SPELLING'S SIGNIFICANCE FOR TEXTUAL STUDIES

Manuscript spellings are rarely taken into account when editors trace the stemma of their text for, if considered at all, they are used to localise the various manuscripts. From an evaluation of two Canterbury Tales manuscripts, Christ Church, Oxford, MS 152 and British Library MS Harley 7334, we seek to establish that this is an important omission in textual studies, for changes in the spelling system of a manuscript can signal whether these represent a change of exemplar and thus can control other methods of determining how many exemplars were used in its production. This conclusion is supported by evidence that changes in the spelling system coincide with changes in these manuscripts' codicology.


Introduction
Typically editors of medieval English texts discuss the genealogy of their text and the possible exemplars that an individual scribe used to prepare his manuscript.For each text, variants among the manuscripts largely determine the number of exemplars, and sometimes that number is extensive.This applies especially to the Canterbury Tales, for in their seminal edition (1940) John Manly and Edith Rickert interpreted small textual variations in the manuscripts as indicative of the scribes' use of many exemplars.However, neither in this edition nor in other analyses of the textual history of the poem was the evidence which spelling offers ex-Nordic Journal of English Studies ploited as a means to detect or reject changes of exemplar.In this paper we consider the spelling system in two manuscripts of the poem, Christ Church, Oxford, MS 152 [Ch] and British Library MS Harley 7334 [Ha 4 ], to test whether this omission is justified.We hope that this discussion will not only add an extra dimension to textual studies, but also that it will be a fitting tribute to Arne Zettersten who has spent much of his career studying the intricacies of lexis and textual transmission, especially in his work on Anerene Wisse.
Spelling has been the subject of scholarly attention since the Second World War, especially through the work of Angus Mcintosh which culminated in the publication of A Linguistic Atlas of Late Mediaeval English (1986).The principal focus of this atlas is to provide evidence through spelling for the localisation of manuscripts, though Mcintosh also suggested that a text's genealogy might be detected through the different levels of spelling in its extant manuscript(s).A scribe's spelling is affected by two factors: his own spelling system and the spelling system(s) of the exemplars) he is copying.A scribe's spelling system is, however, neither uniform nor a constant, for it is affected by such factors as emigration to another part of the country or even to a different country, the training he received, the development of local standards, and his exposure to the spelling systems of those texts he had previously copied.Consequently, a scribe's spelling is likely to develop in at least some respects as he copies text after text, depending upon how much he is affected by the exemplars he copied, their length, and how recently he had copied them.Furthermore, all scribes are influenced by the spelling systems of the exemplars they are currently copying, but that influence varied from scribe to scribe and even within an individual scribe's own output, since the influence of an exemplar is determined by the distance of the exemplar spelling from his own preferred system at the time he is copying it.

95
To enable us to evaluate the value of spelling for textual studies an electronic indexed database of spellings in early manuscripts of the Canterbury Tales was compiled.This followed the general procedure used in A Linguistic Atlas of Late Mediaeval English, which isolated nearly 300 key words exhibiting significant spelling variation.The spellings of each word are grouped under a lemma indicated by its modern English form in capitals.
Thus THEY covers all spellings of this word, such as jey, jei, jay, yå, þey, þei, pay, pai, etc.An analysis of the various spellings in separate tales and the Wife of Bath's Prologue, but not of other links because of their brevity, was undertaken and led us to select from Ch and Ha 4 certain spellings we accepted as significant.The methodology we have employed for the analysis of the spelling systems in both manuscripts focuses on the proportional usage of spellings for common words that occur frequently rather than on unusual or unique spellings.Dialectal spellings such as those words in Ha 4 with the -ud, -us, and -ur endings, which are thought to be Western (Jeremy Smith 1985: 238 and1988: 62), have been considered only for their distribution across the whole text in relation to, for example, -ed, -es, and -er.In some cases we have observed the distribution of spellings for one lemma in the ordered profile and achieved greater confidence in the patterning by isolating the variable feature and tracing it across lemmata at a graphemic level.No scribe is totally consistent in his spelling system and his spellings of a single word vary throughout a manuscript, for it is unusual to find absolute breaks where one spelling, for example, is used exclusively in the beginning and another in the rest of a manuscript.Changes in spelling are gradual and so the frequency of each spelling of a lemma has to be traced and evaluated across the whole manuscript.In what follows we discuss Ch and Ha 4 in turn, considering first each manuscript's codicology and then the evidence provided by its spellings before evaluating the information they provide in our conclusion.To illustrate our argument we have chosen those spellings which re-fleet scribal trends most clearly.Sometimes individual spellings exhibit no clear trend because too many attestations of a given lemma may lead the scribe to be inconsistent in his spelling, especially if his exemplar shows no clear preference.This may explain the conflicting tendencies found in the spellings of WILL and SUCH in Ch, where the scribe seems to prefer old-fashioned spellings but the evidence is not clear-cut.

