University of Virginia Library

II

So far as I know, no general treatment of patterns of authorial revision has been attempted, and it will be well, therefore, before going further,

to investigate what might be called its theory. There are three general types of revision that may occur: simple revision, which may be addition, rejection, or substitution; reverse revision, returning to the original reading; and continued revision, where a new revision is substituted for the rejected one. Combinations of two or all three types are also possible. Simple revision results from a single decision on the author's part (pattern ab), reverse revision from two (pattern aba), continued revision from two or more (pattern abc, abcd . . .), and combinations of types from three or more (simple and reverse: abab, ababa . . . ; continued and reverse: abac, abca, abcb . . .; all three: ababc . . .). In accordance with the concept of simplicity introduced by Hill, we may say that where there is a choice in postulating types of revision, the type resulting from the fewest decisions is, all other things being equal, to be preferred.

Hill has maintained that the usefulness of distributional study is at an end when authorial revision has been established (p. 69), but this would appear to be generally true only when the study is intended as a help in establishing the genetic relationships of a group of texts. If the relationship can be established as radiation, the distributional evidence may continue to be of paramount importance, for by it it may be possible to range the texts in successive order from that representing the earliest recoverable state of revision to that representing the latest, or the reverse. The axioms governing the analysis of this order are as follows:

• Given that the texts are in radiational relationship, a reading found in only one may be taken to have resulted from copying rather than from revision, until the reverse can be proved. This is a restatement of what Greg calls the assumption of universal variation, "the process of transcription is characterized by variation, and it is only in the process of transcription that variant readings arise" (Calculus, p. 8).
• Given that the texts are in radiational relationship, a reading found in two or more may be taken to have resulted from revision rather than from copying, until the reverse can be proved. This is a restatement of what Hill calls the postulate of unity, "it is more likely that unity of manuscripts represents unity in their source than that it represents coincidence" (p. 64).
• Unless there is evidence to the contrary, revisions may be taken to have been clearly indicated in writing (or by marks of deletion). This is a specific application of the assumption behind all textual analysis that there has been no memorial transmission (Greg, Calculus, p. 1, Hill, p. 69).
• The best arrangement of the texts will be that which will require the positing of the fewest decisions on the author's part to explain all the variations.

  ---

  89

  
  
  
  This is a specific application of the concept of simplicity. It is at the same time obvious that such a simplification does not necessarily represent the facts; the author may well have changed his mind more often than was logically necessary to achieve his final revision. There is, therefore, always an appeal open from the distributional to other types of evidence, especially when the distributional evidence is not clear-cut. On the other hand, mere distrust of the results obtained is not sufficient to overthrow them, for unless there is evidence that the author was by nature undecided, we must fall hack on this fourth axiom; its opposite, that the texts should be arranged in such a way as to require the most possible decisions on the author's part, is contrary to reason and experience. As Hill puts it, in another connection, "the more complex [arrangements] 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 [arrangement] was actually false" (p. 92).

As a preliminary to calculating the best arrangement of the texts the variations should be tabulated, using Greg's notation, with Σ: and its subscript and superscript sigla, so that their exact nature may be seen at a glance. Factoring the variations will often facilitate subsequent calculations. Since author's decisions are being measured, all readings found in only one text may be ignored, and since only these variations can occur where there are only two or three texts in question, it follows that the method is applicable only to four or more.

In any single variation, arranging the variant texts together will always decrease the number of decisions that have to be posited, and dividing them will always increase it.

The first step in the calculations, therefore, is to group together the texts with the greatest number of mutual agreements. Normally there will be only two texts in the group, for if there are to be more than two, each must agree equally often with at least two of the others. If two or more mutually exclusive groups are found to occur with equal frequency, that one to which more texts may be added has the priority.

To this nucleus should be added any texts that agree with one of its members more often than with any others outside the group. If there are other texts which agree with the newly added members of the group more often than with any other texts outside it, they too may be added. When the group has been expanded as much as possible, the process is repeated with the ungrouped texts, and continued until all have been arranged in groups. These groups do not indicate any genetic relationship; that is already established as radiation; they are simply a mechanism for clarifying and simplifying calculation.

If there are only two groups, their original nuclei should be placed at the extremities of the arrangement, as this will insure the greatest number of single-decision variations. To secure the greatest number of three-decision variations without lessening the possible number of those requiring only two decisions, that member of the nucleus should be placed at the extremity which agrees most often with texts outside the nucleus, unless it can be placed contiguous to a text with which it agrees more often than the difference between its outside agreements and those of the member of the nucleus having the next largest number. The remaining texts should be arranged so as to secure the greatest number of agreements between and among contiguous texts, for this will give the maximum number of two-decision variations.

