This is an electronic version of Kiernan, “Part Three: The Reliability of the Transcripts,” and the “Conclusion” to The Thorkelin Transcripts of ‘Beowulf’, Anglistica XXV (Copenhagen: Rosenkilde and Bagger, 1986), pp. 97-151.

Part Three: The Reliability of the Transcripts

A complete collation of Thorkelin A and B throws a rather disquieting light over the so-called received text of the poem. Among other things, the collation reveals that Thorkelin A, made by a hired copyist ignorant of Old English and of insular script, is far more trustworthy than B, made by Thorkelin himself. It reveals also that B is not an objective copy of the manuscript, as it purports to be, but an edition-in-progress, which includes a randomly executed conflation of the manuscript and A. Apparently the manuscript was in worse condition after A made his copy than Thorkelin was willing to show in B. In any case, Thorkelin's methods tend to vitiate the evidential value of his own transcript. Agreement between A and B is a useless fact if we cannot be sure whether B is based on A or the manuscript. With a few exceptions, Thorkelin obliges us to relinquish B as a textual authority for the poem.1

Neither Zupitza nor Malone fully explores the reliability of Thorkelin A and B. Zupitza does not comment on their relative merits, and where their readings differ, he merely records in his transliteration (sometimes with emendations and restorations) the more plausible reading. As Malone rightly observes, “this procedure of his could obviously be improved upon.”2 We need to know more about the quality of each transcript before according either one such extraordinary authority. Malone's solution was to make complete collations of each transcript with the part of the manuscript that remained intact. “Systematic comparison of the manuscript text with each transcript,” Malone reasoned,“would bring out the characteristic features of A and B; in particular, it would reveal the mistakes of the two copyists, and when it had been determined what kinds of mistakes were often or habitually made by a given copyist, one would have a set of objective criteria for weighing the accuracy of those readings which now are on record only in A and B, the corresponding manuscript readings having perished” (p. 240).

These two collations led Malone to the conclusion that the B transcript, the one made by Thorkelin himself, was more accurate and reliable than the A transcript. As he puts it in the “Introduction” to his facsimile edition:

The B copyist cannot be trusted when it comes to distinguishing between d and eth; here he made a very large number of mistakes. Otherwise, however, a given mistake does not appear in his text in a large number of cases, whereas A made a great variety of mistakes many times over. This contrast betwen [sic] A and B is in no way surprising. B had a great advantage over A in that he understood, after a fashion, what he was copying. One would expect the B text to be more accurate than the A text, and so it is, apart from d and eth. (p. 25)

The preceding collations lead to very different conclusions, however. Though B in most cases does not make the same mistakes many times over, while A in some cases does, B makes more mistakes, and a greater variety of them, than A. Moreover, B's departures from the manuscript readings are very often emendations and restorations. A, on the other hand, makes a few errors many times over, but with such predictability that they can be confidently corrected even without access to the manuscript, as Rask and Grundtvig admirably showed in the first days of Beowulf textual scholarship.3

The most disturbing results of this new collation are the signs that B frequently relied on A, rather than on the Beowulf manuscript itself, and that many B readings were made when Thorkelin no longer had access to the manuscript. The superiority of A over B clearly begins to emerge as early as the third folio, 132 recto. After proofreading, the A scribe left two errors on the page. He omitted the word mearc r11 (by a kind of haplography with the following word, mære r11) and he wrote cames for caines r15, a mistake the Beowulf scribe also made but later corrected. After repeatedly proofreading his work, on the other hand, Thorkelin left no fewer than fifteen errors on the page, including the same omission of mearc. He first wrote cames for caines, too, but later corrected this mistake he at first shared with A. Moreover, the conjectural reading gi[ga]ntas in the last line shows collation with A rather than with the manuscript, since the partly conjectural -gantas belongs on the verso — a slip Thorkelin would not have made if he had reference to the manuscript. The following preliminary analysis centers on the relative reliability of the two transcripts, because the evidence consistently indicates that scholars have been relying on the wrong transcript, Thorkelin's, rather than on that of his hired scribe.

The A Transcript, Ny kgl. Saml. 513 4°

The new collation of A with the manuscript, in addition to isolating all of A's copying errors in the sequence in which they occur, tells us a good deal about A, the reasons for his errors, and the general quality of his work. Malone believed that “Many of the mistakes [of A] grew out of his ignorance of the language; many others, out of his ignorance of [Old] English handwriting” (p. 5). A's ignorance of the language was actually an asset, impelling him to avoid fruitless guesswork and to restrict his efforts to copying what he could see. With the exception of some metathetic errors,4 very few of his mistakes can be safely attributed to his ignorance of Old English.

Indeed, A acknowledged this limitation to a fault, sometimes not even attempting to reconcile strings of minims into vowels and consonants. He wrote, for example, m for in in drinc and cyning, blunt non- spellings more plausibly attributed to candor than to his belief that Old English might have words without vowels. He no doubt expected Thorkelin to decide whether four consecutive minims should be read as nu, un, ini, im, mi, nn, uu, iiii, and so on. If he often misconstrued the letters, he seldom miscounted the minims.5 With poor or variable lighting, which A surely had to contend with in eighteenth-century London, only an Old English scholar could be expected to distinguish consistently between serifs and hair-line penstrokes in contiguous strings of minims. These non-spellings in A, in any event, are not a major source of confusion for us today. In order to judge the reliability of the two transcripts, we must carefully differentiate between consequential and trivial errors.

Nearly all of A's mistakes, including his problems with minim-strokes, are best explained in terms of his ignorance of insular script, a void that should have been partly filled by Thorkelin before A was left on his own. It is evident that A set to work entirely without Thorkelin's supervision, without even fundamental instructions, for he is patently unfamiliar with insular letterforms like d, eth, g, r, low s, and t, ligatures like æ and ec, and the runic characters thorn and wynn. As a result, insular script confused him at first, and led him to make many trivial and quite needless errors.

The best illustration, in fact, that A's real problem was with insular script, not Old English, occurs where his copy records the change of scribal hands in the Beowulf manuscript. The A copyist eventually learned the first scribe's handwriting so well that his transcript is error-free on the last full page, 172r, of this scribe's section. But on the overleaf, the first page of the second scribe's section, A makes no fewer than twelve mistakes, including several nondescript efforts to mimic the second scribe's strange new letterforms. Once we recognize that A's relevant troubles were paleographic, not linguistic, the nature, frequency, and distribution of his errors can be judged in a clear light. The accuracy of his copy improves rapidly. As might be expected of an untutored beginner, A presumed from a knowledge of his own alphabet that the manuscript he was copying had variant forms for the letters p (thorn and wynn), d (eth), and c (round t), and at first he tended to copy them in ways most familiar to him.6 For this reason, wynns show up as p's, and eths become d's, but the errors are very rarely reversed.7 In the course of his work A becomes an expert in the version of insular script used by the first Beowulf scribe, but new naive mistakes arise as soon as the second scribe's version appears.

Compared to Thorkelin, A made a relatively small variety of mistakes, though he indeed repeated them an exasperating number of times. To use the A transcript efficiently, scholars must of course be keenly aware of A's characteristic difficulties with insular script, but they must also be aware of his overall reliability, including the predictability of his errors. Malone provides long lists of words from A in which p was corrected to wynn, p was written for wynn, p was written for thorn, thorn looks like p; and (with original p's noted parenthetically) in which thorn was written for wynn, wynn for thorn, wynn looks like thorn, and thorn looks like wynn. Without appropriate commentary, the lists give the impression that Thorkelin A is decidedly untrustworthy, riddled with all kinds of errors. What is important about his transcript in this respect is that A confused two unfamiliar runic letters for the familiar p in the first 918 lines of the Beowulf manuscript. For example, of the ninety-eight cases Malone lists (p. 6) where A wrote p for wynn, eighty-eight come from the first 918 lines. Of the remaining ten cases, nine come from the part of the manuscript copied by the second scribe, who himself sometimes makes a p-like wynn.8

Similarly, of the twenty-four instances Malone finds of uncorrected p for thorn, ten come from fol. 141, four from fol. 142, and three from fol. 145v, while all twenty-four occur between fols. 139 and 146, and all before line 918.9 In short, A overlooked a few pages when he went over his transcript to correct p to thorn. Half of the cases cited exhibit what Malone calls “the p-like thorn, a letter form which answers to a thorn in the Cotton text but is more like a p with a heightened vertical stroke” (p. 7). Malone says that “in classifying this letter form as a thorn of any kind I am giving the A scribe the benefit of the doubt.” It seems superfluous, therefore, and potentially misleading, that he lists eighty- two occurrences of the “p-like thorn”, seventy-nine from the first 918 lines of the poem. The three remaining cases after line 918 are not in fact p-like.10 In a sense all eighty-two cases are imaginary, since “a p with a heightened vertical stroke” is a thorn. The A copyist merely refined his version of the letter in the course of his work, as he also refined such letters as d, eth, g, and t, the early forms of which Malone does not itemize.

