THE LATIN ORIGINS OF A BILINGUAL LETTER COLLECTION (SPECIMINA EPISTVLARIA = P.BON. 5)

Abstract P.Bon. 5 preserves the only known collection of ancient Latin model letters, accompanied by a Greek translation. This article argues that the Latin is the primary version and dates the composition to before the early third century. Comparisons with other model letter collections, principally ps.-Demetrius’ Epistolary Types and ps.-Libanius’ Epistolary Styles, locate the text within a wider literary genre. A new reconstructed text is provided in the Appendix at the end of this article.

the work has been further diminished by its characterization as a 'glossary', produced by someone not fully competent in either Latin or Greek.
In fact, most of the infelicities of the transmitted text can be attributed to secondary stages of transmission. Building on previous scholarship, we argue that it originated as a monolingual Latin work produced before the early third century--and perhaps even earlier--which was only later equipped with a Greek translation. One can therefore distinguish between an initial phase of composition and later bilingual adaptation. This would make the original anonymous letters roughly contemporary with other betterknown letter collections from the Early Empire, such as those by Pliny the Younger and Fronto. The paper then explores the significance of this document for ancient epistolography, in particular its connection to two Greek model letter collections, ps.-Demetrius' Epistolary Types (Τύποι Ἐπιστολικοί) and ps.-Libanius' Epistolary Styles (Ἐπιστολιμαῖοι Χαρακτῆρες). The comparison supplements the near total silence of Roman grammarians and rhetoricians about epistolographic theory. 3 In order to make this text more accessible, the Appendix contains a new reference edition of the Latin text. It aims to reconstruct the original state of the Latin collection as far as possible before it became a bilingual document. It therefore differs from, and provides a necessary supplement to, the authors' forthcoming edition of the papyrus text in the Corpus of Latin Texts on Papyrus (CLTP), which presents the complete text in its surviving state. 4 Our citations of the text give the letter and sentence numbers from the Appendix, followed by a parenthesis containing the column and line numbers from Kramer and Cugusi. 5

RECONSTRUCTING THE LATIN ORIGINAL
The high-quality format of the bookroll, which survives in three contiguous fragments, suggests a professional copy. A single scribe produced both the Latin text and the Greek text-in semi-literary minuscule and Greek capitals respectively-on the recto of the document. The verso was later reused for writing two Greek accounts (P.Bon. 38). 6 Based on the script and the documents on the verso, the original document can be dated to the end of the third or to the beginning of the fourth century. 7 Seven double columns survive, with the Latin on the left and the Greek on the right, and the bilingual headings centred between the columns. The use of such double columns for translation is typical of ancient bilingual pedagogy. 8