If there are more than two groups, all the possible arrangements of the groups must be tried, arranging the members of the terminal groups on the principles above, and the members of the inner group or groups so as to secure the maximum number of agreements between and among contiguous texts.

In calculating the number of decisions required, special care must be taken when counting the variations where one or more texts are imperfect, for to count a text as present when it is in fact absent may add two to the number of decisions. Thus, in testing an order ABCDE, to read ΣD:CE as Σ:CE will result in the erroneous count AB/C/D/E. The same precaution must be taken when there are more than two alternate readings in a variation, especially when some or all of the additional readings occur in only one text apiece. It is, by the way, unnecessary and indeed useless to speculate on the readings in the missing portions of imperfect manuscripts; we can work only with what we have, and to refuse to employ the talent that is given us is to remain in willful ignorance.

Once the desired order has been established, distributive study is at an end. To determine which extremity of the arrangement of the texts represents the latest recoverable stage of revision must be left to external or literary study, preferably the former. Dates of copying the texts, even if determinable, are meaningless, as always, for there is no assurance that they were copied from the author's original. On the other hand, reasoning may be possible if one or more of the texts is, or has as a descendant, a printed version. Once a work is in print, interested persons will normally prefer buying the book to copying the author's manuscript, unless the author is known to be still improving it. A printer will normally choose printed over manuscript copy. And an author is not likely to sanction the publishing of an early revision of his book if a later one is already in print. Where external evidence fails, one will normally assume that the more

satisfactory of the terminal texts, so far as he can judge of it, is the later.

Complete proof that the order and its direction are correct is impossible, but the likelihood may be demonstrated pragmatically by showing that the variations fall into consistent patterns and that they suggest or allow reasonable interpretations for a maximum number of individual variations. Included in the interpretation should be a careful investigation of readings found only in the final text, since, the pattern of revision having been established, these readings may no longer be dismissed as unauthoritative.

Once the order and its direction have been established, the texts are, in a sense, no longer in a radiational relationship, for, as Greg points out, a text which has been revised is tantamount to a transcript of the unrevised form (Calculus, p. 8, n. 2). This fact reminds us that while it is especially tempting when dealing with authorial revisions to think of the ancestor of the extant texts as the author's original manuscript, the original is in fact irrecoverable. The author may even have made a new copy each time he revised-if many of the revisions are of the complex kinds it is quite likely that he made a clean copy at least once-and while we have some idea of the alterations of the text as it passed through various stages we can have no certainty that we have recovered all of these or the state of the text before, or, if the author did not publish the work, after revision.

If we think of the revisions as the same as transcripts, the tree which results is identical with one indicating successive variation, but reached in this fashion there are no anomalous variations. Any decision that some of the variants are the result of chance coincidence or conflation will have to be estimated by some other method than distributional study.

It is, therefore, of some interest to examine into the likelihood of the occurrence of the various patterns of revision. The Cambridge manuscript of Milton's minor poems shows for his sonnets, besides a number of simple revisions, some of the patterns aba (XIII, 4, 5; XIV, 9) and abc (IX, 7; XIV, 9), with variants abbc (XI, 10) and abbcc (XIII, 3), and one more complicated, abcb (XII, 10).[4] Milton is more likely to reverse his revisions than to continue them, though on two occasions in Comus we find abcd (351,833); and he often combines the two processes, so that in Comus again besides the common aba and abc we find abac (389), abaca (554),abca (254), abcbc (544), and abcada (a canceled passage following line 4). Pope's penchant for revision is notorious; the editors of the Twickenham

edition of his poems preferred not to try to indicate the variant readings in his manuscripts. To take a random example, if the opening couplet of the Essay on Man, Epistle II, quoted from the Harvard and Morgan manuscripts by Professor Sherburn, be compared with the textual notes in the Twickenham edition,[5] it will be seen that beside two changes of the pattern ab in the first line there is one aba ('then thyself'/'we ourselves'/'then thyself'), another aba in the second line ('study'/'science'/ 'study'), and a more complicated pattern, abcdce at the beginning of that line ('And know the'/'But know the'/'The only'/'Convinced the'/'The only'/'The proper'). Finally, to go no farther than the end of the eighteenth century, the manuscript of Blake's 'The Tiger,' taken with the first edition,[6] shows the pattern ab in lines 3, 15, 16, and 19, aba in lines 4, 5, 6, 13, 16, 19, and 20, abab in line 19, and aabcb in line 15. This cursory survey indicates that the patterns of revision may be of almost any nature, and that it would be unwise to reject any in advance as unlikely to occur.