^{page 258 note 2} I discuss a practical application of this new ‘law’ to the restoration of a recently-discovered papyrus fragment of Calli–machean elegiacs, Antinoopolis Papyrus 113, fr. 1 (b) on pp. 269 ff. below.

^{page 258 note 3} Throughout this article I have used the abbreviation Maas for: Maas, P., Greek Metre (Oxford, 1962; translated by Lloyd–Jones, HughGoogle Scholar).

^{page 258 note 4} Our understanding of the Greek hexameter, for example, would certainly be fundamentally re–established, if not revolutionized, if all known examples were to be analysed on a computer according to their most important characteristics (syllable lengths, caesurae, diaereses, elisions, word–ends, sentence-ends, colon–ends etc.) and then subjected to tests for statistical association amongst any or all of these characteristics. The task would be very tedious, but simple.

^{page 258 note 5} ‘Der homerische und der kallimachische Hexameter’, Göttingcr Nachrichten (1926), 197–229, reprinted and expanded in Wege und Formenfrühgriechischen Denkens (München, 1960), 100–56Google Scholar. Kirk, G. S., Yale Classical Studies xx (1966), 76–104Google Scholar questions the validity of Fränkel's C1 caesura on the grounds that the unit from B to Ci would be ‘unnaturally short’. Kirk does not indicate how long a unit must become before it can be considered to be of ‘natural’ length, and himself concedes (pp. 91 f.) that statistically the C1 caesura is significant. This is the important factor, for the strength of Fränkel's approach lies in its emphasis on normative behaviour: only on this basis can the effect of divergent phenomena be assessed. Kirk on the other hand offers explanations which concern only the historical origins of the phenomena discussed. (More convincing arguments for qualifying Fränkel's conclusions are given by Dale, A. M., ‘Greek Metric 1936–1957’, Lustrum ii (1957), 30–2Google Scholar.) Kirk also questions Frankel's assumption that the metrical cola are potential sense–units: fortunately this question is not relevant to the present discussion, but it must be objected here that Kirk's sample from Iliad 16 is far too small to be of any statistical relevance, and also that in his analysis he falsely assumes that sense-structure in Greek is identical with syntactical structure.

^{page 259 note 1} Also the symbols XP for ‘no colon–end’ (either at B or C₂) and XC₂ for ‘no bucolic diaeresis’.

^{page 259 note 2} This figure is accurate to within ten lines: fragmentary' papyri make an exact figure impossible.

^{page 259 note 3} This includes fr. 384. 59 where the text is lacunose: the line almost certainly contains a Co diaeresis, and although nothing certain can be ascertained about colon–end, Hunt's supplement (see Pfeiffer), even if incorrect, does show that colon–end is not impossible. I have excluded frr. 7. 27, 528. 1 since óδ is almost certainly anaphoric (see below, pp.262 f.). Fr. 536 fails at the bucolic diaeresis, but it seems more than reasonable to assume that there was colon–end there, and I have done so. Fr. 575 is lamentably corrupt, but my observations in this article on the Co diaeresis go some way to resolving the difficulties. The diaeresis at Co almost certainly requires colon–end either before őφις or after αίòλος: since ώστ’ έξ òχεής seems to make sense only if taken with őøις we should probably punctuate after αίòλος and make őøις an object of comparison with the unknown subject of αύχέν †άναύχην† rejecting Pfeiffer' suggested interpretation (i. 402–3).

^{page 260 note 1} Thus, for example, O'Neill, E. G., jun. in his very sizable work ‘The localization of metrical word–types in the Greek hexameter’, Yale Classical Studies, viii (1942), 105–78Google Scholar decided that it was safer to treat all appositives as metrically independent (pp. 108 ff.), but at the same time was compelled to remark ‘they constituted quasi–units, phrases that cohered just closely enough to suggest single words’ (p. no). And H. N. Porter, who followed this treatment of appositives as independent in a very important article (‘The early Greek hexameter’, Yale Classical Studies, xii (1951),17 n. 33Google Scholar), nevertheless also insisted on making ‘semantic’ not ‘phonetic’ units the significant metrical elements (p. 37).

^{page 260 note 2} Fr. 384. 21 may be another preverb in tmesis, but the text is very fragmentary. For the word–pattern cf. Theocr. 3. 21.

^{page 260 note 3} In fr. 75. 62, which has a preverb in tmesis before Co with colon–end at B2, note the delayed particle which supports our treatingas a single word–unit.