Errors and infelicities in the Greek
Despite the suggestive placement of the Latin in the left column, editors have disagreed about the relationship between the versions or the linguistic competence of the composer (s). According to Castiglioni, linguistic imperfections in both versions suggest simultaneous composition. 9 While Kramer accepted the primacy of the Latin text, he concluded that the composer was most likely a speaker of Greek or Coptic, who produced the work relying mechanically on bilingual or trilingual glossaries, a view that has persisted in more recent scholarship. 10 In fact, most of the apparent Latin errors can convincingly be explained as errors of transmission rather than errors of linguistic competence. This was rightly perceived by R. Merkelbach and recently defended by C.M. Lucarini. 11 When these mechanical errors are removed, the Latin text achieves an impressive degree of fluency and syntactic complexity, which generally enables it to stand on its own. By contrast, the Greek text closely follows the idiom of the Latin version, even where this produces Greek that is sometimes unidiomatic and even unintelligible without the Latin.
The strongest evidence for the Greek's secondary status are several errors best explained as misunderstandings of the Latin.  ). The context demands the meaning 'reverence' for ueneratio, but ἐπαφρεδεισίᾳ (read According to Montevecchi (n. 1), the prices recorded on the verso (P.Bon. 38) cannot be later than the first two or three decades of the fourth century. 8 Cf. E. Dickey, 'Columnar translation: an ancient interpretative tool that the Romans gave the Greeks', CQ 65 (2015), 807-21. 9 See Vogliano (n. 1), 210: 'A volte all'autore è riuscito di trovare una più acconcia voce greca, altra volta una latina …, ma ha sostanzialmente, per l'una e l'altra parte, lavorato su lessici; se non erriamo su di un lessico trilingue, cioè coptogrecolatino.' 10 Kramer (n. 1), 109 regarded the following linguistic difficulties in the Latin text as indicative of non-native composition: 'illogical construction' (Spec. 4.1 [= 2.14-18] quod aliter quam meruisse te scimus remuneratus non es a Publio amico tuo; and 4.  11 See Merkelbach (n. 1); before this was published in 1956, some of his emendations were incorporated in Montevecchi (n. 1) in 1953. His opinion has been persuasively defended by C.M. Lucarini, 'Zu den lateinisch-griechischen Musterbriefen (P. Bonon. 5)', ZPE 216 (2020), 73-7, which provides outstanding discussion of several textual problems. The present authors finalized their text before having access to this publication and in some passages reached similar conclusions independently. ἐπαφροδισίᾳ) is only attested with the meaning 'loveliness, elegance' (LSJ). Presumably the translator recognized the root Venus in ueneratio and produced a quasi-calque, ἐπαφρεδεισία, whose meaning is intelligible only in relation to the Latin original.
• 12.4 (= 7.20-2) hi‹c› enim demum speciosus est titulus ( αὕτη [γὰρ] οὕτως εὔμορφός ἐστιν στήλη). titulus here probably has the figurative meaning 'claim to glory or fame, distinction' (OLD s.v. 7; for the collocation, see ps.-Quint. Decl. 9.22 specioso titulo bene uita pensatur 'life is well compensated by a brilliant reputation'), here referring to the act of manumission described in the letter. By contrast, στήλη has the sense 'monument, gravestone' (LSJ). The translator has been misled by the more familiar meaning of titulus as 'tablet' or 'inscription', perhaps not recognizing the figurative meaning or relying uncritically on an ancient bilingual glossary (for example CGL 2.437.50).
Such errors suggest that the translator did not understand the Latin original or was not fully competent expressing himself in Greek. 12 The translator also produced Greek expressions that are idiomatic for Latin but unusual or unattested in Greek. One recurring Latin word is the substantivized suprema, -orum 'last wishes', first certainly attested in this sense in ps.-Quintilian's Declamationes (1.2 neque … patrem suprema sua iuueni iactasse crediderim) and several times in the Digest. 13  Another notable Latinism is the connecting relative, a construction not usually found in Greek. 16 The clearest example of this is 3.1-2 (= 2.3-8) Licinn[i]um amicum tibi carum obisse compertus sum. quem parum memorem obsequi tui fuisse doleo. The construction has been carefully imitated in Greek: ὃν … ἐμνημονευ[κέ]ναι … λυποῦμαι. Perhaps related to this is the author's abundant use of relative pronouns, sometimes to append lengthy subordinate clauses that overshadow the main clause (for example 8.

Mechanical corruptions in Latin
By contrast, most cases where the Greek appears to be more intelligible than the Latin are plausibly explained as mechanical errors of transmission, consistent with prolonged transmission in a Greek-speaking milieu. 18 On this hypothesis, such corruptions would have affected the Latin column only after it was translated into Greek and began to circulate in bilingual form.
The most common type of error is the omission of entire words or word parts, whose presence can typically be inferred from the Greek.
A substantive is needed in Latin to justify the feminine form of the adjectives prosperas and aduersas. A straightforward correction is prosperas <res>, with the noun res omitted by near haplography. • At letter 10 (= 5.25-6), the scribe wrote quod ta[ntu]m quod boluit, set quo[d ] potuit (οὐ μόνον τὸ ἠθέλησεν ἀλλὰ ὃ ἠδυνήθη). The duplication of quod is a scribal error of anticipation, which can be easily corrected from the Greek to <non>.
In two additional cases entire Latin words appear to have dropped out that must be restored based on the Greek: • The first sentence of letter 5 (= 2.26) is missing an infinitive verb in indirect statement, which must correspond to the accusative subject. An infinitive is clearly recorded in Greek (ὠλιγωρηκέναι), which can be used to restore the Latin. 19 Notably, the Latin line here is considerably longer than the corresponding Greek A more complicated but revealing compound error has probably occurred at 4.1 (= 2.22), where the middle syllable of the verb commutet was omitted: nihil ex moribus tuis cum<mut>et parum ingrata sententia (μηδὲν οὖν ἐκ τῶν σῶν ἀλλαξέτω οὕτως ἀχάριστος ἀπόφασις). The papyrus has cum ‧ et, with an interpunct in place of the middle syllable. Conceivably the archetype had com ‧ mutet, with a word-internal interpunct separating the prefix from the verb stem. The scribe's eye skipped over the middle syllable -mut-and he wrote com ‧ et. This was 'corrected' at a later stage to cum ‧ et when it was realized that com alone is nonsensical. This explanation posits at least two stages of error: first a mechanical omission, then a correction by someone whose Latin was good enough to identify an error but without access to the archetype. Alternatively, the archetype may have had the form cummutet, a spelling of the prefix sporadically attested in antiquity (TLL 4.1341.3-8).
In comparison, the Greek text looks relatively free of more than superficial corruption. This is consistent with it being a younger text that was copied and read with greater attention. Characteristic mistakes include the omission of word endings or the accidental insertion of an incongruous particle: 10.1 (= 5.27) καταλελοιπέν<αι>; 11.1 (= 6.8) μέ<ν>. 20 Other peculiarities, such as the hapax legomenon ὑπεικία (3.2 [= 2.7] rendering obsequium) and the occasional omission of a definite article, are more likely to reflect features of contemporary Greek or the methodology of translation rather than textual corruption. 21

