Basic Stemmatology for Students of Early Modern Scribal Anthologies

By Harold Love

The most prolific source of scribally transmitted verse and short prose texts from sixteenth- and seventeenth-century Britain are the scribal anthology and personal miscellany. Anthology is here used in its broadest sense of any collection of short texts falling under a single generic category. An understanding of the transmissional histories of these aggregations and their printed derivatives, as well as being necessary for editors, is also vital for studies of the circulation and communal use of texts.
Sometimes relationships are evident from the order of items, either of whole collections or of sub-collections embedded within them; but more often the student will need to establish the descent of items singly through the analysis of variants. Note that the pattern of descent will often vary from item to item. A full investigation involves tracking down all surviving copies of the item, as is routinely done by editors. Fortunately this information is increasingly available from on-line library catalogues and first-line indexes. Consult James Woolley, ‘First-line Indexes of English Verse 1650–1800: A Checklist‘. The following notes introduce some basic analytic techniques. [Readers of the Adam Matthew ‘English Clandestine Satire’ collection introductory document can skip to section 3.]

1. Practicalities

Manuscripts are inscribed on writing paper, which has a smoother surface than printing paper. Writing ink is water-based unlike printer’s ink, which is oil-based. For the technology of copying, see Michael Finlay, Western Writing Implements in the Age of the Quill Pen (Wetheral: Plains Books, 1990). Palaeography is too huge a subject to be addressed here. Start with Anthony G. Petti, English Literary Hands from Chaucer to Dryden (London: Arnold, 1977); Jean F. Preston and Laetitia Yeandle, English Handwriting, 1400-1650: An Introductory Manual (Asheville N.C., 1999); or Early Handwriting 1500-1700: An Online Course (Cambridge English Renaissance Electronic Service). The short texts that are our present concern usually began circulation as separates containing either a single work or a linked group of related works. Separates might be written on a single leaf, a half-sheet bifolium, or a whole sheet folded as either a bifolium or quaternion.

Separates have historically had a very low survival rate. They would frequently, however, be copied into the personal miscellanies of readers which, being robust, bound volumes, had a much better chance of enduring to the present. Items regarded as commercially attractive might be appropriated by professional scribes copying for sale. As well as producing further separates, some of these scribes vended handsomely written volumes with an excellent chance of survival. One also encounters less carefully inscribed copies of professional volumes that were made by readers for their own use. Some copies in personal miscellanies may also derive from professional separates. Separates and copies in commonplace books derived from separates will show the greatest textual variation since they will normally be several or many copyings apart. Professional copies derived from a single exemplar rarely vary much among themselves and are of value only for what they show about the exemplar. Having reconstructed the readings of that exemplar, the student will need to assess whether it may have been supplied by the author (as sometimes happened) or was simply another chance-encountered separate. Those in printed drollery collections or publishers’ anthologies may equally derive from a fortuitously encountered separate. Note that it was common for scribally circulated texts to be censored and restyled by their print publisher for presentation to their new public. A professional scribe such as Robert Julian might do the same in the interest of making his texts more saleable, but by enhancing their controversial aspects rather than trying to diminish them.

It would be a mistake to assume that the common ancestor of the body of surviving copies would always have been a perfect, authorial text of the work concerned. That would be to confuse the archetype with the author’s original MS: indeed, even an author’s original may have been carelessly written and contain errors. In some cases whole traditions may have arisen from a seriously defective copy, making scribal modification a process of improvement as well as corruption. So the readings of a sophisticated late copy may be polished and plausible without being those of the archetype. There is no reason why a writer should not have made changes every time he or she put a new copy into circulation. This would inaugurate a plurality of traditions, for the same text.

