Caxton’s Editing of the Canterbury Tales

pbsa 108:1 (2014): 41–60 Barbara Bordalejo (Department of Modern Languages, University of Birmingham, Edgbaston, Birmingham, UK, B15 2TT) is a digital humanist and a textual scholar with a background in English Literature. She has published editions of Chaucer’s Canterbury Tales (http://caxton.dmu.ac.uk/) and Darwin’s Origin of Species (http://darwin-online. org.uk/Variorum/). Her interests include digital literature and transmedia storytelling, as well as the transformational impact of digital culture on our ideas of authorship and copyright. She advocates data sharing through creative commons licences without commercial restrictions. Barbara teaches digital literature, new media, and English literature and is a member of the executive of Global Outlook :: Digital Humanities. She is a Senior Research Fellow at the University of Birmingham and can be contacted at b.bordalejo@bham.ac.uk. Caxton’s Editing of the Canterbury Tales

Caxton made what has become a well-known statement. 1 He declared that a gentleman had told him that his first edition of the Tales was inaccurate and that he could provide a better manuscript, one truer to Chaucer's text. Caxton claimed that he corrected the book based on the manuscript with which he had been provided and printed a second edition: 1. Caxton's first edition of the Canterbury Tales (STC 5082, referred to in this article as Cx 1 ) has been dated to either 1476 or 1477. Lotte Hellinga thinks the likely date is the earlier one (Caxton in Focus: The Beginning of Printing in England [London: British Library, 1982], 67-8, 80-1), but Paul Needham dates it 1477 (The Printer & the Pardoner [Washington, DC: Library of the Congress, 1986], 84), following Norman Francis Blake (Caxton: England's First Publisher [London: Deutsch, 1976], 127-8), who later changed his opinion and stated that the date for Cx 1 was "almost certainly" 1476 (The Textual Tradition of the Canterbury Tales [London and Baltimore: Edward Arnold, 1985], 1). The second edition (STC 5088, referred to in this article as Cx 2 ), has been dated to either 1482 or 1483. However the dating of Cx 2 depends on that of Cx 1 , since the book's second edition was, Caxton declares, printed six years after Cx 1 . Thus Needham dates Cx 2 as 1483 (The Printer, 87), while Hellinga's dating of Cx 1 as 1476 would bring the date of Cx 2 to 1482, a date also accepted by Blake (England' s,5).
For I fynde many of the sayd bookes, whyche wryters haue abrydgyd it and many thynges left out, And in somme place haue sette certayn versys, that he neuer made ne sette in his booke, of whyche bookes so incorrecte was one brought to me vj yere passyd, whyche I supposed had ben veray true & correcte, And accordyng to the same I dyde do enprynte a certayn nombre of them, whyche anon were sold to many and dyuerse gentyl men, of whome one gentylman cam to me, and said that this book was not accordyng in many places vnto the book that Gefferey chaucer had made, To whom I answerd that I had made it accordyng to my copye, and by me was nothyng added ne mynusshyd, Thenne he sayd he knewe a book whyche hys fader had and moche louyd, that was very trewe, and accordyng vnto hys owen first book by hym made, and sayd more yf I wold enprynte it agayn he wold gete me the same book for a copye, how be it he wyst wel, that hys fader wold not gladly departe fro it, To whom I said, in caas that he coude gete me suche a book trewe and correcte, yet I wold ones endeuoyre me to enprynte it again, for to satysfye thauctour, where as to fore by ygnoraunce I erryd in hurtyng and dyffamyng his book in dyuerce places in setting in somme thynges that he neuer sayd ne made, and leuyng out many thynges that he made whyche ben requysite to be sette in it, And thus we fyll at accord, And he ful gentylly gate of hys fader the said book, and delyuerd it to me, by which I haue corrected my book. 2 This article shows that Caxton was accurate in his statement about the corrections introduced in his second edition of the Tales, and that he and his compositors were generally far more careful than some modern scholars have acknowledged. Using textual evidence drawn from a direct comparison of the two editions, I demonstrate: that Caxton did more than add and remove lines; that besides those changes, he introduced more than 3,000 alterations that originated in a different manuscript source; and that this manuscript source contained a text closer to the archetype of the tradition (see Appendix A). This evidence is drawn from my own textual research on William Caxton's editions of the Tales, for which completely new collations were carried out.