Even less disturbing than A's natural confusion of wynn and thorn with p are his frequent mistakes of thorn for wynn, wynn for thorn, thorn-like wynn, and wynn-like thorn. It is true that A often confuses the two runic letters, even after line 918, when he learned to distinguish them from p. Yet it would be surprising indeed if he had been confident that the two runes, identical except for the ascender, were two distinct letters rather than variants of the same mysterious one. At one point (A31.17-32.12) he experimented with a wynn that had no ascender at all, but soon began drawing it again with a small ascender (A32.13). A often registered his belief that the two letters were one by producing a hybrid form, drawing both runes with medium-sized ascenders. He was evidently bewildered by the long ascender, three times crossing it like a t on the first page of his transcript (thus confusing it with the abbreviation for “þæt”). In the long run he seems to have lessened his bewilderment by opting for a short ascender. This view explains, at least, why he wrote thorn for wynn twenty-six times in the first 918 lines, but only nine times thereafter.11

When he distinguished them, the A scribe drew wynn with a short vertical ascender, and thorn, of course, with a long ascender. His “mistakes” arise from his regular use of ascenders of variable lengths in between. Many cases Malone cites as mistakes of wynn for thorn have rather long ascenders, and could easily be taken for thorns, or at least for wynn-like thorns. For example:

Figure 1

[Fig. 1: gecyþed A24.1, morþor A35.3, þanan A56.512]

As Malone says, “Often A's letter form may be taken either for wynn or for thorn. In such cases, if the Cotton text has wynn I reckon A's letter a thorn-like wynn; if the Cotton text has thorn, I reckon A's letter a wynn-like thorn. That is, I give A the benefit of the doubt” (p. 8). Where the Cotton text is lost, a knowledge of A's consistent idiosyncracies involving p, thorn, and wynn, like those involving minims, makes confident restorations possible.

In the course of his transcript, A developed a curious problem with the letter r, which in insular script is made with a descender (like p and wynn) and an open loop, rather like n, but normally with a distinct off- stroke, the opposite of p and wynn. The frequency and distribution of his errors involving insular r are significant. Since A had not yet distinguished wynn or thorn from p before line 918, cases of r for wynn and r for thorn should really be counted with A's mistakes of r for p. In this part of his transcript, A often wrote r for p (and for the two p-like runes) and vice versa, but in many cases he corrected his mistakes when he came to realize that insular r must be a separate letter.13 Later, after line 918, he wrote r for wynn forty-six times, but wynn for r only five times. Moreover, once A learned to distinguish wynn and thorn from p, his mistakes confusing r for p and p for r all but disappear.14 From this perspective, A's only significant error involving r was mechanically writing r for p in the first 918 lines, and r for w thereafter. The psychological explanation seems to be that A never became comfortable with insular r, which must have always remained for him a strange conglomeration of several letters. After distinguishing one strange letter (wynn) from p, he merely substituted wynn for p in his later mistakes. His many corrections of all kinds of mistakes involving r suggest that he may have eventually realized what he had done, but his many oversights show too that his later proofreading in this respect was neither thorough nor confident.

The A scribe frequently made a number of other mechanical mistakes, most of which he failed to correct in later proofreading. As in the earlier cases confusing r, wynn and thorn for p, or in the later cases confusing r for wynn, these mechanical mistakes involve letters resembling one another: d and eth; c, t, and sometimes e; and low insular s and f. Malone says that A “found it hard to distinguish between d and eth” (p. 10), but the evidence does not support this view. Of the twenty-three cases he finds where eth appears for the correct d,15 three construe the accent mark over the preceding vowel, a mark that extends over the d as well, for a cross-mark: Figure 2

[Fig. 2: (ádfære 194v17) (gód 182r3) (herepád 180r6)]

A did the same thing with hád 158v10 but later corrected it. Two other cases Malone cites occur where the manuscript was damaged:

Figure 3

[Fig. 3: (gehydde 179v6-7) (mid 190v6)]

Three others very accurately represent the appearance of the manuscript:

Figure 4

[Fig. 4: (geneðde 185v15) (mid 174v8) (weard 187r14)]

One is an eth (wonreðing 194r1);16 and at least one was altered from d by Thorkelin (-wered, A21:10).17 In short, it does not appear that A found it difficult to distinguish between d and eth. Instead, the mistake he committed many times over is that he failed to cross a d to make eth, perhaps even thinking he was making a correction.18

As Malone says, “Confusion between c and t would be expected in the insular hand, and A often falls into this error”(p. 14). A's mistake was caused by the fact that, in insular script, the two letters are alike in their bow from top to bottom, and differ only at the top, which is flat for t and curved for c. A's most common mistake was c for t, which occurs fifty-two times in the collations, compared to thirty times of t for c.19 At the beginning of his transcript, roughly up to line 1300 (A40) of the poem, A imitated the scribal t by making a c, which he then topped with a small bar. He first drew this cross-bar at a sharp angle, but after line 734 (A24.19) he began to make it horizontal, more like the scribal t. After line 1300, A regularly drew his t in one maneuver, beginning with the horizontal bar and ending with the bow:

Figure 5

[Fig. 5: (hete A6.1) (hetes A35.3) (hetende A54.14)]

In view of the evolution of A's t from a sort of hybrid c to a c-like t to a bonafide t, it is plausible that his common confusion of c for t in the first 1300 lines derived from his initial belief that c and t were different forms of the same letter. He came to distinguish them, however, and in fact corrected a large number of his mistakes, though it is not always possible to determine whether these corrections were made by A or Thorkelin.20 At any rate, up to line 1276, he left this mistake nineteen times; up to line 1939, the end of the first scribe's part of the manuscript, he left it only four times. In the second scribe's part he left c for t fifteen times, but here he had to learn new letterforms.21

A similar explanation accounts for the twenty-eight times he wrote e for c, for he drew e by making a c and adding a horizontal cross-stroke. Nineteen of his mistakes were made in the first 1300 lines. Of the remaining nine, three of the c's are easily mistaken for e's in the manuscript:

Figure 6

[Fig. 6: (æghwylc 160v10) (unc 186r6) (-weorc 198v16)]

Two others are conjectural, from the lost edges of the manuscript (hwylc 174r2, ic 176r9), and one follows a catchword in which c was written correctly.22 In short, A effectively solved his early confusion between e and c.

Malone does not adequately account for the thirty-four cases he finds where A wrote f for s.23 He says that “Confusion between f and long s was to be expected, of course, though the two letters are less like in the insular hand than in later hands” (p. 13). The long s Malone refers to is the high caroline s used almost invariably by the first scribe. A does not confuse f with this letter, but with the low insular s used frequently by the second scribe. Insular f and s are made exactly alike, except for the cross-stroke of f. In the few cases where the first scribe used a low insular s, A observantly interpreted the unfamiliar letter as a mistake for the customary f, and crossed it in his transcript:24

Figure 7

[Fig. 7: (eormenrices 156v6) (se 141v3) (swa 143v3)]

Of the thirty-four cases Malone cites, only four come from the first scribe's part of the manuscript, while thirty come from the second scribe's. The A scribe apparently thought he was making a safe correction whenever he copied low insular s as an f. In like manner, whenever he came across a high caroline f, which is very rare in this manuscript, he copied a high caroline s:

Figure 8

[Fig. 8: (cræft 146r2) (frecne 160v1) (wif 157v16)]

A's late mistakes involving s and f, then, actually show his attentiveness to manuscript forms.25

The only other confusion of single letters that appears with any regularity in the A transcript is e for æ. Malone lists twenty-two cases of this mistake, though several come from areas where the manuscript is gone or damaged.26 Another reflects an awkward scribal correction of nes to næs (164v9). The remainder no doubt derives from the prominence of the e-part of the æ ligature, particularly in conjunction with the second scribe's square a, in which the left side is a minim stroke and the top a light hair-line stroke. Another related error of some frequency is a confusion caused by the similar ways A drew the æ and ec ligatures of insular script. Although A writes æ for ec only seven times, Malone notes eighteen additional cases of æ for e, twelve of which “illustrate once more A's difficulties with the ec-ligature.”27 Three additional cases of æ for e accurately represent the appearance of the manuscript:

Figure 9

[Fig. 9: (gestealla 173r15) (leode 198v5) (sercean 189v20) ]

In first case the conjunction of t and e creates a spurious æ, as does the missing top of l before e in the second case. In the third, A misconstrues a scribal correction of æ to e. His repeated mistakes, in sum, are e for æ and æ for ec, not æ for e.

Though his characteristic errors are quite circumscribed there is still no denying that A makes many mistakes in the course of his transcript. Yet as these illustrations suggest, there is normally a reassuring consistency and even an uncanny reliability in the nature of his errors. Old English scholars do not need confirmation from Thorkelin B to show them where A has confused thorn or wynn with p, insular r with wynn (earlier “p”), t or e with c, d with eth, æ with ec or insular f with s. Knowing these characteristic errors, and realizing that A did not even attempt to figure out what were to him meaningless strings of minims, scholars can safely use his transcript to help reconstruct lost words and letters in Beowulf. Since A did not understand Old English and was unfamiliar with insular script, he did not fall into scribal anticipation, or indulge in scribal emendation or restoration, as Thorkelin himself often did.28 If a reading was gone or illegible, the A scribe, unlike Thorkelin, could move on to the next word with equanimity. The A scribe simply tried, with more success than scholars have yet admitted, to make a good facsimile of the manuscript wherever its readings remained legible to him. By understanding the nature of his errors, textual scholars, even without the aid of Thorkelin B, can confidently correct A's evident errors in readings now lost to the manuscript.