Latin idioms misidentified as errors
Furthermore, several passages have been misidentified as errors or interference, when in fact there are good, though rare, Latin literary parallels. These examples provide important clues for dating and locating the composition within a literary milieu.
The first is a case of pleonastic negation. At 4.1 (= 2.23) the phrase parum ingrata sententia (οὕτως ἀχάριστος ἀπόφασις) must mean from the context 'a very ungrateful judgement', as the Greek translation attests, though logically it means the opposite ('a not at all ungrateful judgement' = 'a very grateful judgment'). Merkelbach proposed deleting the prefix in-as a later insertion ( parum {in}grata), and Kramer regards it as Coptic interference. However, such pleonastic negation, specifically before in-compounds, is attested after haud in Sallust (Hist. fr. 4.41 M. haud impigre neque inultus occiditur) and possibly in Livy (32.16.11 oppidani haud impigre tuebantur moenia). 22 20 Lucarini (n. 11), 75 suggests deleting μέ entirely. He also persuasively argues for the deletion of μέν before γάρ at 3.3 (= 2.11) enim. The Greek text is also almost certainly corrupt at 2.2 (= 1.15-17); see n. 31. 21 On the absence of the article in Greek translations from Latin, see Adams (n. 18), 516 and L. Koenen, 'Die Laudatio funebris des Augustus auf Agrippa auf einem neuen Papyrus', ZPE 5 (1970), 217-83, at 232 n. 21. The nature and style of the Greek translation requires more extensive discussion than is possible here. 22 Pleonastic negation is discussed by Hofmann and Szantyr (n. 16), 806 and by E. Löfstedt, The pleonastic construction with parum, used as a polite alternative to haud, can be tolerated as a small logical oversight on the part of the author. 23 In another case involving negation, Merkelbach bracketed the non in a complicated subordinate clause at the beginning of letter 4 (4.1 [= 2.14-18]): quod aliter quam meruisse te scimus remuneratus non es a Publio amico tuo (ὅπερ ἄλλως ἢ ἠξιῶσθαί σε οἴδαμεν οὐκ ἀντεδωρήθης ἀπὸ Πουβλίου τοῦ φίλο[υ] σου). We punctuate the phrase aliter quam meruisse te scimus as a parenthetical remark and translate: 'although-differently than we know you deserved-you were not rewarded by Publius'. Although the deletion of non might produce a less abrupt parenthesis ('although you were rewarded by Publius otherwise than we know that you deserved'), the negation is clearly attested in both the Latin and the Greek. If the non is a later insertion, either it entered the Latin text before the Greek translation was made or, less plausibly, the negation was added later to both the Latin version and the Greek version.
Even more remarkable is a series of unusual genitives: • 2. Greek unambiguously supports the genitive in the last case (τῆς λαμπρότητος). 26 If the Latin genitive is a corruption, it must have occurred before the Greek translation was made, unless the Greek translation was later 'corrected' (προσέρχομαι does not appear to govern the genitive idiomatically).
In fact, all three can be retained as genitives of cause, a rare construction but one sporadically attested in Latin literary texts. 27 For the last passage there is a striking parallel, with the same verb accedere, in Rhet. Her. 4.38.50: utrum auaritiae an egestatis accessit ad maleficium? This is arguably the reading of the archetype and it has been persuasively defended by E. Löfstedt, though some manuscripts insert causa. 28 Likewise, there is a parallel for gaudeo with the genitive in Apuleius (Met. 1.24.7): quid istud? uoti gaudeo, a construction also found in later authors. 29 Another likely genitive of cause in a different construction is also found in the correspondence of Fronto. 30 The transmitted Latin genitives should therefore be retained and not be regarded as incompetent. 31 Furthermore, the occurrence of the same unusual genitive in three different letters suggests unity of style, perhaps even unity of authorship, across the collection.

