Four
Traditional Phraseology in the Odyssey

In this chapter we shall be concerned with applying the findings of chapter 3 to Homeric phraseology, specifically to the traditional diction of the Odyssey . Because so much has been accomplished along these lines since the monumental work of Milman Parry,^[1] I shall not attempt even a brief summary of previous scholarship, preferring to leave that aspect of the history of oral literature research to other works devoted largely or exclusively to that topic.^[2] I choose rather to treat only a few of the most seminal and suggestive of Parry's works, and then only in the interest of opening up the questions to be considered in the body of this chapter. Thus the first section will review a few of Parry's basic premises and some recent responses to the chain of events he began. In section 2 we shall inquire as to what sort of phraseology can be expected in ancient Greek epos on the basis of the tradition-dependent prosody described in the preceding chapter. In the third section we shall turn from theory to analysis, looking at the phraseology of the Odyssey in two test cases: the renowned usage of epea pteroenta , or "winged words"; and a passage of twenty-one lines to be analyzed for formulaic density but also, and for our particular purposes more importantly, for various kinds of formulaic diction. In the final section I shall gather together what has been learned from these two different types of analyses and propose a general theory of the traditional structure of the phraseology.

[1] For the antecedents to Parry's work on formulaic phraseology, see the account by Latacz (1979a); and Foley 1988, chap. 1.

[2] For a general account, see Foley 1985, 3-77; Foley 1988; Latacz 1979a. M. Edwards (1986, 1988) surveys scholarship on the ancient Greek formula in exacting detail.

The Formula: Original Concepts And Developments

As many scholars have noticed, it was not until his "Studies" I (1930) and II (1932) essays that Parry first broached the possibility that his earlier demonstration of the traditional character of Homer's epics must also mean that they were composed orally .^[3] What has not been as clearly noted is that Parry also expanded his claims from the limited arena of the noun-epithet systems in the Iliad and Odyssey to the whole of Homer's diction, and that he did so without the laudable rigor exhibited in his studies of the "traditional epithet" (1928a). His explanation of the term "formula" in the 1930 essay provides an illustration. After defining this phraseological unit as "a group of words regularly employed under the same metrical conditions to express a given essential idea" (1930, in 1971, 272), he goes on to give three examples (Parry's trans.):

("when it was morning")
("he went")
("he said to him")

While all of these phrases can profitably be termed formulas, they are not, like the noun-epithet formulas, of uniform type or structure. The first is a whole-line recurrence, a dependent clause that introduces an action; the second and third are core sentences that combine with subjects that specify their predicates. Moreover, there is a large metrical disparity among the three phrases, which, on the model of the hexameter developed in chapter 3, would be classed as whole-line, colonic, and hemistich formulas, respectively. Parry recognized some of the dimensions of this elaboration in a footnote that precedes his formulaic density analysis (p. 275 n .1):

Formulas, in the strictest sense of the term, may be of any length, but in studying them we are forced to exclude the shorter word-groups, for the following reasons. If we dealt with formulas of all sizes we should have an unwieldy mass of material of varying importance, and it would be impossible to compare the formulaic clement in different poets by means of the number of formulas found in their verse. In the second place, we must set a limit which will shut out any groups of words which are repeated merely by chance, or as the result of their natural order in the sentence. Accordingly I have regarded as formulas, or possible formulas, only expressions wade up of at least four words or five syllables, with the exception of noun-epithet phrases, which may be shorter.

Parry's first criterion for this typology or distinction is subjective: he believed that five syllables would command the hearer's attention and four would not. But the second justification reveals much more about his method and goals,

[3] On the role of his teacher Antoine Meillet and of Matija Murko in the evolution of his thought, see Parry's own remarks in "Cor^[*] Huso" (Parry 1933-35, in 1971, 439-40).

for he argues (ibid.) that "by insisting on four words in a shorter phrase one puts aside almost all chance groups of connective words." Here and elsewhere it becomes clear that Parry's goal is a quantitative profile, a numerical measure of "traditionality" and, he contends, therefore of orality. He sought, in other words, to analyze and to illustrate by example only , and at no time did he contemplate a complete theory of formulaic structure. He is in fact quite forthright on this point (p. 307):

A full description of the technique [of formulaic composition] is not to be thought of, since its complexity, which is exactly that of the ideas in Homer, is altogether too great. One must either limit oneself to a certain category of formulas, and describe their more frequent uses, as I have done in my study of the noun-epithet formulas, or one must take a certain number of formulas of different sorts which can be considered typical.

The goal of his analysis, then, is a sample to be used as a litmus test for the whole work in its much greater complexity, and the limits he puts on the size of the formula are intended to facilitate his quantitative measurements.

As one example of his extension of the concept of formula from the noun-epithet combinations to other elements or phrases, let us examine the system he uses to illustrate formulas of a certain type, "But when X had Y," where X is the implicit subject of g , a verb (figure 2; I have added English translations of the verbs to Parry's diagram [1930, 276]). While this collection of phrases certainly shows a multiformity, it is far different from the particular

Figure 2. Parry's Formulaic System

systemic character typical of the noun-epithet phrases. Here the emphasis is on the substitutable nature of the verb form, with the only constraint on the actual word chosen being its metrical shape. Each phrase has a clear essential idea, but the group as a whole is too wide-ranging semantically to have one core idea behind it; we cannot treat something as amorphous as "But when X does or did Y" as an element equivalent to the noun-epithet formulas.

Again, Parry was quite aware of the differences among formulas and explicitly addressed that variance in his discussion of the phrase alge' ethêke in Iliad 1.2. Noting that this phrase, which takes the classic colonic form of the adonean clausula, is not as extensive as some of his other examples, Parry (p. 309) observes that the shorter phrase

thus belongs to the less obvious part of the technique; yet it would be false to suppose that it is any less helpful to the poet than the longer ones: it is chiefly in the formulas of these shorter types that lie the suppleness and range of the diction, and their usefulness is to be measured by the many different kinds of other short formulas with which they combine.

What emerges from the 1930 analysis, in short, is acceptance of the reality that formulaic diction is in fact a spectrum, that noun-epithet formulas are but one kind of phrase structure within that spectrum, and that the examples summoned are meant to illustrate pars pro toto rather than to assist in assembling a comprehensive theory of formulaic structure. By showing what appears to be utility, Parry reasoned, he has shown that traditional Homeric poetry, by analogy to other traditions, is oral.

The reaction to these ideas comprises an epic tale in its own right (see M. Edwards 1986, 1988), but we may point to three quite dissimilar treatments of the Homeric formula that in various ways depended on and reacted to Parry's ground-breaking theories. In 1967 Michael Nagler published an article advocating a generative, synchronic approach to what he called the "traditional phrase" (see also Nagler 1974, 1-63). In his view the locus of "tradition" was in the preconscious gestalt of associations inherited by the singer, associations which were then mapped onto the unique surface structure of the individual performance. One result of this approach is the elimination, welcomed by Nagler and many others, of the dichotomy "traditional versus original" that has gained such currency in a variety of literatures.^[4] Another, and for our purposes more significant, corollary consists of enormously multiplying the associative echoes of a given traditional phrase, much as connections between and among ideas can be made in myriad ways, often purely acoustically (Nagler 1974, 8ff.). Since, in Nagler's words (p. 26), "all is traditional on the generative level, all original on the level of performance,"

[4] This is especially true in Romance language studies, where the terms traditionalist and individualist (now provided with the prefix neo -) denominate fixed and seemingly irresolvable positions. See Faulhaber 1976; Webber 1986.

the phraseological surface of the narrative is of secondary importance to the mythic ideas that underlie it.^[5] The surface is infinitely complex because ever shifting, and "a comprehensive systematization of Homer's formulaic syntax can never be accomplished"; he goes on to say (p. 28) that "it now seems more probable that exceptionally simple patterns do not represent the real manner of composition at all, but only appear particularly simple because of secondary factors, complicated by a statistically unjustifiable separation of them from the larger class."

Although Nagler's approach differs considerably from Parry's in his concentration on the preverbal associations that make up the tradition, he too understands the phraseology as an enormously complex instrument that consists of many different types of structures. And, in a departure from more conventional theories, he offers the observation that the most common and most obvious repetitions may not be representative of the diction as a whole, but instead are merely the most immediately obvious sort of patterning discernible. From Nagler's perspective, therefore, "traditionality" cannot be measured quantitatively, since on the generative level all is traditional. In essence, Nagler attempts to detach the notion of tradition from a necessary and one-to-one relationship with formula. While this approach gives a much-needed emphasis to aesthetic considerations and uncovers patterns other investigators had not found, it does not bring us much closer to an appreciation of the actual nature of the phraseology. And unless we are willing to follow Nagler completely in his de-emphasis of the structure of Homeric diction in favor of concentration on the associational web of traditional meaning he sees as its generative matrix, we will lack a philological solution to the problem of the phraseology.

Another model for the origin and deployment of Homeric phraseology was put forth by Gregory Nagy in a book entitled Comparative Studies in Greek and Indic Meter (1974; see also 1976 and 1979), which was also discussed in the last chapter. Nagy's theory that from a diachronic point of view formula generated meter and not vice versa, whether or not we accept it in exactly those terms, offers a starting point for a conception of phraseology that takes as its major premise the initial and continuing symbiosis of formula and meter. Certainly, as Nagy himself admits, a metrical norm we abstract from the surviving lines of Homeric epic at some point stabilized and came to serve as a filter for incoming phraseology, but even at that point formula and meter were cooperative and mutually reinforcing. In opening up the diachronic dimension, Nagy also provides a way of understanding how the phraseology is really more like a complicated mix of archaeological strata than a smooth

[5] Indeed, he contends (Nagler 1974, 40), "these meanings and the ways in which they are manipulated seem to be, at least in some cases, essentially continuous with the meanings and processes of myth."

surface of substitutable units:^[6] his chief example, the Homeric formula kleos aphthiton , has roots in Indo-European epic phraseology and is also preserved in Indic verse.

