MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 The Trojans, Statistics, and Milman Parry William Merritt Sale I. Introduction When we examine noun-epithet formulae in all the grammatical cases for the Trojans of the Iliad, we are struck by a remarkable fact: with two exceptions,1 no one formula is repeated exactly more than a few times. This is most evident in the nominative, where all the other characters who occur anything like as often as the Trojans display at least one, and usually several, noun-epithets repeated precisely: the same words, the same grammatical case, the same position in the hexameter. Twenty-six of the familiar Homeric char­ acters repeat a formula at least 10 times; the Trojans, who are men­ tioned more often than 16 of these 26, have no formula at all in the nominative case repeated more than 4 times. Since noun-epithet formulae have come to be regarded as the very staples of Homeric composition, the Trojan deficit-or apparent deficit-requires an explanation. 2 This might appear to be merely a matter of pointing to the noun­ epithet formulae for the Trojans and the others, and counting. But pointing and counting are not enough. For one thing, the Trojans might be defective in the number and occurrences of all their formulae, and not simply in noun-epithets; therefore we must in­ quire whether the Trojans possess as many formulae as the others, and whether these formulae occur as often. For another, the Trojans might possess formulae of a different kind, which Homer employed instead of noun-epithets. Finally, a potential numerical 1 Tprocov bt1tooa~cov and a vocative phrase combining them with the Lycians and Dardaruans that I count as a noun-epithet: see 384 infra. 2 I shall suggest an explanation in these pages, but a full statement must await future publication; the current study is chiefly devoted to stating exactly what it is that the Trojans do not have, and showing that this deficit is significant. 341 MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 342 THE TROJANS. STATISTICS. AND MILMAN PARRY deficit, whether of formulae generally or of noun-epithets, is best tested statistically, and statistics requires a precise statement of what is to count as a formula for our comparisons. Because the Homeric scholar is fortunate in possessing a large number of countable and genuinely comparable data, the use of statistics in Homeric scholarship has a long history. Scott, Parry, O'Neill, and Page are four names that come to mind at once of Homerists who have used numbers. 3 Homer's style is repetitious, and repetitions can be enumerated; Homer's text is long, so that repetitions multiply, and portions can be fruitfully compared with other portions. Students who have a point to make can usually provide a large number of examples and argue that their examples come from the same or comparable populations. And whenever we count and then compare what we have counted, we are engaged in statistics. The use of statistical tests in Classical Studies appears to be recent; but tests merely check, confirm or refute, so that in using them we do no more than take another step along an ancient pathway. In this paper we shall be counting references to the characters and groups of characters in Homer and comparing sets of such references. The theory of this kind of set, which goes back at least to Parry's earliest work, has been developed by Gray, Page, Para­ skevaides, and others.4 Usually such sets have been confined to for­ mulae-formulae for shields or for the sea, for instanc~; in recent work I have extended the concept to include all the references, formulaic or otherwise, made by a proper or common noun, alone 3 J. A. Scott, The Unity of Homer (Berkeley 1921) 84-104; M. Parry, The Making of Homeric Verse. ed. A. Parry (Oxford 1971 [hereafter MHV]); E. O'NEILL. JR. -The Localization of Metrical Word-types in the Greek Hexameter." YCS 8 (1942 ['O'Neill']) 103-78; D. L. Page, The Homeric Odyssey (Oxford 1955) 149-56. 4 D. H. F. Gray, "Homeric Epithets for Things," CQ 61 (1947) 109-21 (=G. S. Kirk, ed., The Language and Background of Homer [Cambridge 1964] 55-67); D. L. Page, History and the Homeric Iliad (Berkeley 1959); H. A. Paraskevaides, The Use of Synonyms in Homeric Formulaic Diction (Amsterdam 1984). For bibli­ ographies of recent work on Homeric formulae in general. see M. W. Edwards. -Homer and Oral Tradition," Oral Tradition 112 (1986) 171-230; J. M. Foley. Oral Formulaic Theory and Research (New York 1985) and The History of the Oral­ Formulaic Theory (Bloomington 1988). A good general account of the poet's technique may be found in M. W. Edwards, The Poet of the I uad (Baltimore 1987) 15-48. MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 WILLIAM MERRITT SALE 343 or in phrases. 