Stages of Transmission
The transmitted Latin text is thus both worse and better than has been previously appreciated: worse, in the sense that it is more deeply marred by mechanical errors; but also better, in the sense that the linguistic competence of the Latin has been consistently underrated. Indeed, once textual errors are removed, the remaining peculiarities can be explained without invoking bilingual interference from Greek or Coptic. 32 26 The Greek is not very helpful in the first two passages: in the first, it appears to be corrupt; in the second, one may read either μνήμη or μνήμῃ (iota ad-or subscriptum is never written in the document). 27 The genetiuus causae is regarded as an internal development of the genetiuus relationis ('Genitiv des Sachbetreffs') by Hofmann and Szantyr (n. 16), 75 (cf. 83); cf. H. Pinkster, The Oxford Latin Syntax (Oxford, 2015), 906. More discussion in E. Löfstedt, 'Genetiuus causae im Latein', Eranos 9 (1909), 82-91 and Löfstedt (n. 22), 1.169. Note also its occurrence on inscriptions: e.g. CIL 10.9 conseruo pietatis fecit and 13.1782 salutis dedicauit. 28 Cf. Löfstedt (n. 27), 86-7, accepted by G. Calboli, Cornifici Rhetorica ad C. Herennium (Bologna, 1993 2 ). Marx prints a lacuna before auaritiae (Leipzig, 1894). By contrast, the Budé edition of G. Achard (Paris, 1989) prints causa auaritiae, following MSS FE. However, other manuscripts insert causa at different places in the sentence (P 2 B 2 V, differently C), which rather suggests a secondary intervention. 29  , the Greek has μνήμη, which we punctuate as a dative with instrumental sense. 32 The case for Coptic interference in the Latin has never been based on extensive evidence; it seems rather to be an inference arrived at by excluding Greek as the author's native language on Where the Greek translation is more intelligible than the Latin text, we can usually suppose that this is because the translator worked from a more accurate Latin text than what survives. Accordingly, one or more stages of copying must have intervened between the production of the bilingual version of the text and the surviving copy. A compound error like cum<mut>et, which most likely involved two independent stages of corruption, strengthens this conclusion. The following three stages of textual history can therefore be tentatively outlined: Stage 1: the composition of the Specimina epistularia as a monolingual Latin text; Stage 2: the state of the Specimina when it was translated into Greek (at this point or soon after the text was arranged in parallel bilingual columns); Stage 3: the state of the combined bilingual text as preserved on P.Bon. 5.
This scheme helps to distinguish 'errors' of composition, which occurred at Stage 1 and therefore reflect the composer's command of Latin, from later errors that crept into the text during its transmission. In only a few cases, where both the Latin text and the Greek text present difficulties, do we suspect corruption occurring between Stages 1 and 2. 33 Most other corruptions seem to have occurred between Stages 2 and 3. The Latin text in the Appendix aims to reconstruct the Latin composition as close to Stage 1 as possible; in practice, however, it is rarely possible to conjecture far beyond what the Greek translator received at Stage 2.

DATING THE LATIN COMPOSITION
As this account emphasizes, the document underwent a series of transformations before it reached the form in which it survives. The composition must therefore be older than the papyrus itself, which dates to the end of the third or to the beginning of the fourth century. But how much older? This section offers several lines of evidence for locating the original composition more precisely in time and social context. That such a collection might be a patchwork composed by different hands at different times is entirely possible. In that case, one might expect a wide range of variation. However, as we will show, similar lexical and stylistic peculiarities tend to recur across different letters, thus favouring a unitary composition. We return to this question in the conclusions (section 3.4 below). [ p]rocesserit, which seems to require an infinitive processisse (see n. 24). We also suspect early corruption at 9.2 (= 5.12) per processorum tuorum, where the use of per with a genitive is unparalleled (perhaps an accusative object was omitted), and in an extremely difficult passage at 12.3 (= 17.13-19). Our text in the Appendix does not attempt to resolve all these difficulties.

Content
The content of the collection implies writers and addressees who belong to a relatively prosperous upper-class Roman milieu. Characteristically Roman preoccupations include: the exchange of beneficia (Spec. 1-2), the naming of friends as heirs and legatees (Spec. 3-11), and manumission (Spec. 12-13). 34 Many of the addressees must be assumed to be Roman citizens since only full citizens could legally be named as heirs. Furthermore, there is a striking reference to clientes (9.2 [= 5.13]), who expect to benefit when their patrons receive inheritances. The patron-client system was a peculiarly Roman institution, for which there was no fixed Greek vocabulary. 35 There is also no unambiguously Christian material in the content of the letters. The reference to fata, possibly personified as Fata, might point towards a pre-Christian background: 3.3 (= 2.11-13) tabulas … suppremorum [h]omines quidem faciunt, [s]et ordinant fata. However, the presence of pre-Christian material is not a reliable criterion on its own. The model letters of ps.-Libanius, for example, which can be assigned with some confidence to the fifth century, continue to show pseudo-pagan elements. 36 By contrast, there is a potentially Christian turn of phrase in the Greek translation of a vexed passage, Spec. 12.3 (= 7.14-16) uti perpetụị[ta]s ̣ uit ̣ae omni uitae tuae spat[iose] detu[r] ('that longevity of life be granted to your entire life extensively'). We understand perpetuitas as referring to 'longevity' or 'continuity' of life, since a wish for immortality would be out of place in the context. 37 However, the Greek translation seems to interpret this as a wish for salvation, using an expression common in patristic authors (ἵνα ἡ διηνεκὴς αἰὼν πάσῃ [τῇ] ζωῇ σου ἐπὶ πλ[εῖ]στον δοθῇ). 38