Richard Janko (1982) also furnishes a diachronic view of Homeric style in his analysis of the diction of ancient Greek epos. By testing Homer, Hesiod, and the Hymns for certain kinds of linguistic archaisms and innovations, he is able to assign a relative chronology to the poems involved and to construct a probable "stemma" for the development of the epos from Mycenaean times onward. Janko's study and mathematical analysis are complex, but the unanimity of his findings is impressive: with the exception of the frequency of n -mobile used to make position, the percentage shifts of various linguistic features are consistent as one moves from the Iliad to the Odyssey , then on to the Hesiodic poems and the Hymns (see p. 200). Like Nagy, he emphasizes the evolutionary nature of the diction; while at any one time it may behave synchronically as a substitution system, diachronically it is always developing^[7] —maintaining some older elements even as new elements enter the phraseology.

These three conceptions of the formula in Homer open up some questions that need to be addressed by both philological and aesthetic studies. First, as Nagler has shown, the "traditional phrase" can no longer be considered simply a metrical unit that serves a useful purpose in composition. Rather, we should conceive of a spectrum of different sorts of phraseological structures, each redolent with manifold traditional associations that arise as the preverbal gestalt, as Nagler calls it, comes to be realized in the varia of the diction. Nagy leads us away from a purely synchronic and deterministic view of the formation of the phraseology and toward an appreciation of the originative and continuing partnership between formula and meter. And Janko also stresses the diachronic dimension of phrase generation and retention, constructing a realistic model for the ontogeny of the composite diction and its chronologically heterogeneous parts. All three of these studies thus point toward a complication of formulaic theory, a refinement of Parry's original

[6] Cf. Peabody 1975; the same concern with diachrony is also apparent in Parry's (1932) discussion of the Homeric poetic language.

[7] Of phraseological development Janko writes (1982, 188-89): "In an oral or mainly oral tradition, especially one with a metre as complex as the dactylic hexameter, formulae are preserved over long periods for reasons of convenience, or even necessity, as an aid to composition. Many formulae are handed down through the generations and preserve archaic forms, some extremely ancient indeed. On the other hand, it is clear that there was much scope for flexibility and originality in such traditions. Formulae are modified, where it is metrically possible, in accord with developments in the spoken vernacular, and such modification is an important pan of bardic technique. Old formulae, rendered incomprehensible or grossly unmetrical by the passage of time, are replaced by more modem expressions. New formulae are constructed after the analogy of older models, and then there will be totally new elements in the poetry, as new themes are elaborated, and ancient formulae bedded in matrices of new words."

approach (intended, we must remember, only to illustrate by example the extent of schematization of Homeric diction) and a step toward an understanding of the phraseology in its complexity and its richness.

A Tradition-Dependent Homeric Phraseology

Before turning to the series of analyses that will make up the third section of this chapter, let us first weigh the evidence developed about Homeric prosody in the preceding chapter and on that basis inquire what we might reasonably expect of a tradition-dependent Homeric phraseology.

As an initial premise, we recall from earlier discussions that the prosody of the Iliad and Odyssey is significantly different from those of the Old English Beowulf and Serbo-Croatian Return Song (the latter in the deseterac meter). Thus, it was argued, the phraseologies in symbiosis with these prosodies cannot be universally comparable; a blanket concept of the formula, for example, fails to take account of the inherent variety of what are finally different natural languages fostering different prosodies. As one manifestation of the principle of tradition-dependence described in general terms in the opening chapter, then, both formula and meter will be tradition-dependent.

The Homeric hexameter, we found, reflected all of the major features of the Indo-European epic verse form, as reconstructed by investigators working in various fields and language families. It demonstrates (1) a quantitative basis, (2) relatively consistent syllabic extent (absolutely consistent in terms of morae), (3) regularly placed caesurae, and (4) what I have called "right justification." Further, this last quality of right justification was seen to be a prominent force in the dynamics of both the inner and the outer metric of the line. To take the latter first, dactyls are favored toward the ends of both lines and half-lines, making these terminal sections more expansive by syllable count and more densely populated by short syllables; this also means that certain metrical shapes will be attracted to these positions and other metrical shapes to other positions. As for the inner metric, it manifests right justification in a longer second hemistich and in relatively more expansive second and fourth as compared to first and third cola. We noted that the principle does not extend to the inner texture of individual cola but that, in general, the hexameter locates the metrically larger elements to the right, or toward the end, of both hemistich and line units.

Let us refer again to the diagram for the inner metric of the Homeric hexameter:

The designations A1 and A2 mark the two possible locations of the first or A caesura, C1 and C2 the two possible locations for the third or C caesura, and B1 and B2 the two possible locations for the mid-line caesura. The juncture points , which indicate common positions for word-end but which do not occur with regularity, are designated by the letters a , b , g , d .

Under such prosodic conditions we may expect first of all a relatively complex phraseology, one that is not easily reduced, for example, to a single kind of formula.^[8] One can at least imagine whole-line, hemistich, and colonic phrases, not to mention groups of lines, as recurrent entities. Second, since each of the hemistichs and cola is to an extent idiosyncratic in length and shape, we may also expect a wide variety of formula lengths and shapes, a variety we can start to appreciate by reference to Peabody's "colonic types" as outlined in chapter 3. Third, the degree of complexity and assortment of metrical shapes should also lead us to anticipate a relatively conservative diction as compared to Old English and Serbo-Croatian phraseology. To put it simply, once a phrase passes muster as part of the hexametric traditional vocabulary, it is unlikely to be quickly or easily supplanted. This conservatism is of course cognate to the thrift exhibited by the Homeric line, a feature, let us emphasize, of Homeric verse in particular and not (necessarily) of oral poetry in general.

Finally, we can in general say that the tradition-dependent qualities of the hexameter, as reflected through quantity, syllabic extent, caesura system, and right justification, will exert specific pressures on the formation and dynamics of the phraseology, pressures that are best documented by example (see "Two Phraseological Analyses" below). Taken together, these pressures and the three expectable qualities mentioned above constitute a set of traditional rules , that is, a collection of ways in which phraseology becomes Homeric and traditional. Since these rules are by definition tradition-dependent, taking their identity from the idiosyncratic properties of the hexameter prosody, they will be applicable only to Homeric epic. What is more, traditional rules are what

[8] Cf. Janko 1982, 277 n .6: "The hexameter's complexity compared with the simplicity of the stichic metres of other traditions seems to have dictated a higher level of formula rigidity (and also of economy) than is found elsewhere." Contending that these qualifies make the Homerie epos seem "more oral" than the comparisons often summoned, he claims that "until a tradition that possesses not only extensive poems covering a considerable period of time, but also a metre as demanding as the hexameter, can be found, the orality of the Homeric tradition will never be approached, though the scop or guslar stand before us and sing. Other epics, be they South Slavic or Saxon, are less formulaic, less thrifty, and less retentive of archaism, and it is the hexameter that makes the difference." While I quite agree with the distinctions he draws and their effect, I must also point out that the absolutism of this statement is somewhat at odds with the principle of tradition-dependence—and thus of pluralism—advocated as the basic theme of the present study. It should also be noted that if one begins by setting up Homer as the unitary example of oral traditional literature, then anything non-Homeric (whether oral traditional or not) will appear, circularly, not to be oral traditional.

differentiate one traditional phraseology from the next; indeed, this is the method we shall use throughout our studies of phraseology, in Old English and Serbo-Croatian as well as ancient Greek epic. From the particular properties of a given prosody we shall be generalizing a set of traditional rules characteristic uniquely of that prosody and therefore of the phraseology with which it is in symbiosis. Analysis of examples from each diction will illustrate the operation of these rules and suggest their importance to formulaic structure. As we turn to that analysis, then, let us keep in mind that on the basis of our investigation of hexameter prosody we may expect to meet a complex and relatively conservative phraseology with units of varying length and internal shape, a diction that responds to and is ordered by a set of traditional rules that are specifically Homeric.

Two Phraseological Analyses

The purpose of the remainder of the chapter is to fill out the general theoretical statements made above by a series of close analyses of Homeric diction. In an attempt to show how traditional rules operate within a dynamic phraseology, I shall look at a rather large and heterogeneous selection of lines from two different perspectives.^[9] The first focus will be provided by the "winged words" formula, long a source of contention among classicists; by examining the occurrences of this phrase in the Homeric epos, we shall determine its status as a formulaic element and try to shed some light on those other elements in the diction with which it combines. In the second part of the analysis, we shall conduct a line-by-line study of a twenty-one-line passage drawn from Book 5 of the Odyssey (424-44), with the intention of illustrating the interplay of traditional rules and formulaic structure.