computer-aided collation of the caxton editions
As part of my PhD at De Montfort University, I made complete electronic transcriptions of both of Caxton's editions of the Canterbury Tales. 3 In order to establish the differences between the transcripts, I 2. Cx 2 , a2 r-v . 3. Barbara Bordalejo, "Caxton's Second Edition of the Canterbury Tales and its Place in the Textual Tradition of the Tales" (PhD diss., De Montfort University, 2003). The transcriptions were published electronically in Barbara Bordalejo, ed., Caxton's "Canterbury Tales": The British Library Copies (Leicester: Scholarly Digital Editions, 2003), CD-ROM, and can also be found freely available online at http:// caxton.dmu.ac.uk/. used Collate, a program developed in the early 1990s for the collation of large textual traditions. Collate in its default mode shows every difference between transcripts, and makes no distinction between variants that might be significant for the establishment of textual relationships (stemmatically significant variants) and those that are not. Any differences in spelling, in word division, or in the encoding, such as the size of an ornamental capital, are also understood as variation. A typical result of a collation carried out with the help of Collate is, for the first line of the "Miller's Tale": In this example, all differences between Caxton's first edition of the Tales, here used as a base for collation, and his second edition are shown. The first word, "Whilom," which presents two different spellings and encodings, appears as a variant. The next item, "therwas" appears as a single word in Cx 1 and as two words in Cx 2 . The only difference in the word "dwelling" is the tail in the final g in Cx 1 . In fact, none of these variants in this first line of the "Miller's Tale" are stemmatically significant. 4 "Stemmatically significant" variants are those that are useful for the establishment of genetic links between different texts. Variations such as those in spelling, encoding, or word division as seen in this first line of the tale are likely to reflect only differences in practice between the compositors of the two editions.
The whole collation had to be read through in order to filter out all instances of variation that were not stemmatically significant. For example, the second line of the "Miller's Tale" contains the following variant: chuf ] gnof Cx 2 Here, the change from "chuf " to "gnof " is potentially stemmatically significant. I use the word "potentially" because it is not possible to decide just by looking at the collation of these two witnesses whether these variants contain information about the affiliations of the source manuscripts. There are three possible ways the variant in Cx 2 could appear: it could be the result of a correction based on a different text; it could be a compositorial error; or it could be Caxton's own invention and likely to be found nowhere else in the tradition. The more variants we find of the first type, the more likely it is that Caxton did as he said and introduced readings from another manuscript. 5 Thus, one has to examine the distribution of these variants in the whole tradition for evidence that might support or deny Caxton's assertion that he used a manuscript nearer to the archetype in preparing his second edition. If Caxton's account is to be trusted, we would expect that the reading "gnof " was present in the manuscript given to him by the unnamed gentleman. If we discover that this reading and many others like it are also the readings of manuscripts that are considered to have been close to the archetype of the whole tradition, then we have evidence that Caxton did indeed introduce readings from a manuscript close to the archetype into his second edition.
sample section: "the squire's tale" Ideally, one would present the complete table of variants between Cx 1 and Cx 2 for evaluation by the reader. However, the amount of data is large, especially if every one of the steps described above is to be followed. Instead of attempting to deal with the complete dataset, 6 I focus on a single tale. 7 Appendix A lists all 132 significant variants between Cx 1 and Cx 2 in the 664 lines of the "Squire's Tale." I give also the occurrence of the Cx 2 variants in ten manuscript copies, transcripts of which were made available to me for collation by the Canterbury Tales Project. 8 Of those 132, 5. My research on the manuscript source used by Caxton to correct his first edition required a further set of collations that included other witnesses. The second set of collations was designed to show whether the readings introduced in Cx 2 could have been drawn from another witness (as opposed to emendations invented by Caxton himself or those that were the result of chance or accident) and whether it was possible to establish the affiliations of their manuscript source. The list of consulted witnesses and their sigils is in Appendix C to this article.
6. See Bordalejo, "Caxton's Second Edition," 117ff, for a full list of the variants in the Canterbury Tales. 7. I chose the "Squire's Tale" because this is one of the tales that Lotte Hellinga refers to as not having a large number of corrections (see below). The data is representative of the number of variants found in Cx 2 and, up to a certain point, it shows the patterns of variant distribution within the textual tradition.
