Where do the Classics come from? Or, the Apparatus Criticus and You.

Max Hardy

“I sing of arms and the man,” said thund’ring Virgil; and all the world knows that he said it. Or do they? How do you know what Virgil said? How do I know? How does anybody know? Take in hand a modern edition of Virgil’s Aeneid – the ‘Oxford Classical Text’ of Sir Roger Mynors, for example – and there you will indeed find printed the words Arma virumque cano, which do indeed mean “I sing of arms and the man”. But on whose authority do the people of the Oxford Classical Text series assert these words to be Virgil’s? And on whose authority do the people on whose authority the people of the Oxford Classical Text series assert these words to be Virgil’s themselves assert these words to be Virgil’s? And on whose authority do the people on whose authority the people – etc., etc., etc. This is less a silly question than it is a silly way of putting a serious question, a question which many have devoted their lives to answering, and which even today continues to perplex us:

How do we know what the authors of antiquity actually wrote?

Why such a question needs asking may not be immediately obvious to you; to me it was less obvious than it was obnubilous. But there is a reason, and it is not a happy reason. Out of all the great works of Classical literature ever written, not one survives in the form of an author’s original manuscript. Papyrus was then the writer’s medium of choice, and, not being a very durable substance, in the damp climate of Europe it tended to rapid disintegration. What survives are mere copies of these originals, and copies of those copies, and copies also of those copies, and further copies still – all connected in one continuous chain, unbroken from the early days of antiquity and extending up to the first light of Renaissance humanism, until the invention of printing eventually took over.[1]

An illustration of a scribe by the 15th-centruy illuminator Jean Le Tavernier (Paris, Bibliothèque Nationale de France, MS Fr 9198, f.19r), recoloured for E.G. Gress’s The Art and Practice of Typography (Oswald, New York, USA, 1910).

Now, as anyone who makes a business of copying things by hand should know, the process is rather a delicate one, and exceedingly prone to error. Imagine an entire book-trade operating along the same lines as a game of ‘broken telephone’.[2] Copyists make mistakes, and copyists copy mistakes. Set several thousands to work copying out all of Classical literature, and you eventually have a bit of a problem on your hands. Even in ancient times the phenomenon of the spelling mistake was a source of confusion and high displeasure. Hear the first-century poet Martial, who has a thing or two to say about it:

si qua videbuntur chartis tibi, lector, in istis        

   sive obscura nimis sive Latina parum,

non meus est error: nocuit librarius illis,

   dum properat versus adnumerare tibi.[3]

“If anything in these pages seems a bit obscure to you, reader, or a little un-Latin, it’s not my fault: a scribe did the hurt in his hurry to count them out for you.”

For a scribe at work in a medieval scriptorium (writing room) the situation was still more dire. Such a man had to contend not only with the often illegible handwriting of his predecessors, but also the inherent difficulties of the language and literature which it was his charge to copy. Latin was no-one’s native tongue in the time of Charlemagne;[4] and those little few who did know the language had not the advantages of dictionaries, standard grammars, and the Logeion app for iOS and Android. Under such conditions as these, errors of transcription were inevitable; and it should therefore be no surprise to you, nor any great reproach to the scribes, if they be discovered muddling the ends of verbs, skipping their eye over paragraphs, or mistaking nouns for their doppelgangers – writing, e.g., deus (“god”) for decus (“glory”), homo (“man”) for humo (“to the ground”), or angues (“snakes”) for aures (“ears”).[5]

Scribal errors were one thing that did the hurt; physical damage was another. Spills of water could make a line illegible; a piece of straw could be mistaken for a wayward colon; pages could fall out, margins could be set fire to, and whole codices could be scrubbed of their contents and recycled into more useful products. One and half millennia of such rough handling was enough to make quite a difference to any text, and many of Greece’s and Rome’s most brilliant works of literature suffered grievously by it. Of the surviving greats, Catullus and Propertius come to mind as having been particularly mangled, but many ‘lesser authors’ were treated far worse – lesser, perhaps, only because the circumstances of their survival were less felicitous.

The first poem in the oldest manuscript of Catullus (c. 1360s, Oxford, Canon. Class. Lat. 30, f.1r).