Formulaic Structure and Epea Pteroenta ("Winged Words")

Ever since George Calhoun's 1935 article, in which he questioned Parry's explanation of Homeric diction according to its usefulness by contending that the "winged words" phrase is employed only when the speaker is in some emotional state, the debate has raged over whether this formula is in fact context sensitive or not. Parry's own rejoinder two years later made the case for a single generic rather than a number of specific meanings, and subsequent

[9] In the analyses to follow, the referent against which formulaic density is measured consists of the Odyssey , with occasional citations of parallel diction from the Iliad and Homeric Hymns (all by reference to Dunbar [Marzullo] [1880] 1962 and Prendergast [Marzullo] [1875] 1971). While I agree with Mark Edwards (1986) that explorations of phraseological multiformity seeking to be exhaustive must take account of variant readings as well, I emphasize that the present study aims not at a determination of orality via formulaic density (if indeed this were possible) or other quantitative goal, but rather at a general demonstration of formulaic structure and the more fundamental role of traditional rules. For this reason I have eliminated variant readings from consideration.

discussion of the matter has neither abated nor reached closure.^[10] In a sense the small body of scholarship surrounding this phrase provides a microcosmic survey of some of the most important issues raised by post-Parry studies of the formula in particular and of traditional structure in general.

Our primary interest in this phrase—and it is in my opinion a necessary preliminary to cogent interpretation of the formula as a unit of meaning—is its role in the diction of the epos. How, exactly, is it used? What sorts of phrases combine with it to form longer units? What status does it have as a formula or as a part of a larger unit? How do metrical norms and traditional rules figure in its morphology and deployment? In order to answer these and other questions, we need to search the Odyssey for the necessary evidence and interpret that evidence in terms of what we have learned about meter and phraseology from chapter 3.

First, we notice that sixteen of sixty-one occurrences of the phrase consist verbatim of the following line:

On this evidence alone, it looks very much as if epea pteroenta comprises part of a larger, whole-line formula that recurs as a single unit. This impression becomes stronger when we take account of the fifteen additional occurrences of "winged words" in lines that vary from the one given above only by insignificant inflections for the gender or person of the speaker. This makes for a total of thirty-one, or slightly over half, of the instances of epea pteroenta as recurrences of the same whole-line formula.^[11]

But the solution to the question of how the "winged words" phrase is deployed formulaically is not nearly as simple as this large group of lines initially seems to indicate. For alongside evidence for a whole-line formula we find evidence for a second-hemistich system, as in the following examples (partial evidence):

	(18.9)
	(22.311 plus 2 ×)
	(2.362 plus 8 ×)
	(15.208)
	(22.436)

[10] See also Calhoun 1935; M. Parry 1937; Fournier 1946; Combellack 1950; Vivante 1975, 1982.

[11] Likewise, in the Iliad twenty of the sixty instances of epea pteroenta occur in the same whole-line formula.

Quite clearly, epea pteroenta also forms part of a hemistich phrase that has a life of its own in the traditional diction; while thirty-one of sixty-one occurrences show one particular, favored combination of hemistichs, this second half-line joins with a variety of partners, some of which are illustrated above. Already we may begin to wonder which identity—the whole-line or hemistich arrangement—is primary.

The picture becomes more complicated when we notice that most of the alternate combinations of epea pteroenta prosêuda involve a first hemistich composed chiefly of a participle together with one or more function words that help to determine the syntax of the line. In the first two examples cited above, this arrangement echoes the most common whole-line configuration both syntactically and metrically, as the first hemistich is organized by the A1 caesura and the pattern followed is [function wds.] [participle] e.p.p . Although we may not wish to accept the idea of a "syntactic formula" (see Russo 1963, 1966; Minton 1965), we must at minimum recognize that the second hemistich has a common association with first hemistich types involving this particular syntax. The association is not as exclusive as that involving a single set of words, but it is well attested, with no fewer than eight different participles found in this position in the Odyssey and nine in the Iliad .

That this syntactic or structural pattern has some validity of its own is proved by the third and fifth examples cited above. In both of these lines the normal A caesura is blocked by what Fränkel would call a "heavy word," that is, a word of such metrical extent that it overruns one or more of the institutionalized word-breaks in the hexameter. Nonetheless, the pattern of function words plus participle holds, and that unit is joined with epea pteroenta prosêuda to form one of the series of lines that fit into the general category. Here as elsewhere the evidence points toward the colon as a normative metrical unit but not a phraseological unit, the smallest consistently defined unit of phraseology being the hemistich. The proof in this case is both positive and negative: the hemistich pattern is realized at the expense of blocking the A caesura. Just this much information, especially when suitably expanded by reference to more examples throughout this chapter, should begin to tell us a good deal about whether we can assign contextual meaning to a single colon or hemistich.

The fourth example given above presents a special case of the same pattern being realized through blockage of the A caesura. Here we might propose a slightly different (because more elaborated) whole-line formulaic system:

[participle] .

This model has the advantage—if it is an advantage—of specifying the function words, although it will restrict our idea of the flexibility of the multiform. But apart from these considerations, note that if the participle employed is epotrunôn , or for that matter any participle beginning with a naturally short syllable, the A caesura is again blocked. Even though the rest of the line is

metrical, and even though other such lines starting with the set of function words kai min are also metrical, this particular combination yields a blocked caesura. Such a situation is not unlike the kind of metrical flaws addressed by Parry in his supplementary thesis (1928b), in that a traditional pattern overrides a metrical desideratum. And once again we have evidence that, while the colon is normatively a metrical unit, as far as phraseology is concerned the hemistich is the smallest important unit.

A similar search of the Iliad yields comparable results. Of the sixty occurrences of epea pteroenta , fully fifty-four combine with prosêuda to make the second hemistich.^[12] Of these hemistich phrases, forty-four are associated with the same [function wds.] [participle] e.p.p . first-hemistich pattern as was encountered above. Once again the whole-line and half-line patterns seem to have lives of their own, and once again the rather frequent blockage of the A caesura by a participle that is also a "heavy word" argues the importance of the syntactic pattern as a traditional unit.

Several hypotheses for the traditional structure of the epea pteroenta phraseology thus arise. First, we can describe a colonic formula as small as epea , which occurs with absolute regularity between the B1 and C1 caesurae; to call this single word a formula, however, would seem quite meaningless, since an "essential idea" is out of the question for a one-word phrase. Here we encounter the initial and most easily resolvable difference between what might legalistically be understood as a phraseological unit and what we might practically interpret as a functional phraseological element.^[13] The next possible level of organization would be the noun-epithet phrase epea pteroenta as a whole; indeed, its recurrence 121 times in the Iliad and Odyssey certainly warrants our inquiring whether it might be the compositional kernel of the lines we have examined. But since this phrase is between a colon and a hemistich long, thus having no normative metrical unit to support it, and since it combines (in elided form) with agoreu- as well as the much more common prosêuda , a better hypothesis would be the hemistich system epea pteroenta [X], where X = one of these two verbs.

So far, then, we have a choice among colonic formula, hemistich formula (all the occurrences of epea pteroenta prosêuda , for instance), and hemistich system—but the list must continue. To these possibilities must be added the whole-line formula, e.g.,

(1.122 plus 15 ×)

[12] The remaining six combine with agoreu - (with elision to pteroent ').

[13] This is not to mention Parry's proscription of any element less than five syllables or four words in length, strictures which, as I tried to show above, were on the one hand subjective and on the other intended to facilitate quantitative explication. It is well to repeat here that Parry excluded noun-epithet formulas like epea pteroenta from these requirements.

and the whole-line system, e.g.,

[X]

where X can be phônêsas('), ameibomenos , and other participles. In fact, we might choose to schematize this system in a structural or syntactic pattern, according to the system [function wds.] [participle] e.p.p .

The problem that presents itself, then, is which of these possibilities is the true characterization of the "winged words" formula and its associated diction. On closer examination we find that while each proposal made above has something to recommend it, each is also inadequate by itself to describe some aspect or aspects of the diction involved. The colonic formula, as already mentioned, has no unitary essential idea; furthermore, it breaks the adjective pteroenta off from the substantive epea . As suggested earlier, and as will be shown throughout this discussion, the colon is a normative metrical but not phraseological unit, so this situation is nothing more than could be expected. The hypothesis of epea pteroenta as the "atom" around which the diction is constructed also proves disappointing, again for reasons already suggested. The hemistich formula proposition fares no better, since it can account for either e.p. prosêuda or e.p. agoreu- , but not both. Should we select the hemistich system in order to explain both of these phrases, then all of the whole-line patterns will be eliminated from the model. Finally, the whole-line formula hypothesis does not account for the flexibility of the line in its many forms or for other hemistich patterns, and the idea of a whole-line system (which, as we saw, can be conceived as either a semantic or a syntactic pattern) also fails to cover the hemistich patterns. Thus, while conventional explanations do bring us close to the dynamics of the epea pteroenta phraseology, none can penetrate completely the complex web of associations surrounding the phrase. What we need is a fresh and complementary perspective that will help to rationalize what seems like a puzzle whose parts cannot be clearly discerned.

Such a perspective is provided by traditional rules , which derive, as indicated above, from the tradition-dependent properties of the hexameter prosody. The spectrum of diction observed in association with epea pteroenta can be explained by recalling the features of the inner metric as developed in chapter 3, features that stem ultimately from Indo-European verse structure. As a first general principle, we recall that the second hemistich is metrically more spacious and conservative, observing as it does the phenomenon of right justification at the level of the whole line. This general feature accounts for the greater phraseological conservatism of the second hemistich as well, that is, for the fact that there is relatively little variation in the second half as compared to the first half of the line. Thus, it should be no surprise that we observe more variation in the first hemistich, as that part of the line accounts for most of the variation in the whole-line combinations.

