Thursday, 18 August 2022

Can we trust mediaeval copies of ancient books?

When you open up a modern copy of Plutarch’s Lives or Tacitus’ Annals, what you’re reading is based on a mediaeval manuscript. It’s a copy of a copy of a copy. Each time a text is copied, there’s a risk of transmission errors. So how much faith can we have in the texts we have? And what kind of faith?

Take Petronius’ bawdy novel, the Satyrica. We don’t have copies written by Petronius’ own hand — we don’t have that kind of thing for any ancient author. Our text for the episode of ‘Trimalchio’s dinner’ comes from a single copy made in 1423, known as codex Parisinus latinus 7989, starting at page 206. That’s over 1300 years after the novel was written. Suppose the text had to be copied every 200 years to survive that long: that means six layers of copying.

So, what should we think of mediaeval copies? Do we trust them, distrust them, or somewhere in between?

The answer isn’t binary, so strictly speaking it’s ‘somewhere in between’ — but it’s very close to ‘trust’. Here’s the standard position:

  1. It is presumed that the manuscript tradition is reliable.
  2. There are good reasons for this presumption.
  3. But it doesn’t take much to qualify or overrule the presumption, in a given passage of text.

Be careful with point 3: it’s easy to get bogged down in caveats. Transmission errors come in all shapes and sizes, and there are many reasons why they happen, so cataloguing them takes a while. That tends to create a false impression. So it’s important to emphasise that reliability is the norm.

We can always distrust a given reading in a given manuscript, but there always has to be some reason for doing so. You don’t ask, ‘How do we know this text is correct?’ You ask, ‘Is there any reason to think this text is incorrect?’

That is, the claim that a textual reading is wrong bears the burden of evidence. The evidence doesn’t have to be strong. But it does have to exist.

Addendum, written a day later: I have come to realise I have overstated my case, and I should have given a much more qualified statement of what ‘reliable’ means. Please also refer to my postscript I have added at the end. The examples I discuss below aren’t representative of all ancient texts: they’re indicative of certain types of texts, in certain genres, after a certain date. And even then, not all of them. The devil is in the details.

So yes, we do trust mediaeval manuscripts, and as I said there are good reasons for this. What remains is to explain the reasons.

  1. Copying, by definition, is a faithful process. The entire point of making a new copy of a manuscript is to produce reliable and conscientious copies. If you copied a computer file from one place to another six times, you’d be disappointed if the final copy turned out to be one bit different from the original. Same thing with manuscripts.
  2. Copying includes error-correction. It’s true that copying manuscripts by hand has a lower accuracy rate than many forms of automated copying. But first, that’s a matter of degree, not a different kind of process. Second, ancient and mediaeval scribes took pains to correct errors when they saw them. Mediaeval scribes corrected obvious errors. Often they copied from multiple manuscripts, so as to compare them. There are cases where a copyist made major alterations — for example, some prudish scribe bowdlerised Herodotos, Histories 1.199, resulting in the chapter being omitted in three manuscripts — but big alterations aren’t normally able to propagate throughout the entire manuscript tradition. Big alterations that did propagate successfully are phenomenally rare. On the scale of an entire chapter, there’s only one known example: the so-called ‘Testimonium Flavianum’, in Josephus’ Jewish antiquities 18.63–64, where a Christian altered or added two chapters about Jesus.
    Note. For discussions with bibliography see for example Olson 1999, 2013, arguing that the passage was added in the 3rd century CE; Goldberg 2021, arguing — implausibly — that the passage is authentic, and that Josephus himself adapted it from a Christian source. I’ve written about it briefly here.
  3. Modern editors have the explicit goals of gauging manuscripts’ reliability and amending errors, and they have a powerful arsenal of techniques for doing so. A short summary of these techniques would do them injustice, because of their complexity and their sensitivity to historical and linguistic context. For a more nuanced picture, below we’ll take a look at how they can work in practice. For a mid-length summary, see Reynolds and Wilson 1991: 207–241; for full book-length descriptions of the methodologies, see West 1973; Maas 1950, 1958.
  4. Where there are doubts over a text, modern editions give full documentation of those doubts. I don’t mean popular translations: the Penguin Classics series won’t tell you about manuscript errors. I mean critical editions. The entire point of a critical edition is to expose flaws, disagreements, and doubts in the manuscript tradition to the reader’s view, to maximise the reader’s understanding of the manuscripts and their genetic relationship to one another. See here for my own guide on how to read these annotations; video tutorial here.
  5. Where it is possible to check the accuracy of the manuscript tradition, its accuracy is high. Accuracy can’t be quantified, because transmission errors are very diverse. (For a taxonomy see Reynolds and Wilson 1991: 222–233.) What we can do is take an illustrative approach. We can check mediaeval manuscripts against ancient copies if we have either (a) an ancient copy of the text, or (b) a quotation in another surviving ancient text. These are always fragmentary, so even though we have thousands of each, they only account for a tiny proportion of the material. Still, that’s plenty to illustrate that mediaeval copies are overwhelmingly very reliable. The expected level of divergence is on the order of spelling variations; the occasional rare word being mistaken for a more familiar word; words being swapped around; and the like.

