University of Virginia Library

TEXTS WHICH ARE PRESERVED IN MORE THAN one form are objects of study.

I. One legitimate aim of such study is the formulation of a consistent way of dealing with variants, and so to make an orderly selection among them, in such a fashion that another student to whom the method has been explained, would be able to make reasonable predictions about which alternatives would be selected by the editor in given situations. This type of study can be called scientific textual study, and is separate from such types of textual study as have for their purpose the improvement or modernization of texts.

II. The study of texts can be pursued along more than one line, or by various combinations of lines. There are four principal approaches.

1. The arrangement of differences in the surviving texts can be studied, without attention to the nature of the differences. This constitutes distributional study.

2. Individual differences can be compared with each other in the hope of discovering the direction of change. This constitutes genealogical study.

3. The physical characteristics of the various versions can be studied in the hope of throwing light on the process of copying. This constitutes external study.


64

Page 64

4. Differences can be compared for the purpose of selecting those which are stylistically preferable, offer the best sense, are free of mistakes of grammar or fact. This constitutes literary study.

III. Each of the above methods of study rests on certain basic assumptions.

1. Distributional study rests on the assumption that it is more likely that unity of manuscripts represents unity in their source than that it represents coincidence. For convenience this assumption will hereafter be called the postulate of unity.

2. Genealogical study rests on the assumption that certain differences are more likely to have been produced by a given direction of change than by another, and the second assumption that differences of this type can be identified by examination.

3. External study rests on the assumption that human or other activity involving physical objects leaves physical effects by which the activity can be traced.

4. Literary study rests on the assumption that an author is less likely to make mistakes than is a copyist, and conversely that an author is more likely to produce a positive excellence than is a copyist.

IV. These assumptions are not all equally probable, or equally necessary.

1. The postulate of unity can not be evaluated in terms of controlled probabilities. Further, coincidence does indeed occur, both in manuscript transmission and daily life. The postulate is therefore an axiom, which like other axioms can be tested only by the partially effective device of reversal. That is, we can make the reverse assumption that unity is more likely to reflect coincidence than unity of source. It is immediately evident that the result of such a reverse assumption would be to group together the most dissimilar manuscripts, rather than the most similar. In actual fact, moreover, we act on the postulate of unity in our daily lives. When two students turn in identical examinations, we do not assume that likeness is the result of coincidence.

2. The genealogical assumptions given above would be reasonably easy to establish empirically if no scribe ever deliberately


65

Page 65
improved his text, but unfortunately the order of correction can reverse that of mechanical change. The assumptions, more generally stated, are contained in those underlying external study, and differ from those of external study only in that they are applied to language material, where those of external study are applied to physical (non-language) characteristics.

3. The assumption underlying external study is the most far reaching of those here given, and is actually the assumption underlying the investigation of all past activity. We act on it so constantly that we scarcely realize it is an assumption, capable of being tested only by reversal—namely by assuming that activity never leaves observable traces.

4. The assumption on which literary study rests is the least likely and the least necessary of the group. It is, however, often the basic assumption of an editor, either explicitly or implicitly held. It has been used as a reason for rejecting the results of bibliographical study by no less a scholar than the late Professor Kittredge.[1] The assumption would be reasonable were authors different from other men (more particularly scribes) in being free from mistakes and stupidities. Unfortunately they are not, as we know if we remember Keats' "stout Cortez."[2] The assumption is not even acceptable if, as with such recent manuscript studies as those of Hitchcock and Wolf, it is made in the more limited form of a means of separating significant variants (mistakes) from nonsignificant variants.[3] Essentially, such an assumption leads to


66

Page 66
circularity. Variants are first judged as right or wrong. Rightness and wrongness is then used as a means of establishing authority, which in turn is used to establish rightness or wrongness. In other words, if it is possible to judge readings as right or wrong independent of authority, then no attention need be paid to authority.[4]

An even more important objection to the literary method is that as yet there is no convergence among critics as to what constitutes literary excellence. It is perhaps to be hoped that such convergence may emerge, but until it does, literary excellence can hardly be a tool in scientific investigation.

I wish to be understood—I am not depreciating literary study, or literary criticism. Both have their own proper sphere, and both are legitimate forms of endeavour. I am merely saying that if the student of texts takes as his aim the orderly and methodical presentation of variants, he does well to ignore literary considerations.

Note. In ignoring literary considerations I am following the practice of the most important of all general considerations of manuscript study, a work to which I owe a heavy debt throughout, The Calculus of Variants, W. W. Greg, Oxford: Clarendon Press, 1927.

V. After the elimination of literary study, the remaining methods of study can be combined in various ways.

1. Study of texts surviving in manuscript usually makes use of distributional and genealogical evidence. This method can be called, for lack of a better name, comparative study.

Note. I am here departing from traditional terminology as used by Greg and other students. The traditional term "genealogical" corresponds to my term "comparative," while "distributional" is not used.

2. Study of printed texts often makes no use of distributional


67

Page 67
evidence, since for reasons which will be supplied, such evidence is inapplicable. On the other hand, modern bibliography makes great and brilliant use of external evidence. Any method which makes use of external evidence, whether or not genealogical evidence is added to it, can be called bibliographical.

3. Some term which would include all three methods, or any combination of them, is obviously needed. This is particularly so, since bibliographers are apt to think of their discipline as different from the study of books in manuscript, and are also apt to complain of the term bibliography because it is likely to be mistaken for the preparation of hand-lists. The matter is perhaps unimportant, but I should suggest biblistics, or if this has too religious a sound, libristics. Both terms are on the analogy of such others as ballistics, linguistics.

VI. It has been stated that the various methods of study can be combined to give complex methods. The various methods can not be combined in a random and inconsistent order, jumping from one to the other at will, unless the results are also to be random and inconsistent. It is then obvious that the student must know the advantages and limitations of each method, and be able to recognize which one he is using. Otherwise he will fail to exhaust a given method systematically before he passes on to another.

VII. In all cases where all three types of evidence apply, the normally best order of application is distributional evidence first, then genealogical evidence, and external evidence last of all.

1. Distribution of variants is an objective and readily verifiable set of facts. Further, if the student is lucky, it can lead to a complete and satisfactory conclusion, needing only checking from other types of evidence.