The evidence of Christ Church, Oxford, MS 152
The codicology of Ch, a paper manuscript from the third quarter of the fifteenth century normally written in quires of twenty folios, suggests that the scribe intended to include a copy of the Canterbury Tales which was to contain only those tales which are today regarded as canonical.But his exemplar(s) presented some problems: the Cook's Tale and Squire's Tale were incomplete.The progress of his copying can be detected by the watermarks in Ch.The Cook's Tale finishes within quire 3 and the rest of that quire was left blank at first.Quire 4 is irregular with only ten folios and with a watermark which is different from the one in quires 3 and 5, but which is identical or nearly identical with the one found in Ch's final quires.In addition, Gamelyn finishes on the last page of quire 4 leaving the bottom third blank.This indicates that, although written in the main hand of Ch, the Tale of Gamelyn, which is not usually considered Chaucerian, was inserted on the remaining blank folios of quire 3 and the added quire 4 after the rest of Ch had been copied.Its later insertion is confirmed by the use of blue ink for initial capitals only in this tale and the Parson's Tale.Hence in Table 1 Gamelyn appears as the last tale.
Whether the scribe left the blank in quire 3 because he was uncertain whether to include Gamelyn or because he hoped he might find a copy of it is uncertain.Quire 12 is also irregular in that the scribe left the rest of the quire blank after the incomplete Squire's Tale.This blank was partly filled by the inclusion of Hoccleve's poem De Beata Virgine which is intro-duced as the Ploughman's Tale.This tale is omitted in our discussion as it was inserted later in a different hand.Some folios in this quire also have a different watermark, but they contain what in the Hengwrt manuscript [Hg], regarded as the earliest extant manuscript, was the Merchant-Franklin link, but was adapted in other manuscripts as in Ch to be the Merchant-Squire link.The anomalous watermarks may reflect some adjustment to the link now present in Ch.We can find nothing further in the codicology to suggest any uncertainty about the tale order during the production of the manuscript, although Ch's order of the tales is unique.
One further point in the codicology may be recorded: the text ink changes from a lighter to a darker shade at fol.l01 v in quire 6.This ink change is the only such change in the canonical tales in Ch we have detected, and it falls at the junction of the tales of the Summoner and the Clerk.Our research shows that the greatest shift in the proportion of one spelling to another between any two consecutive canonical tales consistently occurs at this point.This finding may be illustrated by the variation in the occurrence of crossed h as an alternative to uncrossed h in the table below.
Table 1 -Occurrence by tale of crossed h per thousand characters in Ch.

I.
1 Note The Cook's Tale and all links except for the Wife of Bath's Prologue are omitted as is the Ploughman's Tale; the Tale of Gamelyn is placed last to reflect its late insertion.
In this example it is probable that the scribe at first took over crossed h from his exemplar but soon abandoned this form.However, after what may have been a gap in the copying of this manuscript after the Summoner's Tale, he began again at the Clerk's Tale by following his exemplar more closely with greater use of crossed h, but then abandoned crossed h once more so that it rarely occurs in the latter part of the manuscript.
We can distinguish two sets of spellings in Ch.One set contains exam- Rickert's discussion (1).Before fol.86 v , the final folio of quire 11, the folios either have a head in black ink in large textura or have no head at all, though some contain a scribbled head in dark ink inserted at a later date.
Heads rubricated with either blue or gold paraphs first appear at fol.86 v and continue with certain interruptions to the end of the manuscript.