8. For the "Squire's Tale," I collated Cx 1 , Cx 2 along with Hg, El, Ad3, Bo2, Ch, Cp, Dd, Gg, Ha4, and La. twelve are unique. 9 If we discard the twelve unique variants, likely to be compositorial errors, we are left with 120 variants. This represents a significant change every five lines if we count all 132 places of variation, or one every five and one-half lines if we consider 120. Generally we can talk about roughly one significant change every five lines.

the source of the variants
The sample variants from the "Squire's Tale" show that Caxton did do the first part of what he claimed: he introduced a large number of readings into the text of Cx 2 . We can also assess the veracity of the second part of his statement: that he adopted these readings from a text closer to the archetype of the tradition than was Cx 1 .
The result of generations of textual scholarship on the Tales is a broad consensus about which of the extant manuscripts are likely to be closest to the lost archetype of the whole tradition. Key here is the work of John Manly and Edith Rickert, 10 who built on research by Tatlock and other scholars, and who identified the Hengwrt (Hg) and Ellesmere (El) manuscripts as nearest to their "O," the origin of the whole tradition. Recently, the Canterbury Tales Project has elaborated on this work, demonstrating that a small set of other manuscripts represent independent lines of descent from the archetype, from which they are separated by relatively few intervening copies, and so are likely to preserve archetypal readings. The project labeled these the "O" manuscripts, and for the sections so far analyzed by the project, identified them as Ch, Gg, Ha4, and Ad3, together with El and Hg. 11 Of these, Ch consistently is closest to El and Hg, effectively forming an "inner three" group.
It lies outside the scope of this article to determine the exact place of the source manuscript used by Caxton relative to the other manuscripts of the Tales and hence to the archetype. It is sufficient for my argument 9. Variants that are present in only one witness are non-classificatory from a stemmatic perspective. However, the unique variants in this collation are only unique in the sense that they are not shared by any other collated witness. Once all the transcriptions of the "Squire's Tale" become available, it might be that some variants I refer here to as unique are discovered to be present in other witnesses. 11. Here I present a simplified version of the textual tradition. The individual publications of the project contain fuller information referring to particular sections of the text. that again and again, the reading introduced into Cx 2 is also found in the "O" manuscripts, particularly in Hg, El, and Ch, and for this reason it is likely to have been the archetypal reading. Of the 120 readings introduced in Cx 2 and shared with other witnesses, 100 appear in the "inner three" of Hg, El, and Ch. In contrast, the Cx 1 reading replaced in Cx 2 is very often the reading of the small group of manuscripts labeled as "b" by Manly and Rickert: essentially, He, Ne, and Tc2. This group is characterized by many readings introduced below the archetype, with their number and nature suggesting a long line of intervening copies between the archetype and the "b" manuscripts. Thus, we find over and over-in around a hundred cases in the "Squire's Tale," in thousands of instances across the whole Tales-that the Cx 1 reading replaced by Caxton is found only in the "b" manuscripts, while the Cx 2 reading replacing it is found in most (or even all) of the "O" manuscripts, and indeed in every other line of descent except "b." 12 In the absence of an archetype, one might be hesitant about asserting that one reading is likely to be archetypal or another not, but in these instances, we may be confident that this is the case.