It is, in fact, A's complete independence of Thorkelin, along with his ignorance of Old English and of insular script, that makes him such a valuable witness for the lost readings of Beowulf. A is thoroughly, sometimes painfully, objective. As Malone says, “His method of copying was simplicity itself: he tried to reproduce, letter by letter, what he saw before him” (p. 4), not a bad method for someone attempting to make a facsimile. His objective approach provides an excellent guide to unusual paleographical forms in the surviving portions of the Beowulf manuscript, and so gives us reason to trust his odd readings in the lost portions of the text. Most of A's uncharacteristic, miscellaneous errors are actually meticulous efforts to copy anomalous forms in the manuscript. These objective “mistakes”, while they undoubtedly remind us of A's ignorance of Old English, testify in a more striking and important way to A's reliability as a copyist. His “reliable mistakes” record the unusual appearance of the manuscript in two quite helpful respects: they ingenuously mimic scribal oddities caused, for instance, by false starts, stray marks, and corrections; and they reflect the state of the damage to the manuscript when A copied it.

There are many cases where his characteristic mistakes are justified by the appearance of the manuscript. For instance, a spot of ink on the c-bow creates a spurious e in worca 136r14 (A11.11), while a heavy hook at the top of c and a wrinkle in the vellum creates another one in -rinc 145v14 (A23.15). He was justified in copying f for s in several places: for se 173r20 (A58.14), because of a spot of ink in the manuscript where a cross-mark would go; for sæcce 173v11 (A59.2), because of the strange mark the Beowulf scribe added to his descender; and for cystu(m) 187v8 (A75.5), because of the “cross-mark” created by the top of the following t. Conversely, he was justified in copying s for f in fandod 181r7 (A67.13), because most of the cross-mark has faded; and in efnde 194v15 (A85.17), because of the unusual way the Beowulf scribe corrected -m- to -fn- with an extremely rare caroline f. It is not difficult to see how A miscopied r for w in stow 160r17 (A42.6) and ecgðeowes 187r19 (A74.19), for in both cases the bows are open at the bottom. Other examples are easy to find and indeed could be extended indefinitely. It is seldom a waste of time to compare a mistake in A with the form that inspired it in the manuscript.

The same is true of A's ligaturing of minims, though here it must be admitted that his accurate representation of the manuscript was most likely a matter of chance. It is unlikely in the extreme that in his “mistakes” he deliberately wrote, for instance, m as a separate word (instead of in) nineteen times, as Malone suggests (pp. 11 and 12). It is far more likely that he chose to represent strings of minims in a sensible way, given his ignorance of the language, the condition of the manuscript, and the poor lighting available to him in the eighteenth century. Still, many of his putative errors in ligaturing minims accurately represent the manuscript, either what the scribes themselves wrote, or what weak hair-lines, strong serifs, faded or abraded ink, shine-through, and the like appear to create. Two good examples come from famous cruces in the first scribe's part of the manuscript: A correctly copied manuscript in for what is presumed to be an m in hlitine 154v12 (A35.16), and mi for what is wrongly believed to be ini in wun/dmi 160v4-5 (A42.10). Other cases in the first part of the manuscript where A's ligaturing follows manuscript forms are seemingly ubiquitous.29

The frequency of oddly ligatured minims increases in A's copy of the rest of the manuscript, because the second scribe's heavy, square- formed script tends to obscure the formal connections between minims. Thus there are many cases where A's ligaturing concurs with the manuscript, even though the form is technically incorrect in both places. He accurately copies, for instance, m for in in inne 176v9 (A62.9); in for m in micelne 196v4 (A87.20); n for u in uht 180r18 (A66.17); u for n in demdon 198v17 (A89.16). Perhaps the most obvious evidence that the exemplar, not A's carelessness, accounts for these forms can be seen in the many cases, unparalleled in the first part of the manuscript, where the second Beowulf scribe obscures the ligature between the first and second strokes of h. A is once again copying accurately when he writes (to cite only three cases) li for h in rihta 188r20 (A76.3), ln for hi in hi(m) 184v18 (A72.1) and lu for hi in hi(m) 175v3 (A61.4).

The real virtues of A's work, the facsimile features, remain hidden to us if we compare his mistakes with a printed transliteration of the manuscript instead of with the manuscript itself. Back in the first scribe's part of the manuscript, A's accurate copying of other scribal oddities and anomalies demonstrates how reliable his transcript can be in this respect. From a simple list of some of his miscellaneous errors (for instance, a for o, l for s, d for tl, sc for c, and g for t) one might well get the impression that his copying was signally unreliable. Yet on closer scrutiny these doubts are allayed. He copied a for o in weold 140v1 (A17.11) and mon 154r2 (A35.1) because both o's have straight right sides in the manuscript. He drew l for s in magas 152r6 (A32.12) because the faultily-drawn top of the scribe's caroline s made the letter resemble a cursive l. The way the right side of round t touches the l in the manuscript explains why A wrote d for tl in -setl 153v12 (A34.13):

Figure 10

[Fig. 10: (-setl 153v12, A34.13)]

The A scribe faithfully copied an incipient s in the manuscript when he wrote sc for c in hwylcra 159v14 (A41.14). And he even brings our attention to the crooked bottom of a t in the manuscript by copying g for t in to 161r20 (A43.12).

Another list of miscellaneous errors from the first scribe's part of the manuscript is quite disheartening (n for a, u for a, n for i, dm for d, ec for e, te for æ, g for t, e for æ, ae for a, for æ, oi for i, ea for a) until one looks at the sources of these errors in the Beowulf manuscript. Then it is apparent that they all derive from A's naive, but careful rendering of scribal corrections. Not knowing Old English, he copied what these words look like, instead of what they were supposed to look like. He has n for a in wendan 134r5 (A7.9) because the a is written on top of an original n; u for a in þearf 134r19 (A8.4) because the a is over a partly erased r that now resembles a u; and n for i in hige 135v15 (A10.15) because the top of g (written over an original n) connects with i, creating an n-like character:

Figure 11

[Fig. 11: (hige 135v15, A10.15)]

What seems to be a spurious mdm for d in -bidan 138v16 (A15.9) — is actually A's copy of a partially erased m in the manuscript. The spurious c, ec for e in sceaðan 140v14 (A17.18), comes from the curled e- tongue in the superscript emendation, made by the second scribe (the first scribe makes his ec-ligature in a similar way). The erroneous te for æ in hwæt 142r7 (A19.5) clearly mimics the Beowulf scribe's false start (hwet?). He copied g for t in hruntinge 166r20 (A49.18) because the t was at first a g with its descending loop erased, and e for æ in næs 164v9 (A47.12) because the æ was at first an e with the a-bow squeezed in later below the line. If one ignores, as A did, the dot beneath the line, another case of n for i in side 157r5 (A38.11) is again a faithful replication of the manuscript:

Figure 12

[Fig. 12: (side 157r5, A38.11)]

A cannot be fairly blamed for failing to comprehend the inconsistent use of the signs of insertion and deletion in this manuscript. He consistently copied both the deleted letter and the inserted one. He wrote ae for a in fyrena 149r6 (A28.13), where the Beowulf scribe wrote a above an original e. Likewise, A wrote for æ in mæste 153v5 (A34.9), oi for i in in 158v15 (A40.11), and ea for a in guman 160v12 (A42.15), where the Beowulf scribe used a point to delete o, o, and e respectively. And A wrote ea for a in hraþe 171v20 (A57.1), where the Beowulf scribe used a comma to delete an original e. As we shall see, Thorkelin's scribe is more helpful than Thorkelin is in his representation of scribal corrections.

In the part of the manuscript written by the second scribe, A makes a number of odd mistakes attempting to come to terms with the radically new letterforms. On 172v, the first page of the second scribe's work, he was notably baffled by the curved form of y, unlike any he had previously seen in the manuscript. It is improbable that A immediately noticed the change in scribal hands, and so it is natural that he misinterpreted the strange new cy combination in cyðan 172v4 (A57.14) for the first scribe's ec-ligature, which bore at least some resemblance to it:

Figure 13

[Fig. 13: (cyðan 172v4) (mece 172v2)]

He apparently had similar problems with the gy combination in gyfen (line 11); whatever he wrote was later erased and replaced (almost certainly by Thorkelin) with an ungainly g. A also exhibits on this page some difficulties recognizing ligatures in the less elegant script of the second scribe, and in view of the manuscript forms it is not surprising that he copied ean for em in hem- 172v8 (A57.16) and a nondescript pair of strokes for n in -cende 172v9 (A57.16). The surge in frequency of A's errors at this point in the manuscript is directly attributable to the change in scribal hands.