2. Geneological evidence is less trustworthy than distributional evidence, since as has been said, the direction of change is different in processes of correction from what it is in mechanical copying. Geneological evidence needs, therefore, to be used in accord with conclusions drawn from previously studied distributional evidence. An example of a reading from Philaster may make this statement clear. In Act III scene 1, line 201, appears a set of


68

Page 68
variants, parallesse and parallelless (there is a third variant which need not concern us). Arguing genealogically, one would suppose that the form which drops out one of the two identical series of letters is the younger variant. Yet if we check this genealogical conclusion by the distributional evidence it becomes untenable. The fuller form occurs only in a late and unauthoritative edition, and is therefore a correction which reverses the order of mechanical change. The mechanical change, in other words, occurred somewhere in the limbo of versions of the text which are beyond our reach.[5]

3. External evidence might be presumed to be better studied first, and indeed, for reasons of convenience, it is often at least presented in that position. However there are reasons why at least some of the results of external evidence can be disregarded in manuscript work, if comparative evidence applies.[6] It should not be thought that the results of external evidence can be safely disregarded in their entirety, but since some of them can be thus disregarded, there is a danger of confusing the comparative study unless it is exhausted before an external approach is used.

VIII. The types of evidence are not equally applicable to every situation. The frequency of applicability is the reverse of the proposed order of study.

1. Since all texts must be preserved in some form, external evidence always applies. It is worth noting that the mere statement that a text is preserved only in memory is a piece of external evidence quite worth having, since it leads to a conclusion about the reliability of the text.

2. Geneological evidence can be applied at any point where a text survives in differing form, but not at any point where a text survives in unity, no matter whether this unity is produced by the fact that there is only one manuscript, or because there is unity among multiple manuscripts. To apply genealogical methods at a point of unity becomes the same thing as the improvement of texts.

Note. The application of genealogical methods in the face of unity of the

69

Page 69
manuscripts is inherent in the "common mistake" method which is the classical Lachmannian technique. Thus E. K. Rand, "Dom Quentin's Memoir on the Text of the Vulgate," Harvard Theological Review, XVII (1924), 209, "The first thing for which I look in the apparatus is the presence of errors common to all the manuscripts, and I find . . . that there are no common errors. Our 21 manuscripts, then, do not descend from one faulty archetype." In other words, an editor in Rand's view, is perfectly justified in disregarding all manuscript evidence if he can find fault with it. Rand would almost have been forced to emend "stout Cortez" in Keats' sonnet.

3. Distributional study can be used only in texts which survive in more than one form. It is often assumed that distributional evidence is valuable only if there are more than two manuscripts, but this assumption is not strictly true since when there are only two manuscripts, the editor must at least record the occurrence of unity and difference, and accept the unities.[7]

Distributional studies are powerless to deal with differences unless there are at least three surviving manuscripts, since in the two-text situation no variant has a better chance of having been in the original than another.

Distributional studies, further, are not applicable to all texts which survive in multiple form. Before distributional study can be applied, the student must make the assumption that his text goes back to a single original, and that each version of it has been produced by a unified process of copying. If these are not reasonable assumptions, the method can not be applied. For instance we now know that the variant forms of differing exemplars of the same edition of an Elizabethan book are not the product of unified acts of copying, but are a nearly random arrangement of corrected and uncorrected press-variants. A student who attempted to draw a tree for such exemplars would be in a hopeless situation.

Distributional studies become valueless when any type of evidence shows that the variant forms are the result of material revisions by the author or another. They remain, however, valuable up to the point at which such a conclusion is proved, and often distributional evidence at least contributes to such a conclusion. Thus the Piers Plowman controversy arose because the distributional evidence had not been exhausted, so that it became possible to use genealogical evidence both to affirm and deny


70

Page 70
multiple revision. An example of the effect of persistence in a distributional approach can be constructed with the aid of Meredith's Evan Harrington, known to have been revised by the author. A passage in chapter eleven of that novel appears thus in the early form:
Eyes were on him, he could feel. It appeared that the company awaited his proceedings; . . .
And thus in the revised form:
Eyes were on him. This had ever the effect of causing him to swell to monstrous proportions in the histrionic line.
Were the fact of revision unknown, and were an editor sufficiently unwary, he might produce the following version, on the assumption that anything absent from either version was an omission from the original of both.
Eyes were on him, he could feel. This had ever the effect of causing him to swell to monstrous proportions in the histrionic line. It appeared that the company awaited his proceedings; . . .
In such a conflated passage, every word is genuine in the sense that it was written by Meredith, yet the totality of the passage is false.[8]
Note. J. Burke Severs presents in chapter three of The Literary Relationships of Chaucer's Clerkes Tale (Yale Studies in English, no. 96, 1942) a clear example of this principle. His study is of the manuscripts of Petrarch's Latin version of the story of Griselda. He begins by grouping certain manuscripts which lack material found in all the others. This group he sets up as family d. He then goes on to set up a complete, but tentative tree not only for d, but for all other families. All this he does on distributional evidence. He then turns to a genealogical study of the material lacking in d. It is first of all striking that the text reads equally well with or without the additional material. He next turns to a further distributional fact, and its genealogical interpretation. The fact is that all the manuscripts which belong to d lack an accompanying letter of introduction which tells how the first version of the letter had gone astray, and that Petrarch had therefore found it necessary to rewrite it. From all this evidence he then rightly concludes that the longer manuscripts represent an author's revision, and that it is therefore necessary to draw two trees, one for d, and one for all other manuscripts. Such orderly procedure seems a model of clarity.


71

Page 71

IX. If it can be assumed that the distributional approach is valid, the student makes the following operational assumptions in applying it.

1. Each manuscript before him is the result of copying from one exemplar, though this exemplar is not assumed to be the same for all manuscripts. Mixing of exemplars is assumed not to have taken place unless the variants can be explained in no other way.

2. All manuscripts before him spring from some one ultimate exemplar, present or hypothetical, by one or more steps of derivation. This also will be assumed to be true unless proved otherwise.

X. In collecting distributional evidence the student handles the following types of phenomena:

1. Unity. Whenever the same language symbol occurs at the same place in two or more manuscripts, this occurrence can be called a unity. The term "at the same place" is taken to mean "at the same position in a series of surrounding language symbols, whether or not the surrounding symbols constitute a unity."

Unfortunately "language symbol" is not a transparently clear term, since such symbols are arranged in a hierarchical series of entities, which for our purposes can be defined as letters, words, sentences, and still larger units such as verses or paragraphs. Thus the occurrence of honor in two manuscripts, once with a u, and once without, is a word unity, though not a letter unity. Usually editors find themselves working on the word level, since most of the time the evidence from letters can not outweigh the evidence from words, and since the evidence from words will fully describe the evidence from larger units. It should be understood that I am using a definition of word which would be unsatisfactory in other contexts: namely that a word is any group of letters with a space before and after them, or habitually so written in forms later than the manuscript under discussion if the manuscript does not divide words.