At the level of the hemistich as unit, we notice that the metrically more

expansive elements—the participles, which are often "heavy words"—are "backed up" against the midline caesura (B1). The force of right justification of this sort, as expressed through the formulaic pattern, is so strong that the participles are located in this spot even if blockage of the A caesura results. Indeed, we have noted above that this caesura is blocked quite often, and we may attribute that phenomenon secondarily to the formulaic pattern and primarily to the right justification that serves as the diachronic foundation for that pattern. Correspondingly, the function words at the opening of the hemistich suit the metrically and grammatically less malleable latter part of the half-line to the narrative and syntactical situation, again following the principle of right justification. And it is well to note that these apparently unimportant and most flexible of words in the line actually adjust the whole line to context as well.

To these general observations on traditional rules we can add more specific measurements based on Eugene O'Neill, Jr.'s, tables of word-type localization (1942).^[14] These figures are useful because they reflect the inner metric of the hexameter: against the background of all possible locations for each metrical type are placed the actual percentage occurrences, so that the hexameter's "selection" of favored positions illustrates the influence of the caesura system and right justification.^[15] If we examine the figures for the three most commonly occurring words in this group of associated lines, the statistics shown in table 17 emerge. Like all of the shorter metrical word-types, (the form of epea , with -a making position) is relatively unrestricted in placement; it occurs with some frequency in four of its five possible positions. The favored position at 7 reflects the fit of this word-type with the colonic form marked by the B1 and C1 caesuras, as well as, from the point of view of the outer metric, the

[14] My measurements are based on recalculations of word-types made by taking into account the natural quantity of the line-ending brevis in longo . Using O'Neill's table 29, then, I have re-figured the percentages for each word-type at line-end; these changes of course meant corresponding shifts in percentage occurrence at other positions in the line.

consequent certainty of a dactyl in the third metron.^[16] It is this position that epea actually does take, fully in accordance with traditional rules of word placement.

The metrical type (pteroenta ) also shows some variety in distribution, with almost all of its occurrences coming at one of three positions. In this case our example word, pteroenta , does not inhabit the most favored position, but we should also observe both that the three possibilities are statistically close in percentage occurrence and that the word's actual placement at  hardly constitutes a rare situation. When we add to this reasoning the facts that placement at  offers the advantages of an initial boundary at C1 and encourages (but does not make certain) the desired dactyl in the fifth metron, we can start to see that pteroenta also follows traditional rules. In fact, if epea pteroenta is taken as the single "word" it seems to be throughout the diction examined, and therefore is assigned a composite metrical word-type of its own (), a different and complementary perspective is gained. For this word-type, rare enough that percentage occurrence figures are insignificant, is found at only three positions in the hexameter: , , and 12.

[16] This is also a characteristic of right justification; see chapter 3.

Thus, the noun-epithet phrase as a whole, as a single word, is also well within the operational scope of traditional rules.

The third word, phônêsas , belongs to the metrical type __ __ __, which inhabits slot 5 in approximately one-third of its occurrences. This is a common spot for this word-type, particularly because of the colonic form it constitutes, being bounded on one side by the A1 and on the other by the B1 caesura. The only more common position is also a colonic form which reaches from the beginning of the line to the A2 caesura at 3. In the case of phônêsas , however, we must take into account the influence of the patterns described above; since most of the configurations associated with epea pteroenta favor a participle backed up against the B1 caesura, the pattern—itself based on traditional rules—will to an extent regulate any incoming phraseology and encourage (if not require) placement of all participles in this same position. In this case formulaic phraseology has taken on a life of its own, and, while the birth of the pattern may be traced to traditional rules that are in a sense still operative (if vestigial), it is now apparently the pattern that controls and maintains the diction.

Which, then, is the "truest" characterization of the diction surrounding "winged words"? Is it the colonic formula, the hemistich formula or system, or the whole-line formula or system? The answer offered by traditional rules is clear enough: under their aegis we need not single out any one unit as primary and treat the remainder as derived from that Ur-element. It is quite possible that epea pteroenta was the first unit to emerge as a discernible single element, but we cannot be sure. Nor does it matter. Just as Lord (1970; also 1960, 99-123) has warned against seeking a single archetype for a given story or song in oral tradition, so we should beware of treating phrases in isolation—for, long before the Homeric texts that have reached us were recorded, phrases existed not as counters to be shifted about and manipulated algebraically, but rather as parts of larger units (hemistichs and lines), which were in turn parts of ever larger units (themes and songs). Not only did the singers not conceive of the kinds of detachable elements we locate by virtue of concordances,^[17] but neither was the very phraseology as particle-like as we make it when we analyze the text.

The truest way to conceive of these associated lines is not in terms of either an "atom" of phraseology that attracts various complements or a complex unitary "system" that tolerates variation within its pattern, but rather as a group of lines, hemistichs, and cola associated through traditional rules. This group coheres to the extent that each member was formed first under the

[17] See chapter 2 above, and Lord 1960, 30-67. Apropos of Ben-Amos's (1969) distinction between "ethnic" and "analytic" categories, we may note that our analytic category of the formula is not only exterior to the ethnic categories of the singer's art but in some ways false to the reality of the phraseology.

aegis of these rules and perhaps secondarily according to a formulaic pattern that, having been formed under the same rules, then assumed a life of its own. The "core" of the phraseology is thus not a particular element or pattern, but the traditional rules themselves. Over time, numerous associations arise between cola, between hemistichs, even between and among whole lines; some of these associations become so institutionalized as to become invariably repeated formulas. Other associations remain at the level of multiform diction, all overseen by the same set of phraseological rules but perhaps developing syntactic or semantic links among phrases as well. Still other associations will drop out of the phraseological inventory altogether, so that overall we are left with some older and some newer combinations (cf. Hainsworth 1978).

This means that the sample of evidence from any one body of material that emerges from any single point in the history of an oral tradition can tell us only a limited amount about the history of the diction. That perspective will be further narrowed by the influence of local tradition, the textual medium, and other matters discussed in chapters 1 and 2. What we should therefore seek, and can seek of course only through analysis of examples, is to understand the dynamics of traditional diction through an understanding of the effect of traditional rules on the expression of ideas. While it will be useful to speak of formulas and systems at both hemistich and whole-line levels, we must remember that our analytic nomenclature does not change the fact that traditional diction consists of a fluid and open-ended set of expressions that follow traditional rules and that can also be related (and even created by analogy) according to patterns built on these rules. If no one hypothesis for the fundamental unit in a series of related lines can rationalize the complex web of similarity and variation, it is simply because no one line or unit is ontologically primary. The phraseology does not merely present the possibility of multiformity; it actively is multiform.

Odyssey 5.424-44: Formulaic Structure and Traditional Rules

In what follows I shall continue to compare the explanations of traditional structure offered by conventional formulaic theory on the one hand and by traditional rules on the other. As a sample passage for this exercise I have chosen Odyssey 5.424-44, part of the description of the hero's tribulations at sea after having been advised by the goddess Ino to leave his storm-battered raft. This selection was not made at random, for I wanted the sample passage to fulfill at least two conditions: that it not be an instance of a much-used theme, so that we might avoid too heavy (and unrepresentative) verbatim repetition of formulaic diction;^[18] and that it be as far as possible a sample without any special, unusual structure, so that it would be likely to contain a cross-section of phraseology. We are looking, in other words, for an "average"

[18] Cf. the criteria espoused in Russo 1976.

Homeric passage, one that is neither highly conventional nor (at least apparently) highly idiosyncratic.

The reader will notice that these lines also contain a simile (at 4.432-35), and the question may arise as to whether the analysis of the diction is representative because of the inclusion of such a figure. Again, this choice was quite intentional, since I feel that any theory worth the name should be able to account for all types of phraseology encountered in Homeric narrative, not just this or that subset of the phraseology. While his approach is different from that advocated in this volume, William C. Scott (1974, 136) puts the matter neatly in referring to the similes that occur more than once in Homer: "Consistency in analyzing components of oral poetry demands that the repeated similes be treated as units which were as traditional and autonomous, but also as adaptable, as the basic arming and banquet scenes."^[19] However we interpret the similes, we must agree that they are part of Homeric narrative, and so we must deal with their phraseology.^[20]

Before presenting my formulaic analysis of 5.424-44, let me set the ground rules for my working notion of formula and formulaic in Homeric phraseology, remaining quite aware, as indicated in the first two sections of this analysis, that even the most carefully defined units will fall short of uniform interpretation by all investigators.^[21] Perhaps the most explicit way of accomplishing this definition of units is first to quote the principles followed by Albert Lord in his widely known article "Homer as Oral Poet" (1967a), and then to comment on each principle before presenting certain modifications or additions of my own. Here, then, are Lord's principles (pp. 25-26):

1. Declension or conjugation of one or more elements in the phrase, providing the metrical length of the phrase remains unchanged.

2. Metathesis, or inversion, or, in general, any change in the order of the

[19] Scott (1974, 140) conceives of a kind of storehouse of similes from which the poet could draw: "While it is impossible to identify which similes were taken with little or no alteration from the pool of traditional oral units, the examination of the repeated similes demonstrates that the poet did draw upon such a source at the very least seven times." The most recent full study of the similes by Carroll Moulton (1977) agrees in principle with Scott's approach, although Moulton believes that it is not proven that the similes are themselves oral. He feels, in concert with G. S. Kirk (1962) and others, that Homer was the "monumental composer" who put the poems into their present form, and that an oral tradition was certainly involved at some point in their history. Whether the poems as we have them were oral compositions he leaves an open question.