Interpuncts
One of the text's most striking graphic features is its preservation of several word-internal interpuncts, which separate a prefix from its lexical stem: 1 (= 1.3) in·ertes; 3.1 (= 2.5) com·pertus; 8.1 (= 4.14) [c]on·seruas; and 10.2 (= 6.5) absum·pserat. In one case, it has been suggested that a word-internal interpunct contributed to textual corruption (see section 1.2 above on cum·et, which we restore to com⟨mut⟩et). Interpuncts are also found more regularly at word boundaries throughout the text. Word-internal interpuncts are a palaeographical peculiarity only known from ancient Latin books, which began to be employed less regularly and eventually disappeared over the course of the second and third centuries. 39 Therefore, the scribe of P.Bon. 5 most likely copied these interpuncts from an older archetype. 34 , 1910); cf. e.g. 33.5 νὴ τοὺς θεούς. On this text and its dating, see section 3.1 below. 37 Arguably the first uitae, which looks like a genitive modifying perpetuitas, should be deleted as an error of anticipation. The sense of a long duration of time is well attested for perpetuus (TLL 10.1.1646.10-19) but less clearly for perpetuitas. Alternatively, if uitae is deleted, perpetuitas might refer to the perpetuity of the grant of freedom (TLL 10.1.1636.27-35). 38 The phrase is especially common in John Chrysostom, e.g. Homil. in gen. 36.5 (Migne, PG 53, page 340.29) τὴν ἐν διηνεκεῖ αἰῶνι ἀπόλαυσιν; 40.1 (page 369.5) διηνεκεῖ τῷ αἰῶνι; etc. 39 Cf. Ammirati (n. 6), 32-3, 113.

Lexicon
The lexical evidence also clusters around a date in the first or second century. The substantive suprema, -orum 'last wishes', discussed above (1. Within literary texts, it is found in the ps.-Quintilianic Declamationes (6.14 ut natus … in rebus aduersis praesidium parenti labore atque praestantia solueret lucis usuram).
Lastly, the concessive use of etiam (for etiamsi, discussed at 1.3) is first attested in Columella, Quintilian and Frontinus. .7]). The mere occurrence of such names is inconclusive, since many continued to be used long after the tria nomina system ceased to function.

Onomastics
More instructive is the way in which these names are used: specifically, none ever occurs in the vocative as part of direct address. Instead, direct address occurs exclusively with the vocative frater. 43 This reflects a well-documented chronological trend, in which proper names began to disappear from direct address over the course of the first century C.E., replaced by polite terms such as domine, carissime and frater. 44 In keeping with this development, the Specimina aligns much more closely with the letters in the Vindolanda Tablets and in Fronto than with those in Cicero's correspondence. Of course, it is possible that direct address by proper name was avoided in these letters precisely because they were meant to be used as models.

Prose rhythm
One overlooked feature is the presence of quantitative clausulae at the end of most sentences. To facilitate comparison, we use the scansion methodology outlined by T. Keeline and T. Kirby, who provide valuable comparative data. 45 Of the twenty-six sentences ending in full stops in our proposed punctuation of the text, twenty-three (88.46%) end in identifiable 'artistic' clausulae, of the sort common in literary prose from Cicero to Apuleius. 46 In fact, every letter concludes with such a clausula, where there is no ambiguity about a sentence boundary and where greater rhetorical emphasis is expected. Among the 'artistic' clausulae, the following patterns are found in decreasing order of frequency: seven cretic trochees (26.92% of all sentences), five double trochees (19.23%), five molossus cretics (19.23%), three hypodochmiacs (11.54%) and three double cretics (11.54%). Of the remainder, three are double spondees (11.11%). 47 The twenty-six full sentences from the Specimina, admittedly a small corpus, do permit some tentative comparisons. Strikingly, the Specimina has an overall higher proportion of 'artistic' clausulae (88.46%) than in other, more famous, letter collections. In particular, it aligns with collections that were composed with a view to publication or for addressees with whom the writer was on formal terms: for example Cicero's Ad familiares (75.41%), Seneca's Epistulae (80.92%) and Pliny's Epistulae Books 1-9 (84.87%). By contrast, letters written to intimate friends, as in Cicero's Ad Atticum (69.68%), or to subordinates, such as Trajan's letters to Pliny in Epistulae Book 10 (67.55%), have lower concentrations of clausulae. It therefore seems that the Specimina was written in a more formal and less intimate epistolary style.
The presence of quantitative rhythm is itself significant evidence for the text's date of composition. A new system of quantitative-accentual prose rhythm, sometimes unhappily called cursus mixtus, began to appear in several late treatises of Apuleius (De Platone, De mundo), whose authenticity has sometimes been called into question, and in the works of Minucius Felix and Cyprian, namely around the end of the second 44 Cf. Dickey (n. 42), 44-5. 45 (Cambridge, 1996) 4.2, 7.1, 11.2, 12.3, 12.4; hypodochmiac: 3.2, 11.3, 12.2; double cretic: 2.2, 4.1, 10.2. century. 48 Though this system is not purely accentual like the later cursus, word boundaries and accents place greater constraints on the favoured clausulae. As a result, certain patterns that are freely distributed in earlier authors, such as Cicero, become proportionally much rarer. For example, in a double trochee, an accent on the first trochee is avoided (-́⏑ -×), part of a general tendency to avoid accents separated by only one syllable; and when the final word is itself a double trochee it is usually preceded by a proparoxytone (´× × . -⏑ -×). 49 Several such disfavoured patterns of accent and word boundary are found in the Specimina. For example, two of the double trochees contain an accent on the first trochee (6 respondére laetor; 7.2 [so]lácium fit), and the previous word is not required to be a proparoxytone (8.2 suos [d ]ignum iu[d ]icaui<t>). Two of the three hypodochmiacs do not have the usual word break after the cretic (11.3 desiderandus [e]st; 12.1 con[s]ecu[t]us es). Such patterns are not absolutely forbidden in cursus mixtus, but they become rarer in relation to the favoured clausulae. Accordingly, in so far as there are quantitative clausulae in the Specimina, there is no sign that the constraints of cursus mixtus are yet operative. Though the precise date and circumstances of this transition need more investigation, it seems unlikely that purely quantitative clausulae would be written after the end of the second century.