Points 1 and 2 here are obvious; for point 3 I’ve referred to multiple published expert accounts; and for point 4 I’ve referred to critical editions and my own tutorials on how to use them. So the point that still needs expansion is point 5, checking accuracy against ancient copies.

As I said, the best approach is illustrative. So let’s illustrate. We’ll look at some texts that survived intact via the manuscript tradition, which have also turned up in fragmentary ancient papyri found at Oxyrhynchus, in Egypt.

Example 1. A 3rd century copy of Lucian

The satirist Lucian (or Loukianós) was active in Syria in the 2nd century CE. Here’s a 1961 edition of his Dialogues of the gods, dialogue 10, with the text as transmitted by the mediaeval manuscript tradition, and an English translation.

And here’s a 3rd century papyrus with a fragment of the same dialogue: Oxyrhynchus papyrus 4738, first published in 2005.

Here’s the 2005 publication of the papyrus with a transcription. The papyrus preserves only 18 lines, and only the right hand edge of each line. But this is plenty to see that it’s the same as the mediaeval text, aside from certain standard conventions.

Mediaeval texts and ancient papyri have different orthographic conventions. The conventions used in modern editions were standardised in the Byzantine era, and standardised further in the modern era: punctuation; word-divisions; capitalisation; diacritics; and iota subscript (used in the 1961 edition linked above; ancient copies normally use iota adscript).

Note. Punctuation and diacritics do sometimes appear in ancient copies too, but not consistently, and not as thoroughly. For example, line 3 of this papyrus uses an interpunct to separate the end of Zeus’ speech from the start of Ganymede’s, between [δ]οκῶν and Ἄνθρω[πε].

Setting these aside, there are four differences between the papyrus and the 1961 edition:

  • line 5: papyrus [ε]ξ[ε]ρυηκε; transmitted text ἐξερρύηκε
  • line 6: papyrus μι[ρακιον]; transmitted text μειράκιον
  • line 10: papyrus [εϲτ]ηκει; transmitted text ἕστηκε
  • line 14: papyrus [φ]ηϲ; transmitted text φῄς (elsewhere the papyrus uses iota adscript)

Three of these — ἐξερρύηκε, μειράκιον, and φῄς — are still about orthographic conventions. The mediaeval text uses standardised Byzantine-era spelling, and the papyrus text uses the looser conventions of ancient Koine. They’re phonetically identical to one another. Modern editions normally follow the Byzantine standards.

The remaining difference is in line 10, where the papyrus has [εϲτ]ηκει, and the modern reading is ἕστηκε. There is a subtle distinction in meaning here: one is pluperfect (‘where the statue was placed’), the other is perfect (‘where the statue is placed’). But it turns out [εϲτ]ηκει also appears in one branch of the mediaeval tradition as ἕστηκει: branch γ of the manuscript tradition uses the pluperfect form, while branch β uses the perfect tense. In the 2005 publication of the papyrus, the editor recognises that the perfect form makes more sense, since the rest of the sentence uses present aspect.

So even though [εϲτ]ηκει is attested in antiquity, ἕστηκε is judged to be more authentic. ‘Older’ doesn’t necessarily mean ‘more reliable’. The implication is that the split in the manuscript tradition between branches β and γ occurred in antiquity, before this papyrus was made, within a century of Lucian composing the dialogue.