5 This extension allowed me to calculate the percen­ tage of occasions when a formula is employed to express a given thought-its "formularity." Comparing such percentages is fruitful: for instance, "in Troy" and "from Troy" have a much lower per­ centage of formularity than other place-phrases, such as "in the Greek camp," "to Olympus," "from the battlefield." Homer either lacked, or eschewed, formulaic ways of saying "in Troy" and "from Troy." I suggested that the epic tradition failed to develop such formulae because it did not normally place the narrative action inside the walls of Troy. Homer, who describes many scenes in Troy, failed to use many formulae for "in" and "from Troy'" because he inherited none and developed very few. Whatever the explanation, statistical study can thus expose important facts about the poet's technique-that certain formularities are normal-and about the Homeric text-that on at least these two occasions it behaves abnormally. In the following pages I want to establish an obviously related fact, the Trojan deficit in noun-epithet formulae. Our first task will be to show that the Trojans are not significantly lacking in either the number or occurrences of their overall formulae. To accomplish this, we must first develop criteria for sta­ tistically measurable proper-noun formulae. We then ask whether the Trojans have an appropriate number of different formulae, and discover that they have more than the average number for char­ acters who occur as often as they. We then proceed to measure the formularity of the thirty-eight characters we wish to compare. We anticipate uniform formularity (for this usage, see 352 infra) or in­ telligible divergence: our formulae ought to be compositional tools that the poet applies universally, and if we find deviations we cannot understand, we shall need to re-examine our criteria. As it happens, all the characters except Patroclus display formularities that either approximate 70% or deviate intelligibly. The Trojans are among the characters who appear to deviate, but a ready explana- 5 See w. M. Sale, "The Formularity of the Place-phrases in the Iliad," TAPA 117 (1987 [hereafter "Formularity"]) 21-50, and "The Concept of the Homeric For­ mulae Group," APA Abstracts (1986). The second of these is the preliminary version of the present paper. I make this point to emphasize that Margalit Finkelberg and I, working entirely independently and dealing with entirely different data, have derived similar percentages for formulaic occurrences and similar conclusions from them. See her "Formulaic and Nonformulaic Elements in Homer," CP 84 (1989 [hereafter 'Finkelberg']) 179-87, and nn. 17-19 infra. MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 344 THE TROJANS, STATISTICS, AND MILMAN PARRY tion is forthcoming: their formularity is no different from that of such other groups as the Achaeans and the Suitors. The Trojan for­ mularity is normal. This enables us to take the next step, of showing that another sort of Trojan deficiency does exist. We know that they lack exactly repeated formulae ("'frequent formulae"). We observe that most frequent formulae possess certain qualities largely denied to infrequent formulae, and from this observation we evolve the con­ cept of "'regular formulae" (frequent formulae possessing these qualities). These qualities enable us to determine a minimum number of occurrences required for a formula to be considered regular, and to measure this regularity (the percentage of regular formulae out of all formulaic occurrences) for the 23 of our 38 characters whose overall formulaic occurrence is sufficient to pro­ vide statistically valid measurement). Again we anticipate (except for the Trojans) uniform regularity or intelligible divergence. We find, however, that the regularities, unlike the formularities, do not cluster around any particular percentage; we find instead that for all except the Trojans, regularity varies proportionately with localiza­ tion (the percentage of times that a word falls in that place in the hexameter line in which it most frequently falls). This fact permits us, finally, to pinpoint the Trojan deficit: it is not so much a lack of regular formulae as a lack incommensurate with the metrical properties of their name. It also permits us to rule out meter as the cause of the deficit. In looking for the true cause, we notice another remarkable characteristic of the Trojan formulae: several of their epithets portray them as unfeeling, arrogant, uncivilized-quite unlike the Trojans of our Iliad, very like the Suitors of the Odyssey. Homer as narrator never uses these epithets of the Trojans: they are found solely on the lips of the Achaeans and their gods. Yet the narrator of the Odyssey (whether Homer or another) is perfectly happy to use a similar-indeed often identical-set of epithets for the Suitors, in his own name and often as regular formulae. This points to an explanation for the deficit. Suppose that Homer's predecessors in the epic tradition treated both the Suitors and the Trojans simply as villains and provided Homer with a set of formulae replete with hostile epithets. These would have been his regular formulae, had he continued portraying Troy in the same way. Instead, he changes MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 WILLIAM MERRITT SALE 345 the portrait to one that is far more sympathetic. 