Dating conclusions
Several lines of evidence converge on a date between roughly the middle of the first century C.E. and the beginning of the third. On the one hand, a terminus post quem is provided by several rare words and constructions that are not attested before early imperial sources: the perfect active stem remuneraui (Festus); passive forms of remuneror (ps.-Quintilian and Apuleius); the concessive etiam (Columella); praestantia 'generosity' (Antonine inscriptions, ps.-Quintilian); and the substantive suprema (ps.-Quintilian). The absence of names in direct address fits well with a period from the second half of the first century onwards. On the other hand, there are termini ante quem provided by the presence of word-internal interpuncts, which would have been a palaeographic feature of the exemplar, and the system of quantitative prose rhythm, which is unlikely to have been produced by any writer after the end of the second century. Of course, it is hazardous to rest too much weight on any particular piece of linguistic evidence, and it must be kept in mind that the earliest surviving occurrence of a linguistic feature might be younger by centuries than its earliest occurrence. Nevertheless, it is suggestive that several independent lines of linguistic evidence are consistent with an Antonine date range.

Greek model letter collections
Comparing the Specimina with the two Greek model letter collections that survive complete provides additional information about its structure, audience and rhetorical context. 50 Specifically, it helps to show that the Latin composition was considerably more ambitious than student exercises. This accords with the textual and linguistic discussion, which has shown that the Latin is of higher quality than typically regarded.
As mentioned, only two Greek model letter collections survive in manuscript tradition, both with pseudonymous attributions: ps.-Demetrius' Epistolary Types (Τύποι Ἐπιστολικοί) and ps.-Libanius' Epistolary Styles (Ἐπιστολιμαῖοι Χαρακτῆρες). 51 Intriguingly, the Egyptian Book of Kemit (extant texts from the late second millenium B.C.E.) contains several epistolary formulas, hinting at more ancient, local roots of the genre. 52 A fragment of a Greek epistolary manual on a sixth-century papyrus codex has also been identified (P.Berol. inv. 21190), which appears to be typologically distinct. 53 Ps.-Demetrius is clearly the older of the two, although estimates of its date vary considerably. It is usually thought to have Ptolemaic origins, but later composition or revision is possible as late as the fourth century. 54 An Egyptian origin is undisputed. 55 After a theoretical preface, it discusses twenty-one epistolary genres, each prefaced by a short description and followed by a single illustrative letter.
Ps.-Libanius' Epistolary Types originates in a later milieu, perhaps the middle of the fifth century, under the influence of Libanius. 56 It circulates in two recensions, the first under the name Libanius, regarded as the older and more authentic of the two, and a second under the name Proclus, with a slightly different structure and many Byzantine interpolations, including additional genres and model letters. In ps.-Libanius, the number of genres (χαρακτῆρες) has doubled to forty-one, with seven additional types found only in the Byzantine recension. 57 Nevertheless, the model letters themselves tend to be shorter and simpler than in Ps.-Demetrius, with fewer sentences and less hypotaxis. Usually there is only one model letter for each type, although the Byzantine recension adds multiple, more elaborate and Christianized model letters.
In addition, ps.-Libanius follows a slightly different arrangement than ps. The preface of ps.-Demetrius, addressed to an unknown Heraclides, provides some information on the audience and function of such a treatise. It refers to the haphazard way in which letters are composed by 'those who undertake service for men engaged in politics' (Weichert 1.5-6 [sc. ἐπιστολῶν] γραφομένων … ὡς ἔτυχεν ὑπὸ τῶν τὰς τοιαύτας τοῖς ἐπὶ πραγμάτων ταττομένοις ὑπουργίας ἀναδεχομένων), and it characterizes Heraclides as an educated young man with intellectual ambition (1.13-14). Whereas this treatise supposes a professional audience, there are also suggestions that the composition of letters occurred as a rhetorical exercise, specifically in connection with προσωποποιία. 59