[20] As Moulton (1977, 125-26) points out (proceeding from Fenik 1974, 143-44), there are monologues at 408-23 and 465-73 that form a kind of ting or envelope around our sample passage. Both of these monologues are followed and elaborated by similes (432-35 and 488-91, respectively) which signal Odysseus's "endurance in adversity," a major theme of the epic and a quality that "will lead to his ultimate triumph." This conception of structure may account for the texture of the narrative in our sample.

[21] Cf. Russo 1976.

words in the phrases as long as the metrical length is preserved and the meaning remains unchanged.

3. Repetition of a formula, even if it be in another part of the line from that of the verse being analyzed.

4. In dividing the hexameter into parts one should consider that there may be lines that should be treated as a whole, that cannot readily be broken into parts. Otherwise there may be normally two or three parts to the verse.

5. When a single word is repeated in the same position in a line, it is conclusive evidence in itself for a formula only if the single word occupies the entire part of the line, as happens sometimes with the run-on word or at the end of a line. Otherwise the repetition of a single word in the same position in the line is permissible as evidence only if it is part of a system, which would include the phrase being tested.

To principle number 1 there can hardly be any objection, since morphology is a fact of formulaic life in all three traditions studied in this volume (although they tolerate different kinds and degrees of morphological variation). I would add only the words and texture to length , since a morphological change that also modifies metrical word-type can affect placement and therefore formulaic structure. The same addition is even more important in principle number 2, since inversion and other changes take place relatively seldom in Homeric phraseology without running afoul of the delicate balance of the inner metric of the hexameter. Once again, the Serbo-Croatian deseterac and Old English alliterative line tolerate this kind of change much more readily than the hexameter. The third principle pertains to relatively few cases in Homeric diction, again because of the complex weave of the prosody. Principle 4 proves extremely important, since, as predicted at the outset of this chapter, the prosody allows, or even encourages, the formation and maintenance of units of different sizes. I would add what I consider to be an important modification to this principle, a modification already borne out in our first two analyses: namely, that line and hemistich patterns are not necessarily to be treated exclusively as either the one or the other. We have seen above, for example, that a hemistich formula or system can also participate in a whole-line formula or system, so that in certain cases it is impossible to isolate the "kernel" of a group of related phrases. As for the fifth principle, the idea of making sure that single words have some phraseological context in order to be considered as individual elements in the diction is an excellent one, and we shall have more to say on that score as we go over examples of that phenomenon in the discussion that follows the marked passage below.

Every investigator must make some assumptions about what constitutes valid evidence, and, pace those who seek after the chimera of absolute uniformity,^[22] scholars will disagree about close cases. In order to make the following

[22] Gf. Miletich 1976.

quantitative analysis as uniformly productive of formulas and systems as possible (this goal in turn being an effort to be as open-minded as is practical about the explicative power of conventional formulaic theory), I have resolved the "close cases" in favor of formulas and systems. That is, in those instances which might be better explained by simple word-type localization than by positing formulaic structure, I have for the purpose of the initial quantification called the units "formulaic." In the discussion that follows the analysis, of course, I examine other methods of explanation.

One more assumption should be noted. In concert with what we have learned about prosody and phraseology in chapter 3 and earlier in this chapter, I take the minimal phraseological unit to be the hemistich, not the colon. The latter, we recall, is normatively a unit of meter rather than of phraseology, although the emphasis in that formulation and others must remain on normatively . Colonic "words" and phrases do in fact develop, especially in the C1 fourth colon, where, as we have seen, the metrical extent of the line-part is within three morae equal to that of a (B2) second hemistich. There simply is no easy answer to this quandary, no place to draw the line absolutely without mismarking a reasonably large percentage of the diction. The fact that I choose to draw it here does not mean that I do not recognize colonic formulas, particularly noun-epithet formulas like glaukôpis Athênê Even though under my set of assumptions such phrases are taken as parts of formulaic systems and not as formulas in their own right, the final percentages of "formula plus formulaic" will remain the same. What is more, using the hemistich as the minimal unit of phraseology allows a quantitatively defensible scheme of representation, since we can calculate percentages based on hemistichs and whole lines. Otherwise, calculation would either have to proceed by cola or have to place formulas and systems of various lengths against a single standard. Both alternatives are fatally flawed.^[23]

What follows, then, is a formulaic analysis of 5.424-44 carried out under the principles established and discussed above. The Greek text is underlined to reflect those words which actually occur elsewhere in what I take (with generous definition) to be formulas or systems, with the minimum unit of phraseology understood to be the hemistich. Wherever the structure of a line is such that one could posit either a hemistich or a whole-line unit, I have consistently marked and counted it as an instance of the larger, whole-line

[23] One cannot use cola as a base because not only is the colon normatively a metrical unit, but many cola are much too short to provide a site for formulaic structure (consider the three-mora third colon between a B2 and a C1 caesura, which would have to be assessed for formulaic character if cola were used as a basis for calculation). The other alternative, calculating the percentage occurrence of phrases of various lengths against a uniform standard, is logically untenable.
We also recall that the present analysis has as its goal only a diagnostic examination of formulaic theory, and not a determination of orality on the basis of formulaic density.

unit.^[24] Likewise, if both an exact repetition and inexact, formulaically related phrases occur, I have taken the line or hemistich as a formula without further notation. My literal translation of the passage is appended, along with a summary of the results and supporting evidence for the judgments made about individual lines (at least one comparand for each unit identified).

While he was truning these things over in his mind and in his heart,
Just then a great wave bore him toward the rugged shore.              425
His skin would have been torn, his bones all battered together
Had not the flashing-eyed goddess Athena inspired him.
As he was rushing on he grasped a rock with both hands
And, groaning, held on while the great wave passed over.
And in this way he escaped the wave at first, but suing back again              430

[24] It is important to note that not all lines that show two hemistich systems can support the hypothesis of a whole-line system. Some hemistich patterns are conjunctive , in that they fore a larger whole as well as two smaller patterns, and others are disjunctive , in that they occur together only once in the Odyssey and therefore cannot be shown to fore a whole-line traditional unit.

It struck him rushing on, and cast him far out into the sea.
As when an octopus is dragged from its lair
And the close-packed stones are held in its suckers,
So the skin was torn from Odysseus' hands
By the rocks. And the great wave covered him over.              435
Then, having reached beyond his measure, Odysseus had surely
              perished, miserable,
If flashing-eyed Athena had not given him presence of mind.
Emerging out of the wave, which belched in to the land,
He swam close by, looking toward land to see whether
He could discover the sloping shores and harbors of the sea.               440
But when indeed, swimming along, he came to the mouth
Of the fair-flowing river, the best place presented itself to him,
Free from rocks, and there was shelter against the wind.
And he saw the river flowing out and prayed in his heart.

The list below summarizes the analysis carried out on Odyssey 5.424-44 (see also the formulaic density statistics in table 18).

Discussion of the Passage

In general, it is clear that the second hemistich shows more traditional patterning—both of formulas and of formulaic phrases—than does the first. Although we are not attempting to classify the text as oral or written on the basis of these figures, the hemistichs do reveal this obvious reflex of right justification. In what follows, I shall annotate these quantitative measurements, glossing them where necessary with additional information and offering alternative explanations of phraseological structure where appropriate.

5.424 . Counting this line as a whole-line formula obscures its component parts and their innate flexibility in combination. First, the second-hemistich formula occurs eleven times in addition to 5.424 and its two verbatim repeats. One could also relate another whole-line formula (24.235), , which, with allowed morphological variation, occurs three times in the Odyssey , but this explanation would likewise not take note of the other first-hemistich partners with which kata phrena kai kata thumon joins. Or, alternatively, we might choose to explain this line as one realization of a formulaic pattern [X] , where X represents either function words or the participial structure hedzomenos d ' in the other instance at 6.118. While this hypothesis has the advantage of indicating some of the flexibility associated with this group of related phrases, it too falls short of a complete and synthetic description.

To begin with, we may note that while the second hemistich of 5.424 joins with many partners, the first hemistich is found only in this particular combination. This situation then suggests that, although we shall not be able

to reduce this collection of lines to a single kernel element and a series of elaborations, we should focus our analytical efforts on the second hemistich in order to determine what makes it so consistently realized as a formula in its own right. The word-type localization of thumon offers a clue: the metrical shape generally occurs only 13.8 percent of the time at line-end (position 12), but thumon turns up at 12 in fully 34.3 percent of its 102 occurrences. Clearly, then, some process is interfering with the average distribution of this word.

As we have already learned, Homeric phraseology tends toward "larger words" composed of smaller, individual words. Once admitted as larger groupings, such amalgams function as unitary word-types. Although their individual word-types had to be favorable to combination and localization (or at least permissible) in order for the amalgamation to take place, the composite structure then takes on a life of its own. This is precisely the case with the word thumon , whose thirty-five occurrences at line-end are broken down as follows:

The first three phrases, all of them colonic forms, are composite "words" that act as larger units and obey traditional rules that apply to those units; thus kai kata thumon and hon kata thumon , for example, occupy the most favored location for an adonean shape, position 12. Under traditional rules, which designate position and sequence through right justification and the caesura structure of the inner metric, a group of colonic forms has grown up around thumon . Various levels of fossilization are apparent: hemistich, colon, and shorter phrase. Some of these possibilities are extremely useful for combination with a variety of less fossilized partners, and so we also encounter 5.424 and its entourage of verbatim repeats and more distant relatives. The point is that none of the lines treated is in any sense "archetypal" or "seminal," and yet the similarities are unmistakable. All of them took shape under traditional rules; whether those rules are active in the formation of the line at hand or to differing degrees vestigial because it was under their guidance that a fossilized element must originally have come into being, we must ascribe the traditional structure of these lines—both individually and as a group—to their influence.