6 Homer's Troy is Vergil's Troy and the Troy of most poets since: a city tragically destined to die, condemned by the gods and by the weakness of their own political institutions. The poem is indeed an Iliad, and the harsh epithets are uttered only by Troy's enemies. This explanation can only be adumbrated in what follows. But another theme emerges: regularity, or formulaic frequency, is not only a vital aspect of Homer's technique but also reinforces the fundamental idea of Milman Parry's L 'Epithete traditionelle: the existence of systems of formulae characterized by noun-epithet form, economy, usefulness, and occurrence at certain major metri­ cal cola.? Almost all frequent formulae fall into Parryan formula­ systems. This is a remarkable result and should not be obscured by the fact that we shall be entering several demurrers, e.g. that most infrequent formulae do not fit into Parryan systems. We shall also suggest several important refinements of Parry's concepts, for reasons that will become clear as we proceed. The term "formula" is first given a more general definition than Parry's; we shall then propose criteria for the much narrower notion of "statistically ap­ propriate nominative proper-noun formulae." Parry's contention that the formula expresses an "essential idea" will be restated in terms of Frege's distinction between "sense" and "reference." 8 Parry's "essential" means "what remains after all stylistic super­ fluity has been removed" (MHV 13), and a judgment as to what is stylistically superflous is far too subjective for a fundamental denni­ tion. "Economy" is redefined: metrical overlap is tolerated if the overlapping formulae have importantly differing meanings, or if the overlap is attributable to a generic epithet that is a familiar feature of 6 Whether these changes should be attributed to Homer or to his generation and perhaps its immediate predecessors is not a question that we can answer; but we can say that they were brought about before a new set of Trojan formulae could be developed. When I say "'Homer," therefore, I shall mean "Homer andlor his contemporaries and teachers." 7 See below, 368f. 8 For the term "essential idea" see MHV 13,272; for Frege's distinction see "On Sense and Meaning," in Translations from the Philosophical Writings of Gottlob Frege 3 , edd. P. Geach and M. Black (Oxford 1980), reprinted in Critical Theory Since 1965, edd. H. Adams and L. Searle (Tallahassee 1986) 625-36. A good critique (but in my opinion not severe enough) of Parry's "essential idea" may be found in E. Bakker, Linguistics and Formulas in Homer (Amsterdam 1988) 154- 57. MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 346 THE TROJANS, STATISTICS, AND MILMAN PARRY the poees formulaic vocabulary. The fixed epithet is not called "or­ namental," nor shall we speak of the indifference of the audience to its meaning. And the term "noun-epithet" will now include doubling phrases such as "Trojans and the wives of the Trojans," used in a collective sense. With these qualifications, we can say that our discussion of formulaic frequency entails a remarkable quan­ titative validation of Parryan systems. II. Formularity The entItIes to be counted and compared in what follows are names and naming phrases in the Homeric text. For statistical purposes we group these into sets: "the Trojans," "Achilles," "the Achaeans," "Hera," etc. Instead of maintaining that each set refers to a single "essential idea," the names in each set will be regarded as having one and the same referent (denotation) but not necessarily the same sense (connotation). 9 For our purposes, referents are characters in a given poem, considered either by themselves or as doing, feeling, or saying something. The referent might be Zeus in the Iliad; the sense of a phrase referring to him might be "cloud­ gatherer" or "father of gods and men." Or the referent might be Zeus talking, Zeus angered, Zeus raining. The set has only one personal referent, and contains all the names, formulaic or non­ formulaic, single word or phrase, for that referent. The referent can be plural, provided that the name is collective: "the Trojans," but not "Zeus and Athena." No word or phrase will be included if it is not, or does not contain, a name-either a proper noun or a com­ mon noun used as a name, such as "father, mother, husband, son." No pronouns, no allusions, no implied subjects; for these have a logic, a meter, a syntax, and an aesthetic that require a separate poetic technique. Let us call these sets whose members have one and the same referent "semantic sets." And let us agree to regard a character in the Odyssey as a different referent from a character in the Iliad. We shall