2. The process of copying

It is important not to make limiting assumptions about the nature of the copying process that might have produced any given source. There are several ways of copying a manuscript. While some personal miscellanies may have been copied sheet by sheet, and only then bound up, most were entered into already bound white-paper volumes, and will often have unused blank leaves at the back. Such collections might be in the hand of a collector or secretary, or in a number of hands, sometimes belonging to family members or to successive owners. Scriptorium volumes, on the other hand, were usually written sheet by sheet. Just as when one studies the production history of a printed book, one needs to establish as much as can be inferred about the nature of the exemplar. Often mistakes in a copy will suggest that a scribe was working from an original written in a particular hand that facilitated specific kinds of misreading. Mistakes in the order of text elements may be indicative of structural irregularities in the exemplar. One must try to establish whether the exemplar was a bound book, a body of loose leaves, or a collection of sheets. In anthologies copied from a rolling archive, the exemplar existed as a series of separate leaves or sheets whose order might well change between copyings. In some cases we can follow a process by which elements are progressively removed from the exemplar and others added. We should not assume that scribes always carefully preserved their master copies. Copying might be performed seriatim with each copy becoming the exemplar for the next. This could occur because of a desire to preserve some innovation in presentation or styling or simply through pressure of demand. Note that this is also the practice of print production, with editions normally set from their immediate predecessor. When multiple copies were produced, a form of progressive copying could have been used by which a primary copy of a sheet made by one scribe was passed on to others. Which of them got it first might depend on their individual speed as copyists. This could theoretically lead to a situation in which the textual descent of the resultant volumes changes sheet by sheet. Try to remain aware of such possibilities even if they might have been relatively rare.

3. Recording the evidence

What the student seeks is evidence for the historical sequence of copyings that has given rise to the surviving sources. Experience has shown that this evidence is most likely to be found among verbal (substantive) variants, i.e. those that bring about a change in meaning. These should be scrupulously recorded in their entirety from all known sources for the work and rechecked against them, entry by entry, when the primary collation is completed. An inaccurate or selective record will simply mislead the user. It is always desirable to check apparently anomalous readings in editors’ published collations against their sources. That judgments over what constitutes a substantive variant may differ between editors doesn’t matter much so long as criteria are applied systematically by each one; but the reader does need to know that all examples of any given class of variants have either been recorded or not recorded. ‘Accidental‘ or ‘incidental‘ variations – those affecting spelling and punctuation – should only be included when they might be helpful in determining the descent of sources (e.g. variant spellings of an unusual proper name). Most variation of this kind is so subject to individual scribal whim as to offer no useful basis for reasoning. Including it in a variant list will simply make it harder for the reader to use the list efficiently. The format for recording variation recommended is that used in the The Works of John Wilmot, Earl of Rochester, ed. Harold Love (Oxford: O.U.P., 1999).

In writing variants the researcher will often have the options of using a single phrasal variation or two or more verbal variations. Compare the two examples cited in the Rochester edition:

  • 3 Two] Soon as ~ BLa47, FIS, Ks24, Pt5, WAC, Yo05, Yo40; Wn NLI93, Od29, Oep4; Whenas ~ Orp8; As soon as the ~ RUT
  • Rivall] Royall He36; om. RUT
  • fleets] navies NLI93, Od29, Oep4
  • appearing] appear BLa47, FIS, Ks24, Orp81, Pt5, RUT, WAC, Yo05, Yo40; doe appeare NLI93, Od29, Oep4
  • from afar] in sight BLa47, FIS, Ks24, Orp81, Pt5, RUT, WAC, Yo40; from far NLI93, Od29, Oep4, Yo05

as opposed to

3 Two Rival Fleets appearing from afar] Ab12, BLa22, Ed3/1, Lb54, Od8, SKv69, 80po, 80pf, 91po; Soon as two rivall fleetes appear in sight BLa47, FIS, Ks24, Pt5, WAC, Yo40; Two Royall Fleets appearing from afarr He36; Wn rival navies doe appeare from farr NLI93, Od29, Oep4; Whenas two rival fleets appear in sight Orp81; As soon as the two Fleets appear in sight RUT; Soone as Two Rivall Fleets appeare from farr Yo05

Both versions record the same information at a different level of particularity. The second method gives the more helpful overview to the experienced scholar but the first would be more practical in the early stages of collation. It also permits a quicker perception of the actual groupings. The swung dash (~) is used to economize on repetitions of a word on both sides of the bracket. It can be used irrespective of changes in spelling.

