Five
Traditional Phraseology in the Serbo-Croatian Return Song

The foundation of oral epic in Serbo-Croatian is the traditional phraseology employed by the guslar as he makes his song. In studying this stylized poetic diction, I shall begin with a review of Albert Lord's theory of the formula and an analysis, based on that theory, of selections from two sample texts of Serbo-Croatian oral epic from the Milman Parry Collection of Oral Literature. With this summary and illustration in hand, I shall then proceed to a discussion of how the traditional prosodic structures examined in chapter 3 underlie the spectrum of formulaic diction and how that diction depends for its function and maintenance on a limited set of traditional rules. Along the way we shah look closely at the lines and cola analyzed in the second section in order to determine the range and morphology of traditional diction and to assess its nature and complexity as a linguistic instrument. As elsewhere, my major concern is the faithful, tradition-dependent explication of structure as a basis for aesthetic inquiry.

Albert Lord and the Concept of the Formula in Serbo-Croatian Oral Epic

Just as Milman Parry did for Homeric epos, Lord bases his theory of composition of Serbo-Croatian oral epic on the singer's verbal dexterity in manipulating formulas .^[1] This phraseological unit, which he defines after Parry as "a group of words which is regularly employed under the same metrical conditions to express a given essential idea" (1960, 30), is construed as the sine qua non of oral style—in the sense not only that it is typical of that style but also, and crucially, that it is necessary to that style. For Lord, as for Parry,

[1] All references here to The Singer of Tales (Lord 1960) come from chapter 3, "The Formula." For other studies of the formula in this tradition, see Foley 1985, s.v. SC [Serbo-Croatian].

the primary quality of the formula is its utility to the singer who composes in performance, and throughout his writings he has emphasized the role of usefulness over all other ancillary characteristics.^[2]

One telling formulation of the utility argument is Lord's description of formulaic diction as a language, more specifically as a grammar out of which are generated the actual verses in a song.^[3] Speaking of the systematic variations as well as the more obvious verbatim repeats, he remarks (pp. 35-36) that "in studying the patterns and systems of oral narrative verse we are in reality observing the 'grammar' of the poetry, a grammar superimposed, as it were, on the grammar of the language concerned." Lines are not simply repeated and memorized, then, but are used by the singer and tradition as patterns, and these patterns give rise to a skein of related phrases.^[4] From a functional point of view, the guslar has learned not a set of discrete phrases but a formulaic system that can be pressed into service in a variety of narrative situations, and that will both retain the fundamental design of the system and take on various situation-specific shapes.^[5] Thus, Lord's continuing reference to a grammar of oral poetry explains both straight repetitions and formulaic phrases as the performance products of underlying systemic patterns.^[6]

The grammar generates phrases—lines and cola—called formulas (defined above) and formulaic systems , the latter term referring to units that "follow the same basic patterns of rhythm and syntax and have at least one word in the same position in the line in common with other lines or half lines" (p. 47). Formulaic systems may be unearthed by collecting from a singer's repertoire phrases that satisfy this definition and displaying them as a substitution system. One of Lord's examples of such a group of phrases (p. 48), shown in abbreviated form in figure 4, involves reflexes of a certain verb "to mount" (zasediti ) and various words for "horse." The essential idea, "to mount a horse," remains the same throughout the series, as do both the syntactic pattern of verb followed by direct object and the metrical pattern of six syllables (3-3, filling the second colon). What changes is the nominal variance in the morphology of the verb^[7] and, more importantly because systematically, the specific lexical

[2] For a recent example, see Lord 1981.

[3] Cf. Ashby 1979 on Old French; and Conner 1972 on Old English.

[4] Thus he states (p. 36): "The fundamental element in constructing lines is the basic formula pattern"; he also contends (p. 43) that "there are two ways by which a phrase h produced; one is by remembering it, the other is through creating it by analogy with other phrases; and it may well be impossible to differentiate between the two."

[5] See his explanation on p. 43: "New formulas are made by putting new words into the old patterns."

[6] He argues (p. 33) that "only the necessity of singing can produce a full-fledged formula. The phenomenon of which it is a manifestation arises from the exigencies of performance. Only in performance can the formula exist and have clear definition."

[7] Note that this feature is not formulaic, according to Lord. Formulaic variation at this site in the colon would consist of another lexical choice with the same semantic content.

Figure 4.
Lord's zasediti Formulaic System

designation for "horse." This formulaic system thus answers all of the criteria for a formula except for the exact repetition of the "group of words." Strictly speaking, repetition of any one of the multiple possibilities given above would constitute evidence of a formula, while the multiformity of the group as a whole furnishes evidence of the underlying formulaic system.

These units of traditional phraseology, under the influence of the vocal and instrumental melodies to which the guslar learns to sing them,^[8] comprise a tissue of ideas to be sorted, adapted, and combined in the composition of Serbo-Croatian oral epic. And in keeping with his and Parry's original focus on utility as the functional reason for such a phraseology, Lord (1960, 34) identifies "the most stable formulas" as "those for the most common ideas of the poetry." Those which recur most often, those which show themselves most useful to the singer in performance will, he indicates, undergo little or no change. Thus formulas having to do with (1) common characters, (2) frequent actions, (3) the time when an action occurs, and (4) the place where an action occurs comprise a significant portion of the traditional staple.

In the first case, the names of characters tend to occupy six syllables and to fill the second colon (e.g., Kraljevicu^[*] Marko, beg Mustajbeg licki, etc.), using patronymics, grammatically unwarranted vocative inflections, or descriptive complements to eke out the colonic form; once formed and employed with any frequency, these lexical fossils—like their counterparts the Homeric noun-epithet formulas that Parry used to demonstrate the recurrent nature of that diction—are extremely unlikely to show much deviation. Typical actions, such as speaking to an assembly, mounting a horse, shouting in prison, and the like show somewhat more variation but are relatively constant. Likewise, the time and place involved in an action reveal yet more formulaic variation, since many different times and places must be accommodated in the syntactic and metrical patterns; yet these two latter categories also achieve some stability in the poetry.

[8] The dynamic relationship between melody and traditional phraseology is at best only partially understood. For some discussion, see Herzog 1951 (cf. 1940); Traerup 1974; Lord 1960, 37-38; Bynum 1979, 14-43.

Other parameters also assist in shaping and preserving formulaic diction. Besides the little-studied relationship between melody and the poetic line mentioned above, Lord describes the features of word-boundary pattern, syntactic balance, and a variety of acoustic phenomena, among them leonine or end-colon rhyme, colon-initial rhyme, alliteration, and assonance.^[9] Although he accords to none of these features (illustrated below) the status of prosodic or formulaic determinant, he shows how they function on an ad hoc basis in the composition process.

Word-boundary pattern^[10]

Most frequent syllabic patterns in colon 1: 2-2, 1-3, 4

Most frequent syllabic patterns in colon 2: 2-4, 4-2, 3-3

Syntactic parallelism

Kad tatarin pod Kajnidju dodje,

Pa eto ga uz caršiju prodje ,

When the messenger came down to Kajnidja ,

Then he passed along the main street ,

Leonine or End-colon rhyme

U becara^[*] nema hizmecara^[*] .

For a bachelor there is no servant .

Colon-initial rhyme

Zveknu halka a jeknu kapija.

The knocker rang and the gate resounded .

Alliteration

Na V isoko v iše Sarajeva,

At Visoko above Sarajevo,

Assonance

Još do zore dva puna sa ha ta ,

Even two full hours before dawn,

Once more, these morphological, syntactic, and phonological characteristics are active and relatively frequent aids to the guslar as he makes his song, and

[9] See further Lord 1956.

[10] See chapter 3 in this volume for an explanation of word-boundary patter as and their significance. The examples cited here are from the Stolac songs. Throughout this volume I translate the Serbo-Croatian historical present tense as a simple past.

in that way they have a generic identity in the singing tradition. But they are finally only accoutrements of formulaic diction—means by which that diction can achieve and perhaps maintain a certain stability^[11] —and are not as regular or as fundamental to traditional phraseology as is its prosody or its formulaic structure. Lord describes the characteristics exemplified above as providing a pathway of sound and structure for the guslar as he composes.

Taking account of sound patterns and the like leads Lord to a treatment of the major substructures of the line, the two cola or hemistichs. Understanding the length of the formula as either a whole line or a single colon,^[12] he notes that the more spacious six-syllable second colon typically contains most of the noun-epithet combinations, while the initial four-syllable colon very often contains a verb and begins with a conjunction. As for the relationship between the two segments of the line, he remarks (1960, 42) that "the second half of the line is dependent not only syntactically on the first, but is also to some extent suggested by the sound patterns with which the line opens." In this and other ways, the two cola, metrically unequal in both extent and inner structure as we have seen in Chapter 3, tend to parcel out syntactic duties differentially.

Relying on the ideas of a grammar of oral verse-making and of the primary role of utility as the bases of his concept of traditional phraseology, Lord, again following Parry, performs a formulaic analysis of fifteen lines from one version of The Song of Bagdad by Salih Ugljanin (SCHS 2, no. 1, lines 789-803). The referent for his investigation consists of eleven different songs by Ugljanin, or a total of twelve thousand lines; three of the songs were recorded on phonograph disks, four were recited (not sung) for the records, and four were taken from dictation.^[13] Lord's figures indicate that approximately one-quarter of the (whole) lines and one-half of the cola in the passage are repeated verbatim in the referent; these, then, are the formulas. Yet according to the criteria prescribed, there is absolutely nothing that cannot be shown to be formulaic; every one of the lines and cola can be traced to a formulaic system by reference to the twelve thousand lines used for comparison. Furthermore, he argues (1960, 47) that had he increased the size of the referent and included within it the songs of other singers, it would have become apparent that

[11] Lord puts it this way (p. 57): "A perfectly natural consequence of building passages by syntactic parallelisms and acoustic parterre is that passages so built tend to have a comparative stability, or better, a continuity in time both in the habit of the single singer and, to a lesser degree, in the current of a tradition."

[12] He observes (p. 34) that "the length of the action formula is naturally in part determined by whether the subject is expressed in the same line and by the length of the subject." His analyses of formulaic density (pp. 46, 58-59) reflect the same conception.

[13] In choosing the sample, he notes that he sought to avoid both the stylized proem or pripjeo that customarily begins Serbo-Croatian Moslem oral epic and the more frequently recurring themes in order to test a passage that would have no special narrative reason to repeat as a whole and thus skew the analysis.

"almost all, if not all, the lines in the sample passage were formulas and that they consisted of half lines which were also formulas."