And that accounts for all the divergences. The result is that in its essentials the papyrus contains the same text that we already had. The mediaeval text stands as a reliable indication of what Lucian actually wrote. There is a question over which conventions to follow — should we really use Byzantine conventions for spelling and diacritics, when it’s likely that Lucian used looser Koine spelling and no diacritics? — but that’s a matter of policy, not accuracy.

Example 2. A 1st century BCE/CE copy of Meleager

The poet Meleager (or Meléagros), from Gadara in what is now Jordan, was active in the 1st century BCE. He compiled an anthology of short poems called the Garland. The Garland itself doesn’t survive, but it formed the basis of an expanded collection that survives in a 9th–10th century manuscript, now held in Heidelberg, known as codex Palatinus graecus 23. The collection is known as the Palatine anthology.

Here’s an edition from the 1910s with four epigrams by Meleagros about love: Palatine anthology 9.16, 5.190, 12.157, and 5.152.

And here’s Oxyrhynchus papyrus 3324, first published in 1980. It contains parts of 17 lines. The first line comes from an unknown poem; the rest are the four poems mentioned above. It may be that this is a copy of the Garland itself, but that can’t be certain (see the introduction to the 1980 publication for details).

This time we’ve got more than just orthographic variations. The papyrus shows that there are two actual errors in the mediaeval text. But the papyrus contains an error too. Here are the divergences:

  • line 3 (Anthology 9.16.2): papyrus οι]ϲτροβολουϲι = transmitted text οἰστροβολοῦσι; in the 1800s the editor C. J. Blomfield suggested altering to οἰστοβολοῦσι
  • line 4 (Anthology 9.16.3): papyrus κα]τηρτικεν; transmitted text κατήρισεν; Maximus Planudes reports κατήρυσεν; in the 1500s Scaliger suggested altering to κατήρτισεν
  • line 8 (Anthology 5.190.3): papyrus ανεινται; transmitted text ἀφεῖνται 
  • line 9 (Anthology 5.190.4): papyrus εϲοψομεθα; transmitted text and Planudes ἀποψόμεθα
  • line 16 (Anthology 5.152.3): papyrus ϲο; transmitted text σὺ

In line 3, it’s a modern editor who is at fault. The papyrus confirms that the mediaeval text, Desire ‘striking with a sting’, is correct; Blomfield’s alteration ‘strike with an arrow’ is wrong.

In line 4, the mediaeval text is wrong. In a sense this isn’t news: the mediaeval reading κατήρισεν isn’t a meaningful word. Scaliger suggested correcting it to ‘Desire equipped three bows’, using an aorist verb form. The papyrus confirms Scaliger’s choice of verb, but uses the perfect form, ‘Desire has equipped’.

In line 8, we can’t be certain which is correct. The line is ‘the rudders of my thoughts are set loose every which way’. The variation is between two compound verbs, both meaning ‘are set loose’. Perhaps at some point a scribe changed ἀνεῖνται to ἀφεῖνται to improve the euphony with φρενῶν (‘thoughts’) in the same line.

In line 9, the mediaeval text is wrong. Again we have two compounds of the same verb. The line is supposed to mean ‘Shall we ever again set eyes on tender Skylla?’, but the mediaeval reading ἀποψόμεθα means ‘look away from’; some modern editions already improved this to ἐποψόμεθα ‘set eyes on’. The papyrus appears to correct the text to ἐσοψόμεθα ‘look towards’.

In line 16 it’s the papyrus that is wrong: ϲο isn’t any kind of standard spelling. Perhaps the ancient scribe mistakenly started writing the word σοί (‘for you’) rather than σύ (‘you’).

This papyrus alerts us to a couple of errors in the mediaeval tradition, but it also illustrates the reliability of the mediaeval tradition in other respects. Moreover, the papyrus contains one clear error. Once again, older isn’t necessarily better — but more information always is better.

Example 3. A 2nd century copy of Strabo

The geographer Strabo (or Strábon), from Amaseia in Anatolia, was active in the late 1st century BCE and early 1st century CE. Book 9 of his Geography gives a description of Attica, Boiotia, and Thessaly: here’s an edition published in 1927 along with an English translation. The best mediaeval copy for this part of Strabo is a 10th century manuscript held at the Bibliothèque Nationale de France, known as codex Parisinus graecus 1397.

