Repetitions and their Removal by the Copyists of Greek Tragedy Peter Pickering T IS BEYOND DOUBT that what Jackson in Marginalia Scaenica calls “unconscious repetitions by the copyist” occur.1 (ByI “unconscious” Jackson means little more than “erroneous”; he is not considering changes deliberately introduced by a copy- ist.) I present here the results of an attempt to examine those instances where verbal repetitions found in manuscripts of Greek tragedy are due to scribal error, and to assess some of the circumstances in which such errors appear.2 An unexpected discovery was that besides introducing repetitions scribes can be detected (or perhaps rather some scribes can sometimes be detected) removing repetitions. Editors quite frequently emend away repetitions found in 1 J. Jackson, Marginalia Scaenica (Oxford 1955), Addenda B, 223. Jackson wittily depreciates the value of listing the large number of examples, and does not therefore attempt any systematisation. There are frequent allusions to scribal repetition in editors’ discussion of individual cruces. Repetitions due to the author rather than a copyist are surveyed in my article “Verbal Repetition in Prometheus and Greek Tragedy Generally,” BICS 44 (2000) 81–99. 2 The research was undertaken as part of work for a Ph.D. at University College London. My thanks are due especially to my supervisor, Professor Janko, and to my examiners, Dr Dawe and Professor Carey, for their many helpful comments on my thesis, to Professor Easterling for her comments on an earlier draft of this article, and to the anonymous referee. The collations on which the research was based were: for the alphabetic plays of Euripides, Murray’s OCT (I 1902, II2 1913, III 1907); for Aeschylus Supplices H. Friis Johansen and E. W. Whittle (Copenhagen 1980); for Aeschylus Septem contra Thebas Page’s OCT of 1972 and R. D. Dawe, The Collation and Investigation of Manuscripts of Aeschylus (Cambridge 1964; Dawe reports a number of var- iants relevant to this study which are not in Page, presumably because they are not significant in the constitution of the text); for Sophocles Trachiniae the third edition of Dawe’s Teubner Sophocles Tragoediae (1996); and for Euripides Hip- polytus Diggle’s OCT of 1984: cited below by editors’ names. Greek, Roman, and Byzantine Studies 41 (2000) 123–139 © 2001 GRBS 124 REPETITIONS AND THEIR REMOVAL texts as transmitted, and in doing so must be assuming that those repetitions were the result of copyists’ errors. Sometimes, also, the manuscript tradition presents alternative readings, one with a repetition and one without. Among the considerations an editor must have in mind in choosing which to prefer is whether a copyist has introduced into a text a repetition not originally there or has removed, deliberately or inadvertently, a repetition the author had written. 1.1 Apographs: Erroneous repetitions Any rigorous study must attempt to separate out erroneous repetition by an individual copyist from other sources of error, especially ones due to contamination in an open recension. Only if we can be sure that one manuscript is a pure apograph of another extant manuscript can we be sure that errors in the first are due to its copyist and to him alone. Where the reading of the copy is different, it must be either by mistake (and “mistake” is the right word even in the unlikely but possible circumstance that it produces the original words of the author), or by de- liberate conjecture. There are many apographs in the libraries of the world, but the labour of collating them is not normally undertaken once their totally dependent status has been demon- strated; and even when they have been collated, those collations are rarely published. Fortunately, however, the Euripidean manuscript usually de- noted P (one half being Palatinus gr. 287 and the other Lauren- tianus conv. soppr. 172) was fully collated before it was generally accepted to be totally dependent (in the so-called “alphabetic” plays)3 on the manuscript usually denoted L (Laurentianus plut. 32.2).4 The readings of P are thus reported alongside those of L 3 Cyclops, Heraclidae, Supplices, Electra, Hercules, Iphigenia Taurica, Ion, Helena, and Iphigenia Aulidensis. 