be studying 38 such sets for the 38 members of the Homeric corpus mentioned more than 20 times in the nomina­ tive.10 For the most part we shall be discussing nominative sets. 9 See the references in the preceding note. to The choice of 20 T(otal) O(ccurrences) is somewhat arbitrary. The Chi-square test, which we shall use to determine uniformity and deviance in formularity, is thought not to be reliable if the TO for each sample or character in our study is MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 WILLIAM MERRITT SALE 347 Our :first job will be to decide which members of a naming set are to be identified as formulae. Definitions, though many and various, tend to include three common denominators. There must be repetition of some sort-words, sounds, syntax, or meter. There must either be more than one word, or else a single word or gram­ matical form repeated in the same place in the verse. And the word or phrase must be essential to the overall compositional technique. Formulae cannot be merely repeated references to an event in progress (such as instructions given, then carried out); they cannot be repetitions found only in longer repeated passages not them­ selves formulaic (such as the list of Agamemnon's gifts in Book 9); and they cannot be simply deliberate echoes aimed at some special effect in a given passage or passages. ll Though for many purposes a definition this general may suffice, statistics requires greater precision: since we are counting and comparing, we must know exactly what to count or we shall be most uneasy about our com­ parisons. What we need are statistically viable criteria for "nomina­ tive proper-noun formulae. "12 too low: .. too low" varies according to the average formularity of the group being studied. At 85% formularity, 33 TO is in principle too few; at 50%, we can safely drop to 10. The figure of 20 is therefore too low for some of the average for­ mularities we shall encounter, and unnecessarily high for others; it represents a compromise that will work well most of the time. On the use of the Chi-square test, see D. L. Clayman, "Sigmatism in Greek Poetry," TAPA 117 (1987) 73 n.19, with references. 11 On repetition for special effect, see M HV 272-75, and W. G. Thalmann, Con­ ventions of Form and Thought in Early Greek Epic Poetry (Baltimore 1984) 1-13, on ring-composition in Homer. On the definition of ·formula," see J. B. HAINSWORTH, The Flexibility of the Homeric Formula (Oxford 1968) [hereafter 'Hainsworth'] 33-45; C. Higbie, Measure and Music (Oxford 1990); E. Visser, Homerische Versifikationstechnik (Frankfurt 1987) 1-40; and Bakker (supra n.9) 151-95. Bakker's insistence that phrases function as formulae neatly rules out mere echoing. A good many of my infrequent formulae are evidently formulae because they function as such. Some of them, however, and most of the regular formulae are still, in my opinion, to be thought of as building blocks, or as tools with which formulaic lines are fashioned. Also, Bakker starts with the premise of oral com­ position, which I am unwilling to do; even though I find it the likeliest hypothesis, I prefer not to base the statistical analysis upon it. 12 One might ask why, given the controversy over • formula, " we do not simply replace it with ·compositional unit," or the like. The deficiency of the Trojans would be just as apparent. I resist this procedure chiefly because of my conviction that our quantitative analysis vindicates Parry's formula systems and formula MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 348 THE TROJANS, STATISTICS, AND MILMAN PARRY Parry's definition, "a group of words regularly employed under the same metrical conditions to express a given essential idea" (M HV 13, 272), does not tell us what syntactical structures to count; the "same metrical conditions" mayor may not mean the same part of the verse; and the phrase "regularly employed" ap­ pears to exclude certain phrases that occur only once but which many scholars, Parry included, would call a formula (J!EyaSuJ!oc; 'AXtA.AEUc;, for instance).13 Our criteria must be far more precise than this. The structural formulae of J. A. Russo and others, and M. N. Nagler's formulae generated by unconscious templates, are fascinat­ ing and fruitful ideas, but they do not suit our restrictions. 14 Neither possesses clear-cut semantic boundaries, while our interest in the names of the Trojans necessarily confines us to references to per­ sons. Furthermore, as both authors point out, their concepts do not lend themselves readily to statistical methods. At this stage of statistical endeavor, at least, we must stick to counting signifiers re­ peated either exactly or within exactly demarcated limits. We must aim at rigidity in our criteria; we can even afford to be too rigid, since whatever we leave out will be easy to identify and take up in future study. But we are certain to omit some phrases that others will identify as formulae, and which we would have counted if our goals had been different. types, and that the phrases that will eventually most concern us in this paper are precisely the formulae that most concerned him, especially in L 'Epithete traditionelle. Wherever it seems necessary, to avoid confusion I shall use the longer phrase" nominative proper-noun formulae." 