[25] Some of the occurrences in this category are likewise to be explained as colonic forms, but none is frequent enough to merit individual consideration. A few also exhibit a blocked G caesura, so that they in effect modulate to hemistich units.

5.4.25 . Although the argument can be made that both of these hemistichs are formulaic, we come nearer the true structure by explaining them as instances of word-type localization. The phrase mega kuma , a composite "word" that recurs twice within this very passage (5.429, 5.435), behaves as if it were a shape, with favored positions at , , and 12.^[26] The first two of these placements, both of which result in colonic forms, are favored because of their accordance with the inner metric of the hexameter, and are more basic to the formation of the phraseology than the hemistich patterns for which they serve as the foundation.

The evidence for a system in the second hemistich is even flimsier, with the hypothesis resting on the single line 14.1: . Since there is nothing in the syntax that relates this line and 5.425, we must dismiss the notion that calling this hemistich formulaic leads us toward a better understanding of it as traditional phraseology. What is operative in this case is simple localization, with trêcheian taking one of the two positions (, ) that together account for over 90 percent of the occurrences of this word-type. While we must discard the label of formulaic system for this hemistich, then, we can still see traditional rules playing a dynamic role in its formation.^[27]

5.426 . Here conventional formulaic theory works quite adequately, calling to our attention the existence of the fourth colon in 12.412: . The only drawback to this explication is its positing of a second hemistich that can, for reasons given above, be understood only as formulaic. This account of the line's structure downplays the integral nature of sun d' oste' araxe , which, although only colonic in extent, is a complete syntactic element by itself and from all points of view a unit to which nothing need be added to fill out a larger structure.

In part, this difficulty is due to the stricture we have advocated about the hemistich as the smallest unit of phraseology, and the situation is in part ameliorated by the continuing emphasis of this chapter on the normative—and not absolute—functions of both colon and hemistich. That is, it has been noted that the C1 fourth colon is extensive enough a site to permit the

[26] The actual figures for mega kuma coincide generally with these word-type (WT) localization figures. Seven instances occur at  (WT = 42.7 percent), five at  (WT = 22.6 percent), and one at 12 (WT = 31.7 percent). We may ascribe the relative paucity of instances at 12 either to the small size of the sample or, more likely, to the second-hemistich system apparent in 5.296, 5.435, and 14.315 (which may have helped to fossilize mega kuma in a yet larger structure).

[27] Compare the two occurrences of trêcheian in the Iliad (2.633, 2.717), both of which occupy line-end position (at 12). In both cases the adjective has joined with a proper name to form a composite "word" nearly a hemistich in length (preceded by kai ). The nature of this larger "word" or amalgam is made clearer by the fact that (a) both C caesura positions are blocked by the proper name and (b) the preceding, and presumably optional, kai illustrates the malleability typical of right-justified units.

formation of phraseology,^[28] and that we should not try to fashion an exclusive and unbending rule from what is actually a strong tendency. In the larger context, we must also remember that the diction forms according to traditional rules of word-type localization and inner metric, and that, other things being equal, the principle of right justification favors patterning (or fossilization) toward the end of the line and intra-linear units. In this case that process of fossilization has satisfied traditional rules that pertain to the C1 fourth colon, and a phrase has been created that becomes useful, or so it appears, in composition.

5.427 . This line does not quite as easily yield up its structure to conventional formulaic theory, for behind the convenient approximation of "whole-line system" used for quantitative purposes lies a mélange of interlocking units that do not fall out into additive parts. The second hemistich, for example, can be understood as the full noun-epithet formula thea glaukôpis Athênê or as the shorter version glaukôpis Athênê ; the Odyssey contains thirty-two occurrences of the former and eighteen of the latter, and both phrases combine with a wide variety of partners. Nevertheless, in addition to the lines that allow us to posit a whole-line system [X] , where X stands for têi d ' ar ' in the two additional occurrences, we also find evidence for a first-hemistich system [X] phresi thêke , where X varies widely.

Once again, however, traditional rules help to put the phraseology into perspective. Right justification explains the second-hemistich pair of noun-epithet formulas not simply as bi-forms that fit snugly into the correct spots, but as matched elements that typically allow for flexibility at the opening. Likewise, not only is phresi thêke backed up against the mid-line break with the initial part of the first hemistich open to syntactic adjustment, but as a composite "word" this expression also occupies the position most favored for its metrical word-type. Viewing the two hemistichs from this perspective relieves us of the necessity of wrestling with the complex multiformity of the "whole-line system" and points the way toward understanding 5.427 and related phraseology as the partially fluid, partially fossilized medium that it is.

5428 . Since conventional theory offers a satisfactory interpretation of this line, I shall not pause long over its finer points. Suffice it to note that the first-hemistich formula blocks the A caesura, and that the consequent lack of flexibility must have contributed to preserving this phrase as an integral whole. It is entirely detachable from its present partner, which in turn is a classic example of a system (again with a blocked caesura) that is nonetheless formed under traditional rules.^[29]

5.4.29 . Since the only other occurrence of stenachôn shows no formulaic

[28] Thus Kirk's claim (1966) that the C1 caesura is an alternate mid-line break. See also Kirk 1985, 24-30.

[29] Note that epessumenos is at position 9, the most favored locus for its word-type (76.5 percent).

relationship whatever to this line (although it too occupies position 5), this first hemistich must realistically be classed as non-formulaic. This placement does, however, follow traditional rules of localization (second most favored position, 34.0 percent) and right justification. Moreover, as was apparent from the earlier discussion of epea pteroenta , participles are especially preferred just before the mid-line caesura. The second hemistich, again involving mega kuma , this time shows a truer formulaic system than was discovered in relation to the first-hemistich occurrence of this composite "word" in 5.425. This difference is due immediately to the more extensive C1 fourth colon, and ultimately to the principle of right justification which underlies colon formation.

5.430 . Of the first hemistich all that can be said is that it follows right justification in its general ordering of elements and in the placement of hupaluxe in its most favored position by word-type. In a sense this sequence of words amounts to the antithesis of, for instance, the noun-epithet formula treated above, (thea) glaukôpis Athênê , the former being apparently a "nonce" creation and the latter a fossilized, composite "word." Prior theories would treat these two phrases as diametrically opposed, and there is no doubt of the difference in structure and deployment between the two. But what that dichotomy obscures is the fundamental fact that both obey traditional rules; both phrases take shape under the guidance of word-type localization and other aspects of right justification. The phrases for Athena (really a matched pair as already observed) exemplify the end result of the traditional process, whereby—to put it gnomically—words have jelled into "words" and the rules have become vestigial for individual components, while in the first hemistich of 5.430 those same rules are active and dynamic. Of course, because of its difference in compositional function, the latter sequence is likely to remain ephemeral, and we must always keep in mind that phrases—even if created under the same set of strictures—develop according to their role in the compositional process. Nonetheless, at the source of this development, no matter what its direction or eventual product, we discern the guiding force of traditional rules.

The second part of the line, typically more structured than the first, forms around the placement of palirrothion at position 9, the placement of choice for this word-type (76.5 percent, as with epessumenos in 5.428). Since the only other occurrence of this word (9.485) is found at position 9 but bears no formulaic resemblance to the present phrase, we must conclude that the governing principle behind the second-hemistich phraseology is once again traditional rules and not a hypothetical system (which would at any rate still be based on those rules).

5.431 . On the evidence of three line-initial instances of plêx- , one might perhaps insist on maintaining a formulaic system as the best explanation of the first hemistich. However, the colon structure varies over these examples, as does the syntax, and we may more easily interpret this phrase as the product of a compromise in localization. Position 5 is the only other possibility besides 9

(see 5.430, 5.428) for a word of the metrical shape of epessumenon , and, as noted earlier, participles are favored just before the mid-line break (cf. 5.314). With this common participial shape backed up against the B1 caesura, the hemistich is in effect open to the typical kind of initial modification we have encountered many times before in the hexameter.

Likewise, parallels can be cited that seem to argue for [X] embale [Y] as a system, the most prominent being [X] embale chersin , which occurs three times. But this explanation then runs into the complication of the variety of forms in which line-final chersin participates. One firm handhold in this sea of related but complexly interwoven phraseology is the regularity with which adonean clausulae form. These C2 fourth cola are, in turn, very often based on the placement of spondaic words at line-end (58.9 percent by word-type alone).^[30] To show how embale pontôi can be understood as the product of the same process, I adduce the six Odyssean examples of another adonean—eureï pontôi :

	(1.197)
	(4.498)
	(4.552)
	(2.295)
	(12.293)
	(12.401)

While all six lines share the same line-final phrase, which has itself formed under traditional rules, it is deployed in a variety of ways. The first three examples offer an illustration of right justification and the morphology of formulaic diction, as what amounts to a three-colon phrase based on the adonean shows the typical malleable opening that allows for adjustment to syntactic context. Exactly how flexible this first colon is may be understood by noting that this opening part of the hemistich and line varies the position of eti according to the other function words, which are in turn dictated by the situation-specific usage. The second group of three examples illustrates another kind of morphology, which involves an alternate B caesura and second-hemistich formula. In all of these cases the present participle is, as might be expected, backed up against the mid-line break, and variation, as in the first group of lines, takes place primarily in colon 1, as prescribed by right justification.