Generally only include truly variant elements in your entries and avoid repeating invariant words on both sides of the bracket unless they are essential as cues to distinguish two occurrences of the same word in a line or are inextricably embedded in phrasal variants. When cue words are used it is best if they are regularly the preceding, not the following word. This also gives a simple way of recording lacunae and additional words (Mary had] Mary Mary] ~ had).  Avoid ‘om.’ since it presupposes omission when that need not be the case. When a verbal description is unavoidable prefer ‘Not in . . . ‘ Whenever a variant affects several lines but is not further recorded, it should be bolded (e.g. 12–21 Not in Cb5) so that the reader is fully alerted. Listing can sometimes be signified by embedding a minor variant in pointed braces within a longer one, e.g. ‘Whenas two rival fleets {hosts Cb5 }appear Aa1, Cb5, Dd34′.. Sigla (sg. siglum, i.e. the symbols for sources) can be constructed on any basis. In the present notes, letters of the alphabet and a collapsed version of the library shelfmark are both used. In the early stages of work the reference term or lemma will usually be that of an arbitrarily chosen base text. It may be helpful at a later stage, or for computer processing, to rewrite entries so that the lemma is the majority reading. In a published version the lemma will be the reading selected by the editor for the final edited text.

4. What are our questions?

The evidence is assembled in order for us to make two kinds of judgment. The first is whether certain sources group genetically (i.e. whether they all derive from a single ancestor with distinctive readings). If a particular grouping of sources is consistent over a number of variants the possibility arises that either these sources preserve readings of the exclusive common ancestor of the entire tradition that have been lost in the other sources; or, these sources preserve distinctive variant readings that have arisen in a common sub-ancestor during the course of transmission. It is important that neither of these possibilities ever be lost sight of. Our second kind of judgment concerns the direction of change, i.e. whether one member of a variation can be shown to be the origin of the other. Evidence of direction is often fragile and may be positively misleading. Experience and clear-headedness are vital to a persuasive assessment. As in all aspects of stemmatology, what the scholar offers is likely to be a case rather than a firm conclusion. The reasoning behind a difficult choice must always be explained to the reader.

5. How reliable is the evidence?

Some copyists freely reshaped texts to their own tastes or understandings; but even a faithful and observant scribe would often regard it as a duty to repair errors and to adjust minor aspects of expression, spelling and punctuation. Other changes arise from the psychology and technology of copying. These include anticipation, when a word not yet reached is written instead of the correct one, and perseveration, when a word already written is repeated instead of the correct one. Eyeskips between occurrences of the same word and repetitions of whole phrases can result from the same mechanisms. Scribes may unconsciously replace a word with a synonym or with a word with a similar sound but different meaning. Books on classical and mediaeval scribal practice categorize and explain these alterations. Systematically analyzing one’s own mistakes in writing or copytyping is another good guide.

It is vital to realize that much agreement and disagreement in variation is not genetically indicative and if used as the basis of reasoning will lead to wrong results. Methods of analysis that treat all variation as of equal value as evidence for genetic affinity and judge simply by quantitative measures of similarity and difference can hardly avoid going astray. The reason for this is simple enough. If every change introduced during copying was faithfully preserved in subsequent copies, all variants would be genetically indicative. But this does not happen. Changes may spontaneously reverse to their original. Errors will often be corrected or miscorrected by introducing a replacement reading that was not that of the original. The identical change may be made by different scribes working in different traditions, especially when they flow from changes taking place more broadly within the language. (And language never stands still!) There is a constant pressure to replace unusual or personal forms of expression (the lectio difficilior) with more familiar ones. Scribes may work from more than one copy mingling their readings (conflation) or reactivate a previously memorized reading.