13 On unique formulae see MHV 8f, 312f, and 350f infra. 14 J. A. Russo, "'The Structural Formula in Homeric Verse," YCS 20 (1966) 227; M. N. Nagler, Spontaneity and Tradition (Berkeley 1974). Both structural formulae and pre-verbal Gestalten are surely part of the poet's technique, though the qualms of W. W. Minton must be carefully considered ("The Fallacy of the Structural Formula," TAPA 96 [1965] 241-54). Both are generative, in supplying the patterns according to which many formulaic expressions come into being. For a recent view of the development of oral-formulaic theory from Parry's essentially static concept of the formula through the generative views of Hainsworth, Russo, Nagler, and others, see A. T. Edwards, .. KAEOl: A~8ITON and Oral Theory," CQ NS. 38 (1988) 25-30. On the formula itself as generative, see G. Nagy, Comparative Studies in Greek and Indic Meter (Cambridge [Mass.] 1974) 143. I am certain that all these structures will someday prove to be quantifiable; but we must begin with what we can more readily identify and count. MERRITT SALE, WILLIAM, The Trojans, Statistics, and Milman Parry , Greek, Roman and Byzantine Studies, 30:3 (1989) p.341 WILLIAM MERRITT SALE 349 Since our formulae must include a naming noun accompanied by some other word or words, it is natural to ask what other parts of speech to include along with the noun. 15 If we embrace all parts of speech, provided the phrase is repeated exactly, we both exclude too much (we need room for inexact repetitions) and probably in­ clude too much (does the repetition of "Hpll B' in itself constitute a formula?). But if we were to restrict our data to noun-epithet formulae, we might overlook other kinds of formulae-such as noun-verb combinations-that may function in the place of noun­ epithets for some of the characters. The word 6uJ,1oc;;, for example, occurs in the nominative without an epithet in frequently recurring noun-verb combinations in the Odyssey: 116£A.£ 6uJ,1oC;;, 6 times final; 6uJ,1oc;; ... lCEA.E'\)Et(Ot), 12 times final; etc. If, in studying this word, we were to omit these combinations, the result would be certain to mislead. Now, we shall find that no character in the nom­ inative possesses a noun-verb formula as frequent as these. On the other hand, none of our characters is significantly lacking in noun­ verb phrases, so that we build in no bias. Such phrases (often exactly repeated many times) are very frequent, moreover, with nouns in the oblique cases, both proper and common. I therefore find no good reason to exclude them from our study of the nom­ inative; after all, the most fundamental formula is an extended noun­ verb formula, viz. a whole line that divides into a noun-epithet and a formulaic verb-phrase, such as 'tOY 0' TtJ,1dJ3£'t' E1tEt'tU 1tooaplCllC;; 010C;; 'AXtA.A.£\>C;;.16 Precision demands that we include only cases where the noun is the subject of, and not merely juxtaposed to, the verb. But we are almost certain to be counting some combinations as formulae that others might exclude. On the other hand, statistical need urges us not to cast our nets too widely. For instance, some characters display certain noun­ adverb and noun-conjunction combinations in great abundance. Others lack these almost entirely, and we can readily trace this vari- 15 We cannot count single words used repeatedly at the same place in the verse: see Appendix 1.1 and Hainsworth 35 n.4. In "Formularity" (28) I used the term "minimal formula" for repeated preposition-plus-noun phrases. Let us extend this term so as to include these highly-localized single words, as well as nominative noun-plus-adverb and noun-pius-conjunction phrases. 16 Parry included these noun-verb combinations in his general definition (M HV 13f), giving the whole line ~JlOC; o· ,;ptYEVEta <;, which appears frequently in 6 different places; the formularity is essentially the same. This conclusion is so striking that we should, if we can, subject it to a different statistical test. Since we are asking whether meter has an effect on formularity, we need a way of measuring meter. O'Neill has given us the useful concept of localization, the per­ centage of times that words of a certain shape fall at certain places in the line. Localization is clearly a function of metrical shape, and is thus one possible way of measuring the effect of meter in the construction of a line of verse. Modifying O'Neill's definition somewhat, we can state a word's localization as the percentage of times that it falls in that one place in the line into which it most frequently falls, which we shall call the L-point. 22 'AY