Some strange mistakes of A's (for example, a for ti, h for s, aa for a, fi for N) are properly understood only by reference to the manuscript, since they were palpably induced by the new handwriting. Thus the A scribe read a for ti in gemeting 174r2 (A59.7), because the top and bottom of t touch the i, creating an “a”. In the superscript on 173v4 the s in side (A58.16) looks like a cursive h leaning to the right. The double a shows up in ywan 177v4 (A63.9) because the off-stroke of a connects to the bottom of n, while the on-stroke of n connects to the top of a. The second scribe's capital N in Ne 193r10 (A83.15) was copied as fi by A because the first two strokes of the letter make an insular f lacking its top bar, and the third makes an i. Each of these cases ironically displays A's attentiveness as a copyist along with his ignorance of Old English.

In the same light, many mistakes A makes in transcribing the second scribe's part are, like those from the first scribe's, not really mistakes at all, but ingenuous copies of scribal corrections. Seemingly spurious letters, ch for h in fah 179r9 (A65.5), eo for o in forð 180r14 (A66.14), and ld for d in dæd 192v6 (A82.17), merely retain letters meant to be deleted. The eth after uru(m) 189A(197)r4 (A76.15) is the superscript sign the Beowulf scribe used to show where sceal (in the left margin) was to be inserted. The t before c in aglæcean 187v2 (A75.2) is the vestige of an original g whose bottom loop was erased, but not its top. Other not entirely clear corrections in the manuscript account for other naive mistakes in A. Another partly erased g left what A reasonably interpreted as a t for the first e in forhealden 182v19 (A69.19). Here A probably thought that a t had been written over an original e. On 192r20 the Beowulf scribe first wrote n for u in un and only partly erased the top of n when he made his correction; A consequently copied a for u in un (A82.12). Elsewhere in the manuscript, partial erasures of mistakes by the second scribe explain why A has æ for e in -réc 189A(197)r5 (A76.16), a for u in maðmum 194v3 (A85.11), and p for b in burh 196v6 (A88.1). A's errors required close scrutiny of the manuscript.

A great many of A's apparent mistakes in fact give us a clear idea of the state of the manuscript in the late eighteenth century. These cases often show that the current appearance of the manuscript is at least 200 years old. For example, a dark crease in the vellum explains why A wrote thorn for d in mid 147A(131)r5 (A25.4). A tear that now ruins the right side of d shows why he copied t for d in another case involving mid 196r20 (A87.16):

Figure 14

[Fig. 14: (mid 196r20)]

Another tear in the vellum spoiled the cross-stroke in fæhð 183r20 (A70.10), leading A to transcribe the cross as a macron. Similarly, A omitted the i in m[i]ddan 189r1 (A78.1) because it was obscured by a tear running straight through it:

Figure 15

[Fig. 15: (m[i]ddan 189r1)]

We know, too, that the spots of ink beneath fæþme 172r3 were erased before A copied, for the erasing inadvertently included the descender of thorn, which A then copied as a b (A57.2). He copied r for n in hand 172v1 (A57.12), taking a spot of ink beneath the first minim for a descender. And he reversed his mistake when he copied n for the second r in broðor 188r7 (A75.16), because a hole in the vellum shortened the descender.

Similar cases can be cited showing that shine-through that now disfigures the manuscript must have been shining through the vellum in the eighteenth century as well. For this reason A copied p for o in hors 161r1 (A43.2). On 162v17 the descender of w in wine r17 bled through the page and led A to believe that an r was written over o in þon(ne) (A45.7):

Figure 16

[Fig. 16: (þon[ne] 162v17)]

Even A's f for g in hige 174r1 (A59.6) is understandable in view of the heavy shine-through distorting the g.

In many areas of the manuscript, fading must have been as advanced in the eighteenth century as it is today, to judge by A's misreadings. With the left side of a as indistinct as it is now, for instance, and with the right side of the accent as dark, A would have seen what today resembles a dotted i for a in hál 136v4 (A12.1). In fact, a multitude of A's misreadings are justified by the way fading today transmogrifies a wide variety of letters in the manuscript. It is no coincidence that the same mistakes A made could be easily repeated today by someone else ignorant of Old English and insular script. Some brief descriptions of how fading has altered some of these letters will help make the point: t looks like c in fæste 147A(131)r18 (A25.91) because the left side of the t-bar is faded; the h-ascender is faded in hwearf 181r1, which along with the faulty ligatures leaves something resembling a u (A67.10); the e- head of æ is so faint that æ appears to be a in wæsse 181r14 (A68.1):

Figure 17

[Fig. 17: (wæsse 181r14)]

In two comparable cases, only the front of the e-head of æ is dark in þær 188v1 (A76.4), leaving behind an a and an r with an ascender; and only the top and front of the e-head of æ in hæfde 183r15 is clear, making the æ look like ac (A70.7).

The straight, hairline tops of the second scribe's a's were especially vulnerable to later fading, which when it occurred invariably left behind “u's”: see, for example, eorla 181v21 (A68.16), þa(m) 183r17 (A70.8), syðða 194v6 (A85.12), and maððu(m) 195r1 (A86.1). In a couple of places where fading attacked instead the right side, A was left with distortions he could logically transcribe as e's — once for the second a in aras 186r17 (A73.16), and once for the a in hycgendra 186v21 (A74.9):

Figure 18

[Fig. 18: (hycgendra 186v21)]

Often the fading must be considered in conjunction with the damaged edges of the manuscript, and the gravity (and even the reality) of A's errors must be evaluated accordingly. Who could blame him for writing c for k in kyning 181v19 (A68.15) if the vertical stroke was as faded for him as it is for us?

Figure 19

[Fig. 19: (kyning 181v19)]

Elsewhere, the 7e part of 7eow (i.e., ond eow) 196v3 has faded in such a way that the vertical of 7 and the e-head create a tolerable f (A87.19). A few lines later (v7) some fading on the w-bow similarly explains why A wrote f for w in welan 196v7 (A88.1). Fading together with charred vellum at the end of 180r10 makes the final e resemble an uncial s in healfe (A66.12). More serious damage along the edges explains why A wrote æ for a in fyrnda- 162r3 (A44.6), m for n in sefan 170r15 (A55.3), u for i in riht 181v14 (A68.12), and why he omitted letters in higelac[e] 170r4 (A54.15), [s]iððan 173v1 (A58.14), seolf[a] 196r1 (A87.6), and gre[t]te r12 (A87.12). If the manuscript looked to him pretty much the way it does to us now, it explains, too, why he left deliberately blank the beginnings of such lines as 144v11 (A22.9), 146v5-6 (A24.13-14), and 188v18 (A76.12).

Since a large number of the errors Malone attributes to A in fact come from parts of the manuscript that are gone, a knowledge of A's “reliable mistakes” should help us in reconstructing lost readings. As we have already seen, his reliability helps us understand the state of the manuscript when he copied. In addition to the state of the damage at the edges, it tells us as well that the first page of the manuscript, 129 recto, was in virtually the same condition in the late eighteenth century as it is now. Moreover, his transcript, like Wanley's in 1705, implies that the wear in the lower right corner was most likely caused by use in Anglo-Saxon times, thus supporting the view that the Beowulf part of the Nowell Codex was first used as a separate book.

The A transcript also assures us, by means of its particular errors, that the palimpsest, 179 recto and verso, was in the same relative condition in the late eighteenth century as it is today. Indeed, a mere list of his errors and deliberate omissions provides an excellent guide to problematic readings on this most severely damaged folio. Some scholars have mistakenly believed that the damage is modern, perhaps caused by recent attempts to resuscitate readings on the folio.30 A's transcript lets us see, too, that the first three lines of 180 verso were already rubbed off when A copied and that the last page of Beowulf, 198 verso, was already in the sad state it now is in. A's readings at least provide the limited consolation that the manuscript, apart from the progressive crumbling of the edges before the paper frames of the new binding stopped it, has not appreciably deteriorated in modern times.

One other feature of the A transcript must not be overlooked. There are, of course, many signs of later additions to A. Some of these, like the underlined proper names, the erased interlinear translation, and a few emendations and restorations, were obviously entered by Thorkelin. The later additions made by A himself have been characterized by Malone as later proofreading, the result of a collation A made after copying the manuscript. When Malone first studied A in photostats he concluded that A “seems to have confined his revisions to a few difficult passages.”31 But after he studied the original documents at the Kongelige Bibliotek, Malone changed his mind. He said in his article on “The Text of Beowulf”:

After A had finished copying his text, he went over it, making a multitude of corrections, and in many cases deciphering words which he had previously skipped as illegible. That is to say, A collated his copy with the manuscript text of the poem. We can tell that this is what he did from the fact that the corrections and additions are made with another ink. The words set in the blank spaces nearly always differ from the surrounding text in the size of the letters.32

Malone's comments suggest that A made a systematic collation of his transcript with the manuscript, but this view is not at all supported by the evidence. A's collation was lamentably superficial, confined to a relatively few sections of his text. It appears, moreover, that A never intended to make a full collation. Whenever A deliberately passed over a reading he found difficult to decipher, he left a space far greater than the actual word he omitted. In this way he could later easily locate his omissions without having to make a full collation. A's method seems to have been to leave a gap that included letters he could clearly see, evidently planning to return to them when he had more time to ponder them. Thus blank spaces in A do not necessarily mean that readings in these areas were gone when B copied, only that they were not readily legible. A's work-saving strategy in this regard is a great misfortune for us, for it deprives us of what might have been an extremely accurate transcript of the manuscript.