^{page 261 note 1} Percentages have been rounded off to the nearest I per cent throughout this article.

^{page 260 note 2} The number of examples includes 4. 1312, where περί, though an adverb, has the same status as a preverb and as such is probably responsible for the metrical abnormality (XC2) of òξύταται (see the conclusion above).

^{page 260 note 3} In Callimachus elision may play a part in weakening the CO diaeresis: both the three examples cited above of preverbs in tmesis+XP and the other examples accompanied by the usual restrictions (see below, p. 265, n. 2) have elision (either of the preverb or of a following particle) at the diaeresis. However, the number of examples is too small to draw any firm conclusion, and in A.R., of the 21 examples where a Co diaeresis is weakened by a preceding preverb in tmesis, only 6 have elision. In the other Hellenistic authors dealt with below the number of examples is again too small for significant conclusions, but Arat. 967 does not have elision, the other 5 examples do.

^{page 260 note 4} Throughout this article I have made great use of the χ² test for independent samples. This tests the hypothesis that any difference in frequency of occurrences of the phenomenon under investigation in two samples can be attributed entirely to random effects without the need to assume an underlying difference. The result, the ‘level of significance’, gives the probability of mistakenly or falsely rejecting this hypothesis of no difference. If the level of significance is 5 per cent or less, this means that the chance of such discrepancies between the observations arising from random effects is at most 1/20, and we then reject the hypothesis and may consider the two samples to belong to different classes. If the level of significance is more than 5 per cent it will not cause us to reject the hypothesis, and on this evidence we may group the two samples together.

^{page 261 note 1} For a definition of which see Schwyzer, E., Griech. Gramm. ii. 24 fGoogle Scholar.

^{page 261 note 2} Indeed, Fränkel's remark on p. 147 n. 1 leaves very ambiguous the question whether the third person ó δέ, ή δέ, etc. are to be considered as prepositives; in relation to the Co diaeresis at any rate, it is clear that these are no different from the other first–and second–person anaphoric pronouns.

^{page 261 note 3} Naturally, since ò δέ, etc. inevitably involve a preceding colon–end. There are only three passages in the material covered by this article where a pronoun with no connecting particle might be considered anaphoric: Theocr. 1. 59, 131, neither of which exhibits colon–end, and A.R. 4. 499, which has colon–end at C2. Whereas the Theocritean passages involve the use of τύ to strengthen an imperative, the example in A.R. contains a substantive article, without supporting particle, τούς πείσεμεν (this is rare outside Homer, who does not useτο⋯ς so, but cf. Il. 4. 53 τάς διαπέρσαι: see B. L. Gilder–sleeve, Syntax of Classical Greek, § 528; Schvvyzer, , Griech. Gramm. ii. 20 fGoogle Scholar.).το⋯ς here may be prepositive since it is deictic, just as is the anaphoric article.

^{page 261 note 4} Although it is not in any way incorrect to regard the σέ here as anaphoric to v. 29, prepared by τάς μέν in v. 34, even so the XCo could also be explained as due to the ούκ being retrospective and cohering with οε, not έθλιøεν

^{page 261 note 5} This line illustrates a further important metrical point: as Fräankel, 145 n. 2, pointed out, no break is possible after ãρ’ because of Hermann's Bridge, andmust form a single word–unit. Usually an enclitic cannot absorb the force of a preceding prepositive, but can act only as a bridge. Thus theãρ’, which as a conjunctive particle is enclitic, does not absorb the preclitic force of ούκ which must be passed on to έμελλες. The point is strikingly made with iii. 7 καì πολυωνυμίην ίνα μή μοι Φοîβος έρίζηι: a diaeresis after μοι, a monosyllabic eighth element, would be a unique example of a practice which Callimachus conspicuously avoids (Maas, § 92): however, μοι is an enclitic, and as such does not absorb the preceding μή, and somust form a single word–unit for metrical purposes (which, let it be said, usually reflect sense–structure in examples of this sort: μή q qualifies Φοîβος not μοι).

^{page 261 note 6} The test is significant in A.R. at the 0.1 per cent level, in Theocritus at the 5 per cent level, in Nicander at the 2 per cent level.

^{page 263 note 1} Indeed it is of course a common epic device to place an article in an anaphoric position even though the article turns out to be definitive (cf. Svensson, A., Gebrauch des bestimmten Artikels (Lund, 1937)Google Scholar, Exkurs II, pp. 141–52 who touches on this point): the momentary ambiguity serves to heighten interest, e.g. A.R. 3. 270; SO A.R. 3. 1169, 1361, 4. 454, 1446, Aratus 255, 309, 411, Mosch. 2. 122, also Call. i. 27 (on which see Svensson, op. cit. 59–60), iv. 77, though as an epic device it is very rare in Callimachus (fr. 523 looks like another Callimachean example).