4 Diggle in the preface to vol. II of his OCT says “[Zuntz] controversiam illam diuturnam, quanam necessitudine L et P inter se coniuncti sint, omnino diremit” (p.VI). Diggle refers to G. Zuntz, An Inquiry into the Transmission of the Plays of Euripides (Cambridge 1965) 13–15, where Zuntz describes how a mark in L copied as a colon in P proved to have been a tiny piece of straw. PETER PICKERING 125 in earlier apparatuses, such as that of Murray; a large number of them are also discussed by Zuntz.5 Another, and more recent, published collation of an apograph is that by Friis Johansen and Whittle of Aeschylus Supp. Scurialensis T.1.15, which they denote by E. They published this collation in the belief that the manuscript was not a descendant of Mediceus Laurentianus 32.9; that contention is generally discredited.6 A study of the apparatuses in Murray and Friis Johansen was made.7 All instances where a substantially different word is read in the apograph from that in its exemplar were examined (differences of accent or breathing, or simply of inflection, were passed over, as were all but very striking differences in non-lexi- cals).8 Something like 125 substantial differences were found.9 Table 1 lists the 17 instances where a significant divergence in the readings of apograph and exemplar is certainly, probably, or possibly to be ascribed to the influence of a nearby word.10 5 Zuntz (supra n.4). On p.136 he lists the types of error to which P was prone: dropping single letters, especially consonants; changing or adding letters; writing a wrong but similarly pronounced vowel; and repetition. 6 M. L. West, Aeschyli Tragoediae (Teubner 1990), says in his preface “Quod H. Friis Johansen et E. W. Whittle contenderunt, Md [sc. their E] non a M genus habere sed auctoritate sua aliquid valere, non magis mihi persuaserunt quam aliis viris doctis” (p.XVII) There may, of course, have been a manuscript inter- mediate between M and E. 7 Murray’s readings quoted in this article have all been checked against the published facsimiles of L and P: J. A. Spranger, Euripidis quae inveniuntur in Codice Laurentiano (Florence 1920) for L, and Euripidis quae in codicibus Pala- tino graeco inter Vaticanos 287 et Laurentiano Conv. Soppr. 172 (Florence 1939– 46) for P. 8 With the alphabetic plays of Euripides, despite the total dependence of P on L care is necessary when looking at each particular case, since corrections in L (whether or not by Triclinius) may or may not have been carried into P. 9 Proportionately more divergences were found in Aeschylus’ Supplices than in the alphabetic plays of Euripides; that may be because its text is much more corrupt in all respects. 10 In general the simplifying, but plausible, assumption is made that influence, if there is any, is from the closest word, whether it precedes or follows. What may be an eighteenth instance is cited by L. von Sybel, De repetitionibus verborum in fabulis Euripideis (diss. Bonn 1868) 35. He claims that P originally had tÊxhw for d¤khw in Heraclidae 933 following tÊxhn in 930 and preceding tÊxhn in 935. This is not reported in Murray, nor is it discernible in the 126 REPETITIONS AND THEIR REMOVAL Play & line Euripides L reading P reading Nearby word Heracl. 282 ¥bhn Ïbrin Ïbrin 280 (p) Supp. 374 §w afie¤ eÈsae¤ eÈsebØw 373 (p) Supp. 539 xr!n dØ altered to de› de› 536 (p) HF 74 patØr svtØr s–zv 72 (p) HF 1368 kal«n kak«n kakã 1366 (p) Ion 545 n°ou lÒgou lÒgvn 544 (p) Ion 1453 ¶labew ¶balew ép°balon 1453 (p) IT 503 fyone›w frone›w frone›w 503 (f) IT 637 lãb˙w bãl˙w bal« 635 (p) IT 733 ı tÆnde ˜tan d¢ ˜tan te 730 (p) Hel. 864 fãsgan' bãrbar' bãrbar' 864 (p) Hel. 1186 xroÚw xyonÚw xyonÚw 1179 (p) Hel. 1618 xrhsim"teron svfron°steron s"fronow 1617 (p) Aeschylus M reading E reading nearby word Supp. 334 leukostefe›w neostefe›w neodr°ptouw 334 (f) Supp. 353 bot!ri bat!ri ±libãtoiw 352 (p) Supp. 606 ín #bÆsaimi ín #gÆsaimi ghraiò 606 (f) Supp. 632 g°nei xene› xeoÊsan 632 (f) Table 1: Scribal Errors of Repetition in Apographs11 ——— facsimile; Wecklein (Euripides Heraclidae [Leipzig 1898]) merely reports d¤k in rasura in 933. Sybel supports his contention by the observation that an apograph