In fact, A did begin to make a full collation of his work with the manuscript, but for some reason he stopped proofreading after the first few hundred lines. The extent of A's systematic collation with the manuscript can be inferred from the restorations he makes of lines he accidentally omitted in the course of his transcript. As Malone says, “A omitted 16 passages about a manuscript line each in length” (p. 20). Malone divides these omissions into two lists, one showing omissions of full manuscript lines, and the other showing omissions of parts of two consecutive manuscript lines. What does not clearly emerge from Malone's lists or commentary is that A restored the first three, but no more, of his sixteen major omissions and that Thorkelin began to restore the fourth and restored the fifth of A's omissions, but no more. In other words, Thorkelin took over A's collation somewhere between the third and fourth omission, but did not complete the task. The evidence implies that A did not have the opportunity to finish his own collation. One explanation is that Thorkelin wanted to study the poem or work on his interlinear translation, and so relieved A of this duties at this point. Thorkelin's aborted interlinear translation (ending at A6.9), combined with his aborted collation of A with the manuscript, suggests that he decided to make his own transcript of the manuscript before he carried out his collation of A and the manuscript.

There are many other signs of A's subsequent revision of his transcript, but none of systematic collation of his work with the manuscript after he had finished copying. His proofreading was at best sporadic, and for the most part was confined to the first part of his transcript. Thus nearly half of his later restorations are in the first 340 lines of the poem. Although he made major restorations of inadvertently omitted material early in his work, he overlooked some large omissions, too. Later in the transcript there are no major restorations of inadvertently omitted material, only of words he at first deliberately omitted — which he conspicuously marked by large gaps. In fact, between lines 340 and 1804 the only significant restoration he made in one of these gaps was the word scyldinga at line 464 (140v1). From line 1804 to the end of the poem he makes about twenty additional restorations in gaps he had left behind. The rise in frequency of these later readings should no doubt be attributed to A's realization that he had more problems copying the second scribe's part of the manuscript than the first scribe's.33 But these later insertions should not be mistaken as evidence of a systematic collation in this part of the manuscript.

Just as he was able to locate readings he had deliberately omitted by the oversized blank spaces he left for them, so his later corrections of p to wynn and thorn in the first 918 lines did not necessitate a collation, and in fact help show that a collation was not made in conjunction with them. After he learned to distinguish p from similar insular letterforms, A more or less mechanically corrected his original mistakes by changing the bottom loops of his p's to make them look like the loops of wynns and thorns. The two capital wynns of the opening line were alone done with unusual care. As a result of his more mechanical work, he often changed the loops of what he had correctly copied as p's. Moreover, because he made his corrections hastily and unmethodically, he inevitably skipped over a few considerable sections of text — oversights less likely to occur in a full collation. For example, Malone notes that A failed to correct p to wynn ninety–nine times; it should be noted that eighty–two of these instances occur between lines 425 and 789, a section of the text A cannot have proofread very carefully for this particular mistake. Similarly, A failed to correct p to thorn twenty–three times, all between lines 417 and 685, while seventy–one cases of a p–like thorn occur between lines 410 and 778.

Yet even without a systematic collation, Thorkelin A is far superior to Thorkelin B. Knowing the characteristics of A allows us to restore with confidence nearly 1900 readings now gone from the manuscript. Among these restorations A made about sixty presumptive errors (an error rate of about 3%), but many are easily corrected if we know A's most usual mistakes. There are twelve cases, for instance, where he apparently wrote d for ð; fifteen reflecting his problems with p, thorn, wynn, and r; and twelve where contiguous strings of minims were presumably joined incorrectly. But knowing, too, how often A's unusual errors accurately represent the appearance of the manuscript helps us understand many other putative errors in terms of the state of the manuscript when A copied. For example, he presumably copied b for manuscript l in hil/de (163r3-4), but it is safe to assume that the l had a pronounced off–stroke (like other l's on the page), which heat from the fire might have bowed into a b–like loop. A similar explanation accounts for his c for l in lond (178v12), if the top of the l was burned off. We think he wrote f for p in wearp (164r1), but p would look like f with the right side burned away, just as h would become i with its left side gone (iafað 174v7, A60.1).

If we trust A when he makes uncharacteristic mistakes, we realize that two minims most likely shrank together to form “o” in both “beowolf” (167r17, A51.3) and “helo” (189r15, A78.9), where the last minim must have been lost if the word was helm. Editors assume that his swy...scolde, which he added later in a great gap he left on 159v1, represents an original swylce scolde, but A's dots normally mean that all letters are accounted for. Perhaps fire–damage made the two sides of an a resemble y, and we should be reading swa scolde. Certainly Zupitza's assumption that he wrote ir for op at the end of 180r16 cannot be defended by the collations. Instead, A probably copied ei for the ea–ligature, because the flat top of the scribe's a was no longer visible. Editors are right to read hwearf, assuming that the f was gone when A copied. They should not be so ready to assume that his readings are wrong in wedra (139v1), dyhttig (158v1), sciman (163v2), ricsan (179r4), and sweore/des (183r2–3). All can be defended as Late Old English readings. We might feel more comfortable with them if A had made a systematic collation of his work with the manuscript, but A's draft is still a document that deserves cautious emendation, since his unusual readings consistently tell us something worth knowing about the manuscript.

The B Transcript, Ny kgl. Saml. 512 4°

The collation of B with the manuscript, and the additional collation of B with A, show that B is much less reliable than A. Thorkelin's own transcript is not an independent copy of the manuscript, but a heavily edited conflation of A and the manuscript; it contains numerous emendations and an indeterminate number of conjectural restorations of lost readings. As such, B is really the first modern edition of the poem. Despite his editorial initiative, however, Thorkelin provides no textual apparatus identifying where his record of lost readings depends on the manuscript, on A, or on his own editorial judgment. As a result, his mysterious mixture of emendation and restoration, and of conflation and collation with A, means that we cannot depend on B as a primary source for lost readings in Beowulf.

The two transcripts provide quite different types of data to evaluate, partly because of their differences in format, partly because of the differences in attitude of the copyists. The A transcript attempts to provide a facsimile of the script but ignores the line and page divisions of the manuscript; the B transcript gives us a facsimile of the line and page divisions, but obscures all features of the script. Though B is easier to collate with the manuscript, only A tells us anything about the form of letters from the lost edges. A comparison of their approaches to scribal abbreviation highlights the virtue of A over B in this respect.

Not knowing how to expand abbreviations, A merely copied the scribal macrons, the signs of abbreviation, as they appear in the manuscript. He was very accurate. In the first 918 lines he made the crossed thorn (the abbreviation for þæt) with a round loop, like a p with an ascender, but his version of this scribal abbreviation is in no way confusing. He failed to cross the letter only once (146r8) in 326 occurrences in the manuscript.34 Malone lists fifteen cases where A omitted the macron in other abbreviations. Of these, only ten should be counted as oversights: hwilu(m) 174v1 was later corrected; secgu(m) 141r4 is on a damaged edge; and freog(e), dryh(ten), and herg(e) come from the faded, scorched, and torn final page. The top of the g and the macron are now gone for freog(e), and may have been damaged when A copied; he may have deliberately ignored the unique crossed h in dryh(ten); and he may not have seen the faded stroke over g in herg. More to the point, problems with rare abbreviations in damaged readings do not give a fair impression of A's reliability. For the common abbreviations of final –m and –ne, which occur in the manuscript 368 times, A omitted the macron only eleven times.

Because he thought he knew how to expand the scribal abbreviations, Thorkelin in B mostly interpolated expansions where macrons occur in the manuscript. He was inconsistent, however, and made numerous errors. For example, in dealing with the forty–five cases where the manuscript has a macron for final –ne, Thorkelin failed to expand sixteen times, later added a macron to an inadvertently unexpanded form five times, incorrectly expanded to e five other times, and incorrectly expanded to n eight other times. In fact, he omitted the macron and failed to expand in at least twenty–nine instances, some of which appear to be emendations. Since he emends manuscript eft to æft five times in B (136r7, 136r20, 160r20, 163v11, 178v14), his failure to expand æft to æfter (175r19, 178r10, 186r11) should be viewed most likely as versions of the same emendation. A similar argument applies to his failure to expand of to ofer (198r8, 198r18), since he elsewhere emends ofer to of (160r9).

Beside his frequent failure to expand abbreviations, Thorkelin expanded them incorrectly at least forty–five times. Of these, his treatment of crossed thorn, the abbreviation for þæt, stands in stark contrast to A's careful work. Thorkelin at first thought that a crossed thorn was the abbreviation for þe, and he expanded it in this way eleven times between 129r9 and 132r1. Moreover, shortly before switching from þe to þæt, Thorkelin once used the Icelandic form, þat (130v14). After he learned that the abbreviation actually stood for þæt, he corrected seven cases of þe to þet [both with e-hooks], one to þæt, and failed to correct the remaining three mistakes. The three oversights (two of them on the first page) show somewhat superficial proofreading, while the occurrence of þæt identifies another stage of proofreading.