This demonstration and the accompanying extrapolation from the observed figures to the conclusion quoted above comprise the locus classicus on which scholars from many disciplines have based the often-emulated analysis of formulaic density as a test for orality.^[14] Lord's quantitative investigation of a known oral epic poetry culminated in the proof that this oral epic tradition could be entirely explained in terms of the units Parry had called the "formula" and the "formulaic system." And if one proceeds according to his definitions, there is nothing to gainsay his conclusions for Serbo-Croatian epic: one discovers no exceptions to the rules, no nagging percentage of lines that must be considered "original."^[15] It is then a natural step from analyzing colonic and whole-line formulas to interpreting couplets and even runs of lines as traditional multiform units. If, the reasoning goes, unambiguously oral epic is 100 percent formulaic and 25 to 50 percent (and potentially entirely) straight formula, and if the formula is really the essential "word" used by the singer to make his song, then an epic of unknown provenance that reveals a similar formulaic diction should also be oral. This is the basic credo of formulaic theory, as first postulated by Parry in reference to the Homeric texts and as tested by Lord in the living laboratory of Serbo-Croatian oral epic tradition. Thus it is that Lord can say (1960, 65) that "the grammar of oral epic is and must be based on the formula. It is a grammar of parataxis and of frequently used and useful phrases."

Another aspect of traditional parataxis crucial to the shaping and maintenance of formulaic phraseology is that general area encompassing the phenomena of pleonasm (more properly, "terracing"),^[16] thrift, and enjambement. All three phenomena are involved in the functional redundancy that operates at all levels of oral poetics, and each will bear brief discussion in relation to the formula. Terracing, to be illustrated at greater length in the third section below, "The Spectrum of Formulaic Diction," consists of the repetition in the following line of a word or words employed in an initial line. For example,

Kudgodj ide  knjiga šarovita,
Kudgodj ide , do Karlova sidje.
Wherever  the multicolored letter  went ,
Wherever it went , it came down to Karlovo.

Lord notes that such terracing or pleonasm is related to the "thrift" of formu-

[14] Some of the most authoritative analyses include Creed 1955 (Old English); J. Duggan 1973 (Old French); Webber 1951 (Spanish ballad).

[15] As for "errors," he sees such infelicities as "perfectly normal aberrations" (p. 38) in performance when perceived against the background of tradition; on analogous "errors," see chapters 3 and 10 in this volume.

[16] See Austerlitz 1958.

laic style (after M. Parry 1928a, in 1971, 83ff.), the principle which demands that a traditional diction harbor few if any metrically equivalent phrases for the same essential idea. His discussion of thrift is complex and I shall not attempt to rehearse it here; it will be sufficient to mark that Lord (1960, 53) recommends interpretation of a given phrase selection by considering "not only its meaning, length, and rhythmic content, but also its sounds, and the sound patterns formed by what precedes and follows it."

As for the implications of enjambement, Lord follows Parry in understanding the bound, integral linear unit as the natural expression of the traditional idea^[17] and therefore finds "necessary enjambement," in which the syntactic structure and thought arc incomplete in one line, to be the antithesis of the oral style. Complementarily, total lack of enjambement and unnecessary enjambement, the latter of which describes the case wherein syntax and thought are complete at line-end but are optionally continued in the following line, are understood as the hallmarks of oral traditional composition. He finds proof of this connection in a 2,400-line sample (Salih Ugljanin, in table 19); my own figures, based on one thousand lines of Serbo-Croatian oral epic from two Parry-Lord guslari from Stolac (Mujo Kukuruzovic^[*] : and Halil Bajgoric^[*] : in the same table),^[18] generally agree with his findings and support his conclusions. Although the three singers vary in their relative percentages of unnecessary versus no enjambement,^[19] the totals of these two typical features of oral style in Serbo-Croatian epic are quite close. The primacy of the line as a self-contained unit is secure, at least in Yugoslav and ancient Greek epic.^[20]

The final section of Lord's chapter on the formula is devoted to a description

[17] Cf. Parry 1929. See also Peabody 1975, 118-67; Clayman and van Nortwick 1977 (and, in response to the latter, Barnes 1979).

[18] Kukuruzovic^[*] 1.1-500; Bajgoric^[*] 3.1-500.

[19] As a matter of individual style (a subset of traditional style as a whole), Kukuruzovic^[*] (B) uses many more dependent clauses and vocative phrases than does Bajgoric^[*] (C), while Bajgoric^[*] employs many more couplets and additive runs. These differences account in part for the variance in unnecessary versus no enjambement figures.

[20] But see chapter 6 below on Old English traditional phraseology.

of how the traditional phrase evolved out of ritual into a metrical convenience and compositional device. He views the formula's roots as ultimately religious (1960, 67): "its symbols, its sounds, its patterns were born for magic productivity, not for aesthetic satisfaction." In suggesting this ritualistic origin and in stressing throughout the chapter the compositional utility of the phrase as we encounter it in a guslar 's repertoire, Lord echoes Parry's proscriptive emphasis on function and his admonition against interpreting Homer (and by extension his Yugoslav confrères ) as one would a literary artist fully able to construe his narrative in any style he sees fit to invent for the situation.^[21] How tenable such a position is depends, of course, on the fidelity of the description of formulaic phraseology on which it is founded. To assess the fidelity of Lord's description is the burden of the rest of this chapter.

A Classical Formulaic Analysis of Two Parry-Lord Texts

In order to illustrate Lord's analytical technique and to provide supplementary evidence for later discussion, I have conducted a formulaic analysis of two passages of twenty-five lines and twenty-six lines each, or a total of fifty-one deseterac verses. The passages are drawn from the performances of Stolac guslari Mujo Kukuruzovic^[*] (text 1, Ropstvo Ograscic^[*] Alije [The Captivity. of Ograscic^[*] Alija ], Parry no. 1287a, 1,288 lines, taken from dictation) and Halil Bajgoric^[*] (text 2, Halil izbavlja Bojicic^[*] Aliju [Halil Rescues Bojicic^[*] Alija ], Parry no. 6703, 637 lines, taken from dictation). I selected the first sample virtually at random, the sole criterion being avoidance of major recurrent themes. The second sample, however, was deliberately taken from an occurrence of a common theme in the Return Song, that of "Shouting in Prison,"^[22] in order to test the formulaic density and texture of a passage governed by such a typical narrative structure.

The referent for the formulaic analysis consisted of three additional song-texts, two by Kukuruzovic^[*][23] and one by Bajgoric^[*][24] . All but the last song

[21] See especially M. Parry, "The Meaning of the Epithet in Epic Poetry" (1928a, in 1971, 118-72), and counterpose Parry's own insistence (p. 22) that "it does indeed give a false impression of the character of this language to imply that its creation was, so to speak, a mechanical process" and Lord (1960, 54), on the singer: "The formulaic technique was developed to serve him as a craftsman, not to enslave him." Cf. Foley 1983b.

[22] See chapter 8 below for discussion of this typical scene; for its role in the overall story-pattern, see chapter 10.

[23] Ropstvo Alagic^[*] Alije (The Captivity of Alagic^[*] Alija ; Parry no. 1868, 2,152 lines), taken from dictation; Ograscic^[*] Alija u suzanjstvu^[*] (Ograscic^[*] Alija in Prison ; Parry no. 6617, 2,180 lines), sung for the records. Here and elsewhere I give standard Parry Collection titles, whether or not they seem strictly grammatical. Suppressed inflections are to be attributed to the singers' (compositional) habit of inflecting only the final element of a noun-epithet phrase, a phenomenon in turn explicable by the composite identity of the phrase as a single rec . See further chapter 3.

[24] Zenidba Becirbega^[*] Mustajbegova (The Wedding of Mustajbeg's Son Becirbeg^[*] ; Parry no. 6699, 1,030 lines), sung for the records. The analysis was conducted using a computerized concordance which I prepared at the Computing C. enter of the University of Massachusetts/Amherst. I am grateful to the center and to Robert P. Greed and Joel M. Halpern, who helped me to gain access to the system.

mentioned—that is, four of five texts, or 6,246 of 7,287 lines—are Return Songs, so that the selection of sample and referent generally followed the criterion of genre-dependence advocated in Chapter 1. In addition, both guslari are from the same "local tradition" of the Stolac area in southern Hercegovina, so we may reasonably expect some congruity in their individual singing styles.^[25] In order to put into relief what I shall call the "idiolectal" (individual) and "dialectal" (local tradition) levels of formulaic structure (to be discussed below in the fourth section, "From Prosody to Traditional Rules"), I have purposely limited the Bajgoric^[*] sample and referent to these two songs, only one of which is in the epic subgenre of Return; by comparing the density and texture of the two 25-line and 26-line passages from the two singers, we shall to an extent be able to judge the influence of the size and nature of the referent on formulaic analysis and, further, to test which phrases are shared by the two singers and which are (on available evidence) their own. Of course, figures and minor conclusions would change somewhat if the referent were enlarged or differently composed, but since we are pursuing structural principles as a basis for aesthetic inquiry rather than quantitative analysis for its own sake, we may take this configuration of texts as representative.

The first passage, that from Kukuruzovic^[*] (Parry no. 1287a, lines 1.829-53), details a common enough sequence of events without involving a true theme:^[26] a letter is written and delivered, and as a result a speech, preparation, and journey take place. The sample passage is marked after Parry-Lord practice, with solid underlining to indicate a formula and broken underlining to indicate a formulaic system , according to the definitions quoted above. All lines are analyzed for both whole-line and colonic units, and a literal translation of the text is provided.^[27]

[25] See Lord (1960, 49), on individuals and their local tradition.

[26] Many of these events could be treated thematically by the singer, but he chooses at this point simply to relate their occurrence and not to summon the typical scenes that might be employed to express their essential narrative ideas.

[27] Supporting evidence (examples rather than an exhaustive array of related phrases) may be found in the appendix (pp. 393-94). The double underlining in the following passages and discussion indicates formulicity by colon (upper) and by whole line (lower). Although I have tried to keep this formulaic analysis as much in line with that carried out in the Odyssey passage in the last chapter as possible, tradition-dependent factors make absolute congruency inachievable. As will be shown in the extended discussion of these two passages from the Parry texts, the cola of the deseterac phraseology interact differently from the cola of the Homeric hexameter diction. Nonetheless, I have observed all the formal prescriptions set in chapter 4, counting a given colon (or hemistich) as a formula if a whole-line formula can be demonstrated, or as a system if a whole-line system can be demonstrated (unless, of course, the given colon is a formula in its own right). As we shall see below, the complexity of the diction is better rationalized by traditional rules (which are tradition-dependent) than by conventional formulaic analysis. Inevitably, the consequence of my (often legalistic) application of "formula" and "system" will not satisfy all investigators. I stress, however, that this is a shortfall in the method of formulaic analysis itself, a problem that will be with us as long as we fail to take account of the spectrum of traditional phraseology.

Text 1

When Pero formulated the letter,
Then he found a messenger for the letter.              830
Wherever the multicolored letter went,
Wherever it went, it came down to Karlovo,
And to the hand of the ban's lady.

She looked over the letter, then she laughed at it,
And spoke this word:              835
"Ej! Alija, my foster son,
Since Mustajbeg of the Lika is afraid,
Since he is afraid of you in the Krajina,
If God grants it, when you return here,
I shall give you gifts, son."              840
Let him speak as he wishes—
Let us see what is happening here.
When the fifth morning dawned,
Then Alija prepared his raven horse.
Then he started for white Karlovo.              845
General Pero followed him,
And with him Djulic^[*] . the standard-bearer,
And hajibeg Ograscic^[*]  Alija.
When they said they would take their leave,
Alija went to rocky Kotar,              850
Pero went to Kara Bogdan,
Djulic^[*]  went to the Lika and to Ribnik,
And Alija came down to Kotar.