in Paris has tÊxhw. Sybel’s thesis was that all the (to his mind) pointless and burdensome repetitions (“molestae” in his terminology) found in the received text of Euripides were due to copyists, and he was looking hard for cases where that could be demonstrated. The evidence for P’s reading is not strong enough to use here. 11 There was nothing relevant to repetition in Cyclops, Electra, or Iphigenia Aulidensis. “Nearby word” gives its line-number and whether it (p)recedes or (f)ollows the word it seems to have influenced. PETER PICKERING 127 Most of the divergences do not need this influence to explain them (being the errors in single letters to whose prevalence Zuntz has drawn attention, or errors due to metathesis of b a l - and lab-),12 though several indubitably do. There are in Table 1 six cases where the apograph has cer- tainly repeated whole or part of a previously occurring word instead of the word actually in the exemplar: Heraclidae 282, Hercules 74, Ion 545,13 IT 733, Helena 864, Helena 1618. There are seven cases where a divergent reading in the apograph is possibly due, at least in part, to influence from a previously occurring word: Eur. Supplices 374, Eur. Supplices 539,14 Hercules 1368,15 Ion 1453, IT 637, Helena 1186, Aesch. Supplices 353. There are four instances—three in the Scurialensis of Aeschylus— where the word influencing the repetition is a later one in the same line. Two of these are certain: IT 503, Aesch. Supplices 334. Two are no more than possible: Aesch. Supplices 606, 632. 12 An assumption behind all the work reported here is that scribal errors are not purely random events. However careless a scribe—and by common consent the scribe of P was a particularly careless one—there is a pattern in his care- lessness, which a systematic study would reveal. In this article it is only pat- terns relating to repetition that are studied. 13 The inflexional form is that required by the context, but the word is found in the previous (tetrameter) line. 14 W. S. Barrett, however, Euripides Hippolytos (Oxford 1964) 164–165, draws attention to the general tendency for de› to replace xrÆ in transmission. 15 Although the confusion of kal«n and kak«n is common it may have been influenced here by kakã also at line-end two lines earlier. L writes con- secutive lines in adjacent columns, while P goes straight down the column. If the cause of the error is a mechanical slip of the scribe’s eye, then a scribe copying P from L would be particularly liable to repetition from two lines back. But erroneous repetitions may be due rather to the retention in the scribe’s mind of a word encountered recently. (There is extensive psychological literature on this “priming” effect.) 128 REPETITIONS AND THEIR REMOVAL The conclusion so far is: a copyist is found introducing a repetition into a text from up to three (or possibly seven) lines earlier or from later in the same line.16 1.2 Apographs: repetitions removed Besides the cases of erroneous complete or partial repetition, there are five cases in the alphabetic plays of Euripides17 where P removes a repetition in L within three lines; in three of these the repetition is one that we would recognise as a figure of speech: Heraclidae 27: L ’s ka‹ sÁn kak«w prãssousi sumprãssv kak«w is destroyed by P’s sumpãsxv.1 8 Hercules 1100: the first hand of P destroys ¶sƒze pleuråw §j §moË t' §s–zeto by writing ¶doje.19 IT 669: ¶fyhw me mikrÒn: taÈtå d¢ fyãsaw l°geiw is destroyed by frãsaw in P. In two cases P removes a repetition which we would not recognise as a figure of speech, and which obtrudes: IT 1018: laye›n for labe›n after labe›n in 1016. Helena 739–740: P reads m°llousin for m°nousin in m°nein t' §p' ékta›w toÊw t' §moÁw karadoke›n | ég«naw o„ m°nousin.20 This occasional removal of a repetition does not seem to have been noticed in discussions of the relationship of L and P in the alphabetic plays. There are indeed three other possible exam- 16 Line distances are calculated according to the standard numeration of modern texts. Such a line is not, of course, an exact unit of measurement, since many lyric lines are shorter than trimeters; but the only practical alternative, the word, varies more in length than the line. 17 There is no similar case with the manuscripts M and E of the Supplices of Aeschylus. 