6. Sorting out good evidence from bad

As a result of these distorting factors some groups of sources agreeing in a distinctive variant are not true groups, because the same variant has appeared, fortuitously or through conflation, in sources that do not all derive from the same common ancestor. Others will not be full groups, because a legitimate member of a genetically derived group has lost a defining variant at that point in the text, e.g. through a secondary variation, a reversal to the primary reading, or a lacuna. Other groups are neither full nor true. A variant list for a lengthy text is likely to contain examples of both kinds of deceptive agreement. The full and true group may never be observed but have to be reconstructed from imperfect occurrences. Other groups may turn out to have no genetic basis whatsoever (ghost groups).

Be careful in dealing with incomplete texts or those containing lacunae. Your listing of variants must make clear which sources are witnesses to each particular reading. Some editors, especially of verse, place excessive reliance on lacunae as a means of establishing the derivation of traditions, ignoring the fact that they are often unstable in transmission. The first difficulty is a point emphasised by Sir Walter Greg that if an error has occurred once there is a likelihood that the same error will be repeated independently. So, a predisposing cause in a text may lead more than one scribe to omit a word, a phrase, a line or a couplet. Secondly, if a lacuna harms the sense or metre of a text in an obvious way, it is likely that an attempt will be made at some stage to reconstruct what is missing, either speculatively or through consultation. Even when there is no obvious damage, a scribe’s memory of a previous reading or transcription may prompt alteration. Thirdly what appears to be a lacuna may in fact mark an interpolation, reversing the assumed direction of change. Lacunae are one kind of evidence to be placed beside others. No class of evidence should be either privileged or disregarded. When all agree, we are presumably on the right track. When they disagree there is either a mistake in our reasoning or the source being inspected has a complex rather than a straightforward genesis.

One of the most important skills of the textual scholar is that of sorting out the grain from the chaff among variants. It will soon be found that some categories of change are likely to be more stable in transmission than others, and some much less stable. What these are may vary between traditions but others are broadly predictable. As recently suggested, a gross and obvious error will almost certainly be repaired. It may be useful in defining a true group but is unlikely to remain uncorrected long enough to define a whole group. An easily made but unobtrusive change, such as a switch between synonyms, may survive in all the descendants of the source concerned (a whole group) but be joined in that group by other sources in which the same error has also been made. Other synonym pairs may reverse and re-reverse spontaneously. The most stable kind of change is that which should not strike the copyist as being obviously wrong but which is also unlikely to recur spontaneously in the course of copying.

In arranging the sources into groups the scholar will give most weight to what are judged to be the more reliable kinds of variants within the particular tradition and try not to lay weight on the less reliable kinds or on minor, indifferent variants. The role of this second class is to serve as a control on the first. While individually they may not be trustworthy, taken in sum they should conform to the broad profile determined from the analysis of the more reliable classes of variant. When they do not, the putatively reliable subset will need to be reevaluated.

7. How traditions evolve

Textual traditions are bio-social systems which will usually exhibit a similar kind of life cycle. Early radiation, while the text is topical and sought after, is often rapid and highly productive of variants. It may include episodes of memorial transmission. When a reasonably high population of copies has been created, there is a raised likelihood of comparison between copies taking place, either out of curiosity or in order to remedy perceived errors. Widely distributed texts that are regarded as particularly deserving of social respect usually settle into a ‘vulgate’ version protected from major divergence by communal memory. Really long-lived traditions, such as those of ancient literary and religious texts, will usually be conflated to a degree that makes the construction of a ‘family tree’ of copyings impossible. In this case editors either revert to assessing the originality of readings on a one-by-one basis or resort to the sledgehammer of statistical analysis.

In the case of entrenched vulgate versions, the identification of original readings can only be done on the basis of historical and philological insight. Older critics of Greek and Roman literature prided themselves on their ability to recover readings that had been wholly lost in transmission (divinatio). The student of early-modern texts will sometimes need to make similar judgments; however, early-modern-traditions normally belong to the first of the stages, that of rapid radiation. Conflation occurs but is rarely so radical as to defy analysis. Under these circumstances, and given a rich enough body of variants, it is often possible to undertake the construction of a traditional stemma. This may not be complete, because evidence tends to fade as we approach the exclusive common ancestor of the tradition, but even an incomplete stemma will assist materially in the identification of non-original readings and in narrowing the field over which the student has to rely on personal judgement.