The second sample passage, that from Bajgoric^[*] . (Parry no. 6703, lines 4.10-35), consists of the opening part of the ubiquitous "Shouting in Prison" theme.^[28] The clamor created by the lamenting captives causes the distraught banica to petition her husband the ban for their release; he responds by entering the prison to discuss matters with the troublemakers. The same analytical procedures as were employed with the first passage are applied here.^[29]

Text 2

[28] See chapter 8 for an analysis of this typical scene.

[29] The following passage consists of twenty-six (rather than twenty-five) lines because a natural subdivision occurs after 4.35.

Bojicic^[*] Alija cried out              10
In the white city of Kotar,
Near the hero the ban of Kotar.
Around the young man thirty men of Udbina:
All the young men cried out in prison,
But most of all Bojicic^[*]  Alija.              15
The Turks cried out for a week of days.
They shamed themselves before half the city,
Shamed themselves and caused a great commotion.
When the ninth morning dawned,
It became unbearable for the lady banica,              20
And also for the ban on his porch.
Then the banica spoke to the ban:
"O by your God, master ban,
Give in to the Turks, find some way;
They do not let me sleep at night."              25
The ban jumped up, then shouted angrily:
"Await me, my dear lady,
I shall remove the Turks from the prison."
The ban seized the keys on the chain—
One ring and four keys.              30
Then he went to his prison.
He opened the cover on the prison,
Then he went down into the murky depths.
And when he reached the bottom of the prison,
He found all the Turks in the prison.              35

Formulaic analysis of the two sample passages yields the quantitative profile represented in the following list.

Most important for general purposes is the fact that both passages, the first with a larger and more homogeneous referent, and the second with a much more limited basis for comparison in the work of the singer himself, are 88 to 98 percent formulaic by colon. There are very few cola in either sample (a total of 51 lines or 102 hemistichs) that do not find a formulaic relative elsewhere in the Stolac referent. Furthermore, the Kukuruzovic^[*] passage reveals a higher density of straight whole-line formulas (40 percent as compared to 25 percent) and colonic formulas (84 percent to 50 percent) than do the fifteen lines from Ugljanin examined by Lord. Even the Bajgoric^[*] sample, with a much smaller idiolectal basis for comparison, shows about the same density of colonic formulas (44 percent as compared to 50 percent) as the Ugljanin passage. With virtually universal systemic structure and a high formula count, then, the figures for the Kukuruzovic^[*] and Bajgoric^[*] samples echo Lord's results and support his conclusions.

But there are also some differences between our two passages, and these differences start to indicate how individual styles can exist within more general traditional practice, as well as how referent size and composition can affect analysis. First, the relatively depressed figures for percentage of formulas and the relatively elevated figures for percentage of systems in the Bajgoric^[*] sample may be ascribed in part to the disparity in size (1,667 versus 5,619 lines) and generic make-up (two additional Return Songs versus a Wedding Song)^[30] of the referent. More significant than the formula-system ratios, however, are the nearly identical sums of these figures; no matter how many or how few cola can be established as formulas in a given context, practically all of them

[30] On epic subgenres, see chapter 1.

can be proven to be either formulas or systems. Second, the much larger difference in percentages at the whole-line level argues not for the influence of varied referents but rather for differences in individual style. Bajgoric^[*] seems, relatively speaking, to eschew composition in recurrent whole lines, and especially in whole-line formulas, in favor of colonic units.^[31] While both samples can be shown to be formulaic and traditional at the colonic level, then, the two singers vary widely in the extent to which they depend on whole-line formulaic phraseology. We shall look further into idiolectal characteristics and their significance in the fourth section, but it is well to note here how each guslar , while adhering in general to the traditional Stolac dialect of formulas and systems, nonetheless finds his own modus operandi within that verse dialect.

The Spectrum of Formulaic Diction

The Complexity of the Phraseology

One of the cardinal lessons to be learned from a formulaic analysis such as that just undertaken is an appreciation of the fundamental complexity of the diction under investigation. While it proves possible to categorize each colon and line as a formula, system, or (on the basis of available evidence) neither, one soon comes to realize that this taxonomy inevitably obscures a number of issues and that the sum of these issues amounts to a telling disparity from conventional formulaic theory.^[32] For example, a given colon may be repeated both verbatim and with formulaic substitution, but the solid underlining conventionally appropriate to such a situation epitomizes the verbatim repetition and so masks the multiformity of the phrase and, by extension, the suppleness of the diction as a whole. Or, to take another possibility, perhaps the truly formulaic component of a line is only its second colon, with the apparent repetitive phraseology in the opening section of the line merely a circumstance; nonetheless, if we find that opening section of the line repeated—either verbatim or closely enough to resemble a system—we must, perhaps mistakenly, label both it and the whole line as units of traditional diction. For a third instance, consider the fair number of lines that participate in couplets or clusters. In order to represent these elements of phraseology fairly, we should somehow note that their recurrence is overseen by other than purely formulaic (that is, colonic or decasyllabic) constraints.

Examples could be multiplied, and we shall closely examine many

[31] Compare Lord's finding (1960, 63), in reference to the 1935 and 1950 versions of Aliaga Stocevic^[*] by this same singer, that "Bajgoric^[*] is actually recreating the song with little or no reliance on habitual and frequently sung passages," an example of his general thesis that one can differentiate "individual styles in the epic technique of oral verse-making."

[32] See note 27 above.

individual cola and lines in the next section, but for the moment the crucial fact to be confronted is this: formulaic diction in Serbo-Groatian oral epic dots hot amount simply to a collection of equivalent units but is most faithfully understood as a complex and responsive spectrum of phraseology . While Lord's formulaic theory provides an excellent first approximation of this many-sided, multiform phraseology, it does not explain the many different kinds and degrees of patterned diction. As we shall see, the degree of fixity of the various elements of phraseology is not necessarily a function of frequency of use; certain elements simply achieve a more stable form, whether through their compositional role (e.g., proper name), the influence of sound patterns, or merely a happily unique fit of prosody and essential idea that becomes a staple of the guslar's repertoire.^[33] Building on the firm foundation of Lord's theory, we shah attempt to glimpse the range and depth of traditional phraseology—both its spectrum of structure and its differential evolution—in the individual styles of two singers from the same region of Stolac.

Synchrony, Diachrony, and the Deseterac

From what, we may ask in beginning our inquiry, does this spectrum of diction evolve? How are we to explain its development to the point at which we observe the range of formulaic structure in, for example, the recorded repertoires of Mujo Kukuruzovic^[*] and Halil Bajgoric^[*] ? First, we must observe that Lord's theory, although illuminating certain aspects of the multiformity of the guslar's poetic language, homogenizes its inherent complexity^[34] and deemphasizes the fact that phraseology is not a static collection of items but a dynamic inventory ever in a state of flux or evolution. Such a view opens up valuable perspectives on the relationship among elements, but it does so at the cost of certifying these elements' phraseological equivalence. If formulas are understood as generated by patterns, and the patterns are seen as the source of all diction, then each and every element of diction must be functionally equivalent. More significantly for the history and development of oral literature research, the synchronic model of generation from pre-existent patterns seems to validate the determination of orality on the basis of the formulaic-density test. If we construe oral traditional phraseology as atoms of diction which are in turn the issue of formulaic patterns, then proof of a critical density of atoms and patterns should constitute proof of that particular type of phraseology. From a set of definitions proceeds evidence for a mode of composition.

[33] This does not mean, of course, that any such element necessarily ever becomes a part of another singer's repertoire.

[34] Note that while Lord (1960, 44) contends that formulas "are not all alike either in their genesis or in their intensity of 'formulicity,'" his method of analysis and the conclusions drawn from quantitative evaluation assume the equivalence of phrases. See further his suggestion (p. 43) that "a singer's formulas are not all of the same degree of fixity."

For Lord, then, the synchronic model provides a way of illustrating multiformity in the guslar's language, but it blocks the path to further understanding because it does not account for the observable fact that all formulas are not created equal. Likewise, the same discrepancy among units goes a long way toward invalidating the formulaic-density test for orality; if we are testing for identical phraseological units, and if what we actually have in the texts is a spectrum of decidedly unequal elements, then our analysis cannot bear fruit. In addition to the principles of tradition-dependence and genre-dependence discussed in chapter 1, we thus discover another necessary complication which oral literature research must recognize: the natural heterogeneity of traditional phraseology.

Instead of locating the pivot of formulaic structure at a hypothetical point of balance between patterns and the formulas that seem to depend on those patterns, let us take the investigation a step further to the elemental prosody of the deseterac , the epic decasyllable that supports the diction we are now examining. We recall from chapter 3 that the outer metric, the usual representation of trochaic pentameter, proved illusory and misleading, the result of a flawed set of assumptions. Prosodic structure in the deseterac amounts to the inner metric, the system of cola that comprise the line and the inter- and intra-colonic relationships. In addition to a regular syllable count of ten and a caesura between positions 4 and 5, the deseterac exhibits the Indo-European principle of right justification in a number of features: the four-plus-six inner structure of the whole line; the placement of ictus (emphasis on syllables 1, 3, 5, and 9); three complementary distribution rules and a general limitation on allowed configurations in the second colon; and the shorter-before-longer, highly variable make-up of the first colon. All of these prosodic rules are active in the shaping and maintenance of traditional phraseology in Serbo-Croatian epic—not just as a template for one formula or set of formulas, but as the group of rules within which all formulaic diction must be made and re-made. Thus these rules, as opposed to the assortment of patterns that the synchronic model uses to rationalize formulas, are the most general parameters, the universal guidelines for an ever-evolving Kunstsprache and the supports that assist in preserving phraseology over time.

Apart from the vast disparity between the number of formulaic patterns affected by the two sets of constraints, perhaps the most telling difference between this modest set of prosodic rules and the dictionary of patterns demanded by conventional formulaic theory is that the rules leave the way open for what is prima facie observable in Serbo-Croatian oral epic texts: a true spectrum of phraseology. Indeed, it hardly seems possible that they could do anything else: not only would such basic prosodic parameters admit to the guslar's wordhoard initially inequivalent units of diction (verbatim repeats alongside systems and other patterns), but they would also leave room for differential development. Phrases employed by a singer could answer very

Figure 5.
Generation of Poetic Phraseology

dissimilar descriptions: one might be a six-syllable name-plus-patronymic unlikely ever to change; another might be a verb-direct object phrase "to mount a horse" that admitted a multitude of functional synonyms for "horse" into the system; while a third might amount simply to a particular placement of a given word on the basis of its metrical type. All of these and myriad more kinds of phrases could enter a guslar's repertoire and could evolve toward or away from absolute stability within the range permitted by prosodic rules.