18 In Andromache 462 P very similarly reads pãsxv for the prãssv of all other manuscripts in efi d' §g$ prãssv kak«w, | mhd¢n tÒd' aÎxei: ka‹ sÁ går prãjeiaw ên. 19 ¶doje is not clear on the facsimile. 20 m°nousin is usually emended to m°nous¤ m' following Musgrave. PETER PICKERING 129 ples,21 though since the original repetitions are neither obviously figures of speech nor particularly obtrusive the removal is not striking. It is worth drawing attention in this context to four instances where an alteration or variant in L (that is, not in its copying by P) removes a repetition, whether or not that was the purpose: Electra 311: L and P have éna¤nomai d¢ gumnåw oÔsa par- y°now , following gumnÚn in 308. Triclinius writes na›kaw above (gu)mnåw. This alteration (which is printed by most modern editors, omitting the preceding d¢) is ascribed by Zuntz to Triclinius’s use of another manuscript to correct an error that L had found in his exemplar.22 Kovacs however retains gumnåw , regarding guna›kaw as a conjecture by Triclinius in order to provide éna¤nomai with an object (Kovacs himself does this by emend- ing pary°now to pary°nouw); avoidance of repetition just might have been a subsidiary motive.23 Electra 435: P and, apparently, L in its original state read ·n' ı filãdelfow ¶palle delf‹w. filãdelfow is emended by Tri- clinius to f¤laulow. Triclinius, who was aware of the need for strophe and antistro- phe to respond, would have been attracted by an alteration (coming from Aristophanes Ranae 1317) which secured respon- sion, as well as making better sense. Ion 648–649: In kal«w ¶lejaw, e‡per oÓw §g$ fil« | §n to›si 21 In Eur. Supplices 217 tÚ gaËron d' §n fres‹n kekthm°noi P has xers‹ for fres‹n; but confusion between fres¤ and xers¤ in manuscripts is curiously common, and no influence from frÒnhsiw in the previous line is needed to account for it here. In Hercules 829 ZeÊw nin kak«w drçn P has nËn for nin. These words are easily confused through itacism, and nin in the previous line is unlikely to be relevant. In IT 329 P’s lab"n for bal"n in oÈde‹w tå t!w yeoË yÊmat' eÈtÊxei bal"n is another example of the frequent confusion of the strong aorists of bãllv and lambãnv, and nothing to do with peribalÒntew in line 331. 22 Zuntz (supra n.4) 107. There seems to be little evidence for this other manu- script. 23 P. D. Kovacs, Euripides III (Loeb 1998). There is, incidentally, no repeti- tion in Kovacs’ text, since he deletes 308. 130 REPETITIONS AND THEIR REMOVAL so›sin eÈtuxÆsousin f¤loiw L gives lÒgoiw as a grãfetai variant for f¤loiw. This alteration could have been motivated by failure to realise that f¤loiw is neuter, and the repetition irrelevant. IT 553: In Œ pandãkrutow # ktanoËsa x» ktan"n Triclinius seems to have emended ktan"n to yan"n. This alteration is as striking a removal of a repetition as any noted above. Triclinius’ motive was perhaps however nothing to do with repetition, but that he was reading the line, with Diggle in his OCT, in the simpler way (Orestes has just told Iphigenia that Agamemnon was killed by gunÆ , and she laments the slayer and the slain); Diggle would presumably explain ktan"n as an erroneous repetition by a copyist somewhere in the trans- mission. 2.1 More complex traditions: erroneous repetitions The degree of analytical rigour possible where one manuscript is an apograph of another is not attainable in a more complex tradition, where there is no uncontaminated stemma. But study of the manuscript tradition of other plays should help to re- inforce or refine the conclusions set out above. Three plays were therefore studied—Septem, Trachiniae, and Hippolytus—with the aim of examining every significant variant (that is one where different dictionary words are read, not where there is a differ- ent inflectional or orthographical form of the same dictionary word) and of separating out all cases where repetition might be relevant to the observed difference in readings.24 Tables 2, 3, and 4 below set out the instances in each of the traditions where a difference in readings may with a high degree 