2. Variation. The occurrence at the same place in two or more manuscripts, of differing language symbols, of zero and a language symbol, or of something not a language symbol and a language symbol, can be defined as variation or difference. The sum total of things occurring is a set of variants, and each thing which occurs is a variant.


72

Page 72

XI. Both variation and unity, for purposes of comparative study, must be supposed to contain occurrences of language symbols. Examples of either unity or variation which do not contain language symbols belong rather to external than to comparative study. Things which are not considered to be language symbols can be exemplified by blots, tears, ornaments, etc.

XII. Sets of variants are of differing sorts.

1. Positive sets. These are sets of variants consisting of different language symbols.

2. Variants consisting of a language symbol opposed to its absence. These are variants which include material which may be later defined as being either omissions, or additions. It is, however, important to avoid using the terms addition and omission until a conclusion has been reached. If the student is not using literary criteria, whether given variants are additions, omissions, or must remain undefined, can only be settled by all the distributional and other evidence. To use such a term as addition before that point, is to prejudge the evidence. A handy term which avoids the difficulty is "add-omission."

Note. The general clarity of Severs' exposition of the manuscripts of Petrarch (loc. cit.) is perhaps slightly marred by the fact that he sets up his family d on the basis of "omissions." At worst the difficulty is only one of presentation, since at the end of his chapter he clears up the nature of this omitted material fully. But in reading the chapter for the first time, one is left in doubt as to the validity of his procedure, since if the missing material were genuine, omission would prove that d was a late sub-family with an exclusive common ancestor, whereas if the material had been added later, the exclusive common ancestor would belong to the other manuscripts, not to d, and the common ancestor of d would be the original of all manuscripts. Such a term as that I have proposed would avoid this difficulty.

3. Sets which contain one member which is not a language symbol. These sets can be of significance only if the variants resemble each other physically or otherwise. Thus if one manuscript shows a small stain, and the other a comma, or if one shows an illustration, and the other substitutes a description of the illustration's contents, the conditions are fulfilled.

XIII. The following phenomena do not constitute either


73

Page 73
unities or sets of variants for comparative study, though some may be significant for external study.

1. The occurrence of zero in two or more manuscripts. Obvious as the statement may be, it is impossible to argue that two manuscripts are different because one leaves out that, and the other leaves out which. It is likewise impossible to argue, as emenders do, that all manuscripts of a text are descended from a single corrupt ancestor, differing from an unrecoverable original, because all manuscripts leave out something.

2. The occurrence of non-language objects in two or more manuscripts. Thus the occurrence of the same ornament does not constitute a unity for comparative purposes (however significant externally), nor does the occurrence of two blots.

3. The occurrence of a non-language object at the same place with a language symbol to which it bears no resemblance. For instance, a scribble may occur in one manuscript where the other has a line of verse. Such phenomena become simply add-omissions.

XIV. The first step in actual study is the tabulation of all sets of variants, disregarding unities. Unities can merely reflect the single original of all manuscripts, and so are worthless for showing differences of derivation, no matter how much the student is forced to rely on them in his final critical text.

Note. Elaborate directions for the tabulation of variants are given in the writings of Dom Quentin and the various treatises derived from his work. See particularly, Henri Quentin, Memoire sur l'Etablissement du Texte de la Vulgate, (Collectanea Biblica Latina VI; Rome and Paris: Desclée and Gabalda), 1922, page 227, Quentin, Essais de Critique Textuelle (Paris: Picard, 1926), pp. 63-72. The presentation of such statistical tables has been strongly attacked, but seems to me one of the great virtues of Quentin's method.

XV. When the tabulation of variants is complete, the next step is an attempt to assemble groups of manuscripts. The set of concepts used here are as follows:

1. Derivation. I have refrained from using the word copying, since the two things are different. The distributional student is forced to assume a working hypothesis of copying whenever the arrangement of variants is such as to be compatible with it. Actual copying, on the other hand, is a physical fact, provable


74

Page 74
only by external evidence. Suppose that our distributional student has this situation: manuscript A shares readings with all individual manuscripts, and has no unique readings. Manuscript B shares readings only with A, and has some unique readings. Such a state of affairs is not compatible with supposing that A is copied from B, since then the shared readings of A are left unexplained. The supposition that B is copied from A is not contradicted by this arrangement of variants, and the student will therefore adopt it. It is, however, merely a hypothesis adopted for the sake of describing the relationship of variants, and is in no case proof that actual copying occurred, since a long series of lost manuscripts may intervene. Further, if a new series of manuscripts is discovered whose variants change the relationship of the whole group, the student will be forced to change his hypothesis. Thus in all such situations it is advisable to speak of derivation—defined as a state of affairs compatible with a hypothesis of copying—rather than to speak of copying. The terminology is more important than it seems, since if copying is the term used, there is grave danger of misunderstanding the nature of the diagram in which the comparative student states his results.

Another way of stating the operational hypothesis or procedure here described, is to say that the student begins by supposing, in turn, that each manuscript is derivable from each other, and rejects that supposition each time that he finds evidence to contradict it.

2. Absolute unity. Whenever two manuscripts are absolutely alike in the language symbols contained, they can be treated as one, and it makes no difference which one is chosen to exemplify the pair. This obvious statement is necessary to justify the practise of modern students who often work with photostatic copies rather than originals. The practise is interesting, since to say that an original mediaeval manuscript is copied from a modern photostat is nonsense, but to say that its readings are derivable from those of the photostat is acceptable enough. Once again, derivation and copying are different things.

3. Families, sub-families, lines of descent. Every group of manuscripts which share a group of variants in opposition to those found in others will be spoken of as a family, derivable from some one manuscript, extant or hypothetical, spoken of as an


75

Page 75
ancestor. A sub-family is a family whose ancestor is derivable from a manuscript which is the ancestor of still other manuscripts. A line of descent is a family, or family and sub-family, which share no ancestor with other families except the original from which all manuscripts are derived.

4. Hypothetical manuscript. A hypothetical manuscript is one which must be posited to explain similarities between two or more other manuscripts, extant or hypothetical. The explanation of such unities is the only reason for which a distributional or comparative student can posit such a manuscript. All other reasons are outside the reach of comparative evidence.

Note. This terminology differs slightly from that employed by Greg, op. cit., pp. 2-3. Greg speaks of inferential manuscripts defined as are my hypothetical manuscripts, and then of others whose existence is suggested by non-comparative evidence, which he calls potential manuscripts. The essential difference is that I am denying that the comparative student can ever take account of potential manuscripts, so that Greg's distinction is unnecessary in my approach.