Furthermore, the postulating (and demonstration, see below) of phraseological rules to reflect the laws of prosody in the deseterac makes the role of obvious formulaic patterns clearer. These patterns are unarguably more exacting, but nonetheless nominal, specifications of prosodic rules; in applying strictly to a small but well-defined group of actual lines or cola, they serve as situation-specific intermediaries between universal prosodic rules and certain groups of related lines. To put the matter schematically (figure 5), we can say that formulaic patterns, the molds that Lord sees as the source of formulas, act as linguistic lenses that focus the more general rules. As the diagram indicates, however, not all verse lines can be derived from such patterns; as we shall see in the discussion that follows, there are lines and cola that reveal the operation of traditional prosodic rules without the interposition of formulaic patterns. This model allows us to explain the variant phraseological structures encountered in the spectrum of Serbo-Croatian epic diction as the natural conclusion of a two-level process: lines and cola are all traceable to a few fundamental prosodic rules, and a certain number of lines and cola are further structured by a large number of formulaic patterns.

Of course, there are many more factors to be considered in describing the source, shape, and relative stability of the diction. Some of these include sound-patterning in its various types (assonance, alliteration, different kinds of rhyme), syntactic parallelism, terracing, and thematic focus. While each is a significant force in phrase generation and maintenance, we should keep in mind that all are at least as situation-specific as the formulaic pattern. Each, like that pattern, applies to a limited number of lines rather than informing the traditional phraseology as a whole, and therefore none has even a fraction of the overall significance of prosodic rules. With the understanding that more than one situation-specific feature may help to focus the underlying rules, we may picture the process of creation and preservation of formulaic phraseology as in figure 6. Once again we take account of the possibility that diction can

Figure 6.
Focusing, or Second-Level, Processes

become part of the singer's poetic language without focusing by any of these second-level processes. And while actual lines and cola do appear in Serbo-Croatian oral epic that show no allegiance whatsoever to any of the intermediaries listed above, absolutely none of the 7,286 lines (or their 14,572 cola) examined for the present purpose reveals an important departure from the traditional prosody of the deseterac (see chapter 3). Traditional prosodic rules, the idiosyncratic reflex in the deseterac of ultimately Indo-European characteristics, constitute the primary basis for traditional composition.

As will be illustrated below, prosodic rules and the various focusing effects collectively produce a diction that cannot fairly and accurately be described as one of equivalent units. Some verse lines and cola take their structure only from phraseological constraints that reflect these primary rules; in such cases one is hard pressed to stretch the definitions of formula and system to accommodate the phrases, and solid or broken underlining seems more misleading than informative or analytical. Only by the most literal application of definitions, one that subverts the whole notion of multiformity by claiming genetic connection between phrases only superficially similar, can the categories hold up. And, more than occasionally, even stretching definitions beyond the limits of credibility will not bring elements of phraseology into the fold. Nevertheless, we shall discover relationships of various sorts—some formulaic, some not —that persist in the singers' repertoires examined and are clearly traditional. This spectrum of relationships, whether dependent on none or on one or more of the focusing features, will all show evidence of the operation of traditional rules. From the invariable noun-patronymic formula secured by sound-patterning to the merest, most nominal phrase without any apparent formulaic identity, all elements of diction in Serbo-Croatian oral epic conform to prosodic and phraseological rules. Their own obvious heterogeneity aside, the singer's phrases are thoroughly and fundamentally traditional.

From Prosody to Traditional Rules

In this section we shall selectively re-examine the two passages from Kukuruzovic^[*] and Bajgoric^[*] that were analyzed for formulaic density above. For the purpose of conducting an analysis identical in its core assumptions to that

made by Lord in The Singer of Tales , I employed his definitions of formula and system without qualification. In this way I hoped to present a phraseological profile comparable to his, and in fact the results of my investigation do reinforce his figures. As mentioned above, Lord's notion of formulaic density and all that it implies serve as an excellent first approximation of multiformity in traditional diction. But as I shall document below, we also need to restore a lost complexity to that diction, a complexity that the generic model has eliminated; we need to understand, as explained in the preceding section, that the formula and system are second-level, or focusing, processes and do not represent the most fundamental level of phrase structure. In order to take account of the inherent complexity of the phraseology, then, in what follows I shall forgo automatic categorization of lines and cola as equivalent items in favor of showing just what about a phrase can be considered formulaic. Some elements are manifestly "more formulaic" than others in that they conform to a tighter definition over the sample of related phrases; some are "less formulaic" in that their systemic character is less obvious and more difficult to determine; and some are not meaningfully formulaic at all. Each line or colon drawn from the two samples for commentary and illustration will be examined first on its own terms as a traditional element conforming to phraseological rules and only secondarily as an example of a certain type of traditional diction.

Deseterac Rules and the Diction

The three Indo-European metrical characteristics of syllabicity, regular caesura, and right justification give rise to a number of prosodic features in the deseterac , as we saw in chapter 3. Syllabic count and caesura (or institutionalized word-break) between positions 4 and 5 are the obvious reflexes of the first two, while right justification manifests itself in an interrelated series offeatures:

1. an initial four-syllable colon followed by a six-syllable colon, the more extensive unit following the less extensive one;

2. preference for ictus at positions 3 and 9, then 1 and 5;

3. shorter-before-longer words and accentual groupings (SBL) in both cola;

4. initially accented disyllables (IAD) favored at positions 3-4 and 9-10;

5. medially accented trisyllables (MAT) favored at positions 8-10;

6. a generally greater flexibility in colon 1 (accomplished primarily by stressed monosyllables and by accentual groupings that include proclitics and enclitics) and, correspondingly, a generally greater fixity in colon 2.

Rules 1, 2, and 6 are self-evident and have been discussed thoroughly in chapter 3. Rule 3, the most far-reaching of the other three constraints, stipulates that shorter words be placed before longer words^[35] in both cola, thus reflecting

[35] From this point on I use word to mean accentual grouping (lexical item plus associated proclitics and/or enclitics) as well as the simplex lexical item.

the same principle of right justification that underlies Rule 1. Rule 4, an exclusion to SBL, may optionally overturn the prescribed sequence, however; IAD mandates that initially accented disyllables can (but need not) migrate to positions 9 and 10,^[36] occasionally leaving the second colon with, for example, a four-syllable increment followed by the disyllable. It is important to note that Rules 3 and 4 arc ordered: SBL will determine word order unless an IAD intervenes, in which case the order may optionally be reversed.^[37] For example,

Although both lines are used to introduce a speech, and although both second cola have the same syntactic content (verb plus direct object, in whatever order) and functionally the same semantic content ("spoke" versus "shouted" a "word"), the sequence of elements is reversed. In 1.835, SBL (2-4 pattern) reigns unchecked because rece^[*] is initially unaccented (short falling), while in 1.648, IAD causes the accented form víknu (long rising) to seek colon-end (4-2 pattern). Thus IAD can supervene SBL, but the exclusion is not required, and we shall encounter many lines in which SBL informs phrase structure despite the presence of IAD.

The third and last of the word constraints, Rule 5 or MAT, pertains to those relatively infrequent situations wherein the second colon has a "balanced" configuration of 3-3. In such cases, when SBL cannot by definition apply, the trisyllables will be sorted by the influence of position 9 in the deseterac . Just as with IAD, the word whose lexical accent will coincide with prosodic ictus in the penultimate position will migrate to the final spot in the colon. For example,

The length of the medial syllable of gavrana , as opposed to the short medial syllable of sigúra , secures the position of gavrana at 8-10. And again, just as does IAD, MAT functions as an optional exclusion to SBL; if traditional placement cannot be determined on the basis of syllabic extent, the next set of criteria has to do with correlation of lexical accent (or length) and prosodic ictus at position 9.

[37] Disyllables without initial accent cannot, in other words, assume positions 9-10 and reverse the SBL rule.

As we proceed through examples drawn from the two sample passages, it will become apparent that the normal rules for (prose) word order have a certain force in shaping the sequence of words in oral epic phraseology. But that influence is neither as pervasive nor as fundamental as one might expect. The fact that traditional phraseology usually seems to follow conventional prose word-order does not mean that poetic diction takes its cue from that convention, since there are numerous instances—as many as half the lines in some passages—that depart from the supposed prose norm. When we balance this observation with the fact that there are absolutely no important departures in those same samples from the six rules sketched above, we begin to understand the ontology of influences in verse composition. First and most fundamentally, the guslar composes according to traditional rules; then the influence of conventional word order enters the process. To put it aphoristically, poetic word order tends toward the conventional prose norm within the limitations of traditional rules.

Examples of Traditional Composition from the Stolac Guslari

Mujo Kukuruzovic^[*]

1.829 . The opening line of the Kukuruzovic^[*] passage offers a first insight into the complexity of traditional phraseology:

The Parry-Lord notation indicates that (1) the first colon is a formula, (2) the second colon is a formula, and (3) the whole line is a formulaic system. What that notation does not indicate is that both cola are also systemic and that the entire line allows more substitution than the underlining can suggest. We have had to solidly underline the first colon because it appears intact elsewhere in the referent, but, as might be expected, many names and simple nouns other than Pero can appear at positions 3-4, among them Tale, Mujo, Huso, and the nouns momak (young man) and beze (bey).^[38] As for the second colon, it tolerates considerably less variation, but the essential idea remains intact in such phrases as knjigu preucijo (he composed the letter) and knjigu nakitijo (literally, he decorated the letter).^[39] These two colonic systems can occur together, as here, or separately; in the latter case, different whole-line essential ideas are involved.^[40] Thus, even a glance at the phraseological context

[38] The option beze is actually a vocative singular used as a nominative singular subject; this "poetic inflection" allows the word beg to take the disyllabic form appropriate to the formulaic system.

[39] A usually consequent essential idea, represented by such phrases as knjigu prifatijo ("he seized the letter"), may also be construed as related to this system under the narrative aegis of letter-writing and -reading phraseology.

[40] E.g., 1.110: Svakome je knjigu nacinijo ("he formulated a letter to each one"); 6.1239: Kad je Mujo rijec: razumijo ("when Mujo understood his words").

reveals a greater multiformity of colonic (and therefore whole-line) structure than is apparent from simple underscoring. Furthermore, we have evidence for two distinct but concurrent levels of organization; each colon can participate in other whole-line phrases and thus has a life of its own outside this particular whole-line system.^[41]

This much we can demonstrate without venturing far beyond conventional formulaic analysis. But how do traditional rules come into play? And do they in fact rationalize this more complicated view of formulaic structure in 1.829? In general, we seek to determine the applicability of SBL and the relative fixity of cola 1 and 2, since right justification manifests itself chiefly in these two phraseological characteristics.^[42] In colon 2, which permits a narrower range of systemic substitution than colon 1 (the semantic constraint in colon 2 is more limiting than a mere exchange of names), SBL demands the configuration knjigu nacinijo , or 2-4 by syllable count, and that is the sequence that obtains both in this instance and throughout the system.^[43] Although conventional prose word order would be *nacinijo knjigu , SBL prescribes the opposite sequence for traditional poetic diction. Nor can the IAD exclusion reverse SBL to echo the prose convention, since the initial syllable of knjiga is unaccented and short.