There is, in fact, one case of an original þæt on 130r17, before Thorkelin switched from þe to þæt. This case, too, derives from a later stage of his proofreading, however, for all but the first word of this line is written on an erasure. Thorkelin's change from þe to þæt lets us date the correction sometime after he stopped using þe to expand crossed thorn. Even after Thorkelin learned to make the correct expansion, however, his treatment of this abbreviation remains far less reliable than A's. In one case he conflated his expansion with the following word, ðær, creating the nonsensical reading þæd ær, which he later “corrected” to þæð ær (154r2); in another case he mistook crossed thorn for a “7” and expanded it to and (151v2); in another he made a very late restoration, undoubtedly based on A, by expanding to the contemporary English spelling “that”; and in yet another he invented an abbreviation by copying ða as ðæt a (151r15), perhaps because the eth was unusually large.

Malone concludes on general principles that Thorkelin's acquaintance with insular script and Old English enhanced the reliability of B, while A's ignorance of them diminished the reliability of his copy (p. 25). Yet as B's treatment of abbreviations begins to illustrate, this assessment needs to be reconsidered. We have seen, for example, that A consistently wrote p for wynn in the first 918 lines of his transcript and that he overlooked many mistakes when he attempted to correct these early mistakes. Likewise Thorkelin, after correctly using w on the first five pages, consistently wrote v for wynn throughout his transcript. Later he too attempted to correct the v's to w's, but failed to make the correction 165 times, over twice as often as A failed to correct p to wynn. Both mistakes are relatively minor: A's error innocently represents a p–like runic letter as a p; B's represents an Old English spelling in an Icelandic way. In other words, while Thorkelin's scribe tries to follow the manuscript, Thorkelin knowingly changes it.35

Other mistakes peculiar to the B transcript are similarly induced by Thorkelin's linguistic roots in Iceland and Denmark. As Malone points out, the many mistakes involving d and ð arose from features of eighteenth–century Icelandic and Danish orthography (p. 23). He also notes that in the cases where Thorkelin wrote þa for adverbial ða “he was doubtless influenced by his native Icelandic” (p. 24). Malone's observations here should also serve as a pertinent warning about Thorkelin's attitude as a copyist. In addition to the use of v for w and þa for ða, and the case of þat for þæt, other pseudo–Icelandicisms can be seen in such spellings as kynne (for cynne 149v17), bearni (for bearm 129v11), siðan (later corrected to siððan 140v5), þæim (for þæm 138v5), and rather often þonn (for þonne).36 One of his pseudo–Icelandicisms, wundini goldi (MS wundmi golde 160v4–5 — for wundungolde?), even found its way into modern editions. We want to be able to trust the B transcript for lost readings in Beowulf, to believe (among other things) that his spellings accurately reflect manuscript spellings. Knowing that his mistakes regularly arise from extraneous orthographies does not inspire confidence. The accuracy of Thorkelin's spelling makes a great deal of difference for those areas of the Cotton text that are now gone.

As these examples suggest, the B transcript is less reliable than the A transcript in large measure because Thorkelin, for various reasons, thought he understood what he was copying. If his knowledge of Icelandic led to some errors, his knowledge of A led to far more. Having studied A first, he was irresistibly tempted to edit as he transcribed, capitalizing perceived names, expanding scribal abbreviations, normalizing and otherwise altering manuscript spellings, and emending difficult readings. For these reasons, errors in B are usually subjective, based on Thorkelin's understanding of the text, while those in A are objective, based on A's perception of the physical manuscript. As a result, Thorkelin's mistakes are more varied and unpredictable than A's. These unreliable features of the B transcript are easy to document by comparison with the surviving parts of the manuscript. Yet we are mainly interested in the reliability of Thorkelin's record of lost readings, which by definition cannot be documented. Nonetheless, Thorkelin gives us plenty of reason to worry about his reliability here in view of his demonstrable lack of reliability in the preserved text. A's later restorations may contain mistakes, but they are undoubtedly based on the manuscript; B's must always remain doubtful at best, since they may be based on A, the manuscript, or mere conjecture. Moreover, though B clearly displays several strata of proofreading, most of them are based on collations with A, not with the manuscript.

Thorkelin made broad, varied, and long–term use of the A transcript. There are many clear signs that his initial study of Beowulf began in his scribe's copy. It was obviously Thorkelin, not his hired scribe, who at various times in A dotted the i's in strings of minims, provided hyphenation for what he saw as unnatural word divisions, and underlined what he thought were proper names. Thorkelin's copyist had no knowledge of these matters. Hyphenation with a single line (e.g. DE–na A1.1–2, bear–me A2.3), a double line (wul–dres A1.10, weoro–da A3.1), or a curved line (scyl–dingas A2.16, þis–ne A3.9) occurs regularly on the first seven pages of A, but thereafter becomes relatively scarce. An isolated case of underlining occurs on the third page (elan A3.2, perhaps noting the misspelling helan in Wanley), but otherwise underlining of names only occurs between pages 10 and 41.

The hyphenation and underlining is good testimony of Thorkelin's early efforts to understand what A had copied. Equally good testimony of Thorkelin's efforts to improve the quality of his hired transcript appears between pages 2 and 17, where he inserted material that A had omitted. In one case he supplied a reading from Wanley in the margin (medo, A3.5), which, however, he later erased. In four other cases he restored omitted material by reference to the manuscript: eah– in the margin (A7.1); þæim (for indistinct MS þæm) in the space A left for it (A15.2); the beginning of A's omission, heorotes hrof hea(rd) under helme þ[æt] he on (A15.12), which he erased for some reason; and all but ic in A's omission, siððan (þ)a fæhðe feo þingode . sende ic (A17.14), which he did not erase. However, he made no further attempts to improve the A transcript on the basis of the manuscript.

In addition to all these signs of Thorkelin's scrutiny of A, there is, of course, his aborted interlinear translation, which he wrote and later erased on pages 2–6 of the A transcript. The limited scope of the translation suggests that at the time he had not fully decided what he wanted to do about Beowulf.37 In other words this shortlived enterprise, like the hyphenating, the underlining of names, and the restoring of omitted material, presumably derives from an early stage, before he made his own transcript. Indeed, this evidence seems to tell a quiet story about his ultimate decision to make a second copy of the poem. The hyphenating on the first seven pages and the restoration of medo on page 3 was needed before the translating on pages 2 through 6 could well begin. The subsequent restorations, based on the manuscript, and the underlining of the names, prepared the way for more translation. Finding several omitted lines at this stage (A15.12, A17.14, A23.10, A24.18), Thorkelin perhaps lost faith in the A transcript and so decided to make a second copy himself.38

Whenever he erased his aborted translation, Thorkelin failed to erase the Latin word ita (beside Old English eac A4.6) and did some minor damage to the A transcript, leaving A with more errors than he actually made. For example, heah looks like hean (A2.15) and guð r12 like gud (A2.16) because Thorkelin accidentally erased some ascenders along with his interlinear translation. He restored the cross–stroke he erased in morð (A5.14), but only the first of the two strokes he erased in ðryð (A5.11). Conversely, Thorkelin created an ascender by failing to erase a descender from his translation, leaving the false impression that A copied þæs for wæs at 132v13 (A5.8). Perhaps his most careless erasure was when he deleted A's interlinear correction, nænig (A6.9); his most zealous was erasing his own restoration of medo (A3.5).

Many indications of Thorkelin's use of his scribe's work appear as well in Thorkelin B. If we may judge by the considerable coincidence of error in A and B, Thorkelin regularly used A in tandem with the manuscript, presumably because A was easier to read and perhaps preserved readings that had since crumbled from the burnt edges of the manuscript. We can see the seeds of Thorkelin's conflation of A and the manuscript in the capitalized proper names in B, which in the manuscript are not immediately recognizable as names, since they are uncapitalized and often written in more than one part. In A, these names were naturally copied as they appear in the manuscript, uncapitalized and in the scribal word-divisions such as hroð gar and beo wulf. By underlining the proper names he recognized in A, Thorkelin highlighted words he would later capitalize and word-divisions he would later unite in his own transcript. He inevitably invented spurious names, overlooked legitimate ones, and changed his mind back and forth along the way. For example, he underscored byrelas 155v3 in A but not in B, ecg-wolde 164r4 in B but not in A, and sige-munde 149r3 in neither A nor B. Reflecting similar uncertainty, he underlined wealafe (153v9) in A and later in his own transcript, toying with the chance that he had found a name like Wiglaf, but then he never capitalized the word.39

Thorkelin's creation of proper names in Beowulf also helps us see to what extent he comprehended what he was copying. As late as 20 January 1807, the year he finished the first version of his edition, he responded to inquiries from his patron that he was finding the work much more difficult than he had anticipated. Indeed, twenty years after acquiring the first transcript he still reported that he had arranged only half of the poem in verse, and had translated only a quarter of it. He understood Anglo-Saxon prose, he explained, but pointed out that the poetry, like Icelandic verse, uses a different language, discourse, and way of thought.40 After this edition-in-progress was destroyed in the British bombardment of Copenhagen later in the year, Thorkelin needed another eight years to edit it again. His final Latin translation shows that he was still having insuperable problems with the language of Beowulf. But since the B transcript was in large measure Thorkelin's first edition of the poem, his understanding of Old English at the time he was copying has some bearing on the accuracy of his copy. His repeated capitalization of proper names like Beowulf and Hroðgar, when viewed in isolation, appears to display his knowledge of Old English. When viewed with repeated mistakes in capitalization, however, these changes seem far less trivial, displaying instead a readiness to make uninformed emendations in his transcript.