Quire 20 has a rubricated head only on its final verso, and quire 30 a rubricated head on three of its pages.Heads are scribbled on the remaining pages of quire 20 and on three pages of quire 30 whose remaining pages have no head; quires 27 and 28 vary between no head at all and a scribbled one (2)."Narrats", marking the beginning or the narrative part of a tale, in the same style as the rubricated heads occur ten times in total from quire 12; but "narrat" is scribbled in the margin of fol.227 in quire 30 marking the Monk's Tale.There is no "narrat" to the Squire's Tale in quire the tales of the Man of Law with its endlink and of the Squire were copied in immediate succession, although it was recognised that this might not be their final order in the manuscript.Smith (1985Smith ( : 241, 1988: 62) : 62) has noted that a change in spelling possibly coincides with the change in ordi-Nordic Journal of English Studies natio at fol.86 v .Our research into the variations in spelling within Ha 4 confirms, independently of the codicological analysis, that this is the case: the pattern that we see in the spelling data indicates that a change of exemplar has taken place after quire 20 and that quires 12-19 were copied later than quires 1-11 and 20, for when taking the whole text into consideration and allowing for a process of "working in" at the beginning of both stints, we can distinguish two spelling systems from the proportional usage of spellings for common lemmata.
Table 2 illustrates the distribution of selected spellings extracted from our ordered scribal profile.Among the spellings that differ significantly in their proportions between the two stints are 'cowde', -ake(þ, 3), nat, seylseyä, jey, two, þeer(e), and þou characteristic of quires 1-11 and 20, against 'couje ', -aki(p, 3), not, say/sayd, jay, tuo, per(e), and pow characteristic of quires 12-19 and 22-38.None of these spellings is especially significant dialectally.We have also included in Table 2 the spellings of FIRE, LIFE, and OWN (adjective), though these lemmata do not occur in quire 20, for the forms fuyr(e), lif and oughne are practically exclusive to the second stint.Other spellings which are not included in Table 2 but which distinguish the two stints are ey-in AIR, -ey p-in HIGH, and the endings -ud, -us, and -ur characteristic of the first stint, compared with ai(e)r for AIR, a-and to-forms of BEFORE, eo-in EARTH, y-in EVIL, and -ie-in HEAR characteristic of the second.Finally, some graphemic evidence confirms the division, since tailed d, for example, is more common in every tale in the first stint than in any tale in the second and is especially characteristic of the Miller's, Reeve's, and Squire's Tales.Whereas the Ch scribe in his first set translated some spellings in his exemplar, Ha 4 's scribe was influenced by the spelling systems of both his exemplars.Single citation marks surround a type of spelling.

Conclusion
Spelling may not be useful for textual purposes unless it shares certain boundaries with features of a manuscript's codicology, for the interpretation of a spelling profile without outside support is complex and liable to error.This is because a scribe's spelling is never totally uniform, and it is that variation that allows us to deduce how many exemplars the scribe had.It is necessary to consider the entire text, preferably a text of some length, and to start from a high number of lemmata since our findings are bolstered when patterns are found to recur.It is helpful, as with the Canterbury Tales, to have a text which is extant in many manuscripts as that allows comparison of spelling profiles across other manuscripts since that may give us some help in determining what base a particular scribe started from.
Our analysis of Ch indicated that there were two sets of spellings in the canonical tales and that Gamelyn had a quite separate set of spellings.The occurrence of crossed h in Gamelyn shows that the scribe could reflect the spelling of his exemplar extensively even when his procedure in the rest of the manuscript was to change them gradually to uncrossed h.This suggests that Gamelyn was copied from a separate exemplar with relatively many examples of crossed h or, at best, that the Ch scribe's exemplar contained Gamelyn in a spelling which did not match the rest of that exemplar, thus causing the scribe to adopt a different strategy for that tale.Of the two major sets of spelling, one revealed a fairly rapid process of translation, by which the scribe abandoned the spelling of the exemplar to his preferred forms.But this process occurred twice in an almost identical way suggesting there was a break in the copying of Ch.The second set of spellings shows the scribe gradually adopting some spellings characteristic of Hg in preference to his own.That means he sometimes translated what was in his exemplar and at other times adopted spellings found there.Paradoxically, the existence of these two contrary processes strongly suggests the scribe used a single exemplar, for they show a consistency of approach which argue against his use of more than one exemplar.
This conclusion is directly opposed to that proposed by Manly and Rickert (1940, 1: 85-91, 2: 487) and accepted by Owen (1991, 77-9), for they saw Ch as a manuscript copied from numerous exemplars and arranged in a unique tale order, although Manly and Rickert did suggest that its text was the end result of several layers of copying (1940,1: 90).Nevertheless, they also recognised that Ch preserves early material high in authority, and they describe its exemplars as "independent" in some tales, and in others as variously affiliated with such "good" early manuscripts as Hg, British Library MS Additional 35286, Ellesmere, and Cambridge Univer- Scribes later in the fifteenth century may have had more than one complete manuscript to act as exemplar, but the presumption is that they usually worked from a single exemplar except possibly for some special material like the Tale of Gamelyn, though that may have existed in a separate booklet.