In turning to colon 1, we notice at once that this phrase behaves differently from the latter part of the line. While it can be said with reason that the range of disyllables seeking positions 3-4 is certainly wider than that of verbs that maintain the essential idea of the second colon,^[44] there is more to the difference between phrases than that. The element Kad je [name/noun] is simply less fixed in form than Knjigu [verb "to compose"]; to put it another way, the Kad je [name/noun] system is more a nonce formation that recurs because of the confluence of traditional rules and a very generic "When" clause than a formulaic system that is obviously a part of the guslar's repertoire. The proof of this "nonce" character of colon 1 consists of the tremendous variety of line-types that begin with these words. If the colon 1 phrase has any essential idea at all, it is so general and adaptable as to be more a result of traditional constraints producing a generic idea in much the same form than a true formulaic system.

[41] The same can be said for narrative themes or typical scenes, some of which occur in different sequences and different subgenres and thus demonstrate lives of their own outside the text under examination. See chapter 8.

[42] I set aside the recurrent syllabic and caesural constraints as obvious, and see IAD and MAT as optional exclusions of SBL, as explained above.

[43] In fact, the four-syllable verb form (usually in past tense third-person singular or plural) following its direct object becomes an institutionalized feature of deseterac verse and shows evidence of existing as a grammatical pattern (based ultimately, of course, on traditional rules).

[44] As in many other lines, the greater number of elements in this position are initially accented, so the name/noun tends toward 3-4 by convention.

If we step back a moment and have another look at the whole line, the variegated spectrum of traditional diction begins to come into relief. In essence we have a very focused, tightly organized system (colon 2) combining with the non-systemic issue of traditional rules (colon 1) to form what may be termed a whole-line system. The combination in no way represents the integration of two equivalent elements to synthesize a third. Compositionally, the core of the essential idea behind this line lies in the second colon, and the first colon—quite typically, as we shall see—is the shaping device that suits the core idea to its situation-specific position. Under the influence of right justification, the more flexible first part of the line has adapted the more fixed latter portion to its present purpose.

As we go further and start to gain a sense of what the spectrum of traditional diction actually means, it will become more and more apparent how subtle and dynamic an instrument the guslar's diction is. Conversely, it will become less valuable, as examples multiply and more complexity is introduced into the model, to return to diagrams that are limited in the ways in which they can symbolize different degrees of formulaic character or dependence on other second-level focusing processes. For the moment, however, we can profit from interpreting 1.829 in terms of figure 7. This description of the source and evolution of 1.829 more faithfully represents the singer's (more accurately, the tradition's) task and the nature of the phraseology, first as an essential idea takes shape within the constraints identified as traditional rules and then as it is in part focused by a traditional formulaic system (which itself operates under traditional rules). The singer does not merely mesh two prefabricated units but rather adapts the second-colon "word" to the particular form needed at this juncture in the narrative. The poet is no mere assembler of parts; his compositional task is nothing less than true poiesis .

1.831-33 . A few lines further on in the Kukuruzovic^[*] passage we encounter a series of decasyllables best considered as a unit:

The underscoring indicates that all six cola (actually five, since 831a and 832a are identical) have independent lives of their own outside the whole-line configuration in which we find them, and also that the first two lines exist as combinations of these cola elsewhere in the Stolac referent. The third line is apparently a unique combination of colonic phrases repeated elsewhere verbatim. As an initial profile, then, we have two lines with both colonic and whole-line structure and one with colonic organization only. But the situation proves more complex than that approximation, since the cola themselves are of various sorts. The phrase knjiga šarovita , for example, turns out to be a typical second-colon formula in its relative fixity; it tolerates no substitution whatever,

Figure 7.
Generation of Line 1.829

with morphological adjustment the only change observed in seventeen additional occurrences. As a characteristic first-colon formula, Kudgodj ide can be modified by occurring either as a phrase complete in itself or as a verb phrase in need of a noun to serve as its subject; the two instances in 831 and 832 make this point. Additionally, do Karlova sidje constitutes one specification of a narrowly substitutable system associated with the first colon by the leonine rhyme on ide/sidje , and the final line consists of a second-colon formula that itself varies little and a first-colon phrase that always demands a dative complement to fill out its meaning and syntax.^[45]

In short, lines 1.831-33 offer us more evidence of how dissimilar cola (and lines) can be in their compositional structure. Although by adhering strictly to Parry-Lord orthodoxy we can posit formulas and systems corresponding to these lines and cola, such labeling takes us only part of the way toward their most fundamental make-up. Once again it is clear that the synchronic model of combinations of equivalent elements will not faithfully describe the compositional process. And added to the complex variety of phraseological units manipulated by the singer is another complication: parallels from elsewhere in the Stolac texts illustrate that these three lines form a "cluster" or "run" that has a unitary dimension of its own, and further that this cluster owes its integrity and stability in part to the institutionalized terrace in 1.831-32. Consider one of the thirteen recurrences of this three-line unit:

The first line derives from the same system as the opening line of the cluster under examination, only this time the second-colon formula (again one that allows no morphological variance or substitution) names a person instead of

[45] We may even question the characterization of A na ruke as a colonic formula, since without the second colon the essential idea is incomplete.

a letter to be delivered. The second line likewise stems from the system that generates Kudgodj ide, do Karlova sidje at 1.832, with the expectable modification of place from Karlovo to Ribnik in order to suit the story's action. The third line, although quite different from the A na ruke [person, dat.] system, serves the same general purpose by specifying the person to whom the character or letter mentioned earlier is directed.^[46] In this case the idea-sequence of (a) x goes, (b) x goes generally to a place, and (c) x goes specifically to a person takes shape as a single unit expressed over three lines and held together in part by a terraced repetition of Kudgodj ide . Since the three lines memorialize and express the idea-sequence traditionally, we would do well to recognize that in this case three lines constitute one unit with a single essential idea. While there can be no doubt that some of the lines and cola that comprise the sequence individually have their own essential ideas, the cluster as a whole takes its noetic shape from the unitary nature of the traditional idea that underlies it.

And in accord with the shape-shifting that consistently characterizes oral traditional phraseology—and which is diachronically the source for its evolution within limits as well as synchronically the source of its usefulness as an idiom—this cluster of three lines can be expressed without the terrace and leonine rhyme contributed by the middle member in the sequence.

With reference to the first occurrence analyzed (1.831-33), we can see immediately that lines 1 and 3 of the two clusters derive from the same systems, the only variations consisting of a change of the person to whom the letter is directed and the insignificant substitution of the preposition u for na . In the second line, however, the poet has adjusted his phraseology to accommodate a syllabically more extensive place-name. But while the actual phraseology has shifted, the essential ideas behind both the second element (x goes generally to a place) and the idea-sequence as a whole have not changed in the least. Even the verb sidje remains as a traditional specification of the traditional idea.

One more avatar of this same cluster will offer a final perspective on its morphology and suggest the variance between different singers' idiolectal forms of the same unit. This instance is drawn from the Return Song dictated by Halil Bajgoric^[*] .

[46] Note that this person, the beg of Ribnik, occurs here in the dative usage that typifies the system.

The second line in the sequence closely resembles the middle line seen twice above, this time with Udbina substituted as the place-name rather than Karlovo or Ribnik. But there the phraseological comparison ends; lines 1 and 3 bear no formulaic relation whatever to the first and third members of the cluster as examined above. Nevertheless, these same apparently heterodox

lines do flesh out—in an idiolectal way—the fundamental sequence of x goes, x goes generally to a place, and x goes specifically to a person. The unitary essential idea persists through the multiformity of the diction and even crosses the border of idiolect; the unit maintains its identity and epitomizes its traditional message.

Beyond the cluster formation, formulaic structure, and other second-level focusing processes singularly applicable to these phrases, lines 1.831-33 reveal the same elemental dependence on traditional rules that characterizes not just certain lines or passages but all deseterac phraseology. In 1.831 normal word order obtains in the first colon, but right justification in the form of SBL reverses the expected order in colon 2; just as in 1.829, the initially unaccented knjiga is subject to a poetic metathesis. In 1.832 two forces combine to determine phrase configuration and to insure the stability of the whole-line system. The verb sîdje will tend toward positions 9-10 via the IAD exclusion, yielding a 4-2 syllabic series (since the proclitic preposition do joins its object Karlova in a single accentual group). The leonine rhyme on ide/sidje helps to maintain the system by providing a phonological bridge between cola. In the final line of the cluster we have SBL reinforcing normal word order in the first colon (1-3 syllabic pattern, with na proclitic) and a balanced second colon that may well be the product of analogy to banova djevojka , whose long medial syllable would seek the strong ictus at position 9 (MAT). On these traditional rules are built phrases that exhibit an intricate variety of second-level compositional processes—formulas and systems, leonine rhyme, terracing, and clustering. None of the phrases, however, owes its most fundamental allegiance to these more obvious features; the aegis under which all of them occur and recur is that of traditional rules.

1.836 . Another example of the limitations on description imposed by the synchronic model of formulaic structure is provided by this ubiquitous line-type:

It so happens that this entire line is repeated seven more times in the Stolac corpus, but that fact—and the solid underlining that represents it graphically —turns out to be largely irrelevant. The true nature of the whole-line phrase lies in its systemic properties, which demand that the second colon not change at all while the first colon supplies the name upon which the address is built. Once again colon 1 recurs without colon 2, but moj na mjestu sine never occurs outside this whole-line system. So we discover versions of the system, an idiolectal item that is apparently not a part of Bajgoric's^[*] repertoire, with the names Alija (8 ×), Djulic^[*] (5 ×), and Ograscic^[*] (1 ×).^[47] This combination and shaping of two quite different cola nonetheless rests on the steady foundation of traditional rules: colon 2 is firmer and more limited in scope than the more flexible and adaptive first colon, and SBL orders the first phrase (1-3 or 4), while IAD (sîne ) causes a modification of normal word order in the second (1-3-2).^[48] In short, the line is best and most faithfully interpreted as a right-justified whole-line system that takes shape under traditional rules.^[49]

1.840 . The multi-leveled process of composition and the rationalizing perspective afforded by traditional rules are again evident in this line, although conventional formulaic analysis does not give the whole picture:

At perhaps the most superficial level, this line or one of its relatives serves as the final line in a speech in fourteen of twenty-one occurrences. Beyond that function, however, each colon is much more flexible than the solid underlining would cause us to expect. The phrase Ja cu^[*] tebe and its morphological variants prove common enough—indeed, inevitable enough given traditional rules— to beg the question of whether it deserves to be considered a formula or system at all; even if we accord it that status, though it has no demonstrable "essential idea," we must recognize that the phrase is by itself incomplete and in general much less stable and tightly defined than many other cola studied above. Alongside this typical initial phrase stands the colon that completes the verb construction (cu^[*] ... [infinitive]), and, again typically for a second colon, its variation is more controlled. If the vocative case noun (in this instance sine ) is understood as comprising part of the system, thus yielding a template like [noun in vocative] darovati , we shall find only four recurrences, three with sine and one with beze . If we are willing to broaden the essential idea to permit

[47] Another form of the system without the first-colon specification occurs in 6.2064; Jes', Boga mi, moj na mjestu sine ("Yes, by God, my foster son").