Some examples of mistaken capitalization in his transcript will help put this seemingly minor error in a clearer light. On 133v9 Thorkelin did not understand the compound gifstol (“gift-throne”), and so created a person named Gif, who stol gretan moste (“had to approach the throne”). Similarly, at line 15 of the same page, he did not see a compound in wigweorþunga (“honor to idols”), and apparently interpreted Weorþunga as a tribal name of those who wordum bædon. On the next page the compound swanrade (“swan-road” or “sea” 134r18) is broken when guðcyning ofer Swan rade (presumably “the war-king trampled Swan” or rode a horse named Swan).

Later, new heroes emerge, as when “the brave Gylp then spoke” (gespræc þa se goda Gylp, 145r20), or when “Ecglaf's son spoke about Gylp” (sunu Eclafes on Gylp spræc, 151r16). Some cases of capitalization can best be explained as marking new sentences, but there is no reason to think that Thorkelin was merely capitalizing nouns. Verbs become flesh in Healdan gyf þu (“Healdan, if you..,” 156r7), and a couple of pages later, where Thorkelin emends healden to Healfden (apparently a name for one of Hroðgar's sons — 157r8). In the next line the second son appears: her is Æghwylc eorl (“here is lord Æghwylc”). More complicated still, depending as it does on an emendation of -ceare to cearfe and a failure to see the proper name Grendel, is Gud cearfe grendles modor ides Aglec (“Gud cut up the mother of a grendle, the woman Aglec,” 157v15-16).

While making his transcript Thorkelin may or may not have interpreted these spurious proper names in the ways suggested above.41 The point is not at all to ridicule his pioneering efforts to understand a difficult poem in a language he was only beginning to learn. It is rather to emphasize that, partly because he had studied the poem in A before he made his own transcript, he was ready and willing to make emendations as he copied. His transcript should therefore be viewed more as an edition-in-progress than as an objective transcript.

In studying Thorkelin's corrections and alterations of the A transcript, we must distinguish various stages of proofreading. We have seen that in his early proofreading of A ,Thorkelin attempted to improve its accuracy, presumably because he had not yet made his own copy of the poem, and perhaps at this point had no intention of making it. Yet one can discern in Thorkelin's use of his scribe's work a crucial change in attitude, from an interest in A to an interest in B. This shift in purpose becomes apparent where Thorkelin stopped restoring omitted material in A, and is highlighted where he erased some of the restorations he had made. His later changes in A are not based on the manuscript, and seem geared to improve the appearance of B instead of the accuracy of A. The most troubling manifestation of this new purpose appears in a few cases where Thorkelin altered the A transcript so that A's readings agreed with his own. Perhaps Thorkelin anticipated Danish readers of his edition, like Rask and Grundtvig, consulting the two transcripts in the Kongelige Bibliotek. To be sure, he advised later readers on its title page that B alone was “accurately transcribed.”

There are a number of words in A that Thorkelin clumsily altered to agree with B, and a few that were more skillfully altered, as well. It is easy to see that Thorkelin inserted -ga- between gi- and -ntas (A4.16), but hard to say whether he or A changed e to a in -ntas. A simpler case comes from 139r19. Except for the capital N, Thorkelin copied niceras correctly, but later altered the reading to Inceras. The A scribe had copied m for ni, but Thorkelin changed m to in by dotting the first minim. Similarly he changed A's fealh to wealh (A38.1) and swin to swinc (A40.3) so that these readings would agree with his emendations. Sometimes he emended A but rejected the emendation in B. See, for example, his changes of þeah to þleah (A8.5), fæges to fægest (A46.5), stonc to stond (A67.5), and probably weorð to sweorð (A47.5). Thorkelin is probably responsible for erasing the b in A's b ... am (A64.13, 178v11); the manuscript reading — now “.am” — seems to have been damaged already when Thorkelin copied “..m,” which he changed to “.am,” and finally to “bam”. He doubtless changed mm to nn by erasure in hlæmmum (A64.15, 178v15), since A, in view of the manuscript, would have no reason to make the change.

Another complicated case of emendation of A occurs on 154v2. When he came across his error, hlawelan for manuscript hlawe hafelan, Thorkelin was obviously collating with A, not the manuscript. He marked his confusion by writing hlaf æ above the misreading in B, and by altering A's correct reading, hafelan, to hlafelan (A35.11) instead of properly correcting his own mistake. This complex error resurfaces in Thorkelin's edition as hlawe/Hlaw ælan (translated acervum; I Congerium omnem, p. 85). A final example comes from 149r6: though fitela is perfectly clear in the manuscript, and both A and B copied it correctly, Thorkelin later emended it to wite la in both transcripts, and eventually translated Buton wite la as Eheu impunia in his edition (p. 68). Examples like these emphasize that we must be especially sceptical, in view of Thorkelin's methods, of any lost readings in A that have been tampered with, even the early corrections of p to thorn or wynn, since Thorkelin apparently made some of them.

On the other hand, Thorkelin is not always culpable in the most suspicious circumstances. A salient example is the guilty-looking, damaged reading nægling, the supposed name of Beowulf's sword (189A[197]r20). The A copyist deliberately omitted the word when he copied the page. Later, in a different ink and a seemingly different hand, someone inserted nægling, the reading of B. It seems unlikely that A, who did not understand the text at all, would single out this reading for special attention later. Conversely, it seems certain that Thorkelin not only would be particularly interested in it, but would have known the sword-name “Nagelung” from the Icelandic Þiðreks saga. However, the possibility that the name of Beowulf's sword is a brilliant invention of Thorkelin's can be rejected, for Thorkelin did not recognize the sword-name in Beowulf. In fact, he thought it was a shield, and accordingly translated the phrase Nægling forbærst as Scutum dissectum est in his edition.42

While he frequently undermined the reliability of A by emending it, Thorkelin nonetheless depended on A to help him complete his own transcript. The difference in format between A and B, which at first glance would appear to prove that Thorkelin did not depend on A, discloses instead some probable instances of conflation. At the beginning of 153r6, B has nu git over erased selest (from line 7), while at the beginning of line 7 he has selest over erased bi- (from line 8). These consecutive corrections indicate that Thorkelin may have written the first words of some lines before he wrote the lines in full, a strategy that would facilitate the use of A in tandem with the manuscript. The evidence of conflation of A and the manuscript is twofold. The crucial distinction is between Thorkelin's dependence on A while he was supposedly copying the manuscript and his dependence on A afterward, while proofreading his transcript. In both cases, disturbing signs of conflation appear in the coincidence of error in A and B.

Since Thorkelin's hired scribe copied the text continuously, word for word, ignoring line and page divisions, he obscured the places where manuscript lines and pages begin and end. When Thorkelin paid too much heed to A, and not enough to the manuscript, a line of text in B does not correspond exactly to one in the manuscript. There are thirty- six cases where Thorkelin either misplaced or began to misplace a reading, not counting the deliberate cases on 198 verso, where lack of space was the determining factor. The misplacement of text on 142v13 and 145r6 was probably done on purpose, as well, to separate it from the fitt numbers, while -stanes (189v18) may have been written on the wrong line to keep track of the proper name, Wihstanes, which Thorkelin also capitalized. The remaining thirty-three cases, however, were all apparently done inadvertently, sometimes in the course of copying, sometimes in the course of later proofreading.

On three occasions (129r20, 132r20, and 138r20) while proofreading Thorkelin placed a reading on the wrong page, as well as the wrong line, a virtually inexplicable error if he was depending on the manuscript, since he would have had to turn the page to see the reading. On 138r20 he pencilled in tiges, which also appears where it belongs on 138v1, but in a different ink from the rest of the line, so here we have to deal with two stages of proofreading. On fifteen occasions43 he misplaced readings while copying without ever correcting his mistake. This phenomenon can be explained, of course, as momentary inadvertence while copying the manuscript alone, but it would have been more likely to occur if Thorkelin was concurrently referring to the A transcript. Thus, on 134v6 the manuscript line ends with myr-, while the line in A (8.9) ends with myrcu, the same way the line ends in B. It seems both reasonable and prudent to suspect, therefore, that Thorkelin was momentarily influenced by A in such cases, that he was subconsciously creating a conflation in B.