24 Variants that turn only on interjections or on the presence or absence of anadiplosis have not been covered. For the collations on which the studies were based see supra n.1. Some manuscript variants turning on repetition have been gleaned from Sybel (supra n.10); they are not mentioned in Diggle’s appara- tus, presumably as being what Diggle calls “manifestos singuli codicis errores” (I p.xiv). PETER PICKERING 131 line Page, OCT Variant Nearby word In line Variant found in 46 kataskafåw katafagåw taurosfagoËntew 43 (p) P 86 ÙrotÊpou Ùr(r)oktÊpou ılÒktup- 83 (p) all but Igggg M 121 érh˝vn érge¤vn ÉArg°Ûoi 120 (p) DDDD 126 pÊlaiw pãlaiw pãlƒ 126 (f) M ac 240 tarbosÊnƒ tarboskʃ bÒsketai 244 (f) Y 267 eÈgmãtvn Ùlugmãtvn ÙlolugmÚn 268 (f) Y 333 értitrÒpoiw értidrÒpoiw »modrÒpvn 333 (f) RRRR gggg rrrr M s.l. 333 »modrÒpvn »motrÒpvn értitrÒpoiw 333 (p) UUUUBBBBHHHHRRRRggggrrrr var in OQ 350 értitr°feiw értibr°feiw br°montai 350 (f) CaPV 359 yalamhpÒlvn yalamhpÒllvn pollã 360 (f) C 386 fÒbon lÒfon lÒfouw 384 (p) Y 410 stugoËny' timoËny' tim«nta 410 (p) Nd 497 ÖArei êyeow ¶nyeow 497 (p) DDDD ac 787 pikrogl"ssouw pikronÒmouw sidaronÒmƒ 788 (f) RRRR gggg rrrr 881 dvmãtvn dÒmvn dÒmvn 880 (p) HKNAVY RRRR gggg rrrr 906 t°low m°now m°nei 902 (p) Y, var in BDDDD 928 teknogÒnoi paidogÒnoi pa›da 929 (f) Q, P s.l. 1048 x"ran pÒlin pÒliw 1046 (p) Y Table 2: Scribal Errors of Repetition in Septem 132 REPETITIONS AND THEIR REMOVAL line Dawe Variant Nearby word In line Variant found in 143 payoËsa pãrei paroËsa 141 (p) AUY 182 kék ka‹ ka‹ 182 (p) R 240 eÈxa›w eÈkta›' eÈkta›a 239 (p) AUY 339 §f¤stasai §p¤stasai §pistÆmhn 338 (p) all but Lpc K A s.l. Y s.l. 558 fÒnvn fyÒnvn fy¤nontow 558 (p) Zo 673 maye›n labe›n labe›n 670 (p) L 744 may$n par$n pãr' 744 (p) P.Oxy. 1805 ac 947 prÒter- pÒter- pÒtera 947 (p) all but LT 965 fore› fvne› ÙjÊfvnow 963 (p) K 1008 épole›w épole›w m' épole›w m' 1008 (p) Lac 1212 genÆsetai fyonÆsetai fyÒnhsiw 1212 (p) ZgZo 1273 pãntvn yanãtouw yanãtouw 1276 (f) A Table 3: Scribal Errors of Repetition in Trachiniae of plausibility be ascribed to erroneous repetition of, or in- fluence from, a nearby word. The summary of the manuscript evidence in them does not lay claim to completeness—correc- tions and grãfetai variants may be ignored; the sigla used are those of Page, Dawe, and Diggle respectively. It must always be remembered that there are very many variant readings in texts without there being any word in the vicinity at all like the in- truder. The tables must be read with that caveat. PETER PICKERING 133 line Diggle Variant Nearby word In line Variant found in 57 bl°pvn efisor«n efisor« 51 (p) P.Sorb. 2252 107 xre"n ye«n ye«n 106 (p) Vpc test. Sybel et Wecklein 255 muelÚn filiaw fil¤aw 254 (p) P.Sorb. 2252 271 §l°gxous' §nn°pousa §nn°pein 271 (f) M 301 pÒnouw lÒgouw lÒgoiw 299 (p) O 303 §t°ggey' §pe¤yey' pe¤yetai 303 (f) scholia in several MSS. 520 tÒkƒ t°knv t°knon 517 (p) L 628 spe¤raw yr°caw yr°caw 628 (f) gnomologium E 630 futÚn kakÚn kakoË 629 (p) OVCDELP MMMM gggg rrrr 658 patr¤ kakã kakÒw 654 (p) M 688 dØ de› de› 688 (p) OV; CD omit 776 dÒmvn drÒm- bohdrome›te 776 (p) BCDEP 1038 ka‹ oÈ oÈk 1038 (p) A 1039 eÈorghs¤& eÈork- ˜rkouw 1038 (p) DE & 2nd- hand in BAVL etc. 1089 fug!w xyonÒw xyonÒw 1087 (p) D teste Sybel 1323 par°sxew ¶neimaw/¶meinaw ¶meinaw 1322 (p) DLPHE Table 4: Scribal Errors of Repetition in Hippolytus 134 REPETITIONS AND THEIR REMOVAL 2.2 More complex traditions: repetitions removed It is important to see if the tendency found in the apograph study for repetitions to be removed in transmission is peculiar to the scribe of P in the alphabetic plays or if it is a more wide- spread, if little noticed, phenomenon. The manuscript tradition of Septem has only one compelling example: 1005: dokoËnta ka‹ dÒjant' épagg°llein me xrØ where Q has m°llont' for dÒjant'. The scribe may have thought the context demanded a future (just as Headlam conjectured dÒjont'), or been affected by -ell- in the next word. It is never- theless remarkable that these feelings should have counteracted the influence of dokoËnta two words earlier. The tradition of Trachiniae is much more interesting. Despite sharing with Zg the introduced repetition in 1212, and intro- ducing one of its own in 558, the manuscript Zo appears to remove repetitions on as many as five occasions, on one with the approval of modern editors: 187: ka‹ toË tÒd' ést«n µ j°nvn may$n l°geiw In the light of the tendency the following cases will demonstrate in Zo, it is tempting to see ¶xeiw for l°geiw as a way of varying from lÒgon in 184. 