5. Original. This is the manuscript, hypothetical or extant, and presumed to be single unless proved otherwise, from which all extant and hypothetical manuscripts are derived. It is important to understand its nature. Since all that we know of it is ultimately derived from extant manuscripts, the original can have no characteristics not found in some at least of the extant manuscripts. The original, therefore, must be presumed to be the last version of the text produced before copying of the extant manuscripts and their ancestors began. Any number of lost versions may intervene between it and the author's original, which is forever out of our reach.

Note. This view of the original has been held by many of the ablest students of texts, e.g. Greg, loc. cit., and John M. Manly and Edith Rickert, The Text of the Canterbury Tales (University of Chicago Press, 1940), II, pp. 40-41. The view is also in accord with the view of the nature of starred ancestral forms held by modern linguists in the closely similar activity of linguistic reconstruction. For all that, the view is not universally held, necessary as it would seem to be. Note the explicit statement of Wolf, loc. cit., that the aim of his study was to recover the author's original. What I have called the original is what is called the archetype in Lachmannian textual criticism. I am, of course, denying that the "original" in the Lachmann sense is recoverable.


76

Page 76

Well known as such statements about the nature of the original may be, they are still necessary since they close the door to much speculative tampering with texts. If the author's original is out of our reach, how much further out of our reach the author's intention—what he meant to write, but did not—must be! Yet, for instance, modern editors still defend Theobald's famous "'a babbled of green fields" on the ground that this is what Shakespeare must have meant to write.[9] I am not, of course, denying that Theobald's words are infinitely superior to those of the old text, nor that I should experience a sense of loss in a reading text which did not adopt the emendation. Yet it remains true that the emendation is without authority.

XVI. Distributional evidence can establish the fact of derivation from a common ancestor. When, however, a single group of manuscripts is being considered in relation to the rest, agreement merely offers proof of some common ancestor, and does not define the relationship of that ancestor to manuscripts outside the group showing agreement. Thus if A and B agree 99 times against the rest, A and B have a common ancestor, but this ancestor may be derived from one of the other manuscripts, hypothetical or extant, or may even be the ancestor of all texts. Only the sum total of distributional evidence for all families offers any conclusion, or possibility of conclusion, about the exact nature of derivational relationships.

Note. It is exactly this sort of situation which faced Severs (loc. cit.), in his development of family d. The eight manuscripts which he set up as family d agreed 13 times against all others. All but two of these agreements consisted of add-omissions absent in d. Not until all the distributional evidence had been surveyed, and then only with considerable appeal to genealogical evidence, could Severs conclude that in effect the d manuscripts were descended from O, and that all others were descended from a later revision.


77

Page 77

XVII. Normally some distributional evidence will be contradictory. Suppose that manuscript A shares 99 readings with B and C, which form a group that has previously been defined as exclusive. There are 4 readings which A shares with D and E, also previously defined as exclusive. For the neatness of manuscript diagrams, it would be convenient if such difficulties did not arise, but all manuscript experience leads the student to resign himself to them.

XVIII. The postulate of unity was given in the form "it is more likely that unity among manuscripts reflects unity of source than that it reflects coincidence." The form of the postulate carefully avoids denial that coincidence—whether caused by convergence of change or by contamination—can and does occur. Yet if coincidence is less likely, it ought to occur less often, even though there is no accurate way of measuring what the proportions of occurrence should be. All that can be said is that the greater the disparity in frequency, the more likely it is that the smaller group represents coincidence. The proportion of 4 to 99 would probably trouble no wise editor. The closer the frequency comes to equality, then, the more suspicious the conclusion becomes. As a result of the fact that frequency of agreement and disagreement is a check on conclusions as to which readings reflect unity and which reflect coincidence, it becomes important for a student always to present a statement of the total distributional evidence in statistical form.

XIX. Distributional facts settle some questions about the nature of derivation. Thus if in a group of more than two manuscripts having a common ancestor, all manuscripts have unique readings, and all manuscripts have shared readings in no constant arrangement, these facts are compatible only with derivation of the type known as radiation, in which all manuscripts are derived directly from the ancestor with no intervening sub-ancestors.[10] On the other hand, if the relationship is constant, so that the shared readings of A are always shared only with B, the facts can be taken to be incompatible with radiation.


78

Page 78
Note. For our purposes, it makes little difference whether the student works directly with an arrangement of manuscripts thought of as proving or disproving a given kind of derivation, or if he works with the Quentin concept of "intermediaries." The Quentin term is a useful tool, but eventually arrives at essentially the same result as is stated in this paragraph. After the intermediaries have been studied and interpreted in terms of derivation, it is again shown that distributional evidence can prove the existence or non-existence of radiation, but can not settle the exact type of derivation if radiation is non-existent. The Quentin term "intermediary" can be briefly defined as a manuscript from a group of three related manuscripts chosen for comparison, of such a character that the other two manuscripts do not agree against it. Thus if manuscripts A, B, and C constitute a family, Quentin would study the agreements and disagreements among them. If A and B agree in some readings against C, and B and C agree in some readings against A, but A and C never agree against B, B is then defined as an intermediary. But the statement that it is an intermediary is not the same thing as saying that it is the ancestor, since it may be the ancestor of the other two, may be the middle term in a straight line derivation, or may be in still other relations to A and C. The concept can, however, be used to establish radiation. Suppose that there are four manuscripts, A, B, C, and D. If A is intermediary in every combination of three, then all manuscripts are in radiation from A. If the relationship is not radiation, then the most that determination of the intermediaries can do is to place some limitation on the possible relationships. Thus if the family of three is in a straight line derivation and B is the intermediary, as above, either A or C can be the original of the other two, but B can not. Much the same sort of limitation, however, is arrived at by other means of studying distribution, as will be shown later. The wise student will, then, use whatever means of comparative study offers him the best possibility of reaching conclusions. The Quentin method is developed in the works by him already cited, but is perhaps most clearly explained in Severs, "Quentin's Theory of Textual Criticism," English Institute Annual 1941 (Columbia University Press, 1942), pp. 65-93.

XX. Distributional facts do not settle all questions of the nature of derivation. As stated above, distributional facts can establish radiation or the absence of it, but if radiation is absent, distribution can not establish the nature of derivation more narrowly. That is to say, distributional facts make the presence of sub-families evident, but will not demonstrate which groups are families and which sub-families, unless in turn the ancestral manuscripts can be shown to be derived by radiation.