Instead of insisting on a formulate system for the second hemistich of 5.431, then (and instead of insisting on one or another formula for lines involving eureï pontôi ), I would advocate explaining both situations via traditional rules,

[30] A relatively large percentage (in fact the highest) for such a short word-type—that is, for a word-type that can, at least theoretically, take so many different positions in the hexameter.

which favor the formation of an adonean "word" to occupy position 12.^[31] Widening the focus to take account of the entire hexameter instrument, in lieu of limiting interpretation to a single word-group produced by that instrument, allows us to credit the poet with a more flexible, multiform linguistic medium, a medium well suited both in its malleability and in its referentiality to the twin goals of compositional aptness and traditional art.

5.432 . At first sight there seems to be nothing traditional about this line. The lack of demonstrable formulaic expressions in either hemistich can be attributed at least in part to the fact that this is the first line of the famed "octopus simile." And this in turn means not only that the vocabulary to be employed here and throughout the simile will be far less familiar than in the straight narrative part of the poem, but also that this particular simile, unassociated with any others by Scott (1974), stands alone and unconnected with the rest of the Homeric poems. It is in short no wonder that we find no evidence of patterned diction in the concordance.

But if the words themselves are not repeated, the rules under which they combine to form this and all other lines are ever present, and this unique line thus provides a good test case for seeking traditional structure through the influence of rules rather than through the phraseology which is their issue. We must first acknowledge hôs d ' hote as one possible realization of the initial multiform partner of the hôs ... hôs frame that marks all similes in Homer, with its opposite number to be found at the beginning of 5.434. From that point on, this line is purely the creation of right justification and traditional rules, as all three word-types take their most favored positions: poulupodos at 5, thalamês at 7, and exelkomenoio at 12.^[32] . There are no signs of agglomeration of these elements at all; no caesurae are blocked, no formulas or systems disturb the word-type distribution, and each word behaves individually according to its type. In comparison with other lines examined in this section, 5.432 seems to be a line formed according to traditional rules but without much probability of multiformity. It seems, in short, like a simile.

5.433 . This line is not much more productive of classically defined formulas or systems than the one preceding, and for many of the same reasons. Nevertheless, we do notice the hapax legomenon kotulêdonophin backed up against the mid-line caesura as would be expected; in this case, the blocked A caesura is of no great import because a word of such extent must block

[31] On adoneans, cf. Nagler 1974, esp. 5-9; and Ingalls 1972. Note also that while by word-type we would expect only 58.9 percent of the occurrences of pontôi to be line-final, in actuality fully thirty of thirty-seven, or 81.1 percent, are at position 12. This disparity can be traced to deflection of the normal distribution by the patterned phraseology that we have examined, in which pontôi is fossilized as part of a composite "word."

one of the caesurae no matter where it is located.^[33] Of the many examples of echontai and its relatives in the Odyssey , none really matches the syntax of the second hemistich, and so the interpretation of that section of the line as a formulaic system is legalistic at best. What we can say about the latter part-of the line is that echontai and its morphological kin (ech - plus ) occur 105 times in the poem, with fully 91.4 percent of them at line-end, in harmony with word-type localization. This placement then associates with numerous other words in different combinations to form a wide variety of patterns, not all of them classically formulaic, that cannot be reduced to sets of units.^[34]

5.434 . This line also shows signs of being "nonce" diction, that is, phraseology composed according to traditional rules without discernible roots shared by the rest of the Odyssey diction. The hypothesis of a first-colon system rests on petrêisi 's recurrence at  in 5.156, an interpretation made somewhat more attractive by the fact that it is also preceded by a preposition there, forming an identical accentual grouping. But we must also contend both with the repetition of pros petrêisi in line-initial position at 9.284 and with the reality that word-type localization of the composite form (proclitic plus petrêisi or ) is the most basic determinant of position at .

Traditional rules and right justification are even more obviously behind the formation of the second hemistich, which has no possible formulaic relatives whatever. The adjective thraseiaôn may block both positions for the C caesura, but it does take the position most favored for its word-type (51.7 percent). And the importance of considering words in their accentual groupings is illustrated by cheirôn , which in position 12 seems generally to go against localization rules; although the spondaic word is at 12 fully 58.9 percent of the time, cheirôn occurs there in only 12.5 percent of its instances.^[35] In the present case we must look to the accentual grouping apo cheirôn , or , which turns up at position 12 every one of the forty-four times it is employed in O'Neill's sample of one thousand lines. Thus both hemistichs show signs of being "nonce" diction, but still diction that depends on traditional rules.^[36]

[33] In effect, of course, the "word" is a hemistich long because pros is proclitic.

[35] Again, as with pontôi , the reason for this deflection is fossilization of the individual word within a larger pattern, which then obeys rules appropriate to the larger word-type.

[36] It is worth mentioning in passing that a blocked caesura such as the one in 5.434 very often marks phrases that are at one of two extremes, either highly formulaic (usually straight formula; cf. the usage of trêcheian in the Iliad as documented in note 27) or "nonce" diction.

5.435 . The hypothesis of a system underlying the first hemistich is much weaker than the positing of a similar pattern behind the second hemistich. The evidence for the former rests with two other occurrences of hrinoi , with different inflections, in line-initial position; in these cases the differing inflection is part of a different syntax, and so the hemistich pattern is not truly systemic. At the same time, however, hrinoi occupies the second most favored position for its word-type and, with apedruphthen , itself in the most favored placement for , shows right justification of the metrically more extensive element. As noted above, the mega kuma phrase does seem to form part of a bona fide system (see discussion of 5.429).^[37]

5.436 . The opening part of this line provides a clear example of how the categories of formula and word-type localization can become blurred. For the first hemistich we can adduce the following comparand (6.206): . We can then posit the hemistich system [X] dustênos , where X may represent a number of function words. Legalistically, this is a formulaic system, but does it really function as such? The same phenomena, which do not include an essential idea, can be accounted for simply by word-type localization (the shape occurs 54.4 percent of the time at 5). Such a right-justified arrangement would be further favored because of the malleable opening it features. Whether we actually include this hemistich in a list of systems, then, is finally of little consequence; even formulas and systems, as we have seen, are of widely varying types, flexibility, and so on. It is more important to recognize that, on the spectrum of traditional Homeric phraseology, this hemistich pattern falls just at the cusp between a phrase due entirely to traditional rules and a true system.

And while the second hemistich may also seem to have recourse to a system involving huper moron , we best construe the half-line by noting the confluence of two localizations: Odusseus at line-end (, 92.9 percent at 12) and huper moron , a composite "word," as the third colon ( 95.6 percent at 8). One probable method for the construction of the hemistich would be the establishment of this form of Odysseus's metrically variable name at line-end, with a verb forming the adonean clausula and the third-colon "word" filling back to the B2 caesura.

5.437 . The noun-epithet phrase glaukôpis Athênê (treated above in the discussion of 5.427) is the only true formulaic element in this line, although the rest of the diction naturally follows traditional rules. Word-type localization and right justification order the placement of individual words throughout.

5.438 . In an attempt to stretch formulaic theory as far as possible, I counted both hemistichs in this line as formulaic. Of arguably greater significance,

[37] Cf. kuma kalupsen ("wave covered," 5.353) and, by analogy of sound, kôm' ekalupsen ("sleep covered," 18.201), which appear to be related to the present phraseology not through formulaic character but rather through traditional rules.

however, is the fact that kumatos and ereugetai are placed in the most highly favored positions for their respective word-types. Also entering the picture is êpeironde , which occurs eight of ten times at 12. Other associations among these words are lacking.

5.439 . Comparative evidence yields other examples of nêche at line-beginning and of parex at the end of an A2 first colon, but not of these two occurrences together. The only replicable part of the first hemistich is the phrase es gaian , which functions as a single "word" both linguistically (proclitic plus object) and compositionally; by word-type its preferred position is as a colonic form at , and seven of eight instances are placed in that spot.^[38] This positioning is thus traditional, whether we choose to view the hemistich as a system or not.

The participle horômenos is found four more times at 8, but this regularity seems to be due more to localization of the generic word-type than to systemic structure involving this particular word. A better possibility for a system is the incomplete phrase ei/ên pou epheuroi/-ô at line-end, as also employed in 5.417 with complement in 5.418:

I underline to indicate that not only does the last colon of 5.417 match the equivalent part of 5.439, but that the following lines are identically the same (5.418 = 5.440). In other words, what we have in this instance is a formula five cola long—a combination that supersedes the usually formidable line boundary with the aid of an institutionalized (necessary) enjambement. To the spectrum of units of phraseology we now add an element larger than a single line, to accompany the line, hemistich, and colon units.^[39]

5.440 . For quantitative purposes, and because conventional formulaic theory has no provision for units longer than a single line (unless it be a run of an indeterminate number of lines), I ranked 5.440 in the category of "whole-line formula." As demonstrated above, however, 5.440 is really the latter part of a formula that begins with the adonean clausula of the preceding line.

[38] The sole exception occurs at the secondary word-type position (). Since localization figures indicate an expected relative distribution of 52.4 percent for  and 37.8 percent for , we can interpret the deflection of those statistics by es gaian as a sign of formulaic structure.