330–331: mhd¢ prÚw kako›w to›w oÂsin êllhn prÒw g' §moË lÊphn lãb˙ The line is read thus by Dawe and by Easterling,25 and (with oÔsin and lãboi) by Lloyd-Jones and Wilson in their OCT of 1990. But êllhn is in Zo alone. LKR have lÊphn twice; AUY lÊphn … lÊphw ; Zg and T have loipØn … lÊphn , which is a grãfetai variant in Y and (as loipe›n) in U. Easterling describes the line with the repetition of lÊphn as “obviously corrupt,” with Zo’s êllhn removing the main difficulty. Davies says “the 25 P. E. Easterling, Sophocles Trachiniae (Cambridge 1982). PETER PICKERING 135 lÊphn … lÊphn offered (or presupposed) by most manuscripts is that common phenomenon, an error of anticipation.”26 lÊphn … lÊphn is no doubt wrong, but these editors give the im- pression that êllhn was the reading of which lÊphn was a corruption; more likely the exemplars from which Byzantine scholars and scribes were working had lÊphn twice; loipØn and êllhn for the first lÊphn and lÊphw for the second were differing attempts to make sense of it. Zo’s mild aversion to repetition has led it to a good conjecture (supported by the fact that all MSS have oÔsin (or oÂsin) with paragogic n, which is unmetrical before l). But it almost certainly is a conjecture, like Schenkl’s kainØn , and F. W. Schmidt’s dipl!n for the second lÊphn. 472–473: §pe‹ se manyãnv ynhtØn fronoËsan ynhtå koÈk égn"mona; This is an effective polyptoton, which Sophocles must have writ- ten and which no modern editor would think of destroying. But for ynhtå Zo reads Ùryå. Even if this started life as a gloss, whether on ynhtå or on oÈk égn"mona , its displacement of ynhtå is odd. 965: pò d' aÔ fore› nin … Zo’s fron°ei can be seen as an attempt to get away from the fore› of LAUYT and the f°rei of the common parent of Zg and Zo; there is f°rei in 967 and f°retai in 968. 1182–1183: UL …w prÚw t¤ p¤stin tÆnd' êgan §pistr°feiw; HR oÈ yçsson o‡seiw, mhd' épistÆseiw §mo¤; Picking up a word in a stichomythic line is very common, and in- deed both natural and rhetorical in dialogue. But for épistÆseiw Zo reads épeiyÆseiw. 26 M. L. Davies, Sophocles Trachiniae (Oxford 1991). 136 REPETITIONS AND THEIR REMOVAL There is one separate example: 963: In all manuscripts but the Triclinian T j°nvn is preceded by j°noi ; Triclinius’ correct emendation was no doubt made to secure responsion with the strophe rather than to remove the repetition. The phenomenon is also found in the tradition of Hippolytus: 797: Haun. (Hauniensis 417) has d¤kh for tÊxh at the end of this line, with tÊxhw at the end of 801.27 984: The natural kaloÁw is replaced by polloÁw in six manu- scripts (VCDELP), thus avoiding repetition with kalÚn in 985 —not a repetition Euripides himself would have shunned. 992: In the apparently figural and Euripidean pr«ta d' êrjo- mai l°gein | ˜yen m' Íp!lyew pr«ton …w diafy°rvn , VCDELP replace pr«ton with prÒteron. 1178: P has f°rvn for ¶xvn following ¶xvn in 1177. 1311: grafåw ¶taje for grafåw ¶grace in A is a striking avoid- ance of figura etymologica. 3.1 Conclusions: erroneous repetitions The analysis above helps answer the question how far the influence of one word on another can be shown to extend. Common sense suggests that influence would diminish over distance, and that that distance would be less where the influence is from a later word than where it is from an earlier word (since a word once read might persist in a scribe’s mind indefinitely, while his eye is unlikely to run far ahead of his pen).28 Data would be useful, for instance in assessing claims that a word read in all manuscripts is a corruption due to the 27 This and 1178 are reported by Sybel (supra n.10), whose concern was repetition, and by N. Wecklein, Euripides Hippolytus (Leipzig 1900 [Teubner]) but not by Diggle. 