[48] The more customary order would be *moj sine na mjestu (1-2-3, with na proclitic), but mjestu is not initially accented.

[49] That is, to see 1.836 as the melding of two equivalent units is simply reductive. In fact, I would question the formulaic character of colon 1, since it seems to be merely the inevitable way to fill out the whole-line pattern via traditional rules; see further the initial colon of 1.829.

Figure 8.
Traditional Phraseology with darovati

the substitution of adverbs, adverbial phrases, and direct objects at positions 5-6, the list of referents quickly grows in length and complexity.

The clue to a proper perspective on this complexity consists of concentrating on the most fundamental structure of the lines concerned, that is, on traditional rules. Figure 8 represents the source of and genetic similarities among various formulas and systems related to the second colon of 1.840. The infinitive darovati tends toward colon-final position via the SBL rule, a clear example of right justification, and into the colonic "word" template thus formed enter a variety of disyllables, including two vocative nouns (sine , "son," and beze , "bey"), an adverb (dobro , "well"), and an adverbial prepositional phrase (s vrancem , "with a dark horse"). Although one can argue that the disparity in syntax precludes calling all of these phrases members of the same formulaic system, at least by the classical definition, there is no denying their relationship through traditional rules and the "word" template that results from the application of those rules.^[50] Whereas the usual shortcoming of conventional formulaic analysis is the collapsing of phraseology into an oversimplified taxonomy, thus obscuring the differences among elements of diction, in this case the conventional approach does not penetrate far enough to illuminate a highly traditional (if not actually formulaic) association.

1.843 . Both Kukuruzovic^[*] and Bajgoric^[*] use variants of the system underlying this line as a diurnal marker to segment the progress of the narrative:

The underscoring indicates that both cola combine with other partners in otherwise unrelated lines, the former as a system and the latter as a formula.

[50] Note that SBL takes precedence here, the optional exclusion by IAD (with sîne and s Vráncem ) not occurring. This situation is due most immediately (that is, synchronically) to the [x] darovati frame, but ultimately (that is, diachronically) to the hierarchy of traditional rules: IAD only optionally supersedes SBL.

In fact, as we might expect, the Kad je [x] system does not exist in any meaningful sense outside this whole-line system, since its shape is the inevitable issue of traditional rules and not a demonstrable full-fledged pattern in its own right. Equally typically, the second colon turns out to be a stable core idea, "the morning dawned"; there are no other related uses of jutro and no other uses at all of the verb osvanuti in the entire Stolac referent. Further, since jutro osvanulo does not occur in lines wholly unrelated to 1.843,^[51] we can best understand the line in question as a whole-line system, or a traditional "word" one verse in length. This line thus exemplifies another principle of traditional composition: units—or, as I prefer to call them, traditional words —are of different lengths as well as of different texture and stability. Far from being mere counters simply moved into appropriate position by the poet composing in performance, they comprise a vast and complex network of phraseology that epitomizes essential ideas in forms subject to the dictates of traditional rules.^[52]

1.845 . The second colon of this line provides another example of an essential idea epitomized optionally in a colonic form (as we shall see below, it can also take a whole-line form):

The phrase do Karlova bila recurs verbatim and can thus be labeled a formula by the classical definition, but it is also a member of a colonic system that substitutes other place-names into the middle position; we find instances of Janok and, with a shift of preposition, Ribnik as well. Should it prove useful to the Stolac singer, he also has available a whole-line version of this colonic system, Do [place-name] grada bijeloga , employed in genitive and prepositional inflections with the cities Karlovo, Kanidza, Janok, and Koluto. Depending on the narrative and phraseological situation, the guslar may turn to either the colonic or the whole-line form of the essential idea, the only difference between them consisting of the metonymic noun-epithet phrase "the white city." Kukuruzovic^[*] does just that, using both versions (or lengths) of the idea, while Bajgoric^[*] consistently turns to the longer form only, whatever the situation.^[53] Thus the two singers' idiolects sort the multiformity of their traditions differently, but still consistently within formulaic patterns and ultimately within traditional rules.^[54]

[51] Line 1.1157, Dvades-peto jutro osvanulo ("The twenty-fifth morning dawned"), is a borderline case and makes the argument for considering jutro osvanulo a formula per se.

[52] Briefly stated, colon 1 follows normal word order reinforced by IAD (pêto ), and colon 2 is organized, again in agreement with conventional order, by SBL (jutro is initially unaccented).

[53] See, for example, line 4.11 of the Bajgoric^[*] passage.

[54] The adjective bíla provides an IAD to override the SBL rule and reverse the normal word order (4-2, with do proclitic). The SBL rule orders colon 2, and the IAD (grâda ) in this case does not supersede it, again yielding a reversal of customary order.

1.847 . Lines 1.846-48 enumerate, in a fashion very common in Serbo-Croatian epic, the characters who will undertake an action, in this and many other instances a journey. The middle member of this short passage well illustrates typical differences between the compositional functions of the first and second cola:

Both cola arc apparently formulas in their own right, recurring elsewhere with other partners but combining uniquely in this line. However, an examination of the referent reveals that colon 1 is only one version of a multiform (cf. A do njega, I kod njega ; with morphological adjustments as well) that is incomplete in itself, a very generically defined system that seeks a complement to perform its enumerative function. For its part, the second-colon complement, here Djulic^[*] bajraktara , just as characteristically permits no change other than morphology. In other words, the relationship between these cola is of the common type that includes a second colon which fixes on one of various possibilities and tolerates no systemic change and a first colon which tends to vary systematically or, from the point of view of traditional rules, inevitably. Compositionally, colon 1 adjusts the grammar and position of the line to bring into play the semantic and informational focus of colon 2. This kind of combination reflects the effect of right justification at every level, most obviously in its characteristic colon structure and texture and the formative role of traditional rules.^[55]

1.849 . One final example from the Kukuruzovic^[*] passage will demonstrate the formative role of traditional rules and of the second-level process of leonine rhyme, as well as furnish an instance of idiolectal phraseology:

While viewing colon 1 as a formula may be viable according to the classical definition, that perspective leaves unexplained the fact that it can occur either as a colonic clause complete in itself (in which case application of the label formula seems more justified) or as an incomplete prelude to a whole-line system. In this example what binds the cola together and contributes a certain degree of stability, in addition of course to the continuous syntax, is the leonine rhyme of the two aorist verbs in colon-final position (in -oše ). Functionally a feature of the poetic language (since it does not figure to any extent in contemporary standard Serbo-Croatian), the aorist tense is not seldom employed in this way as a syntactic/phonological bridge between cola that tends to help preserve phraseology over time.^[56] In addition, then, to the force of SBL in

[55] Colon t forms an accentual group of four syllables (proclitic-enclitic-disyllable) that admits variation most readily at positions 1-2, while colon 2 follows SBL (2-4).

[56] Numerous examples could be cited from the Stolac referent, such as kad rekoše da pokajitiše (6.513) or Kad rekoše da se kumovaše (1.139). See further Stankiewicz 1973.

both cola and the formulaic pattern that underlies the whole line, the colon-end syntactic balance and rhyme in aorist tenses provides traditional structure. And within this structure, Bajgoric^[*] varies slightly but noticeably from Kukuruzovic^[*] by employing a traditional linear "word" that substitutes a second coordinate phrase for the dependent clause used by Kukuruzovic^[*] :

Such idiolectal variance within the scope of traditional rules amounts to the individual's signature on the unwritten document of oral traditional style.

Halil Bajgoric^[*]

In looking further at selected phrases in the passage from Halil Bajgoric^[*] (4.10-35), we shall again be concerned with description of the complexity of the oral traditional idiom, and specifically with understanding the role of traditional rules and of second-level processes in generating phraseology. At the same time, we shall continue to point out examples of the singer's personal idiolect versus instances of the traditional dialect shared by both guslari . By extending the analysis from Kukuruzovic^[*] to Bajgoric^[*] , we not only generalize our results beyond a single poet's repertoire but also offer a more finely articulated perspective on the inherent complexity of the phraseology as a firm basis for the eventual establishment of an aesthetics suited to the art of oral traditional verse-making.^[57]

4.12 . Early in the Bajgoric^[*] passage we encounter a line reminiscent in its typical texture of many verses by Kukuruzovic^[*] :

As the underlining indicates, the colon od Kotara bane recurs in other combinations as well as in a whole-line association that we may tentatively call systemic. The evidence for this system is a single comparand found only in this same text:

In fact, the second colon itself is systemic, with the ban of Janok (od Janjoka bane ) as its referent, and the whole line seems more the result of traditional rules than of a linear pattern. The situation is thus a familiar one: the core of the linear phrase is a second-colon system of tight definition and limited

[57] I repeat the caveat that the referent for the Bajgoric^[*] Return Song is considerably smaller than that for the three Return Songs by Kukuruzovic^[*] , for the reasons stated above. This difference will result in fewer comparanda for the elements of diction used by Bajgoric^[*] , especially, of course, for those phrases limited to his idiolect.

multiformity which is shaped to fit the narrative and phraseological situation with the aid of a "nonce" first colon based on traditional rules. The SBL constraint marshals the phraseology of both cola—reinforcing the normal word order in colon 1 (1-3, or 4 with U proclitic) and causing a reversal in colon 2 (IAD, bâne , 4-2 with od proclitic)—while the second-level process of a formulaic system assists in maintaining stability in the second part of the line. Right justification forms the basis for the composition of the verse, both in the relationship between cola and in the intra-colonic texture of the phrases.

In contrast to Bajgoric^[*] , Kukuruzovic^[*] develops three additional whole-line patterns using od [Janjoka/Kotara] bane . As might be expected, one of these consists of a complex of lines introducing speeches, among them the following:

All of these initial cola, the latter two related formulaically, also combine with other noun-epithet formulas, the introduction lines in Serbo-Croatian epic being many and various. A second category involves the opening colon A da vidiš ... ("But you should have seen...") or its near-relative, but again this fairly fixed initial phrase finds completion in any number of noun-epithet sequences. A third category, or. traditional word, one which happens to include only od Janjoka bane and no other colonic name in the present sample, is

Apart from its usage in 4.12, then, the second-colon system has a life of its own and recurs in three quite different kinds of lines. A smooth-surface, synchronic model for such phraseology will not suffice; the rule of thumb for traditional diction continues to be its heterogeneity and complexity.^[58]

4.15 . At times the distinction between essential ideas and their expression proves less than absolutely discrete, as this apparently simple line of specification illustrates:

[58] Also worthy of mention is the fact that Kukuruzovic^[*] evolves an alternate and idiolectal formulaic phrase for the idea expressed in od Janjoka bane : the colonic form mlad janjocki bane ("the young ban of Janok"), which is not found in Bajgoric's^[*] repertoire. As with the grada bijeloga expression discussed in reference to 1.845 above, the added concept of mlad is metonymic and not to be considered as forming a different essential idea.