8. Identification of groups

In identifying groups we are only interested in those containing less than half the number of sources. The rest comprise the undifferentiated pack. As explained earlier, a consistent minority group preserves either the reading of the exclusive common ancestor of the whole tradition or that of a subancestor within the tradition. Do not assume that a small group, or even a single source, may not be the one that preserves the reading of the ancestor of the whole tradition. A group is potentially identified whenever a number of sources, being less than half of the overall body, agree consistently and meaningfully in readings not found in other sources.

In investigating potential groups it will always be helpful to reformat the list, using a simple macro, so that agreements are placed in alphabetical order of their component sources. E.g.

  • Ab12  11 feel] fell
  • Ab12  32 pour] power
  • Ab12  60 ] The] Thy
  • Ab12, Ca39, DA, HA-uncorr, HA-corr, Is45, Lb54, Np40, SKv69 52 example in] Example led
  • Ab12, Ca39, HA, Lb54, Is45, SKv69           30 devils] Mischeifs
  • Ab12, DA, HA, Is45, Lb54, Np40, SKv69  31 those] these
  • Ab12, DA, Lb54, Np40, SKv69       17 the wide world] the world had
  • Ab12, DA, Lb54, Np40, SKv69       18 or’t] . . . else it
  • Ab12, DA, Np40, SKv69      14 And] When
  • Ab12, HA-uncorr?     6 noble] nobles
  • Ab12, He36, Orp73   26 E’en] Even
  • Ab12, Lb54, Np40, SKv69   36 Are snatched by sudden blasts] By Sudden blasts are Snatch’d
  • Ab12, Lb54, Np40, SKv69   58 does] do
  • Ab12, Lb54, Np40, SKv69   58 scorn] hearts
  • Ab12, Lb54, Np40, SKv69   47 soft] own
  • Ab12, Lb54, Np40, SKv69   59 their hearts] themselves
  • Ca39   44 Or] And
  • Ca39   20 battered] battering
  • Ca39   45 pale] cold [and so on]

Note the recurrence of Ab12, Lb54, Np40, SKv69 as a consistent group both in its own right and as an element in larger groups. Begin analysis by examining all the consistent pairs and assessing them for reliability (some may well be fortuitous). Then move up to threes and larger groups until a number is reached that is one less than half the total number of sources. Reliably attested groups should also occur as sub-elements within larger groups and as part of the ‘pack’. Contradictions and conflicts will invariably emerge and will need to be mediated. Often this takes the form of a renegotiation of the respective claims of higher-level and lower-level agreements.

At this stage of our investigation we make no judgments about the direction of change. Our diagrams are laid out along a horizontal axis to indicate this. It is only at a later stage, or when we encounter unresolvable difficulties, that we need to ask which variant was the origin of the other. At this point the information is rearranged vertically in the conventional tree form.

9. Intermediation

The formal logic underlying the analysis of groupings is rigorously presented in Sir Walter Greg’s The Calculus of Variants (1927). There is much to be learned by carefully working through this short but demanding book and mastering its terminology, which however is only sporadically used in these notes.
The principles of distributive logic are very simple ones. To begin with sources are either terminal or intermediary. In the arrangement below A and D are terminal and B and C are intermediary.


A terminal text can be identified by its possessing variants not shared by any other (like Ab12 in the list above). If new readings were introduced by each copying, those of intermediary texts would generally be carried over into the copy. Thus the possession of unique readings indicates that the tradition contains no copy of that particular source. There may be practical circumstances under which this rule does not strictly hold but it is a sound general guide. If a copy was so faithfully made that it introduced no verbal changes whatsoever, it would be a bibliographical duplicate and could be disregarded. Whenever our analysis shows clearly that a source is wholly derived from another surviving source we dismiss it from further consideration.