For example, in line 50 of "Squire's Tale," the reading "And so this kyng" is replaced in Cx 2 by "Thys Cambuscan," which is in agreement with eight of the ten collated witnesses, including all of the "inner three." Something similar occurs in line 202, where "grekis" is substituted by "troye" in all ten collated witnesses. These variants are all stemmatically significant, and the vast majority of them appear to represent the replacement of a non-archetypal reading by an archetypal one. Moreover, many of the introduced changes significantly affect our reading of the text. An extraordinary instance is line 379, where Cx 2 reads "foure" in place of Cx 1 "ten." The full line as it appears in Cx 2 is: That in the Ram is foure degrees up ronne The reading "ten" is found widely in the tradition: in the "b" group (whence it makes its way into Cx 1 ), and in the very numerous "cd" group. "Foure," the reading accepted by all modern editors, is found in just five of the ten witnesses I collated, but that five includes the "inner three" 12. Thus, the Cx 1 reading "chuf " in line 1 of the "Miller's Tale" discussed above appears also in the "b" manuscripts Ne and Tc2 (and in Ii, Se, and Ld2, apparently by coincidence), while "gnof " is in all six "O" manuscripts, as well as in Manly and Rickert's "a," "c," and "d" groups: a total of forty-three of the fifty-four witnesses and every line of descent except "b." along with Dd and Gg. Notable too are nine cases where the Cx 2 reading agrees with El against Hg, usually with a few other witnesses: l. 15, "strong" El alone; "and strong" Cx 1 Hg l. 53, "feste so" El Ad3 Dd En1; "feste" Cx 1 Hg l. 194, "they demede" El Gg Ha4 Ht; "demede" Cx 1 Hg l. 231, "wyth his" El Gg; "for his" Cx 1 Hg l. 290, "me" El Gg; "you" Cx 1 Hg l. 419, "and" El Gg; "of " Cx 1 Hg l. 447, "loue" El Gg; "Ire" Cx 1 Hg l. 491, "hard day" El Gg; "ilke day" Cx 1 l. 502, "that I ne coude" El Gg; "that no wight wolde" Cx 1 I observed above that 100 of the 120 non-unique Cx 2 readings were found in Hg, El, and Ch. Here, nine of the twenty not found in the "inner three" are found in El, commonly with Gg, an affiliation that can be observed elsewhere in the Tales. Thus, even where the manuscript source deviates from the "inner three," it commonly agrees with readings found in other manuscript combinations close to the archetype.
Another striking example in the "Squire's Tale," suggesting derivation from a manuscript close to the archetype, is the variant in line 289 (variant 63 in Appendix A), "haue that / soupen al by," in which the Cx 2 reading, "supen al by," is supported by different lines of descent in the manuscript tradition. A similar conclusion can be reached from lines 631 "plastris / salues" (129 in Appendix A) or 277 "lokyngis / lokyng and dyssymylyngis" (60 in Appendix A). 13 By examining the distribution of the variants in Cx 1 and Cx 2 across the whole textual tradition of the Tales, we find similar results in the texts of other verse tales. For example, line 65 of the "Miller's Tale" in Cx 1 reads "Tarselid with grene," while Cx 2 reads "sylk" instead of "grene." The variant "grene" here is restricted to the "b" group and El. In this case, not only has Caxton replaced the reading in Cx 1 by a very different one in Cx 2 , but there is no doubt that he was following his source manuscript in this instance. It appears that the reading "sylk" was likely to have been present in the archetype of the whole tradition: hence its appearance in Hg, and all of what we call the "O" manuscripts except El, representing an uncertain number of independent lines of descent, and in all other lines of descent except the "b" group.
Occasionally, what are usually regarded as the best witnesses to the text appear to agree in error. This is the case of Hg and El in line 1067 13 The Hengwrt and Ellesmere readings are the result of eyeskip from line 1065 ("This is thy doghter, which thow hast supposed"). The Cx 2 reading, "purposed," is attested by at least six independent lines of descent. Moreover, it is the lectio dificilior and, for that reason, likely to be trivialized to the more modern form "disposed." Although the move from "purposed" to "disposed" can be explained as a trivialization, a hypothetical move from "disposed" to "purposed" is very unlikely to have happened. The best explanation here is that "purposed" is the archetypal reading, and might have stood in the exemplar or exemplars copied separately by Scribe B into the Hg and El copies-and both times the scribe made the same mistake and wrote "supposed" not "purposed." In line 484 of the "Wife of Bath's Prologue" we find that a nonsensical reading was introduced in Cx 2 ("I made hym of the same wode a troce"), while Cx 1 had the word "croce" (which makes complete sense given the context Although the reading "troce" is nonsensical, it is shared among several witnesses-Cx 2 , Hg, Ad3, Ch, Ad1, Hk, Pn, and Wy, including four manuscripts Hg, Ad3, Ch, and Hk identified as "O." Peter Robinson has suggested that this reading originated in an error that was present in the archetype. 14 The above examples show the changes Caxton introduced derive from a manuscript from a different branch of the tradition than that used for the first edition, a manuscript likely closer to the archetype. The revisions introduced by Caxton amount to almost 3,000 instances of changes at the word level. These can be seen in Appendix B, which presents a summary of changes in each section of the Tales.