An example may make the statement that distributional facts


79

Page 79
can not offer a complete solution unless the relationship of groups is radiational, somewhat clearer. Suppose that A and B agree 99 times against C and D where C and D disagree with each other. C and D agree against A and B 101 times where A and B disagree with each other. These facts can be taken as incompatible with radiation. That is to say, A and B have an ancestor which we can call x, and C and D have an ancestor which we can call y. The possibility that x and y are descended by radiation from the original, O, can be disregarded as uneconomical (as I will show), but two other possibilities remain, namely that x is derived from y, and that y is derived from x. No further arrangement of readings within ABCD will establish one of these alternatives. If, for instance, there are some readings where B and C agree against diversity in A and D, the most that can be said is that this suggests that in these instances the readings of x and y were like those now found in B and C. Further, if there are some readings where B and C agree against unity in A and D, we are forced to say that one set of agreements is the result of coincidence, and have still learned nothing about the relationship of x and y.
Note. It is this sort of difficulty that Joseph Bedier perceived and made use of as one of his most telling arguments against the whole of comparative study. In his La Tradition Manuscrite du Lai de l'Ombre (reprint from Romania; Paris: Champion, 1929), pp. 51-52, occurs this possible tree for the Lai: In this tree r is a manuscript posited merely for sake of argument, not actual or hypothetical. Bedier sets it up as differing from w five times only, and makes the further condition that in all of these readings it agrees with z. It is thus, in Quentin's terms, intermediate between w and z. Bedier then goes on to say that all the relationships expressed by the tree above would be equally well expressed by this alternate tree, with r made the archetype.

80

Page 80
In spite of the fact that these two trees as drawn by Bedier are open to some criticism, he is quite right in his main statement, since it amounts to no more than saying that distributional evidence will not settle whether it is w or r which equals O. To settle this question he would have had to give us the five readings, and let us decide on genealogical grounds which tree was better.

If, however, we discover a new group of manuscripts which have a common ancestor z, we can now compare the readings of x, y, and z, and may be able to establish that contrary to our earlier necessary assumption, all three are derived by radiation from O. We should not, of course, be troubled by the fact that new evidence upsets a previous hypothesis, but should on the other hand be thankful that our problems are now fully settled. If, however, the relationship of x, y, and z is not radiation, there will be a constant grouping of two (xy, xz, or yz) and the only economical hypothesis is that one of the members of the constant group of two is identical with O. But there is again no distributional evidence which will identify for us which of the two manuscripts is O; in other words, we can not define distributionally which is derived from the other.

XXI. The above discussion of ancestral relationships has implied the existence of a tool which has not yet been defined. This is the reconstructed or ancestral reading, which is the result in individual instances of the postulate of unity. The purpose of setting up ancestral readings is twofold. First, only by setting up such readings is it possible to examine whether the readings of possible daughter manuscripts are really compatible with a hypothesis of derivation; and second, the setting up of ancestral readings reduces the number of variants with which the student


81

Page 81
must deal, so that when ancestral readings have been set up in successive stages, the student can eventually begin to set up the readings of O by a relatively simple process of selection. The reconstruction of readings on distributional evidence can be reduced to a few type situations.

1. If there are only two manuscripts, the ancestral readings can be given only where the manuscripts agree. This statement holds also for any situation where more than two manuscripts have been previously reduced to two ancestral manuscripts.

Note. It should be pointed out that an ancestral reading is simply the reading of an ancestral manuscript, and need not be the reading of a hypothetical ancestral manuscript. In the two-text situation, as will develop later, ancestral readings are commonly the readings of one of the extant manuscripts.

2. If there are more than two manuscripts in radiational relationship, the ancestral readings can be given where all the manuscripts agree, and also where some of the manuscripts agree against diversity in the rest.

3. If the group concerned is a sub-family, its ancestral readings can be given where all the manuscripts of the sub-family agree, or wherever one or more manuscripts of the sub-family show a reading supported outside the sub-family at a point where the other members of the sub-family show divergence from each other.

4. In all other situations where reconstruction of readings is possible, some reliance must be placed on genealogical or other non-distributional evidence.

XXII. When distributional evidence has been exhausted, the student must proceed to genealogical evidence. Geneological evidence can be summed up as the examination of the content of sets of variants for the purpose of answering the following generalized question: Is it more likely that A is derived from B, that B is derived from A, or are the probabilities equally balanced? Readings where the change seems equally likely to have occurred in either direction can be called reversible, those in which the order of change seems more likely in one direction than the other can be called non-reversible. The first task of the student is to separate the reversible readings from the non-reversible. Thereafter the reversible readings are to be disregarded.


82

Page 82

XXIII. It is not my purpose to summarize the nature of nonreversible readings, since bibliographers who deal with such entities can be expected to know much more about them than can a strayed linguist like myself. There are, however, two types which I should like to discuss.

1. Instances where a non-language object varies with a language symbol, over a bridge of similarity between the two. A brilliant piece of genealogical reasoning which illustrates this type of variation is Brusendorff's conclusion that the unique reading in one manuscript of Chaucer's Purse "that of eye lowness hath no peer" can be derived from the usual reading "yellowness" by the bridge of an ancestral manuscript which read "yelowness" with accidental obliteration of the first l. That is to say the non-language characteristic of faintness of a letter was interpreted as a meaningful space between words.[11] Similar would be the not infrequent instances of variation between a smudge and a mark of punctuation.

Note. Probably the locus classicus of genealogical evidence of this particular type is to be found in the manuscripts of the Dialogues of Epictetus. One manuscript (Bodleianus, Cod. Graec. Miscellanei 251) has a greasy stain which covers text at the beginning of chapter 18 of the first book. All other manuscripts have a blank corresponding exactly to the number of letters covered by the stain, or present the letters which appear on either side of the stain, without recognition of missing material. In consequence editors have recognized Bodleianus as the archetype of all known manuscripts. The course of the reasoning leading to such a conclusion is clear. It is more easily believable that copyists of manuscripts derived from Bodleianus should have come upon the stain and been unable to decipher the letters which it covered, than that the copyist of Bodleianus should have worked from a manuscript containing a blank, and have then concealed his inability to fill in the blank by putting text into it and covering it with a stain which made the text illegible. For the material on this manuscript family consult Joseph Souilhé, ed. and trans., Epictète Entretiens, Livre I (Paris: Société d'Édition "Les Belles Lettres," 1943), p. lxxii.