Oxyrhynchus papyrus 3447 is a copy of book 9 dating to the early 2nd century CE. Like the ones discussed above, it’s fragmentary. Unlike them, it’s a collection of many fragments: there are 103 fragments in total. It was first published in 1982.

We can’t discuss every variant reading individually here. Instead, here’s the editor of the papyrus counting the different categories of variants:

  • a number of variations in the spelling of proper names
  • three textual errors in the papyrus (all minor)
  • seven variants ‘of uncertain value’
  • four variants where the previously known text appears to be wrong
  • five places where ‘the reading of the papyrus is uncertain, but probably different from that of the [manuscripts]’
  • ten places where the papyrus agrees with the mediaeval text, and disagrees with another ancient copy, the Vatican palimpsest (Vaticanus graecus 2306 + 2061A).
  • two places where the papyrus agrees with the Vatican palimpsest, and disagrees with the mediaeval text

The four variants where the papyrus is right, and the previously known text is wrong, are as follows:

  • fr. 14.i.9 = Geography 9.2.35: a quotation from Homer, Iliad 2.507, given correctly in the papyrus as οἵ τε; incorrectly in the mediaeval text of Strabo as οἱ δέ; in Homer the correct text was always known to be οἵ τε; both variants have the same meaning, ‘and they who’
  • fr. 14.i.20 = Geography 9.2.35: papyrus οὐδ’ (‘nor’); transmitted text οὔτε (‘neither’; a 19th century editor had already corrected this to οὐδὲ)
  • fr. 14.ii + 15.i.7 = Geography 9.2.36: papyrus μὲν τοίνυν (‘well then, in the first place’); transmitted text τοίνυν (‘well then’)
  • fr. 38.9 = Geography 9.5.17: papyrus καὶ ταύτην (‘both this’); transmitted text καὶ (‘both’) + lacuna (it was always known that there is a gap in the text)

In regard to the disagreements between the papyrus and the 5th century Vatican palimpsest, the papyrus editor warns against the idea that they reflect a split between branches of the manuscript tradition. Rather, it’s that the Vatican palimpsest has more inaccuracies than other copies. This was something that had already been noticed by the modern editors of the palimpsest, which was first published in 1956.

In summary

It can be a bit overwhelming when you look into the nitty-gritty of ancient copies and their variations from mediaeval texts. The upshot is:

  1. Modern editors really know what they’re doing, and their expertise in sorting out the correct text deserves a huge amount of respect.
  2. Mediaeval copies are very accurate, with only minor discrepancies from their ancient counterparts.

Now, having said that, there are situations — or rather, literary genres — where we do expect much more discrepancies. Some ancient texts weren’t copied as such, but instead went through recensions and reworkings.

This is the case with, for example, the Aesopic fables. We don’t have any of the Fables as originally composed by Aesop in the 6th century BCE. No modern edition pretends that we do. What we do have is reworkings, reframings, retellings, by a variety of people in different periods: Greek and Latin prose versions attributed to Dositheos, Libanios, and Phaedrus; poetic versions by Babrios and Phaedrus. Sokrates himself is reported to have composed a verse version of one of the Fables. And there are several mediaeval fable compilations, which are different again.

Another example is the Alexander romance, a heavily fictionalised version of Alexander’s life, which exists in numerous recensions. There are four recensions in Greek ranging from 3rd to the 9th centuries; two in Latin; two in Armenian; and others in Hebrew, Syriac, Arabic, and Ethiopic. Some of them are closely related to one another, others less so.

In cases like these, it genuinely is futile to expect a mediaeval copy to be a close replica of an ancient text. But there’s an important difference between these and the examples we looked at above. When we look at a literary text attributed to a specific author, the text turns out to be very stable. It’s when a text gets excerpted, reorganised, or fully recomposed: that’s where the process of copying breaks down, and it becomes something quite different.

Postscript, a day later

I’ve realised that I greatly overstated my case. The three examples given here aren’t representative of all ancient texts: they’re indicative of certain types of texts, in certain languages, and certain genres, after a certain date — and even then, not all of them.