^{page 263 note 2} Only 4 of these 16 passages (none of them those with XC2) involve elision, which is therefore probably not to be considered as a possible influential factor (cf. above p. 261 n. 3).

^{page 263 note 3} The Callimachean passages omitted, in addition to the exception noted above, are: frr. 262, 682, ii. 31, iv. 212, E. 1. 5, E. 42. 5. All have C2 diaeresis and usual colon-end.

^{page 263 note 4} e.g. the sequence of Co examples in Theocr. 1 runs: 11, 12; 40; 62, 67; 78, 82, 83, 86; 102, 103.

^{page 263 note 5} Kirk, G. S., Yale Class. Stud. xx (1966), 76 n. 2Google Scholar, defends the 1,000-line sample usually used: ‘Homeric colometry is remarkably consistent.’ This is simply untrue: individual books may differ to the extent of 200 per cent. Also, as can be seen here, the Iliad and Odyssey are very different from one another.

^{page 265 note 1} The Homeric poems are so large that sampling is inevitable; however, I have taken samples much larger than those normally used, and the results should be that much more reliable. The sample sizes were chosen by taking one book and analysing the data, adding a second book and re-analysing the data, etc., and adopting a sample several books larger than the point at which the series of accumulative data showed definite consistency.

^{page 266 note 1} See Fisher, R. A. and Yates, F., Statistical tables for biological, agricultural and medical research (London and Edinburgh, 1948 ³), Table XII on p. 56.Google Scholar

^{page 266 note 2} See Fisher and Yates, op. cit. 13 (‘Other Transformations’) and Table XII on p. 56.

^{page 266 note 3} The most surprising feature among the early works is that the Iliad and Odyssey are completely at variance: the confidence limits do not even overlap, being 2–9 per cent for the Iliad and 9–21 per cent for the Odyssey. Cf. junO'Neill, E. G., ‘The localization etc.’, Yale Class. Studies, viii (1942), 130–1Google Scholar: ‘The Odyssey varies notably from the Iliad, and this variation is usually in the direction of a majority of the Alexandrians. This is true of Hesiod also, but less consistently and less markedly.’

^{page 267 note 1} Mr. Thomas Gelzer informs me that in Musaeus also many apparent deviations from usual metrical practice are the result of anaphora and antithesis. See also below, p. 268 n. 2.

^{page 267 note 2} Two occur in Id. 17, where the XC2 can easily be represented as reinforcing the sentiment of the word which bridges C2 (vv. 44, 56); in the sixth case, 22. 82, the XC2 has no obvious relation to the sense. As regards the first section of Id. 15 Mr. Parsons has kindly drawn my attention to ‘another deliberate oddity’ in this part of the poem in the unusual avoidance of the Bucolic Bridge: this too was no doubt aimed at producing a rough colloquial effect (v. Maas, p. 94).

^{page 267 note 3} X2 is significant at the 0.1 per cent level.

^{page 268 note 1} With both x2 is significant at the 1 per cent level.

^{page 268 note 2} It is worth pointing out that Theocritus stands even closer to Callimachus than appears at first sight. 12 examples are to be found of Co+XP, of which only one (16. 58) has no exceptionally close relationship bridging the diaeresis; of the others seven (1. 40, 102, 2. 161, 6. 4, 33, 7. 85, 22. 201) bridge Co with Adjective + Noun, e.g. 1. 40, 2 (2. 137, 6. 39) with a forward-looking adverb immediately before Co, 1 (1. 103) with Predicate+Verb. 26. 2 has anaphoric repetition (see above, p. 267):

^{page 268 note 3} Mannered only in so far as studied formulation in Callimachus sometimes makes for excessive elaboration; the extra restrictions certainly do not stifle the rhythm of Callimachus’ poetry as they do in, e.g., Nicander. For example, Nicander's use of colon-end with Co betrays a degree of rhythmic monotony not found in Callimachus. Thus:

^{page 268 note 4} Cf. a similar method of explanation for certain metrical characteristics in Quintus Smyrnaeus, Callimachus, and Nonnus in Wifstrand, A., Von Kali. zu Nonn. 37–9. 39 ff., 48 ff., 52 f.Google Scholar