2. If a set of variants consists of a non-existent and an existent language form, and if further, the non-existent form can be explained as the result of a known type of mechanical error such as


83

Page 83
eye-skip, haplography, or the like, the existent form can be presumed to be the earlier, unless the distributional evidence proves otherwise.
Note. A typical, if somewhat extreme, example of a non-existent form fully explainable by a series of mechanical errors, is cited by Havet, op. cit. p. 25. A single palimpsest manuscript of Cicero has the startling form COSOSNECVTVTVTVSEGVTVS . Part of this multiple error, resulting in a form which is both unique and scarcely conceivable as an existent Latin form, is due to misinterpretation of abbreviations ("coss. consecutus," i.e., "consulibus consecutus"), and part to the well known mechanical error of perseverence. It is no wonder that Collomp (op. cit. p. 9) remarks that this scribe must have been overcome by sleep. The reading of this manuscript is unique, so that an editor would be justified in rejecting it out of hand. Were it, on the contrary, supported in a large number of manuscripts, the distributional evidence might force the conclusion that the mistake was in the original. The presence of a known type of error can, of course, be used to decide the direction of change, even if all the several forms are existent, and offer perfect sense. An example of this sort of conclusion is found in Rand, (op. cit. pp. 257-58, quoting Quentin Memoire p. 477, and Genesis 19,8). In this passage the variants, all of which offer sense, are as follows: sub umbra culminis mei (9 mss.)
sub umbraculum culminis mei (2 mss.)
sub umbraculum tegminis mei (5 mss.)
Quentin regards "umbra culminis" as derived from "umbraculum tegminis" by the omission of "-um teg-." Rand properly derives "umbra culminis" not from the source suggested by Quentin but from "umbraculum culminis" by omission of "-um cul-," which can be accounted for by eye-skip, where the omission suggested by Quentin can not be correlated with such a type of known error.
The caution given above is necessary, since the mistake may have been in O, and the existent form produced by later correction; the relation of the readings parallesse and parallelless cited earlier is an example of this sort of reversal. Nonetheless, it seems a safe assumption that the number of mechanical mistakes will exceed the number of conscious corrections—at least my own experience with secretaries seems to bear it out.[12]

The terms existent and non-existent used above need some


84

Page 84
explanation. I have purposely avoided such terms as right, expected, or understandable, since to use them would be to reintroduce the literary criterion which I have attempted to avoid. Further, a form may obviously be existent without being understandable, as Professor Menner has shown with the unintelligible "lad hem bi lag mon" of Gawain and the Green Knight.[13] To use the term non-existent, however, necessitates an assumption which is inescapable, even though we know it to be false. To say that a form is non-existent in Middle English actually supposes that we know all about Middle English, when in fact we do not. We are, in short, using negative evidence which is always dangerous. There is, however, nothing we can do about it except to be sure that we have made a diligent search before we say that a given form is non-existent.

3. The result of these statements is this: I regard individual pieces of genealogical evidence with distrust, and am firmly convinced that they should be used only after the distributional evidence has limited the alternatives. Also, genealogical evidence should be presented statistically, since some of it will be contradictory, as was distributional evidence. If the non-reversible readings have been well chosen, however, the group of them which is most frequent ought to give trustworthy evidence of the direction of derivation.

XXIV. At the end of all his labors, the editor presents his results in the shape of a family tree, leading back to O, whose reconstructed readings he will follow in his text. I should repeat that a tree is a description of the relationship of readings found in manuscripts, and ought never to be understood as a statement that A was copied from B. It merely states that the readings now found in A are derivable from readings now found in B, after examination of all the extant evidence. The fact that the readings now found in B were in historical fact taken from a now irrecoverable manuscript is quite irrelevant to any study of readings and their distribution. I am laboring the point, since


Errata

Page Errata

Errata verso

Page Errata verso

85

Page 85
students seem sometimes unnecessarily timid in the face of external evidence, particularly of chronology. Even such a soundly courageous student as Germaine Dempster, in arguing for a certain type of descent, finds it necessary to say that the date of copying of extant manuscripts is not out of keeping with her view of the descent studied.[14] Actually an editor who has derived A from B should be quite unmoved by external evidence that B is on twentieth century paper, while A is on mediaeval vellum.
Note. The irrelevancy of date of copying for conclusions about the derivation of manuscript readings is a product of the fact that all families of manuscripts must be presumed to have been decimated. The date of copying is, however, relevant in drawing up a tree for printed books which are preserved in multiple copies, so that decimation is much less likely. It is probably a failure to realize the necessary differences between the decimated families of manuscript work, and families presumed to be complete, which is responsible for the fact that the Lachmannian type of student regularly assumed that the oldest manuscript was the best. An example of the way in which decimation can affect a tree is the following from Havet, op. cit. pages 8 and 27. Havet cites a print of the Fables of Hyginus of 1535, which can be shown to be a direct copy of a ninth century manuscript which now exists only in fragments. Not only is such a time-spread a clear indication of the unreliability of guesses about authority based on date of copying, but since the ninth century manuscript is incomplete a student would be justified in deriving the readings of, say, an eleventh century manuscript from readings now found only in the print of 1535. The clearest statement of this principle is to be found in the Quentin Essais, p. 103, in answer to one of his many critics. ". . . 'on sera étonné,' écrit-il, 'de voir une généalogie où le Ms. Laud, du IXe siècle [et autres] . . . dérivent de Leg qui n'est que du Xe siècle. . . .' La vérité, c'est que mes généalogies s'appliquent non aux manuscrits eux-mêmes, mais aux types de transmission du texte qu'ils representent." An astonishing and most misleading misunderstanding of the conditions under which date becomes a good guide is in E. K. Rand's attack on Quentin, already cited, on pages 208-11. Quentin had performed the interesting experiment of copying out a Latin passage twenty-two times, labelling the copies from A, the original, to Z, the copy farthest removed from it. This known tree he then used as a test for the validity of his methods. Rand attempted to show that the classic "common mistake" technique gave surer results far more quickly than did the Quentin system. To do this he assumed that manuscripts labelled with letters late in the alphabet were late and corrupt, manuscripts labelled with letters early in the alphabet were older and more authoritative. Not unnaturally

86

Page 86
he was able to reconstruct Quentin's experimental tree very quickly, since under these circumstances a mistake was readily and correctly defined as any reading which appeared for the first time in a manuscript labelled with a late letter. Such an attempt, however, is little more than a copying of the tree already given, and can not be accepted as a refutation of Quentin's theories. Rand had strangely failed to recognize that he was working with a family known to be complete, an assumption which would never apply in actual manuscript work. One can not wonder at the tone of slightly aggrieved patience with which the good Dom describes this treatment in his Essais, page 55.

XXV. A manuscript tree is a diagrammatic description, and is to be judged like other instances of description in science—by the same criteria that we apply, for instance, to a description of phonemes. It follows, therefore, that like other descriptive statements, a tree is non-unique. That is to say, some other description might also contain and present the facts and be quite different in shape from our own diagram. To say that a tree is non-unique is not, however, to state that all trees are equally good. Two trees are equally good only if they differ at points where choice is truly arbitrary, and in such instances the general similarity between the two should be very considerable.