Eventually, from all this copying, miscopying, cutting, splicing and pasting there emerged a vast multitude of Classical manuscripts dating from every epoch. Some authors appear in only one or two of these MSS (as the standard abbreviation goes); others can lay claim to hundreds. Like witnesses to a crime (the crime of gradually debasing the words of a great author, I suppose), each manuscript has its own version of events. One may give Virgil’s words as Arma virumque, “arms and the man”; another Arva virumque, “fields and the man”; a third Aram vinumque, “wine and the altar”. Here enters the editor. Their task is to select out of all these different ‘readings’ (often called variae lectiones in Latin) those which, in their judgment, are most probably the words of the author. This is largely a game of informed guesswork – what you might call a “science conversant with literature”.[6]

It is also in great measure an art, especially that province of criticism which deals in the reconstruction of ‘corrupt’ text – text which the manuscripts furnish in a state so utterly nonsensical that its author could not possibly have written it so. A good example of ‘conjectural emendation’ – that is, restoration by informed guesswork’ – may be taken from the first book of Manilius’ Astronomicon, a long poem on Roman astrology. One important manuscript presents verses 423–4 of Book 1 in the form †esurcione Iuppiter ipse | quod poterat non posse timens, “Jupiter himself – then gibberish – fearing he cannot (do) what he once could.”[7] esurcione is no Latin word I’ve heard of, nor one that the formidable critic A.E. Housman had ever heard of either; in 1903 he conjectured, quite brilliantly, that esurcione was a debasement of the words eguit Iove, and that the whole phrase really meant “Juppiter stood in need of himself, fearing that he could not do what he once was capable of.” All later editors accept this correction into their texts, taking it for a moral certainty that Manilius once wrote what Housman’s brain long after divined.

Few will contest editors’ judgments in this matter. But what happens when their judgment is not so assuredly sound? What if they get it wrong?  Editors have an answer to this question, and they give it in the form of the apparatus criticus, or what you might call in English a ‘critical apparatus’. Literally, the apparatus criticus is an ‘arsenal of judgment’; it is also a place for editors to record all the relevant evidence on which their text of an author is founded: i.e., the testimony of the manuscripts whose evidence matters, and the later corrections of critics. Editors give this information as a tool for the reader’s use, that he or she may consider editorial decisions in the full light of the evidence, and consult his or her own judgment in places where the text does not appear to be sound.

The text that inspired the OCT series (1898–): Ingram Bywater’s edition of Aristotle’s Nicomachean Ethics (1890).

The apparatus is eminently a tool for the people (not the editor), but it is not made to seem so. Different publishers have different policies regarding how fat or thin their so-called ‘app. critt.’ should be, and naturally editors have their own opinions. The original tagline of the Oxford Classical Text series – “brevique adnotatione critica instruxit,” “and furnished with a short critical commentary” – promised brevity where other publishers proclaimed exhaustiveness. Compare how the latest editor of Homer’s Iliad for the German-based Teubner series described his handling of the text: “recensuit, testimonia congessit Martin L. West”, literally “edit [and] pile up with evidence did Martin L. West.” A quick squiz at the apparatuses of the OCT and of West’s Teubner will be enough to see the difference:

The apparatus of the Iliad is a special case, not least because it is the oldest literary text we have in Ancient Greek; others are much simpler. Let us take for our example the manuscript tradition and apparatus criticus of Manilius’ Astronomicon, famously edited by Housman in five instalments (1903–30). The complete text of this poem is established upon the testimony of three manuscripts which editors consider to be of ‘independent authority’, which is to say they are the only three manuscripts we have that are not known to have been directly or indirectly copied from another manuscript still in existence. Each therefore matters to the editor, as each stands at the end of a long chain going back to the author’s own text.[8] For the sake of time and space, editors have collectively chosen to denote these manuscripts by ‘sigla’, individual letters in Greek or Roman script. For the Astronomicon these sigla are L, G, and M, standing respectively for the Codex Lipsiensis (“manuscript from Leipzig”), Codex Gemblacensis (“MS from Gembloux”) and Codex Matritensis (“MS from Madrid”).

Here things begin to get a tad metaphorical. Editors like to talk about manuscripts as if they partook of the nature of families. (Technical discussions about manuscripts thus have a peculiar tendency to sound rather like a family drama.) So, a manuscript like L may be said to be a ‘parent’ (i.e., a manuscript from which others have been copied), as well as a ‘sibling’ (i.e., a manuscript related to another derived from the same ‘parent’ as L). As it happens, L really does have a sibling, and it is G: both L and G are thought to have derived from a common ancestor no longer in existence, and which editors denominate α. (Greek letters are often used for ‘conjectural’ MSS, or MSS regrettably no longer in existence.) M is thought to be the sibling of α, and thus the ‘uncle’, so to speak, of L and G. You can draw a family tree (or stemma) of this arrangement like so:

The manuscript O (sometimes also ω or Ω) is the letter typically given to the archetype, or what may be defined as ‘the latest common ancestor of the known manuscripts’.[9] The point of reconstructing these various affiliations is not only to produce a pretty diagram, but also to help the editor adjudicate between variants. Imagine, for example, that in the sentence currum ascendi (“I hopped into my chariot”) M gave the first word as currum, but GL instead gave crinem (“hair”). The agreement of GL enables you to infer (with practical certainty) that their parent α also had crinem: it is vanishingly unlikely that the scribes of L and of G both made the same mistake, at the same point in the text, of writing crinem for currum.