10. Turning groups into trees

A reliably tested pair, both of whose members are terminal texts, implies the existence of a lost intermediary parent, which we insert into our diagram using a siglum that indicates its hypothetical status. Using this principle we can attempt to build a non-directional diagram, moving up from pairs to larger groupings and in the most favourable circumstances to the whole body of sources. However, evidence that seems to point towards a particular inferred intermediary may well be subject to other interpretations. Let us assume that three terminal texts group consistently A:BC. A little thought will show that this distribution could be the consequence of one stemma with no intermediary (i.e. with A, B and C each independently descended from their common ancestor) or three different ones with one intermediary (i.e. with a subancestor for either AB, AC or BC). With four terminal texts dividing AB:CD there would be no solution with no intermediary, two possible solutions admitting one intermediary (i.e. with a subancestor for either AB or CD), and a further three with two intermediaries, and so on as the number of texts increases. The key point, once again, is that agreement may result either from independent preservation of the reading of the ancestor of the whole tradition (or indeed from any higher level subancestor) or from exclusive common descent from an immediate subancestor. A choice among these possibilities cannot be made by formal means alone. Instead judgment must be reached via some other avenue, usually one reliant on external evidence or a qualitative assessment that one reading is likely to be original and the other derived. This introduces an unavoidable element of circularity into textual reasoning, but one that in a tradition of any complexity will have its inbuilt checks and balances, as provisional resolvings of conflict prove to be either productive or unproductive when analysis proceeds further.

11. Intermediation spectra

Sources can now often be arranged along an intermediation spectrum, similar to that given above for A, B, C and D, except that there would also have to be space on the line for the hypothetical subancestors of those texts which branch from the main axis. A record should be made of the variant readings which define each boundary along the spectrum. The same information can also be presented in the form of Venn diagrams. In that case the key defining variants should be noted on either side of the boundaries of each internal element of the diagram. However, as long as our stemma remains non-directional, and therefore without a designated point of origin, the issue of derivation is deferred.

12. Dealing with conflation

In any real-life tradition the student is likely to encounter the phenomenon of a source wanting to be in two groups at once. The usual cause of this is conflation. Conflation arises when a source incorporates the readings of two or more different exemplars either in combination or sequentially. Sometimes it is prompted by an exemplar in which the variant readings of a second source have been written in the margins. Early modern scribes might also be textual critics! It frequently happens in such copying that the more important and prominent readings correspond to one text (A) while the bulk of relatively minor variants are faithful to another (B). In this case we are dealing with a copy of B that was selectively corrected from A. Or a scribe may begin by incorporating copious variant readings from A and subsequently pay them less attention. In these two cases it is relatively easy to divide the source into two textual strata located at two different points of the diagram; however, a source that is the product of more than one conflationary episode may well remain inscrutable. In some cases there may be genuine doubt over which is the conflated and which the unconflated source. Conflation in which a copy is begun from one exemplar and completed from another is much easier to resolve. For a method of assessing more involved cases see Colin Flight, ‘Stemmatic Theory and the Analysis of Complicated Traditions’, Manuscripta 36 (1992), 37–52.

13. Determining direction

Once we have enough evidence to produce a convincing non-directional stemma for our tradition, we need to consider the evidence for the direction of change. In practice, as mentioned, we may well have to do this much earlier in order to overcome local ambiguities and contradictions but it is helpful if this movement from horizontal to vertical orientation is delayed as long as possible. The exclusive common ancestor of the whole tradition can either be one of the surviving texts or a hypothetical intermediary. This intermediary can in turn stand at any hypothesized node or on any line connecting nodes within the diagram.

A variety of criteria can be applied to determine its location. The most important are those which allow us to say that one reading has occurred as a consequence of an another, and may be philological, codicological, palaeographical or contextual. No part of textual reasoning is more dependent on knowledge of the habits of particular scribes and the idiosyncrasies that characterize particular genres and periods of writing. What persuades in one tradition need not do so in another. Chronology has to be invoked with caution. An old scholarly principle says recentiores non deteriores, or to put it another way, a manuscript written in 1580 need not be any closer to the origin of the tradition than a manuscript faithfully copied in 1780 from a lost manuscript written in 1560. A source is only as good as its exemplar.