My research on Caxton's editions of the Tales indicates that it is very likely he wrote corrections from this better manuscript into an unbound copy of his first edition 15 that then was the copy for the second edition. 16 The conflated text that resulted from the corrections made by Caxton remained for years the most widely available text of the Tales. 17 My analysis confirms that Caxton was accurate in his preface. When preparing his second edition, he did have access to a manuscript nearer the archetype than his first edition, which textually belongs to the "b" group. He introduced many readings taken from this manuscript-thousands of them, in every part of the text except the prose-throughout the whole length of the Tales.

alternative views on caxton as a printer
However, my suggestion that Caxton actually did in his second edition what he said he did in its preface disagrees with several recent studies. Norman Blake edited all Caxton's prefaces and other original writings and wrote a widely used monograph devoted to Caxton. Here is what Blake says of Caxton's second edition and the preface to it: Although he [Caxton] expressed concern for the text of his first edition, this may have been little more than publisher's talk. He was not sufficiently worried that he felt he had to produce a completely new text. He thought it sufficient to make one or two minor adjustments. This somewhat cavalier attitude has by not means disappeared from the publishing trade today, so we should not be surprised to find it in Caxton. Few people today take a publisher's "blurb" at its face value, and there is no reason to do so with Caxton's prologues, which serve the same function. 18 (London: Deutsch, 1969), 104; England's, 99; Thomas Dunn, "The Manuscript Source of Caxton's Second Edition of the Canterbury Tales" (PhD diss., University of Chicago, 1939).

See Norman Blake, Caxton and His World, The Language Library
16. Hellinga maintains that only two compositors were at work during the printing of the second edition of the Canterbury Tales (Hellinga, Caxton, 61). In my edition of Caxton's Canterbury Tales, I put forward an argument for three rather than two compositors.
17. See W. W. Greg, "The Early Printed Editions of The Canterbury Tales," PMLA 39 (1924): 737-61, on the matter of how the four fifteenth-century printed editions of the Canterbury Tales relate to one another.
Two major points can be drawn from this quotation: first, Caxton made only "one or two minor adjustments" to his second edition of the Tales; and second, he was not truthful in his prefaces because this is the way in which publishers have always behaved. Blake's wording ("Few people today") implies that Caxton was conducting himself like a modern publisher, writing a "blurb" to advertise his books, and thus the current distrust of modern publishers' claims can be applied to Caxton. One can trace the influence of Blake's criticism of Caxton in the writings of other scholars. For example, Beverly Boyd seems to be defending Caxton from accusations of lack of knowledge of the text, and she seems to relieve him of any blame when she points out that he never claimed to be a scholar: His [Caxton's] knowledge of the texts of Chaucer's works was not that of an exegete of his own time, much less that of a present-day editor of Chaucer, though the printer himself never claimed such skill. In no case do we have copies or precise records of Caxton's sources for these books. Within these limitations the evidence is strong that he followed his texts closely except in the second edition of The Canterbury Tales, where he can be faulted for a serious mistake in judgement when he tried to correct the first edition from a manuscript obviously of a different textual tradition, though the circumstances were not altogether his fault. 19 When Boyd refers to Cx 2 , she states that "he can be faulted for a serious mistake in judgment when he tried to correct the first edition from a manuscript obviously of a different textual tradition." It would be unfair to condemn Caxton for conflating texts from different recensions and for not recognizing this fact, given that the genetic groups for the witnesses of the text have been widely accepted as accurate only after Manly and Rickert's edition published in 1940: to suggest that Caxton should have had this knowledge is imposing an anachronistic judgment on a case that might be better served if approached synchronically. Lotte Hellinga continues the same line of observations in her description of Caxton's second edition. Hellinga's assertion that Caxton made only a small number of textual corrections in thirteen of the Tales and virtually none at all in ten of the verse Tales appears to support Blake's argument : Caxton's main concern when editing the text was to add lines of verse that had been omitted in his first edition (to a total of 277 lines) and to take out or replace lines which he now rejected as spurious (to a total of 89 lines . . . ). He made a small number of textual corrections, partly derived from his manuscript source and partly independent, in 13 of the Tales Indeed, Hellinga's argument appears the more forceful because of the detail she presents: most of the changes (she suggests) were to add or remove whole lines (adding 277, removing 89), 21 with few other changes. Included in her list of tales left "practically untouched" is the "Squire's Tale." Yet, my analysis of the "Squire's Tale" suggests that Caxton introduced some 120 changes into the second edition, drawn from the "better manuscript": approximately one change every five lines. Indeed, my analysis of Caxton's changes over the whole edition, presented in Appendix B, suggests that Caxton made some 2,900 changes in the 19,000 verse lines of the Tales, or a rate of one change every six lines.