Yet, since M and α are siblings, one cannot say for a surety whether their (lost) parent O had currum or crinem. All you can say is that either M or α inherited the true reading, and the other made a mistake. The only way for the critic to decide what stood in O, and to reconstruct what we trust Manilius wrote, is simply by considering the sense produced by each alternative: as “getting into a chariot” makes markedly more sense than “getting into a lock of hair”, very probably the poet wrote currum, not crinem.

With this method in mind, let us turn to consider a real-world example: the beginning of the third book of the Astronomicon, lines 1–4. (The text printed below is Housman’s.)

in nova surgentem maioraque viribus ausum                      

nec per inaccessos metuentem vadere saltus

ducite, Pierides. vestros extendere fines

conor et ignotos in carmina ducere census.

2 vadere G : invadere LM | saltus GL2: saltos M | 4 ignotos Housman : dignos LM : indignos G : indictos Gron.| carmina M : carmine GL | census Unger : cantus O

“As I rise to fresh heights and venture a task beyond my strength, fearlessly entering untrodden glades, O Muses, be my guides. To widen your domains I strive, and to bring new treasure into song.”[10]

Unpacking the arguments of these curious footnotes (what we’ve been calling the apparatus criticus) is simpler than it may at first appear. Each entry is separated from the others by one or two vertical strokes |, and each corresponds to a specific word or collocation of words in the text. You can tell which line an entry refers to by the number at the front of it. Within each entry are contained the variant readings of manuscripts and suggestions of critics, all separated by colons, and ordered from left to right according to how probable the editor considers them. Thus the entry “2 vadere G : invadere LM” translates to “in line 2, the manuscript which I call G gives the reading vadere, and I as the editor consider this word as more probably what Manilius wrote than the reading given by manuscripts L and M, invadere”. Hence in Housman’s text, printed in the main body of the page, vadere stands, while invadere appears only in the apparatus.

The start of Manilius’ third book in a 15th-century manuscript (which also contains the medical writer Serenus), written in Ferrara, Italy, by Pellegrino Agli in 1461 (Boston (USA) Public Library, q Med. 20).

Manuscripts are usually distinguished by bold letters; italics are reserved for the conjectures of recognised human beings. Thus “census Unger : cantus O” means “the archetype O had cantus (“chants”), but personally I prefer Mr Unger’s correction, census (“treasures”), for it gives better sense to me.” As paper is a precious resource, and as the editors of Classical texts are acutely conscious of the ecological ills that accrue to wastefulness, not everyone mentioned in an apparatus is permitted to have their names recorded in full. The price of fame in the world of text-editing is to have one’s surname clipped by a dot in the app. crit.: thus “Gronov.” in line 4 stands for “Gronovius”, i.e., Johann Friedrich Gronovius. Among other common abbreviations are Heins. for one of two different people (Daniel and his even more ingenious son Nicolaas) named Heinsius, Bent(l). for Richard Bentley, and Lach. for Karl Lachmann.

Now here may be a good place to test an editor’s judgment against our own common sense. The first question we meet with in this apparatus is that of whether to print vadere (“to proceed”) or invadere (“to enter upon”) in line 2. That is not an easy problem to solve, and Housman’s judgment on this point may perhaps be faulted. For one, “to proceed through inaccessible groves” sounds rather weaker than “to venture upon” the same. invadere is assuredly the more striking word, and the more suitable to this proem’s vaunting tone. Housman’s choice of vadere may also be mistaken in one other important respect: it ignores what we know of Manilius’ stemma, and ignores what we know of how to use it. Consider the readings again in this format:

This is not a simple case of “M has invadere, α has vadere; pick one or other at your pleasure.” M certainly does have invadere; whether α had it is in doubt, and therefore whether O had it is much less assured than if both branches were in equal disagreement. Consider it this way: if O did have vadere, then two manuscripts (ML) must have each independently innovated the reading invadere by mistake.But if O originally had invadere, then only one manuscript (G) need have produced an innovation (by writing vadere). As the second of these two hypotheses involves the fewer assumptions, a prudent editor would say it is the more probable of the two, and thus the better explanation to adopt – you know, Occam’s Razor and all that.[11] To such an editor Housman’s preference for vadere may seem perplexing. Sure, the fact that the archetype O had invadere is no guarantee that Manilius actually wrote it (for many MSS must have passed between Manilius’ original and the archetype), but given that the manuscripts are our first and only evidence in this matter, why distrust them here?