The interpretation encoded in the underscoring has broken najviše into its two constituent morphemes, the superlative prefix naj- and the comparative root više , in order to show that multiformity in a traditional word need not respect what we customarily define as a lexical unit.^[59] Thus comparanda for 4.15 include such lines as

Notwithstanding the usually slight differences in connotation of the root word (the final example being equivalent syntactically but semantically exactly opposite), the syntactic pattern is identical throughout, and the general idea of the four colonic phrases is certainly consistent enough to support terming these lines members of a formulaic system. But we come yet closer to the truth when we interpret the whole line as composed of that initial colonic system— incomplete by itself and requiring a focus in a new phrase—and a characteristically more strictly defined second colon. Most often that partner phrase will be a noun or a noun-epithet formula of little or no flexibility.^[60]Bojicic^[*] Alija thus completes the colonic system with the customary invariable phrase, and, while we have little evidence for calling 4.15 a whole-line system by the classical definition, we misinterpret the line if we do not take into account the association of the first-colon formulaic phrase with the more stable focus it requires for completion of its function.

In addition, it is well to note the idiolectal character of Bajgoric's^[*] traditional "word," if we may in some way see this line as a unit. For he uses only najviše (1 ×) and najprije (2 ×) in positions 2-4, while Kukuruzovic^[*] employs only najprvo (1 ×) and najpotlje (8 ×). As for the second colon by itself, once again we have an example of a phrase that combines with numerous different partners to form lines otherwise unrelated to this one, and once again some of these lines are integral units in themselves. In effect, then, the colonic element Bojicic^[*] Alija offers us evidence that traditional words can serve two, or even more, masters: they can exist both as self-contained units to be pressed

[59] Of course, there are many examples of multiformity in which a colon is treated as a word—such as noun-epithet sequences that add inflections only to the end of the colon-final lexical unit—but formulaic character of the sort observed in 4.15 is much rarer.

[60] In addition to those examples cited above, we find instances of od Orašca Tale, Pero generale , and beg Mustajbeg licki .

into service in association with phrases such as A naj [-više/-prije] and as permanent parts of whole lines repeated verbatim and themselves understood as (larger) units. As in the case of Homeric phraseology, the spectrum of diction as evidenced by the Stolac material contains both cola and whole lines of every description, and it is this richness that is the guslar's traditional inheritance.

4.16 . The description of the Turks' noisy wailing in prison presents another example of the heterogeneity of traditional phraseology:

The Stolac concordance shows that the simplex cmil- (from cmiliti , "to cry out, lament") occurs in one of three formulaic systems and nowhere else:^[61]

System 2 is a well-defined whole-line pattern whose syntax and acoustic patterning (cmili/bila ) contribute to its stability. System 3, much less tightly organized, admits a wide range of first-colon noun phrases as subjects; while this pattern is surely not the same sort as system 2, it would be shortsighted simply to call it a colonic unit and ignore its connection to the initial colon. As the underlining suggests, the system that lies behind 4.16 takes a variable subject (either suzanj , "prisoner," or Turcin , "Turk") and one of two second cola, both denoting a period of time. The verse thus has a whole-line identity, a linear pattern. But, as in other examples we have considered, the second colon also has a life of its own in traditional diction; both za nedjelju dana and tri bijela dana appear in other, unrelated combinations.^[62] Once again we have evidence that a phraseological element—typically a relatively more sharply defined and more stable second colon—can exist both as part of a whole-line pattern and as a fully formed unit with an independence signaled by its compositional adaptability.

Of the three systems involving cmil- , the first two are employed by Kukuruzovic^[*] only and the third by Bajgoric^[*] only. In addition to these idiolectal categories of phrases, even the choice of numbers in the variable colonic

[61] The word order in the first colon of system 1 can be reversed, as in 4.16, for reasons discussed below (note 64).

[62] Examples include Od dnev' do dnev' za nedjelju dana ("From one day to the next for a week of days," 1.1261) and Ev' imade tri bijela dana ("There passed three white days," 6.1492).

pattern [x] bijela dana helps to characterize the individual singer; for Kukuruzovic^[*] the number is always three, for Bajgoric^[*] always two. In fact, some other analogous phrases for duration of time, all of them second-colon formulas and nearly half of them constituting part of many different linear arrangements, are also idiolectal, as table 20 shows. To further illustrate the heterogeneity and complexity of the phraseology, we note that only examples A and D are strictly colonic and reveal no involvement with whole-line patterns of any sort. These are also the only two formulas of this category in Bajgoric's^[*] analyzed repertoire.^[63] Formulas F and G, at the other end of the spectrum, occur without exception in whole-line arrangements. The remainder of this category turns out to consist of phrases that can combine with a number of initial cola and, like za nedjelju dana , can exist as units complete in themselves or as participants in larger units. This kind of multiformity, with units that comprise one or more different types of traditional phrases, is characteristic of all levels of oral epic structure in Serbo-Croatian.

The fundamental traditional rules that govern the formation of the second-level system underlying 4-16 prove quite simple. As is typical of the more flexible first colon, the word order—here reversed from the conventional prose norm—could go either way: both words are disyllables and both have initial accent.^[64] The phrase za nedjelju dana (4-2 with za proclitic), which follows conventional sequence, is ordered by the IAD dána .

4.17 . Often a particular phraseological pattern proves dependent on the rules that govern the phraseology of the deseterac and, at the same time, looms larger than the classically defined formulaic system. Line 4.17 exemplifies such a pattern:

Polu grada rezil ucinili ,

[63] We may recall that figures for different levels of formulaic density as presented in section B also suggested that Bajgoric^[*] : depended more on colonic and less on whole-line patterns than did Kukuruzovic^[*] .

[64] Both show IAD: cmíle, Tûrci . In the three additional examples of the same phrase system, cmíili (IAD) takes second position at syllables 3-4 in deference to suzanj and Tùrcin , neither of which is initially accented.

While the first colon has no formulaic relatives anywhere in the Stolac referent, colon 2, with its essential idea of "to bring shame," is repeated elsewhere verbatim and thus qualifies as a formula. But, as in most cases, this first approximation does not penetrate to the functional core of the phrase, and we must look to the multiformity of the second colon in order to understand its compositional structure. The most faithful representation of the pattern on which the latter section of the line is based would be [x] ucinijo, -ili, -ila, -iti , etc., where x , the direct object of some form of uciniti ("to do" or "make"), is a Turkish loan word. Under the influence of the Turkish language, especially during the Ottoman Empire, the poetic idiom admitted many such lexical items and enclosed or fossilized a large number of them in what amounts to a traditional "word." To put it another way, this very system amounts to diachronic evidence of the influence of Turkish vocabulary on Serbo-Croatian deseterac diction.^[65] What was diachronically an influence has become a synchronic stylistic habit as construed traditionally in a pattern much larger and more pervasive than most formulaic systems; many variant ideas, rather than one essential one, are imaged in this pattern, but the integrity of the phrases as belonging to one "word" is beyond question.

The [x] vciniti element presents an interesting morphology, as documented in the following list:^[66]

[65] Cf. Skendi (1953) on the influence of the Albanian language and epic phraseology on Serbo-Croatian epic.

[66] The source for the meaning of the Turkicisms is Skaljic^[*] (1979), with my English translations.

On this pattern, well attested in the analyzed repertoires of both guslari , Kukuruzovic^[*] idiolectally erects a related "word." In effect, he turns the unit to yet wider usage by including some non-Turkicisms, that is, words of Slavic etymology that did not enter the poetic language during the Ottoman occupation or later but were part of the earlier language stock. As opposed to 145 total occurrences of this colonic "word" with Turkicisms over the whole of the Stolac referent, we find only nineteen instances of the same pattern with a native noun or adverb in the substitutable position:

Clearly, the phrase pattern was developed for and continues to serve words of Turkish origin, but Kukuruzovic^[*] has formed phrases by analogy with the older pattern, that is, neologisms that take their structure from a traditional unit.^[67]

The influence of traditional rules is manifest in the SBL organization of the second colon of 4-17 and of all its Turkish and Slavic relatives; furthermore, the 2-4 syllabic grouping throughout the 164 examples in the Stolac referent constitutes proof that the phrase pattern and not the individual instance is primary, since none of those instances which feature IAD have changed the already reversed SBL word order. In fact, and significantly, traditional rules also marshal the ostensibly unformulaic phraseology in colon 1. The IAD rule sorts the balanced 2-2 configuration by exerting pressure to align grâda with the third and fourth positions at the end of the colon. Thus, even if Polu grada cannot be shown to be formulaic, it can still be understood as traditional, for it follows the rules from which all second-level features of phrase generation take their cue.

4.18 The very next line furnishes another example of what amounts to a nonformulaic but still fundamentally traditional phrase:

Ucinili i uzabunili.<<<

[67] I judge the pattern to have arisen from use with Turkicisms because of their much greater number and variety in the repertoires of both singers. On the basis of the Stolac referent, I further posit that Kukuruzovic^[*] used eight additional phrases constructed on the pattern of the system; although he may have learned these eight phrases elsewhere and thus not have been responsible himself for their "creation," they are nonetheless a feature of his idiolect on the basis of available evidence.

Here again the underlining is misleading, because it depends on a single recurrence of ucini- in initial position:

Moreover, the instance of ucinijo in 6.1170, just as ucinili in 4.18, derives not from a formulaic phrase but rather from the second-level process of terracing; in both cases the verb ucini- repeats the final element of the preceding line. As for the second colon, although uzabunili is a hapax legomenon in the Stolac referent, we can see that sound-patterning, another second-level process, also helps to hold the line together. Not only the syntactic balance of two past plural verbs and the consequent leonine rhyme in -ili^[68] but also the assonance of uc - and uz - participate in an aural network that binds the cola into a linear whole. In respect to traditional rules, we should remember that leonine rhyme depends on the relative prominence of colon-end positions (3-4 and 9-10), so that the -ili sound pattern is finally a reflex of right justification.

4.22 . Like many other examples we have considered, this line finds its focus in a relatively well-defined second-colon system tailored to the narrative situation by a first-colon phrase of much more flexible texture:

The second-colon system, best understood as a noun in the dative singular (indirect object) followed by govorila (also including other inflections for gender) that recurs thirteen times in the analyzed referent, has a life of its own outside this particular combination. As a formulaic phrase it also follows SBL, with the syllabic configuration 2-4.^[69] At the same time, the apparent system [x] banica , where x stands for Tad, A, Pa , or Kad , is more a direct reflex of traditional rules than a systemic unit; according to right justification, it varies only at the beginning of colon 1 and assigns the three-syllable item to positions 2-4. Once more we view the typical difference of pattern and formulaic character between the two cola of the deseterac . In this case, however, the whole-line structure is supported by another second-level focusing process, that of function—for 4.22 is one of a number of lines used as introductions to speeches, and, just as in the Old English and Homeric epic traditions, these kinds of verses gain a stability due at least in part to their compositional role.