For balance, the closing discussion of literary genres where there’s more variance should have been much longer, and should have had more emphasis.

For example, our texts of Attic drama are significantly further away from the original plays. There are several reasons for this:

  1. There’s a much bigger time gap between the surviving papyri and the original texts.
  2. We know that the ongoing tradition of theatrical performance led to increased interpolations and substitutions in the text (actors adding their own lines, reorganising scenes, etc.).
  3. Orthographic and editorial conventions make a much more significant difference than in the case of Roman-era texts. The pre-Hellenistic classics suffered much more intrusive editorial practices in antiquity, and we don’t know all the details.
  4. Aischylos didn’t use the Ionic alphabet. That’s an alteration probably dating to the 4th century BCE. Aristophanes’ late plays, by contrast, possibly did.

So there are many intermediate situations — in between reliable replicas of the original texts, on the one hand, and the extreme variance I described at the end. In these intermediate situations, we can’t talk in terms of having ‘replicas’ of the originals. At the same time, it’s still reasonable to say that there’s a pretty good degree of resemblance with the lost originals.

All pre-Hellenistic literature goes into this intermediate category. We cannot possibly guess what the earliest written texts of Homer looked like. Even in Roman-era texts, there are cases where there’s more substantial variance than I talked about above. For example, the extant papyri of Josephus show a substantially higher degree of variance. Then there are texts that underwent ongoing remodelling, such as the Bible.

In short, I should have been much clearer about what the three examples I looked at above are representative of. There are many individual cases where much greater caution is needed.


  • Goldberg, G. J. 2021. ‘Josephus’s paraphrase style and the Testimonium Flavianum.’ Journal for the study of the historical Jesus 20.1: 1–32. [DOI]
  • Maas, P. 1950. Textkritik, 2nd edition. Leipzig. [Internet archive]
  • —— 1958. Textual criticism (= translation of Maas 1950). Oxford. [Internet archive]
  • Olson, K. A. 1999. ‘Eusebius and the Testimonium Flavianum.’ Catholic Biblical quarterly 61.2: 305–322. [JSTOR]
  • —— 2013. ‘A Eusebian reading of the Testimonium Flavianum.’ In: Johnson, A.; Schott, J. (eds.) Eusebius of Caesarea: tradition and innovations. Washington, DC. [Center for Hellenic studies]
  • Reynolds and Wilson 1991 [1968]. Scribes and scholars. A guide to the transmission of Greek and Latin literature, 3rd edition (see also 4th edition 2013). Oxford. [Internet archive]
  • West, M. L. 1973. Textual criticism and editorial technique. Oxford. [Internet archive]


  1. "The Turkish geographer Strabo"...
    You knew this would get a reaction didn't you...? 🙂
    Describing Strabo as Turkish is a bit off, especially for a scholar of your caliber..
    One is to understand that his writing is in the Turkish language...? or that his geography is that of the Ottoman Empire...? he must have had some amazing clairvoyance for that...!!
    Please do revise/edit, it is laughable as a description.

    1. It just gets a bit laborious writing 'in what is now Syria', 'in what is now Jordan', 'in what is now Turkey'. You may well be right: I'll sleep on it, and will probably decide in the morning to follow your advice.

    2. May I then suggest writing their birthplace as then called, i.e Amaseia, Pontus for Strabo or Samosata for "Syrian" Lucian...?
      Your readers are of a high enough level to appreciate the accuracy of calling some place by its original and correct name.

    3. I decided a bit sooner, as it turns out. You were perfectly right to warn me that there's no need to be inflammatory.

    4. Thank you...
      Scholarly inflammatory would be OK but not this kind.

  2. An excellent Article! χάριν coι ´οιδα την μεγιcτην! Might you get it posted in evangelicaltextualcriticism.blogspot ? I look forward to reading through all your articles here! ειρηνη και ευλογια!

    1. Thanks, but that really wouldn't be appropriate: that site is intended for 'the perspective of historic evangelical theology', and I'm definitely not coming from that position.

      In any case, I've already come to realise that I overstated my case. I need to add some caveats to the 'in summary' section. The examples I discuss here are much too narrow to be representative of all ancient texts; there are plenty more intermediate genres and authors where we certainly don't have 'replicas' of the original texts, even if we're a lot closer to the originals than we are with the extreme cases I mentioned at the end. (For reference, I'd certainly include the texts in the Bible in this intermediate category.)