It might also be noted that the preference shown by Ch's scribe for some older spellings such as nat indicate that the development of a standard English spelling system was not as straightforward as some scholars like that of crossed h and includes: tailed vs plain d, tittle vs 0 on any graph, crossed double I vs uncrossed single I in ALL, SHALL, and WILL, Tironian nota vs and, avs o before w+consonant, and her vs here irrespective of meaning.All show a reduction of the former for the latter spelling in the two stints starting with the General Prologue and Clerk's Tale respectively.These spelling shifts reflect the process of "working in" which often takes place at the beginning of a stint of copying, and both ultimately result in the final part of each stint showing a more or less identical spelling system in these features.This congruence of spelling with codicology argues against a scribal change of exemplar at any point during Ch.Rather, it points to an interruption of some duration in the progress of copying before the Clerk's Tale was begun that led the scribe to repeat the process of progressive translation found in the early part of the manuscript.A second set of spellings indicates a shift in the proportion of spellings for a common lemma throughout the roughly 20,000 lines in Ch and includes not vs nat for NOT, the synthetic superlative final -0 vs final -e, and saydlsaid vs seydlseid.The shift from the former to the latter spellings is gradual, but continuous across the text ink change, though irregular between successive tales.Their slow and irregular movement argues for these spellings becoming accepted gradually by the scribe, for a spelling such as nat, which is characteristic of Hg, is unlikely to have been part of his own system given the accepted dating of Ch.Worlds of Words-A tribute to Arne Zettersten 99 The evidence of British Library MS Harley 7334 Ha 4 , a vellum manuscript, is written in one ink throughout by a single scribe, who also wrote Corpus Christi College, Oxford, MS 198 [Cp] of the Canterbury Tales.Both may be pre-1400, though which came first remains disputed.Its sole contents are the Canterbury Tales including the Tale of Gamelyn.It originally consisted of 38 quires, though quire 21 is now lost.The standard quire consists of eight folios, though there are three exceptions.The first is quire 9, a quire of six containing the conclusion of the Tale of Gamelyn.This tale begins at the top of fol.59 r , the third folio of quire 8, after the Cook's Tale finished prematurely at the bottom of fol.58 vwith the couplet A4413-14 on a single line, thus omitting eight lines of this tale.There is no rubric between the Cook's Tale and that of Gamelyn, but a scribble probably in another hand in the lower right-hand corner of fol.58 v reads Icy come[n]cera le fable de Gamelyn.Gamelyn finishes 27 lines down the verso of the last folio of quire 9, leaving the last lines blank, and the Man of Law's Prologue commences on the first folio of quire 10.So it is probable that Gamelyn was inserted in the manuscript later, though the occurrence of six blank folios in quire 8 suggests the scribe expected to get something to fill them.Quire 19 is the second anomalous quire.This quire was reduced from a quire of eight to one of six by the removal of two folios between fols.l46v -147 r .The two folios were probably removed to prevent the occurrence of empty folios in the manuscript because there was not enough text to fill them, for the Squire's Tale which was now to follow in quire 20 was already written(Manly and Rickert 1940, 1: 224 and Blake   1985: 68).Fol.l48 v , the final folio of quire 19, is ruled to receive the 38 lines of text characteristic of Ha 4 , but a line is left blank on either side of both the one-line explicit to the Merchant's Tale and the two-line rubric after the Merchant-Squire link[E2419-40 and Fl-8].In addition, the explicit to the Merchant's Tale is the only explicit after fol.86 v that is not immediately followed by an incipit.It is significant that the Host names the next narra-tor as Sir Squire at Fl, although Sir is otherwise found only in those manuscripts such as Hg that have Franklin rather than Squire.The blanks are a device to fill up the frame, for no spacing surrounds other rubrics in Ha 4and the text has been spread out generously to make this page seem complete.The third anomalous quire is the final one of two folios which contains the end of the Parson's Tale and Chaucer's Retraction, which complete the poem.Manly and Rickert (1940, 1: 223)  distinguish two styles of ordinatio in Ha 4 , which they attribute to two rubricators, the first being responsible for quires 1-11 and 20 as well as a few folios of quire 27.This allocation reflects the following features which we have developed fromManly and 20 (3).Ha 4 is ruled throughout to receive rubrics within the text-space.At first the scribe includes an explicit only in the text ink in a one-line