Note. The non-uniqueness of manuscript trees was the basis of Bedier's famous attack on comparative study of manuscripts, from which he concluded that the only defensible method of editing was to select a single good text and depart from it as taste and judgment dictated. The attack is found in its fullest form in La Tradition Manuscrite du Lai de l'Ombre, previously cited. He here attempts to reduce all investigations of authority to an absurdity by drawing no less than eleven different trees for this single text. It is true that all of these trees would explain the arrangement of readings, but Bedier made no attempt to judge between them, in spite of the fact that many of them are rather obviously unlikely.

The relative worth of alternate trees can be judged by the criteria universal in descriptive science; completeness, consistency, and simplicity.[15] The first two of these are relatively simple to apply, and are universally acknowledged in manuscript work; the third will need special discussion.

1. A tree is incomplete if it leaves any body of readings out


87

Page 87
of account, or if its author has failed to compare the readings of any manuscript, extant or hypothetical, with those of any other manuscript, extant or hypothetical.

2. A tree is inconsistent if the same evidence, or type of evidence, is used to give contrary results. A startling example of inconsistency is pointed out by Mrs. Dempster in discussing the tree of Manly and Rickert for a part of the Canterbury Tales.[16] For a part of a sub-family Manly and Rickert use the presence of unique readings in each manuscript, and the absence of readings shared with other individuals within the group, to establish that the manuscripts are in radiation. Yet there is a smaller group of manuscripts which belong to the sub-family which they set up as a sub-family within the sub-family, in spite of the fact that this group has no constant body of readings necessitating the setting up of an ancester for it. Manly and Rickert meet the difficulty by assuming that the ancestor of the smaller group which they set up must have copied the ancestor of the whole family without making any mistakes. Thus they have used unique readings and the absence of shared readings to establish radiation, and used exactly the same set of relationships to set up an unnecessary sub-family.

XXVI. The criterion of simplicity needs special discussion both because it has not been explicitly used in manuscript work, and because it will be useful to work out the definition in such a way as to give a simple operational procedure. Simplicity will be assumed to be inversely proportional to the number of hypotheses involved in a given explanation. That is to say, if two explanations equally well fit the facts, that which does so with fewer hypotheses will always be judged preferable.

Note. While it is true that (to my knowledge) simplicity has never been explicitly used as a criterion in manuscript study, it has, curiously enough, been at least once explicitly denied. Bedier (op. cit. pp. 66-67) says of some of his less probable trees, "Certes, supposer un O2, puis un O3, c'est aller en compliquant toujours davantage. Mais ni la simplicité d'une hypothèse n'est un gage de sa justesse, ni sa complexité n'est une marque de sa fausseté." Such a startlingly unique assumption goes far to explain how Bedier could revolt against nearly a century of literary scholarship.


88

Page 88

1. A basic hypothesis which equally underlies all trees can not be used to compare one tree with another, unless there is a difference in the number of times the hypothesis is used. At the outset, then, it is necessary to point out that all manuscript work assumes the existence of a single original, which we have called O. In other words, the minimum number of manuscripts with which a student must deal is always one more than the extant number. Even in trees which do not end in O, the student has assumed its existence, but has made the further conclusion that O is identical with an extant manuscript. At the moment, then, it is important to note that the existence of O is not a special hypothesis inherent in only some trees.

2. A second general assumption is, of course, that manuscripts assume their present shape by copying. In other words, no tree can be drawn without showing lines of derivation. But each such line of derivation is a separate hypothesis, and the general statement of simplicity given above means that the tree which shows the fewest lines of derivation is the simplest.

This fact can be used to express simplicity by a simple device of scoring. Suppose that our arrangement of readings for three extant manuscripts is such as to permit either of the following trees: Scoring each line as one, the first tree has a score of three, while the second tree has a score of two. Both trees have assumed that O exists, but the second tree has been simplified by making the separate existence of O unnecessary. In other words, other things being equal, whenever an extant manuscript can be identified with O, a simpler tree results, and this simplicity will always be reflected in the number of lines of derivation.

3. Another type of hypothesis made in trees is that of the existence of hypothetical ancestral manuscripts. Except for O, such hypothetical ancestors are not equally necessary for all trees, so that their presence or absence becomes a measure of comparative simplicity. Further, the setting up of a hypothetical manuscript


89

Page 89
other than O is a double hypothesis, first of the existence of one more manuscript, and second of derivation from it. Thus in our scoring device, we will count any line of derivation connecting a manuscript with a hypothetical manuscript other than O as having a value of two. Suppose that our tree can be either of the following: The number of lines of derivation is equal, but the simplicity is not. Actually the first tree has dealt with four manuscripts (the three extant ones and O), the second with five (the three extant ones, O and x). Our scoring device reflects this by giving the first tree a total of three, the second a total of six.[17] The first tree is then preferable.

XXVII. Such a measure of simplicity, rough though it may be, gives the student a powerful tool in dealing with situations which have often been thought to be hopelessly ambiguous.

1. In any two-text situation, or any situation which has been reduced to a two-text situation, either one manuscript must be descended from the other, or both from O. If one manuscript is descended from the other, the resultant tree has a score of one; if both are descended from O the tree has a score of two. Thus unless there is some evidence which positively contradicts the hypothesis, the student will assume that one manuscript is descended from the other, and proceed to genealogical evidence to decide which way the derivation goes. Later evidence may show that his hypothesis was false, but that danger is inherent in all decisions based on evidence. In the mean time, his decisions have been consistent and orderly, and so have a greater probability of being right than if they had been inconsistent and disorderly—unless,


90

Page 90
of course, one is prepared to defend the horrifying position that there is no order at all in human experience.
Note. The statement here given constitutes an answer for one of the major objections raised by Bedier in the work previously cited. He surveyed a large number of trees drawn up by scholars using the Lachmann method. He found that a large majority of them, almost amounting to unanimity, were two-forked trees. Such a state of affairs, depriving the editor of the opportunity of bringing supported readings to bear against unique readings, could not, he felt, be either the result of chance or of falsification on the part of scholars. He concluded therefore that the almost unanimous choice of two-forked trees reflected a weakness in the method itself. If the criterion of simplicity is applied, the student is under a genuine obligation to resolve all two-forked trees, so that Bedier's objection would be met.

2. In practice this statement means that a tree of the type will always be rejected in favor of one of the types given below, which identify O with an extant manuscript. It is for this reason that it was stated earlier that in a situation where both AB and CD had ancestors, it is inadmissible to draw a tree in the form: Such a tree is one which reduces a group to a two-text situation, wherefore it will always be assumed that either x or y is identical with O.