Hellinga does not state from where she derived the account of the changes made by Caxton in Cx 2 , and particularly the apparently convincing listing of exactly which tales were and were not changed, which goes beyond anything said by either of the two scholars Hellinga here cites, Norman Blake or Thomas Dunn. However, it appears that her opinion derives from a misreading of Dunn. Dunn devotes his entire third chapter to listing all cases of "unique variants" in the second edition in the each tale, showing that there are rather few unique readings in the thirteen tales that Hellinga says had "a small number of textual corrections" and very few or none in the ten tales that she says "remained practically untouched." Instead of tales that have very few variants, Dunn is listing only the tales where there are very few or no "unique variants" (variants found only in Cx 2 ). The great majority of variants indeed do also appear in other witnesses to the Tales, and we expect this, given Caxton's declaration that he took the variants from another manuscript and thus were current in the tradition. Dunn is fully aware of this, and goes to on to report and analyze in Chapter 4 all the variants between the two 20. Lotte Hellinga, Catalogue of Books Printed in the XVth Century Now in the British Library (London: HES & DeGraaf Publishers, 2007), 11:132. 21. The total number of lines present in Cx 2 but not in Cx 1 is 324, not 277: it appears that Hellinga is here following Dunn, who gives the number 277. editions that appear in other selected witnesses (Hellinga does not refer to this chapter). Both Dunn's Chapter 4 and my own work show clearly that, contrary to Hellinga's opinion, every verse tale was substantially corrected. 22 My research shows that in the 2,909 lines (counting only the short version of link 30) that were present in Cx 1 , at least 19,105 changes were introduced that significantly altered words; this means there were changes in about 15 percent of lines. No addition or deletion or substitution of complete lines has been taken into account here. 23 If we add to this the fact that there are 244 "added" lines, with another eighty lines replacing eighty-two in Cx 1 , that thirty-one lines were deleted, and that there are fourteen instances of line rearrangement, it could hardly be said that Caxton did not revise his first edition. Whether these changes were the result of a business plan to sell more books or whether they were made to oblige a rich patron is not as important here as establishing that Caxton was truthful in his preface and demonstrating that the variants introduced in the second edition are the only extant evidence of what must have been a text extremely close to the archetype of the textual tradition.
The studies by Blake, Boyd, and Hellinga present a view of Caxton as careless and unreliable. If widely accepted, this would have a powerful effect on scholarship: one thinks particularly of the debate over the merits of Caxton's edition of Malory against the Winchester manuscript. In the most recent comprehensive survey of the Winchester and Caxton Malory texts, William Matthews accepts Blake's characterization of Caxton as driven by commercial imperatives but rejects (as does this study) Blake's assertion that Caxton made only "haphazard and minor changes," according to his own analysis of the changes made by Cax- 23. This number is the result of my initially computer-aided but ultimately manual collation. I have checked these results with VBase for the "Miller's Tale" and link 1 and discovered that I had the tendency to underestimate variants by around 20 percent; that is, the completely automated collation detected more variants than my manual one. At least in part, this can be attributed to the thorough correction that the collations and transcriptions underwent before finally being published. ton in Cx 2 . 24 In his introduction to the posthumous publication of Matthews's analysis, Robert L. Kindrick concludes that "We are forced by the evidence to assume Caxton was serious in his interest in a good text and that he told the truth in his prologues." 25 My research reinforces these conclusions. This study of the agreements between the readings introduced in the Caxton second edition and their consistent occurrence in manuscripts believed close to the archetype of the Tales affirms that Caxton did not just make many changes (as Matthews shows) but that he did have access to a "trewe and correcte" copy of the Tales, and that he made extensive and consistent use of this copy as the source for the revisions in his second edition. Unequivocally, Caxton was not careless in his preparation of his second edition of the Tales and we should take his claims in its prologue as truthful.