space, but by the Reeve's Tale he modifies this strategy by including the explicit in the margin.An explicit and incipit in rubric in a two-line space becomes the norm with the junction of the Man of Law's Endlink and the Wife of Bath's Prologue on fol.86 v , except there are unfilled spaces for ru-brics in quires 27-28 and 30 where directions to a rubricator are scribbled in the margin.There is no rubric in quire 20; the rubric after the Merchant-Squire link occurs at the end of the preceding quire as described, and the Squire's Tale fails to finish within quire 20.In other words, quires 1-11 and 20 share one style of ordinatio against the other quires with quires 27-8 and 30 fitting the second style although their rubrication is incomplete.In addition, the signatures of an original series which have survived cropping mark the first eleven quires, including the irregular quire 9 with the end of Gamelyn.In other early manuscripts of the poem like Hg and Cp the Squire's Tale comes immediately after the Man of Law's Tale and this is where one might have expected it to appear in Ha 4 .But in a change of order it was decided to insert it after the Merchant's Tale.In Hg there is no link between the Man of Law's Tale and the Squire's Tale, though a blank between them was left suggesting uncertainty as to how they would be joined together.Cp, however, contains the Man of Law's Endlink which acts as the link to the Squire's Tale with the Squire named as the next narrator at B 1 1179.A version of this endlink appears in Ha 4 , omitting B : 1175 and B^lSó^O, and the next narrator is named as the Summoner at BU179, although his tale follows in no extant manuscript.No rubric or space for a rubric marks the beginning of the endlink, but it is made to fit into the prologue-tale-prologue pattern by its explicit, in the second ordinatio style, which announces the end of the Man of Law's Tale.This sequence and organisation suggest that the Squire's Tale was originally designed to follow the Man of Law's Tale in Ha 4 .If so, then the evidence we have considered of the junctions of quires 11-12 and 19-20 suggests that

sity
Library MS Gg.4.27.The excellence of Ch's text is confirmed by Peter Robinson who assigns it to his O group -that group of manuscripts which individually stand in an independent, close relationship with the archetype of the tradition(Robinson 1997).It is extremely unlikely that the Ch scribe would have access to the complete text of the poem in the shape of numerous early exemplars in the latter part of the fifteenth century, a point which argues in favour of the scenario we propose in this paper.A scribe at this late date, especially if he worked outside the London-Westminster area, as has been suggested(Manly and Rickert 1940, 1:    90-1), is unlikely to have had access to many exemplars.The position of Ha 4 is different.Once again the spelling system fits well with the codicology of the manuscript, but its interpretation is different.In Ha 4 , which Manly and Rickert similarly saw as a manuscript copied from numerous exemplars and arranged in a unique tale order, we detect the existence of two spelling systems, which we feel represent two stints of copying from different exemplars.The first stint consists of quires 1-11 and 20, and the second of quires 12-19 and 22-38, for quire 21 is no longer extant.Both stints represent continuous copying because quire 20 was copied before quires 12-19.But whereas Ch is a manuscript from the third quarter of the fifteenth century, Ha 4 is an early manuscript and probably written before Chaucer's death in 1400(Blake 1997).So what is meant by exemplar for a Canterbury Tales manuscript at this period is very different from what it means for later manuscripts.There is growing evidence to suggest that some pre-1400 manuscripts were being copied at the same time in the London-Westminster area and it is possible that booklets or at least tranches of text were passed among scribes, possibly after passing through Chaucer's own hands during which process he made alterations to the text and the order of the tales (Blake forthcoming).The allocation of the Squire's Tale to a position later in the poem is one such change made en courant.The tranches of text themselves may have been prepared for Chaucer by different scribes as the poem was being composed and taking shape.The codicology and spelling systems in Ch and in Ha 4 , therefore, coincide to suggest a considerably simpler and rather different textual history than that advanced by Manly and Rickert on the basis of their study of numerous textual variations among the manuscripts.The very earliest manuscripts may have been in production at the same time and the scribes exchanged parts of the text with one another, though this process was clearly supervised by someone, almost certainly Chaucer himself.

Table 2 -Distribution of selected spellings in Ha 4 .
Tale and all links except the Wife of Bath's Prologue.