    2. Thank you for your comment.
      I didn’t assume that you were coming from an evangelical position!
      Why would you include the texts in the Bible - or, say, here, the Greek New Testament - in the intermediate genre, “where we *certainly* [my emphasis] don’t have ‘replicas’ of the original texts”? I note, however, that you seem to include such texts as being “a lot closer to the originals” than other texts; but is it your position that the GNT texts were not copied without interruption, so as not to qualify as ‘replicas’, but had to be reconstituted ‘dē novō’ following interruption(s)?


  4. This is very helpful. Thank you for this work.

  5. Not directly of this post, but maybe...
    Could I please have you view on this :
    I trust your opinion more than the website. Thank you.

    1. I'm going to take it that the examples cited on that page are accurate because checking every one of them would eat up a lot of time today! I wouldn't be surprised if some of them turn out to be the word χρηστός, but let's let that slide, because there's an essential point that the brief 'explanatory hypothesis' section is unaware of.

      That point is that in Roman-era Greek, ι, ει, and η were phonetically identical. (As they still are in Modern Greek; though it took a few centuries longer for υ and οι to join the club.) And epigraphic and literary evidence give us copious examples of Roman-era writers freely swapping between the three of them.

      I mean for example things like spelling Αἰνείας as Αινηας, or Ποσειδῶν as Ποσιδων. There's literary evidence too: a Neronian-era poem by Leonides (Palatine anthology 9.355) uses an isopsephy which relies on spelling μίμημα as μειμημα (that is, add up the numeric values of the letters, and the two couplets add up to the same total; it doesn't work if the ε is omitted). Above in this piece, I mentioned that the Lucian papyrus spells μειράκιον as μιρακιον.

      This variation is so frequent as to be regarded as standard. Variations between χριστ- and χρηστ- require no further explanation.

      (Though of course there is further explanation, since pre-Roman orthography also had χρηστός as a legitimate word, adding extra motivation for orthographic variance.)

      The appearance of both Chrest- and Christ- in Latin texts requires the extra step of transliteration. Greek-to-Latin transliteration had well defined traditional equivalents for this vowel -- ι, ει > i, and ε, η > e -- which fully motivates the different transliterations of χριστ- and χρηστ- even though they were phonetically identical. We even find both forms in a single passage, in the case of Tacitus, Annals 15.44 (a Christian scribe normalised the spelling chrest- to christ- at some point, but the original text is legible).

      I was going to end by saying that the real surprise is that we don't find the spelling χρειστ-, which is also phonetically identical -- but then I took a moment to check and it turns out we actually do have that spelling too! For example in these inscription: 1, 2, 3, 4. So there's no mystery at all.

  6. What about those texts whose transmission relies on a single surviving copy? Are there ways to confirm that what we have is a good copy?

    1. Not to confirm, no, but with Roman-era texts the default presumption is going to apply. There needs to be some kind of reason to treat a reading as unreliable.

  7. An excellent article. However calling Strabo "Turkish geographer" seems a little anachronistic.

  8. Hi Peter, Thank you for these examples. It's always helpful to look at specific cases. In what could be understood as a footnote to your postscript, I've summarized some related conclusions about the text of Thucydides here:

    1. Older authors like Thucydides are a much more challenging area, as I realised after writing the original form of this post!

      I wonder what you think of the potential role of metagrammatism: it's very easy to imagine an early 'editor' causing more substantial alterations to the text if we suppose that older Attic authors had to be transliterated into the Ionic alphabet -- though given the date of Thucydides, it's perfectly possible he used Ionic letters all along.

    2. Well, thanks for teaching me a new word. I'm not sure; it's not a topic that I have though a lot about. But it is an interesting question. There is a kind of fun example adduced in this article:

  9. Between what's linked here and the satire on the City of God Lucian seems like a really funny guy.

  10. Do you have any thoughts on the ancient practice of 'publishing' books, especially poetry, by reading them aloud in public venues, libraries, basilicas. theatres, and odeums? I haven't read anything good on this practice. Maybe the subject for a future blog?