Housman felt he had a reason to distrust them. To his knowledge, and to the knowledge of the writers of our dictionaries, the verb invado is not anywhere found in employment with per (“through”) when it has the sense of “to venture upon”, and the standard usage of Latin verbs of motion with the prefix in- is not such as to suggest that invado could ever be so employed. Or is it? Housman himself admitted the example of intravit levior per corpora somnus, “lighter sleep entered through the body”, a verse from the pseudo-Virgilian Culex (v. 206). And who’s to say that Manilius could not have innovated this particular use by analogy, or that further examples of invado + per could not have existed in other Classical texts now lost to sight?

Housman at play in the garden of his friend, Percy Withers, near Bicester, Oxfordshire: this picture was taken around 1930, at the time he completed the commentary on Manilius that occupied half his lifetime.

The conduct of Housman’s mental operations in this matter – that is, adopting into his text what I believe to be the error of a scribe instead of the reading of the archetype – is an instructive example of how ideology may operate even in the arid territory of textual criticism. You might say that his approach here is that of a ‘standardiser’, of one who would prefer his Classical authors to conform always to accepted rules of grammar, not to abandon them on a whim. The only reason why invadere + per does not appear in the Oxford Latin Dictionary (1968–82), or in the monolithic Thesaurus Linguae Latinae (1900–?), is because Housman chose not to print it here. Had someone with a more liberal view of Latin poetic usage edited Manilius, perhaps these books would contain such an entry. That is not to say that editing is an affair of mere taste or literary preference; canons of grammar and metre have a central place in criticism along with stemmata. But dogma and doctrine partake of criticism as well: editors have their styles as well as authors.

So, is all that now reason enough to read Virgil with an apparatus? If you have the time and inclination, why not? It gives you power to hold editors to account for their decisions, and may even afford you the opportunity, like it did Housman, to restore the words of an author yourself.

To read a Classical text without an apparatus has been justly compared to the experience of sitting in the back of a car while another controls the wheel. In the fullness of time, come the development of the self-driving car, this simile may eventually lose its force; but the apparatus will certainly not have lost its relevance to those who are fascinated about what ancient writers actually wrote.

Max Hardy is about to become a DPhil student in Classics at Trinity College, Oxford. He writes mainly on the text and interpretation of Latin poets like Tibullus and Statius.

Further reading

For a list of abbreviations in the apparatus criticus, see Antigone’s list here.

For a general survey of principles and practice, see M.L. West’s Textual Criticism and Editorial Technique (B.G. Teubner, Stuttgart, 1973). For a recent discussion of how digital technology is changing the landscape of text-editing, see Tom Keeline’s article “The Apparatus Criticus in the Digital Age,” Classical Journal 112 (2017) 342–6. And, in reply, S. Douglas Olson’s “Further notes on the Apparatus Criticus,” Classical Journal 114 (2019) 330–44. Finally, there is an excellent video series on many aspects of textual criticism and manuscript transmission, produced by the Digital Latin Library available on YouTube. Here’s Robert Kaster holding forth on the app. crit.

A textual cricket gets to work.


1 That is, speaking roughly, from the eight century BC up to the fifteenth century AD.
2 This game may be better known to British readers as ‘Chinese Whispers’.
3 Epigrams 2.8.1–4.
4 That is, in the eighth to ninth centuries AD.
5 It can happen to any of us. Once, in a passage of Lucan’s Bellum Civile (2.43–4), I happened to read the words numina saeva (“cruel divinities”) as nomina sacra (“sacred names”, a technical term for the abbreviation of such holy words as deus or dominus). If I had been a scribe in the age of Charlemagne, this mistake would probably now survive in several manuscripts.
6 Thus A.E. Housman, The Confines of Criticism. The Cambridge Inaugural, 1911 (Cambridge UP, 1969) 16.
7 The enveloping † – the ‘obelus’ – denotes textual corruption in the transmitted text.
8 Briefly put, the logic behind that thinking is roughly this: if manuscript A is a copy of manuscript B, and you have both manuscripts to hand, then you can forget about using A to reconstruct the text of your author. A will have no more truth in it than B, but will certainly have more errors, so B is the only manuscript worthy of your and your readers’ time.
9 NB: the archetype is very rarely the same as the author’s original manuscript; usually several stages of copies (and therefore corruptions) have happened between the production of the author’s manuscript and of the archetype.
10 This is the translation of G.P. Goold, Manilius: Astronomica (Loeb Classical Library, Cambridge, MA, 1977).
11 Entia non sunt multiplicanda praeter necessitatem, “things should not be multiplied beyond what is necessary.” With that in mind,this introductory discussion ignores the complicated problem of textual “contamination” (where unrelated manuscripts influence each other ‘horizontally’ across the stemma): this very thorny problem is not one I wish to venture upon here!