1969, 39-70; Robinson 1985.

TABLE 15. Basic Correspondences Among Metrical Theories
	Sievers	Pope	Cable	Creed
A	/x|/x	as Sievers	1\2/3\4	aa
B	x/|x/	(/)x|/x\	1/2\3/4
C	x/|/x	(/)x|/\x	1/2\3\4
D1	/|/\x	as Sievers	1\2\3\4	dg
D2	/|/x\	as Sievers	1\2\3/4	db
E	/\x|/	as Sievers	1\2\3/4	gd

systems can be discerned (table 15). These, then, are the allowed half-line patterns, presented in each of the four major theoretical forms.^[105] They may be rationalized into measures, redivided into isochronous units with initial lyre strokes, or derived from a four-position series, but they all prescribe roughly the same permitted sequences of stresses for verse-types in Beowulf . Compositionally, this collection of verse-types will constitute the metrical foundation for our discussion of formulaic structure in Old English, with, synchronically speaking, the wide variety of actual lines developing from these patterns via the generative rules of resolution and ramification.

If these patterns constitute the prosody of Beowulf at the level of the half-line, what of the whole-line unit? On the basis of computer studies undertaken to analyze the meter of a machine-readable text of the poem,^[106] I have been able to prescribe three favored line patterns that, taken together, account for over 90 percent of the metrically recoverable lines of the poem.^[107] These line-types are the most commonly used combinations of the verse-types listed above, and represent the Beowulf poet's "choice" of patterns from among all possible combinations of verses. Table 16 indicates the make-up of each of the three paradigms in terms of all four metrical systems.

Paradigm 1, which accounts for 54.7 percent of the metrical text of Beowulf , consists essentially of an A verse followed by either a B or a C verse, the equivalent notation in Creed's system being aa followed by or .^[108] Like

[105] I do not, of course, mean to imply absolute equivalence among the four theoretical presentations, for there are, as indicated above, important differences in the criteria according to which they were formulated and the kinds of explanations they are intended to provide. But it is well to notice that, at the level of describing permitted verse-types, all four systems agree in their essentials.

[106] See note 99 above. I take this opportunity to thank David Woods, George Maiewski, and Dorothy Grannis for their assistance in programming.

[107] The metrical edition, prepared from the manuscript facsimile by Robert P. Creed and myself, eliminated about 2.3 percent of the poem on paleographical grounds.

TABLE 16. Metrical Paradigms in Beowulf
	Creed	Sievers	Pope	Cable
Paradigm I
Verse 1	aa	A	A	A
Verse 2	or	B or C	B or C	B or C
Paradigm II
Verse 1	gd or dgdb , bd	D1, D2, E	D1, D2, E	D, E
Verse 2	either I.1 or 1.2	A or B/C	A or B/C	A or B/C
Paradigm III
Verse 1	either I.1 r I.2	A or B/C	A or B/C	A or B/C
Verse 2	repeat	repeat	repeat	repeat

all verse-types and line paradigms, this abstract pattern can be filled out with a variety of syllabic complements, such as

(Bwf 483)

or

(Bwf 346)

—lines which would of course be scanned slightly differently in the Pope and Creed systems, with the B and C verses beginning with a stress taken on the instrument during a vocal rest and the fourth primary stress lowered to a secondary. Also, because the Old English alliterative line demonstrates a half-line as well as a stichic prosody, the paradigms may be reversed; even with the possibility of reversal, however, the prescriptive nature of the paradigms remains exacting. For paradigm 1, verse metathesis simply means B or C followed by A, or followed by a a :

(Bwf 1850)

The possibility of inversion argues implicitly for an associative relationship between the two verse-types; although the first and second verses arc not interchangeable at the level of phraseology, since the second half-line cannot tolerate double alliteration, at the level of prosody the half-lines or verses do seem to be interchangeable parts of the larger whole which combine according to paradigmatic rules.

The second of these line patterns, paradigm 2, combines a D or E verse with either half of paradigm 1, that is, either with an A type or with a B/C type:

Once again, the order of half-lines may be reversed, as in

Together these two versions of paradigm 2 cover another 24.7 percent of Beowulf . The third paradigm is also a recombinant form of the first line-pattern, consisting of either verse-type from paradigm 1 taken twice, that is, either AA or two successive B/C half-lines. Examples include

and

Half-line reversal naturally does not enter the picture here; paradigm 3 is the pattern on which 14.9 percent of Beowulf is founded, for a three-paradigm total of 94.3 percent.

The combination of and interrelationships among the verses comprising these line-patterns or paradigms indicate the specific prosodic texture of the Beowulf text that has survived to us.^[109] In conducting our investigation of phraseological structure in Old English epic, we shall be able to proceed directly to the metrical underlay by referring to the compositional habits of the Beowulf poet in precise and rational terms. But the most significant findings to have emerged from this analysis may well be the most general. First, what we have in Paradigms 1, 2, and 3 is a group of metrical formulas ;^[110] the Old English alliterative line as we have it in Beowulf consists not of a colon-based, quantitative meter with tradition-dependent features arising from the particular tradition's expression of right justification, but rather a verse-based, stress meter which figures itself forth in a set of multiforms. Second, the nature of these multiforms—in particular, their quality of reversibility—reveals that the poet composes in whole lines with verse substitution, that, in short, his making of the poem is a two-level process.

[109] We cannot extend the analysis any further, for example in an attempt to distinguish individual from general tendencies, because we lack the kind of comparative material necessary to the investigation. The 3,182 lines of Beowulf unfortunately stand alone as our only real example of epic in Old English.

[110] Cf. Cable's notion of "melodic formulas" (1974, 96ff.).

Coda

Over the next three chapters on traditional phraseology I shall be applying the insights gained in this treatment of comparative prosody to the ancient Greek, Serbo-Croatian, and Old English epic traditions. As we proceed it will be well to recall that while numerous parallels can be drawn among the three phraseologies, with productive results, we must also remember their individuality. Milman Parry's theory of the formula was, after all, based firmly on determining the "same metrical conditions" that made possible the recurrence of elements of diction, and we have discovered in the present chapter that each epic tradition has an idiosyncratic, tradition-dependent prosody that we may expect to exist in symbiosis with a correspondingly idiosyncratic and tradition-dependent phraseology . As noted in earlier chapters, and as will be evident throughout the volume, the double focus of comparison and contrast must inform the study of structure in oral traditional epic.