91

Page 91

XXVIII. Of the three text situation Greg has said:

. . . it is clear that, where three manuscripts only are concerned, no merely formal process can throw light on the relationships between them. Either the readings will be all divergent or else the variants will be of the type . . . where one group is represented by one MS only . . . and since, in the latter case, the reading of the single divergent manuscript may always (theoretically at least) be unoriginal, it will never be possible to establish a common source for any pair of manuscripts to the exclusion of the third. Given three manuscripts, therefore, it is impossible either to prove or to disprove independent derivation. This fact, which I call the ambiguity of three texts, we shall find meet us at every turn of the discussion. . . .[18]
The measure of simplicity gives us a way of dealing with this situation which no longer leaves it ambiguous. We shall partly accept Greg's statement, partly deny it.

1. There are two types of tree involving a common ancestor for two out of three manuscripts. They differ only in that the second has identified O with an extant manuscript. Both trees are of course types, so that the labels attached to given branches are of no significance. There is a slight difference in simplicity between these two, since the first has a score of seven and the second one of six. The fact that the second is slightly preferable is, however, irrelevant, since the choice is between these two and still simpler types.

The simpler types are


92

Page 92
The first of these has a score of three, and the last two a score of two. By Greg's own statement the more complex trees can offer no explanation of the relationship of readings which is not equally well given by one of the three simpler trees. Therefore, the more complex trees are always unnecessary, and will never be used by a consistent student, no matter if there is a danger that later evidence might show him that his assumption of a simpler tree was actually false.

2. If the three-text situation were truly ambiguous, it would be further impossible to decide between the simpler trees, that is to say, the type employing a separate O and the two types in which O is identified with an extant manuscript. This question also can be resolved.

3. The tree which has a separate O is less simple, it is true, than the other two trees, but it is also true that distributional evidence can establish the fact of radiational derivation. A generalized example will make this clear. Let A, B, C represent the three manuscripts. Let a represent supported readings, and b represent unique readings. We can thus represent a situation in three differing text positions thus

       
A   B   C  
a   a   b position one 
a   b   a position two 
b   a   a position three 
If, in this situation, any one of the three extant manuscripts is taken to be the ancestor of the other two, one of the three text positions is unexplained. Thus if we draw the tree in the form the readings in position three must represent coincidence, which is contrary to the postulate of unity. The same statement would apply to any tree placing either B or C as the ancestor. Thus in the situation where there are always two readings against one, but the grouping is not constant, the only reasonable conclusion is radiation from a hypothetical ancestor.

93

Page 93
Note.. An example of the way in which the procedure I have been trying to outline modifies the results of previous methods can be drawn from the following short Latin poem, to be found in Heinz Pflaum, "Sortes, Plato, Cicero satirisches Gedicht des dreizehnten Jahrhunderts," Speculum, VI (1931), 499-533. There are three manuscripts only, and at page 525 the editor gives the following Lachmannian tree: The reasons for this tree are clearly given on the same page: "N and P bilden eine besondere Klasse, da sie, gegenüber der richtigen Lesart von T, an vier Stellen charakteristische gemeinsame Fehler haben . . . während T nirgendwo in einem unzweifelhaften Fehler mit der einen gegen die andere übereinstimmt. T is durchaus die beste Handschrift, obwohl Str. 35 in ihr fehlt; P, die jüngste Hs., is ziemlich stark verderbt." The tree above gives a stage beyond that recoverable by comparative study, since it separates O and OI. Second, it draws a needlessly complex tree, as the discussion of the three text situation has attempted to show. I give the first verse of the critical test, with variants, as follows: Sortes pre consortibus currit in consorcio;
in equis et curribus non est dispensacio.
Platonis in manibus sonat disputacio
et de sortis cursibus longa demonstracio.
In line 2 T has canibus, corrected to curribus, the reading of N and P. In line 4, P has corsibus, N and T have cursibus. In line 4, N has longa, T and P have magna. I am not attempting to draw a complete tree for this poem, but it is striking that the first three variants exactly illustrate the scheme drawn for a three-text situation compatible with radiation from O. It is curious, finally, that the editor in printing longa in line 4, has adopted a reading which his own tree disproves.

4. If, on the other hand, the readings are constantly arranged in a fashion which can be generalized thus:


94

Page 94
       
A   B   C  
a   a   b  
a   a   b  
a   a   b  
or the readings are all different in a fashion which can be generalized by letting x, y, and z represent unique readings, thus:        
A   B   C  
x   y   z  
x   y   z  
x   y   z  
the hypothesis of radiation from a hypothetical ancestor becomes unnecessary, and the student will adopt the simpler type of tree. As to which of the types of tree employing no hypothetical ancestor he will choose, distributional evidence will offer him no help, and he must base his choice on genealogical evidence.
Note. The first of the two schemes given in this paragraph places some limitation on possible trees, just as the Quentin "intermediary" does. The ancestor can be either A or B but not C, if the copied manuscripts are independently derived. If the derivation is in a straight line, either A, B, or C can be the ultimate ancestor, but C can not be the middle term. The second scheme, which represents a relatively rare state of affairs, places no limitation on the possible trees, except that radiation from O is an unnecessary hypothesis.

XXIX. It is, I believe, of no use to carry this analysis of the effects of measuring simplicity into situations involving more texts than three, since such situations will resolve themselves into larger radiational groups, two- or three-text groups, two- or three-text groups mixed with larger radiational groups. In any case, it is easy to carry the suggestions I have made into situations involving more manuscripts.

XXX. A brief summary of what has been done may be useful. I have first of all tried to follow Greg and Quentin in setting up a method of dealing consistently with variants, making the minimum use of variants preclassified into right readings and wrong readings. In this I differ from the implicit assumptions of many editors, and the explicit assumptions of such students as Hitchcock and Wolf. Second, I have tried to make a few basic assumptions,


95

Page 95
in themselves as simple as possible—indeed sometimes so simple as to be obvious—and see how far it is possible to carry them. Third, I have broken with Greg, whose view of a tree is that it must be unique before it can be regarded as free of ambiguity. Instead of accepting uniqueness as the only criterion, I have stated that trees are often necessarily non-unique (as Greg well knew), but that in instances where a simple tree and a complex tree both explain the arrangement of variants, the student can choose the simple tree. To do so will be as acceptable as if the simpler tree had been fully proved, since to do so will achieve the aim of distributional study, which is to set up an orderly and consistent method of dealing with the extant variants, subject always to the reservation that when more variants are discovered, the results will have to be re-examined.