In situations (the majority) where neither the content of groups nor the direction of change can be determined with precision, the scholar must use experience and informed judgment to choose a solution, but always making it clear to the reader where and how personal judgment has been exercised. It is an editor’s duty to supply a reading text. Disputable readings in that text should be discussed in the explanatory notes to the edition. The text as such is never anything more than a frozen version of an editorial argument or assemblage of arguments.

14. Use of bioinformational algorithms

Scholars hoping for simpler and quicker solutions have sometimes made use of algorithms developed by bioinformatics researchers for analysis of the historical descent of genetic mutations. There can be no objection to this provided that the textual critic is fully aware of the forms of logic employed by the algorithm and their likely limitations, which are often severe. A 2003 paper by biologists Charles H. Bennett, Ming Li and Bin Ma using a sophisticated ‘relatedness’ algorithm to analyze the derivation of a collection of chain letters is well worth reading for the insights it generates from looking at old problems in a new way (‘Chain letters and evolutionary histories’, Scientific American, June 2003, 64–9). The authors appreciate that the point of origin of a tree can not be located through distributional reasoning alone and show ingenuity in their qualitative judgements, though not all these would have been valid for more complex verbal traditions. The weaknesses of phylogenic analyses when applied to literary traditions generally include the following. In the first place, all judgements of genetic contiguity are purely numerical. (Bennett et al. assess it by the extent to which two texts can be shrunk by a file compression program!) Secondly, all differences between texts of the work are treated as possessing an equal degree of genetic indicativeness; nor is it acknowledged that some are bound to be counterindicative. Thirdly, a good deal of information that literary editors would regard as significant may simply be discarded as too difficult to encode. Fourthly, all branches in the tree have to be binary whereas in literary copying it is common for a source to leave multiple descendants. Finally, the frequent problem of what to do when a source wants to be at more than one point in a Lachmannian stemma never arises because it is simply removed on a headcount or probabilistic basis by the software. Far from wishing to erase conflict, literary editors will see their most important task as one of locating it and then resolving it by the exercise of judgement. Formal analysis of agreements is only one part of their responsibilities: the other, no less important, is the attempt to discover the reasons why particular changes came to be made, which involves an assessment of human behaviour and motivation. Bioinformational analysis may well be of value in arriving at a range of preliminary hypotheses, but these can not be relied on to contain the best solution.

15. Conclusion

Stemma construction used to be presented as a mechanical, rule-governed process, but nothing could be further from the truth. Neither is it in any meaningful sense mathematical. Its few, relatively simple operational principles – intermediation and the linking up of sources via hypothetical intermediaries – are rather, as Greg pointed out in 1927, a department of formal logic. There is nothing wrong with those principles: when applied to a full and genetically indicative data set they will produce persuasive results. There is rarely any difficulty in establishing the relationship of printed sources! The problem lies in the data, or at least that part of it that is supplied by textual variation. Because it is often non-indicative or counter-indicative, it will always be subject to assessment and interpretation. Moreover, often it is absent exactly where it would be helpful. Nothing dictates that a scribe will make an error where an indication of grouping or direction is most sorely needed. The analyst’s progress will usually be abductive rather than inductive or deductive. What is presented is an argument that has to be assessed like any other argument and is always open to being superseded by a better one. The most valuable skill may well be that of imagining possible explanations. That skill is acquired from experience of the way texts and words behave in transmission. One learns how to determine the genetic relationships of sources by continually performing the kinds of analysis that have just been described. For that to happen in the course of a busy scholarly life presupposes an active curiosity about the ways in which written texts behave as members of bio-social systems, moving forward adaptively through time in the turbulent medium of language. The more one looks into such questions carefully and yet imaginatively the better one will get at answering them.