28 In a particular case there may be another explanation of a repetition from a word which follows in modern texts—there might have been a dislocation of lines at some time in the transmission so that what is now a following was once a preceding word. PETER PICKERING 137 Distance in lines Influence precedes Influence follows 0 13 8 1 11 4 2 8 0 3 5 1 4 2 1 5 0 0 6 1 0 Table 5: Distances between Erroneous Repetition and Apparent Source of Influence influence of a nearby word and should be emended. Table 5 brings figures from Septem, Trachiniae, and Hippolytus together with those from the apograph study.29 It appears from this table that most errors in transmission that could be attributed to the influence of another word ( i.e. that could be described as “unconscious repetitions” by the copyist) are within a few lines of the apparent source of the influence.30 Overall, influence from preceding words is much commoner than influence from subsequent ones, the only ap- parent exception being partial repetitions in Septem (where a word is not replaced, but corrupted by a later one, often to a 29 Excluding those described as “possible” rather than “certain” in the apo- graph study. Since many manuscripts are being looked at in the studies of the three plays, while there was only one apograph in each of the apograph studies, there are likely to be more examples found in each of the three plays than in any play of the apograph studies. 30 There is danger of circularity here; the nearer a similar word is to the error, the more its influence suggests itself as the source of the error. 138 REPETITIONS AND THEIR REMOVAL non-existent or totally inappropriate word).31 Where the puta- tive influence is a subsequent word, it is usually very close to the error. Only Septem 240 (tarboskÊv for tarbosÊnv with bÒske- tai four lines later) and Trachiniae 1273 (yanãtouw for pãntvn with yanãtouw in 1276) are plausible examples of influence over an interval of more than one line. 3.2 Conclusions: removal of repetitions The study of differences between manuscripts other than direct copies found enough convincing cases of removal of a repetition during transmission to support the surprising findings of the study of apographs. These cases are not evenly distrib- uted among manuscripts. In the Hippolytus tradition four out of five are found in, and only in, manuscripts of the VCDELP group.32 In all but one Trachiniae case the manuscript without the repetition is Zo; it is striking but probably co-incidental that Zo is, physically, the same as P for Euripides,33 where as shown above there were at least five instances of the removal of a repetition read in L. What emerges from all this is a discernible tendency in some manuscripts of tragedy to remove repetitions that must have been written by the poet. I cannot, at present, explain this. It is not a natural error like repeating a word one has just come across, and it seems at first sight unlikely that Byzantine scholars, let alone scribes like the one who copied P, had the aversion to repetition of nineteenth-century scholars like Sybel, 31 Such a corruption is almost always in a single manuscript, whether in its text or as a variant; the corruption probably does not spread because a copyist is likely to notice, and do something about, a nonsensical word in his exemplar. The apparent concentration of such corruptions in Septem may be an illusion, produced by the comprehensive nature of Dawe’s collation. 32 V is, according to Barrett (supra n.14) 74, closely associated with CDE in that part of the play where the instances are found. 33 That is, the early-fourteenth-century manuscript that was divided into two portions, Palatinus gr. 287 and Laurentianus conventi soppressi 172. PETER PICKERING 139 and made occasional emendations accordingly.34 It would be worthwhile looking at the manuscript traditions of other Greek authors to see if any similar tendency is observable there also. September, 2001 3 Westbury Road Woodside Park London N12 7NY U.K. pe.pickering@virgin.net 34 The Hellenistic scholars Zenodotus and Aristophanes, according to the scholia vetera (III 600 Erbse), read ka‹ megãlouw for émbros¤ouw in Iliad 14.177 because that adjective occurs three other times in nine lines.