Comparing letter collections
Compared to the genres in the Greek letter collections, those in the Specimina show a tendency to greater generic elaboration. 60  Furthermore, the Specimina illustrates each subgenre with multiple letters, ranging from three (de exiguis legatis suasoriae) to six (gratulatoriae hereditatium acceptarum). ἤμην εἰς Ἀλεξάνδρειαν) and 'ships in public service' (9.3-4 οὔτε … πλοῖόν ἐστιν εὐπορῆσαι πάντων εἱλκυσμένων πρὸς τὰς λειτουργίας). 56  Both of these Greek letter collections contain two elements that are not attested in the surviving portion of the Specimina: first, a theoretical preface of some kind, which in ps.-Demetrius also includes a named dedicatee; second, descriptions of each epistolary genre that complement the model letters themselves. In ps.-Demetrius, these descriptions directly precede each model letter; in ps.-Libanius, all the genres are described in one section, followed by all the model letters separately. 61 Conceivably, these two elements are standard features of ancient model letter collections and were also part of the complete Specimina. If so, the genre descriptions must have been placed in a separate section preceding the model letters themselves, as found in ps.-Libanius.
The two epistolographic 'super-genres' securely attested in the Specimina are also found in both Greek collections but with some terminological variation. What are called suasoriae (συνβουλευτικαί) in the Specimina corresponds to the τύπος συμβουλευτικός in ps.-Demetrius (Ep. 11 = 7.3-11). It is also found in an interpolated branch of the older recension of ps.-Libanius (Ep. 42 = 34.1-4). 62 The gratulatoriae There is also a loose resemblance in content between this Byzantine letter and one of the corresponding Specimina letters, in so far as both stress the mutuality of benefits from friendship. 63 If the word was not coined independently, which is conceivable, it suggests a distant connection: perhaps the Greek translator of the Specimina was familiar with a lost epistolographic source also known to the Byzantine redactor; or perhaps the Specimina itself was known in some form to the early Byzantine epistolographic tradition.
The genre of the first two letters in the Specimina does not survive, and their fragmentary state makes it difficult to reconstruct. One attractive possibility is that it corresponds to the τύπος ἀπευχαριστικός of ps.-Demetrius (Ep. 21 = 11.14-12.2). 64 This is called εὐχαριστική in ps.-Libanius (Ep. 6 = 23.5-8). In the Latin tradition no clear terminological equivalent is attested, but possible names include gratifica, gratatoria, or eucharistica. 65 This identification is made plausible because of the overall resemblance in content and tone between the uncertain reconstruction of the second letter in the Specimina and the corresponding model in ps.-Demetrius (23.5-8). 66 To sum up, there are some general resemblances between the Specimina and the two Greek model letter collections, particularly in terms of genre, since all three of its 'super-genres' (suasoriae, gratulatoriae and-though the precise Latin terminology is uncertain-perhaps gratificae) are attested in the Greek collections. There are also a few very distant similarities in content, though nothing that can be called a direct borrowing or allusion. At the same time, the Specimina differs strikingly from any Greek collection in its degree of elaboration, showing multiple subgenres (at least of gratulatoriae) and multiple model letters for each. The content too shows considerable independence from any known Greek collection. Even if it was loosely inspired by a Greek collection that no longer survives, the density of references to Roman content shows that the anonymous composer was working with considerable artistic freedom.

Modified private letters as models?
While some ancient model letter collections appear to have been composed as such, others were produced by adapting pre-existing correspondence for pedagogical use. The latter method is illustrated by two Greek private letters that survive as part of a Ptolemaic administrative collection copied onto the verso of the Eudoxus papyrus (P.Paris 63). 67 Their function as models is suggested by their personal content, which differs from the surrounding administrative material, their careful punctuation and the absence of a named addressee or date. Some of the content of the letter has also been omitted, which the copyist indicated by writing μετὰ τὰ λοιπά (144.46). 68 At around fifty continuous lines, each of the letters is much longer than any letter in the other model collections, and the content seems both too specific and too complex to fall under the heading of a single epistolographic genre. 69 The best explanation for these features is that they were originally genuine letters, later adapted for use as models within the Ptolemaic chancellery, specifically by removing the prefatory and concluding sections and the omission of less relevant passages. A similar method can be detected in other sources. Dioscorus of Aphrodito in the sixth century appears to have copied several letters into his archive for this purpose. 70 Books 6 and 7 of Cassiodorus' Variae contain anonymous formulae granting administrative positions, which originated as actual letters. 71 Later examples can also be found, such as a tenth-or eleventh-century anthology of Alcuin's letters, in which the names of some addressees have been replaced by their generic titles or simply by illi. 72 The distinction between tailor-made model letters and those adapted from real correspondence is a delicate one. Genuine correspondence could be written with a view towards imitation, as in Pliny's Letters, and model letters might be composed out of intertexual segments recalled from 'real' correspondence. Nevertheless, the Specimina clearly aligns more closely with ps.-Demetrius and ps.-Libanius than it does with the Paris letters, both in terms of the relative brevity of its model letters and through the lack of reference to external content. Of course, it is possible that the letters had a much more complex prehistory before they were brought together in the Specimina, but there is nothing in the surviving portion that indicates this.