[68] Cf. the leonine rhyme and syntactic balance of two aorist-tense verbs in the Kukuruzovic^[*] passage (line 1.849).

[69] Only three of these 13 reverse the 2-4 order, all of the reversals involving IAD or MAT (Zlâti, mâjci , and Halîlu ).

4.26 . This line also functions as one of introduction to a speech, but in this instance idiolectally:

Whether the first-colon system is deployed as a sentence complete in itself (four occurrences) or as part of a whole-line phrase (four occurrences), all examples in the referent are from the Bajgoric^[*] texts. Characteristically of first-colon formulas, this phrase shows a good deal of flexibility in the length of the "word," in its combinations with second cola, and in its own variability.^[70] When Kukuruzovic^[*] wishes to express the same essential idea as in the first colon of 4.26, he turns to one of two dialectal systems of his own:

In their particular ways, all of these cola and lines follow basic traditional rules, but the resulting phraseological variety testifies that these same rules leave room for a singer to develop and maintain individual habits of composition.

The latter portion of 4.26, pa srdito vice , contains a middle element that proves a hapax legomenon in the Stolac referent, but we still have sufficient information to declare the colon a traditional phrase. For even if we chose not to define an involved system pa/a/i [adverb/adv. phrase] vice , so that lines like A sve beze pa iz grla vice (6.143) can be viewed as comparanda, traditional rules will govern the formation of the colon; that is, the IAD vîce seeks colon-final position and reverses the SBL order (4-2, with pa proclitic). That IAD is more fundamental than the second-level focus contributed by the proposed formulaic system can readily be seen by observing that, leaving aside the four formulaic comparanda, all eight additional occurrences of vice in colon 2 are at positions 9-10 while ten of twelve in colon 1 are at positions 3-4.^[71] This line thus often a clear example of how traditional rules shape second-level processes: underlying the more immediately recognizable systemic patterns of the diction are rules that depend ultimately on the principle of right justification. In a sense, traditional rules are as current as the guslar's most recent performance and yet as ancient as Indo-European versification.

[70] Examples of its morphology include Skoci Djulic^[*] , hizmet mu ucini ("Djulic^[*] leaped up, he served him," 3.180); Skoci momak na noge lagane ("The young man leaped up to his light feet," 3.19); Skoci Mujo i za glasa vice ("Mujo leaped up and shouted in a loud voice," 4.333).

[71] The two exceptions to the rule are in fact not exceptions. One of them, A vice ms buljubaša Mujo ("And commander Mujo called to him," 4.299), uses enclitic binding (and normal enclitic placement) to construct the first colon (SBL, 1-3), while in the other, Vice bane cetiri telala ("The ban called four criers," 6.806), the initial colon is sorted by the other IAD (báne ).

Summary

Through the foregoing analyses and examples I have attempted to show that the oral poetic language of Serbo-Croatian epic is a much more complex and sensitive instrument than has heretofore been realized. We have seen, in short, that all elements of phraseology—all traditional "words"—are not created equal. The wide spectrum of diction includes some units that are always repeated exactly, some that undergo only morphological change, so. me that admit regular substitution and can be meaningfully described as formulate systems, and some "non-formulaic" cola and lines that are best viewed as the inevitable issue of fundamental traditional rules. Nor does the traditional expression always assume one canonical length: units of colonic, linear, and multilinear dimensions have been shown to populate the Stolac referent. In the face of this heterogeneity and complexity, we have argued, it proves only logical to recognize the limitations of the concepts of formula and formulate system, and to understand that synchronic approximations that go far toward making evident the characteristic multiformity of the guslar's poetic language can also obscure its variety and richness. Our philological and critical imperative is consequently quite clear: we must come to appreciate this variety and richness as the basic character of the compositional idiom if we are to make informed and faithful aesthetic judgments about the works of art it figures forth.

Specifically, we need to concentrate on traditional rules as the most fundamental and pervasive influence on the making and re-making of diction. As illustrated in chapter 3, the deseterac verse form preserves the ultimately Indo-European features of syllabic count, caesura, and right justification. This third feature is expressed idiosyncratically in the decasyllable in a variety of ways: the initial shorter colon of four syllables followed by the longer one of six; preference for ictus at positions 3 and 9, then 1 and 5; shorter-before-longer arrangements of words and accentual groups in both cola (SBL); initially accented disyllables favored at positions 3-4 and 9-10 (IAD); medially accented trisyllables favored at positions 8-10 (MAT); and a generally greater flexibility in colon 1 and correspondingly greater fixity in colon 2. This combination of rules (and not simply the poor approximation of a trochaic pentameter and associated patterns) underlies the formation of all deseterac phraseology, no matter what more superficial process may aid in its creation and maintenance. In other words, while formulas and systems explain the poetic idiom inexactly as a dictionary of quantitatively equivalent but qualitatively different paradigms, traditional rules account for the entire spectrum of traditional phraseology—every line of every song-text.

And just as we see these rules in action in every line of every text, so we also see their formative influence on the series of second-level processes that further focus the expression of essential ideas in the poetic tradition. To put it

hierarchically, none of these focusing features can operate except within traditional rules; since each of the second-level features pertains to a very small percentage of the lines in a given sample, each can affect its limited constituency only if it works within the set of constraints that determines the shape of all lines in the wordhoard. At this second level, the formula and system are the most pervasive structuring devices in the poetry, but we must remember that these two terms refer to an enormous "dictionary" of actual units and that even the most generative of single phraseological patterns oversees a relatively modest number of cola and lines. In fact, even when a guslar makes a "new" phrase by analogy to a pattern he already knows and uses, he does so under the immediate influence of that pattern but within the much more far-reaching code of traditional rules. As we have seen, formulaic structure is a significant aspect of diction, and analysis by formula can lead us to understand the multiformity of oral epic phraseology in numerous interesting and valuable ways; nonetheless, formulaic theory is at its most powerful only when we also take into account its basis and continuing context in universally applicable traditional rules.

Other second-level or focusing processes also assist in the creation and maintenance of traditional diction, among the most important of which is sound-patterning. In Serbo-Croatian epic, assonance, consonance, and alliteration play large roles, often stitching together words, cola, phrases, lines, or even extended passages. The primary site for rhyme proves to be colon-end, that is, either at 9-10 in two successive lines (relatively less frequent) or at 3-4 and 9-10 in a single line (also called leonine rhyme and considerably more frequent). Both forms derive from the stress-emphasis at colon-end, which is in turn a reflex of right justification. Thus institutionalized internal rhyme constitutes another example of a synchronic echo of an ancient poetic characteristic.

Added to these focusing processes are the stylistic features of terracing or pleonasm, which actually promotes unnecessary enjambement in Serbo-Croatian, and of syntactic balance or parallelism, which can take myriad forms either within a single line or from one line to another. We may end this list, necessarily incomplete,^[72] with the feature of thematic focus, whereby, as in the sample text from Bajgoric^[*] , narrative structure to some extent marshals the deployment of certain phraseology.^[73] But I must stress that while all of

[72] The list is necessarily incomplete because second-level processes are both dialectal and idiolectal. This discussion is intended to cover the major features of traditional diction as a whole and not to treat exhaustively any important features in the analyzed repertoires of these two singers. It remains for future investigators to add to and qualify these beginning (but hopefully fundamental) remarks.

[73] I emphasize that this last principle affects only a limited number of phrases, since many lines—formulas and tightly defined systems among them—occur independently of thematic structure. See chapter 8 below; cf. G. Nagy 1979, 2-4 and passim.

these processes, from formulaic structure through the thematic constraint, participate dynamically in the shaping of diction and therefore in the act of oral traditional epic composition, all are decidedly second-level features in that they affect only limited parts of the much larger body of deseterac phraseology. All of them operate on traditional diction that takes its archetypal shape from rules that in turn echo the prosody of the epic decasyllable and ultimately that of Indo-European verse.

One more step remains to be taken in our assessment of traditional structure in Serbo-Croatian phraseology, and in taking that step it is well to remember how our investigation has proceeded from establishing the general tenets of Indo-European prosody to a tradition- and genre-dependent view of the deseterac^[74] and on to the two levels of phraseological organization—traditional rules and a series of second-level focusing processes. Rules and processes as powerful as these, generating at the first level all lines in the Stolac referent and shaping at the second level much smaller but still significant groups of verses, must necessarily be generic enough to order the expression of a vast panorama of traditional ideas. For this very reason, they will also leave space for a guslar to place his personal (if largely ephemeral) signature on the wordhoard by making or re-making idiolectal phraseology employed, as far as can be determined, only by him. For the mediocre singer, the more general local dialect of the poetic language shared among the singers in his community will be the phraseological support used to buttress by far the greatest part of his individual performance. But the talented singer will depend somewhat less on the lingua franca of verse dialect and at least slightly more on his own idiolect.^[75] From the perspective described and advocated throughout this chapter, a guslar's idiolect consists of expressions unique to an individual and yet constructed in accordance with traditional rules; we shall see more evidence of the individual's signature on his tradition when we study thematic structure in chapter 8. Thus, to the lengthy lists of tradition- and genre-dependent factors that make for complexity and suppleness in the compositional idiom as employed by any given singer, we can add the "singer-dependent" phenomenon of a traditional idiolect.

What do we gain by tracing the guslar's words from Indo-European versification forward through the maze of compositional devices described in this chapter? And what do we make of the result—a complex, heterogeneous, ever-evolving collection of inequivalent elements overseen by rules and processes no singer ever consciously imagined? Briefly stated, what we gain, apart

[74] Note that Lord has laid the basis for a tradition-dependent perspective on phraseology: "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" (1960, 31; emphasis added). See also Lord 1986c.

[75] On a similar point, see Bynum 1981, esp. 72-77; cf. Lord 1960, 49.

from a philologically sound profile, is a foundation for aesthetic inquiry that is firm because it is faithful to the language and poetics of Serbo-Croatian epic. By formulating rules for the phraseological events we perceive as lines, we begin to restore a lost complexity to oral traditional diction; in effect, the point of view advocated in this chapter, and for that matter throughout the volume, allows us to "re-complicate" poetic composition, to take it out of the arena of lockstep simplicity and back to the realm of language—the most complex of human abilities and arts. On the basis of the observations made and the examples provided above, it should be more than obvious that the guslar is no mere assembler of pre-fabricated parts, no automaton mindlessly spinning lines of verse from a limited selection of movable counters.^[76] Rather, the Yugoslav oral singer is fully a poet who has come to be able to speak his native poetic idiom, with a fluency determined by a combination of his own talents and the bequest he has received from tradition. If we can restore this natural quality to traditional diction by understanding it not as a patchwork of remnants but as a highly complex weave accomplished by blending many different colors and textures according to fundamental rules of order and pattern, then we re-admit the possibility—even the necessity—of oral poetic art in its archetypal sense: the individual poiesis of tradition.

[76] That this mechanism is not what Milman Parry and Albert Lord had in mind is apparent as early, as Parry's M.A. thesis (1923), with its remarkable analogy of Greek traditional verse to Greek sculpture.