[39] Since these two instances of the phrase are only about twenty lines apart in Book 5, and since no other examples are to be found in the Odyssey , the issue of proximity and its effect on the formation and maintenance of phraseology must arise. From the point of view of traditional rules, there is nothing to prevent the temporary existence of certain combinations that are not preserved beyond the moment of their immediate usefulness, just as there is no reason to doubt that certain combinations—perhaps by virtue of their wider, more generic applications become fossilized and are employed again and again. Of course, there must be, within a given poem or sample of material, the narrative opportunity for a given phrase to occur and recur, so the element of utility in a given narrative is another dimension to be considered.

The present line also reveals some unusual inner structure that leads to understanding its formation under the aegis of traditional rules. In the first hemistich, the combination of êïonas-te and paraplêgas , both metrically heavy elements but both in their second most favored positions according to word-type, leads to blockage of the A caesura and even of the mid-line break. That this metrical infelicity is offset (and from a compositional—that is, synchronic—point of view perhaps caused) by what follows is supported by the C1 phrase at line-end. To this fourth colon we adduce two examples, the first from the Odyssey (13.195) and the second from H. Apoll . (24):

These examples establish both line position (including in both cases a blocked mid-line caesura) and the integrity of the C1 phrase as a traditional pattern. Typically, this final unit is the most defined one, the earlier part of the line being constructed according to a compromise between localization tendencies. With the longer formula stretching from 5.439-40, then, we discern one component unit that also exists in other combinations, with the whole set of phraseology governed by traditional rules and right justification.

5.441 . The first hemistich may be interpreted as one realization of the system [X] potamoio , where X may be any of a number of function words, in the same fashion as the first hemistich of 5.436. But word-type localization explains the situation just as readily without having to resort to a pattern for which no essential idea can be demonstrated. Placing potamoio at , up against the B2 caesura, allows a malleable initial section of the hemistich and thus produces an arrangement that has many of the compositional benefits of a formulaic system, short of the word association. The phrase kata stoma seeks position 8 (the localization figure is 95.6 percent) in 18.97 as well, but the lack of a match in syntax makes a system a less satisfactory explanation than traditional rules.

5.442 . On the basis of 7.281, , one can posit a whole-line system underlying 5.442. In order to be cautious and to respect the possibility that the accumulation of function words in the second colon was fortuitous, however, I settled on calling the first hemistich non-formulaic and the second a formula, for the purposes of the quantitative analysis summarized above in table 18. In fact, an investigation of eeisato (favored 95.6 percent of the time at position 8) and associated diction reveals that none of the explanations from formulaic theory is entirely correct, for each fails to account for a significant part of the related phraseology.

I choose to represent the interrelatedness of this diction in a kind of stemma, as shown in figure 3. Traditional rules provide for the localization of eeisato at 8, either with or without a pattern. Those lines which on available evidence seem to lack further pattern are 2.320 and 22.89:

Figure 3.
Eeisato Diction

We also observe a core element aspaston eeisato grouped around the mid-line caesura and either unrelated to other elements, as in 5.398, , or part of a whole-line formula different in some respects from 5.442 (7.343 and 8.295),

Then there is in addition the possibility of the whole-line system mentioned above, which would include 5.442 and 7.281.

This collection of related lines illustrates how formulaic diction of various sorts can evolve from simple word-type localization, or in other words how traditional phraseology can develop from traditional rules. Not only the preferred placement of eeisato but also the positioning of the other involved words is governed by localization and right justification, and it is on this basis that formulaic diction takes shape. Establishing formulaic structure is an important, even invaluable part of the study of traditional phraseology, but in order to be fully understood that structure must be interpreted against the background of traditional rules, which govern all Homeric lines.

5.443 . According to the program described above for the quantitative analysis, I have counted 5.443 as a whole-line formula, based on the exact recurrence at 7.282. As with 5.439-40, however, we discover that the phraseology reaches beyond a single line:

(7.281-82)

Once again the unit of phraseology is longer than the single hexameter line, and once again enjambement—on this occasion "unnecessary" enjambement—assists in the extension of the phrase. Within this larger pattern, we also find evidence for a second-hemistich system, as suggested by 6.210 and 12.336:

Both the longer and the shorter patterns follow traditional rules of word-type localization and right justification, with one exception: the line-final anemoio . The expected figures (average occurrence) for words of the shape , together with the actual figures for this word, at positions 12, , and , are, respectively, 31.7 and 84.6 percent, 22.6 and 15.4 percent, and 42.7 and 0.0 percent. The preponderance of occurrences at 12 may be explained as the result of a two-step process typical of the development of diction in the epos. First, position 12 is one of the three preferred placements for this word-type, so there is no difficulty with the simple admission of anemoio to the position at line-end. From that point on, this particular word has become caught up in a number of traditional phrases, or composite "words," which behave as larger units. In addition to the formulaic patterns already mentioned (which account for four occurrences of anemoio at 12), these patterns are is anemoio (three occurrences) and hama/meta pnoiêis anemoio (four occurrences). Thus the deflection of word-type localization stems from formulaic structures in which the given word has been fossilized, structures that are formed in concert with traditional rules governing individual word-types but that later on come to obey the rules appropriate to composite "words."

5.444 . Although for the sake of quantitative analysis both hemistichs were counted as systems, the latter one more clearly fits the criteria than does the former. The hypothesis of a first-hemistich system rests on the line-initial position of egnô in half of its occurrences, a statistic just as well explained by localization alone, especially since no recurrent essential idea can be discerned for the first hemistich. The second hemistich is formed around the adonean hon kata thumon plus a preceding verb, with localization of euxato (see the discussion of 5.424).^[40]

Conclusion

As predicted by the prosaic studies of chapter 3 as extrapolated in the second section of this chapter, the diction of the Odyssey cannot fairly be reduced to a single type of phraseological pattern but h most faithfully understood as a spectrum of phraseology . In the section just completed we encountered traditional

[40] We may see ring-composition in the echo of 5.424 (kai kata thumon ) and 5.444 (hon kata thumon ).

structure at the levels not only of hemistich and whole line, but also of colon and multi-line unit. The colon was shown to be normatively a unit of meter rather than of phraseology, mainly because of its relative shortness but also because of the vagaries of the caesura system; nonetheless, traditional phrases do occasionally form at the colonic level, chiefly in the C1 fourth colon, which is the most extensive metrically and therefore the likeliest site for association of individual words and eventual production of a formula. The smallest normative unit of phraseology was discovered to be the hemistich , at least for purposes of quantitative analysis; at this level we had examples of everything from an invariable formula to a system that could barely qualify as such. The whole line also proved to be a site for exactly repeated and formulaic diction. And finally, the multi-line unit , of which we have had two examples in the passage from Book 5 of the Odyssey , illustrated how traditional association could cross the boundary of the line—not as a run of individual lines (the proper subject of chapter 7 on thematic structure) but as a phraseological unit extended by enjambement.

This spectrum of phraseology is the natural result of the Homeric hexameter, and should be confronted in all of its complexity rather than reduced to simplex elements that may be more easily quantifiable but are not true to the reality of the diction. The "common denominator," as it were, of this complexity is furnished by the universally applicable set of traditional rules appropriate to (because derived from) the prosody of the hexameter; these are primarily word-type localization and right justification. No matter how intractable phraseology may seem from the point of view of formulaic theory, these rules provide a way to explain the traditional structure of the line without having to resort to the necessarily partial explanation of formulas and systems. In this sense, diction can be understood as traditional even if classically defined formulas and systems cannot be demonstrated, since the rules are the primary laws under which all phraseology—even the formulas and systems themselves —comes into being.

In the end it seems best to conceive of Homeric diction not merely as a patchwork of ready-made phrases (no matter how flexible or ideal)—first since much of the diction simply does not fit that model and second because that model does an injustice to the reality of a phraseology that is developing and dynamic—but rather as an ever-evolving inventory of unequal parts under the dominion of traditional rules. At no time will this diction be "complete" in the sense that all of the parts are somehow contained in the vessel of tradition and ready for deployment. Conversely, at no time will the poet be without a large assortment of ready-made phrases, however different those phrases may be from one another. As recent studies have shown (esp. Janko 1982), there is a good deal of change within traditional diction in the Greek epos, and the Serbo-Croatian comparand (see chapter 5) helps us to imagine how formulaic and thematic repertoires must also have been affected

by local traditions and even by the personal habits of individuals in ancient Greece.

Traditional rules offer a way out of this quandary by focusing on the root causes of traditional structure in Homeric diction, rather than exclusively on certain aspects of the diction per se. Under the aegis of these rules, certain cola, hemistichs, lines, and multi-line units fossilize (just as a noun-epithet combination fossilizes), and others do not. Perhaps the frozen bit of phraseology is particularly useful for naming a certain character whose name blocks a caesura even in its most favored localization; perhaps a particular combination proves so useful that not one but many formulas and systems are founded on a single composite "word." Or perhaps, as in the case of the phraseology employed to express the octopus simile, there is no evidence that any parallel units whatsoever exist. In all of these cases, and in the ones situated between these poles on the wide spectrum of Homeric diction, the rationalizing factor is traditional rules. These rules provide the matrix through which all phraseology must pass in order to enter the tradition; they will demand right justification, word-type localization, and observance of the hexameter's inner metric. Having once entered, they may evolve in many different ways—they may form composite phrases, formulas, and systems (which in turn may skew localization figures for individual words, as we have seen); they may remain in their original form, at least for a time; or they may never be used again. The most fundamental point is, however, that whatever the life cycle and growth of such diction, it was formed under traditional rules.