Conclusions
The study has aimed to establish a few basic parameters about the origin and nature of the Specimina, so that it can be investigated as an epistolary composition in its own right rather than as a bilingual glossary or an impromptu exercise. First, the basis of the collection appears to have been a document composed in Latin before the first quarter of the third century C.E. 'Composed' rather than 'assembled' or 'adapted' because the comparative evidence suggests that they were written as epistolary models rather than as letters that were retroactively adapted for pedagogic use. As compositions they show rhetorical training, as evidenced by the prose rhythm, and a willingness to elaborate the generic system found in Greek manuals.
Second, there is no unambiguous evidence of second-language interference from Greek or Coptic. When the multistage transmission history of the text is correctly understood, the strongest evidence for imperfect Latin at the composition stage disappears. That is not to say that the text contains only the purest Ciceronian Latin or that no textual difficulties remain (see Spec. 2.2, 9.2, 12.3), but there is no strong evidence that the composer spoke Latin as a second language.
Third, assuming that what survives is only a fragment of a larger collection, it would have been longer and generically more elaborate than any surviving Greek model letter collection. Indeed, given the wide range of date estimates on ps.-Demetrius, it is not entirely out of the question that the Specimina is the earliest witness to this genre that survives in either language. It presupposes an educated and relatively well-off Roman audience who could participate in the exchange of financial favours and testamentary bequests. Though it is not itself a legal treatise, there is a striking affinity to legal topics, including inheritance and manumission. Such topics might have been especially 1.2 n. 19. The second and third letters contain headings which suggest that the relationship between author and addressee is of more interest than their identities: 3.20 Ν[οτά]ρ(ιος) ἐμπαίδευτος uacat πρ(ὸς) ἑταῖρον. Cf. Fournet (n. 10), 61-3. 71 Cassiodorus' formulae are official grants that bequeath specific honores to the recipient, with the name formulae suggesting their legal character (cf. TLL 6.2.1117.53-60 'i. q. institutio magistratuum, dignitatum'). Thus they differ from the kind of model letters found in the Specimina. Cassiodorus claims that he has produced these formulae on the basis of letters he had written to real people (Var. praef. 14). 72 C. Chase, Two Alcuin Letter-Books (Toronto, 1975), especially 8-11. appealing in a Greek-speaking milieu to an audience of new citizens after the Constitutio Antoniniana in 212. Furthermore, there are several hints of unitary authorship, such as recurring vocabulary (remunero, suprema) and recurring constructions (genitive of cause). It would be normal for a collection such as this to be changed and adapted over time, with new genres or letters added, and the language lightly modernized, as perhaps occurred in the case of ps.-Demetrius. Nevertheless, nothing in the evidence we have examined suggests creative interventions into the Latin text by multiple hands at widely disparate periods. This tentatively favours speaking about a single anonymous epistolographer responsible for the surviving Latin text.
Ideally one would like to assign a name to the anonymous composer of the Latin original. There are indeed several 'known unknowns' within the history of Latin epistolography, writers of letter collections that have not survived in direct transmission but are mentioned and occasionally quoted in other sources. 73 Among the most prominent of these is Varro, known to have composed Latin and Greek letters and a treatise Epistolary Issues (Epistolicae Quaestiones). 74 Brutus was also highly regarded as an epistolographer, and Greek letters circulating under his name were known to Plutarch. 75 The language of the Specimina tentatively speaks against such an early date, but it stands as a representative of a larger, mostly lost tradition of Latin literary correspondence that circulated in the Greek-speaking portions of the Roman East. This Appendix tentatively reconstructs what the Specimina might have looked like before it was converted into a bilingual format. We also propose here a new system of reference according to units of sense and structure rather than the lines and columns of the document used by Kramer and Cugusi (we give columns and line numbers in parenthesis). Leiden conventions are used to indicate supplements and deletions, and every departure from the transmitted text is indicated in a selective apparatus criticus (an extensive apparatus criticus and translation are found in CLTP).