Almost as often Thorkelin committed or began to commit the same mistake but then corrected himself. The corrections emphasize that line correspondence between B and the manuscript was an important feature of his transcript as far as Thorkelin was concerned, no doubt because it would later help him locate in B (without having to refer to the manuscript) possible trouble spots, the lost and damaged edges of the manuscript. At any rate, he later corrected many false starts, normally by pointing, crossing out, or erasing the misplaced readings.44 These corrections, of course, illustrate ultimate recourse to the manuscript. A possible misplacement that has some bearing on the text occurs at the start of line 2 on 161v, where B has “f...” while A indicates that the manuscript is gone. Thorkelin may have first written the f  from fuslic on the wrong line. This explanation would account for the odd occurrence of an isolated f  where otherwise A and B seem to agree the vellum was gone.45 Some other cases look like false starts, too, but changes in ink identify them instead as later mistakes, based on hasty proofreading with A (see gr- 139v14, me 140v7, -feng 145v12). Thorkelin twice was able to restore text he had omitted (ge- 176v9 and mundum 196r20), but since he placed them on the wrong lines they may well be based on A too. On 153r9 a slash-mark shows that he knew that wæs was misplaced, but on 174r12 the same mark has the opposite effect, indicating that denes, rather than nes, belonged on the next line. Here the mark may be recording either the possible occurrence of a name (“of the Dane”) or a lingering case of conflation with A.

The only credible reason why Thorkelin would use A in tandem with the manuscript, making in effect an impromptu conflation of A and the manuscript instead of a direct transcript, is that the manuscript had deteriorated from A's constant handling of it in the course of making his transcript. This is not mere conjecture. As Malone observes (p. 4), we date the transcripts relative to one another, and in fact name them A and B, based on our observation of this progressive deterioration. But there is reason to believe that Thorkelin suppressed much of the actual decay of the manuscript in B, perhaps because he forgiveably wanted his own transcript to be more valuable than A, for his own peace of mind if not to impress prospective scholars at the Kongelige Bibliotek. His transcript provides signs of this progressive deterioration. The edge of the manuscript on 189r12-14, for example, must have been in far worse shape when Thorkelin copied than the final copy in B would suggest. He omitted final e in healde r12, inserted and deleted spurious þæt r13, incorrectly completed the line with þeoden (instead of þeo), and failed to expand dryht(en) at the end of r14. The bottom corner of 189r seems to have been gone, too, for both A (r17-21) and B (r16-18). The readings are accordingly suspect in B.46

Thorkelin has left us some circumstantial evidence that the manuscript was in worse shape when he copied than his transcript would seem to indicate. The evidence comes from three leaves in B (corresponding to fols. 165-167), which Thorkelin curiously displaced in his transcript before he had it bound. These pages are still out of order in the facsimile edition published in 1951, but they were finally put in the correct order in the manuscript in 1983. Malone draws our attention to the disorder in Thorkelin B at the start of his commentary, but does not attempt to explain it. He says,

three leaves of the B text, answering to fol. 165, 166, and 167 of the Cotton manuscript, six pages in all, somehow got out of place and now stand between the leaves answering to fol. 160 and 161. The reader of the B text, then, when he has finished B's copy of fol. 160 verso, must skip six pages to find fol. 161 recto, and when he has finished fol. 164 verso he must turn back 13 pages to find fol. 165 recto. When he has read fol. 165-167, he must skip eight pages to reach fol. 168. After that, all is plain sailing.

At the bottom of page 64a, containing Thorkelin's copy of 160v, someone has pencilled in a note that also alerts the reader to the disorder. The note, barely visible in the facsimile but easy to read in the manuscript, simply states that the following three leaves are in the wrong place.47 How these pages ended up where they did may well reflect the state of the Beowulf manuscript when Thorkelin was proofreading.

As we have seen, the binding of Thorkelin's transcript of Beowulf was not done in Denmark. Hence the disordering of the leaves must have taken place while Thorkelin was still abroad. The evidence suggests, moreover, that Thorkelin himself rather than an incompetent bookbinder misplaced the leaves before binding and while he still had access to the manuscript. His transcript is mainly copied on two-sheet gatherings, but in this case Thorkelin apparently made a mess of the last leaf, for he excised it from the original gathering, leaving a half-sheet and a bifolium for 165-167.48 It is easy to explain, then, how Thorkelin's copy of these three folios got out of place in his transcript. After cutting away the spoiled leaf he carelessly misplaced the remaining leaves before instead of after the gathering for 161-164. But what led him to misplace them here?

We can explain the blunder by positing that the first lines of 161r and 165r were in the same damaged state when he was collating as they are now. Today, all that can be seen of 161r1 is:

Figure 20

[Fig. 20: ...(þ)a wæs hroðgare hors ge...]

The A scribe copied the line as spræc þawæs hroðgare hprs ge bæted, showing that it was still intact when he copied.49 But when Thorkelin made his transcript, he copied the line as þæs Hrodgar. hors ge bæted. He later added sprec [with e-hook] (for MS spræc) and corrected va to þa and þæs to wæs. The original readings suggest that sprec [with e-hook] probably derives from A, not the manuscript; that the ascender of þa was already gone for Thorkelin and so looked like wa (spelled va in B); and that wæs looked like þæs to him because the crumbled edge of the vellum seemed to have ruined an ascender.

Folio 165r1 is in better condition than 161r1. It reads:

Figure 21

[Fig. 21: ... ceorlas þaðe mid hroðgare onhol(m)...]

For this line Thorkelin A has, ceorlas þaðe mid hroð gare onholm wliton, again showing that it was completely intact when he copied. Yet when Thorkelin himself made his copy, he wrote ceorlas þaðe mid Hrodgare on holme.... He later erased the spurious e in holme, and added wlitom (m for n by mistake, perhaps induced by holm) at the end of the line. At a superficial glance, lines 161r and 165r look alike in the manuscript (...þa... hroðgare...ho...). Thorkelin may have enhanced the resemblance if the spurious e in holme derives, as seems probable, from what now remains at the end of 161r1 — hors ge. In short, with what is now left of the manuscript readings for 161r1 and 165r1, Thorkelin could think that 165r belonged after 160v, as long as he centered his attention on the first lines of 161r and 165r. If the edge of the manuscript had already crumbled away in both places, he could have thought that ceorlas (not spræc) was the lost reading at the beginning of 161r1, and wlitom (not -bæted) the lost reading at the end of the line.50

The manuscript on 161r and 165r was undoubtedly in worse condition for Thorkelin than the final versions in B suggest. If many of the brittle, charred edges of the manuscript had already crumbled away when he copied, Thorkelin would have inevitably resorted to the A transcript and, on occasion, continued referring to it even where the manuscript was intact. Indeed, it would be quite surprising if he did not have the A transcript by his side while he copied the manuscript. After all, A provides us with an extremely useful guide to manuscript forms, and there is no reason why Thorkelin should not have used it for this purpose as well. There are repeated clues that many of his errors stem from fleeting glances at A while copying the manuscript.

The coincidental errors of A and B can be explained by assuming that Thorkelin naturally referred to A whenever the manuscript presented difficulties, and from time to time even where the manuscript was intact. As we have seen, this hypothesis helps account for the many times he began or ended a line with the wrong word. It also helps explain some odd coincidences. At the end of 130v20, for example, the Beowulf scribe for some reason used an unusual form of punctuation, two points and a comma, which the A scribe accurately copied. Since Thorkelin in his copy represents this punctuation with five points, his customary way of showing an elipsis in the text, perhaps he was here depending on A: without the manuscript he could misinterpret A's punctuation as a conventional sign of a gap in the text. It is also suspicious that, like A, Thorkelin includes the missing fitt number XXXVIIII on 191r12.

Unintentional conflation might explain other coincidences, such as why Thorkelin often miscopied a scribal correction in the same way as A. In one case the Beowulf scribe changed on to in by underdotting o and writing i as a superscript (158v15); A naively copied oin, and Thorkelin repeated the error in B. He was presumably influenced by A here for he earlier shows that he understood the scribe's use of under-dotting in his correction of most to mæst (153v5), which A copied as moæst. Indeed, Thorkelin later realized his blunder in oin and deleted i, thus recreating on, the form the Beowulf scribe changed to in. There can be little doubt that the manuscript was not at hand at the time he made his “correction.”

Whether or not he was actually copying A, Thorkelin nonetheless misinterpreted in the same way other scribal corrections in the manuscript. Both A and B disregard the scribe's correction of madmam to madmum (194v3). They also miscopy manuscript horde as hopde on 190r21, and on the verso they copy p for w in sceawode at line 10. On 192r20 both have a for u in unswiðor. Thorkelin surprisingly came to adopt A's misinterpretation of the punctus delens as a sign of insertion, and so misread with A four (or five) additional scribal corrections. The doubtful case comes from 157r5, where the Beowulf scribe corrected wide to side in an innovative way, using the descender of wynn as the base for caroline s, the bow of wynn as an i, and pointing the original i. Characteristically ignoring the point, A faithfully copied snde; Thorkelin followed his example, but wrote u for n to provide the needed vowel (for illustrations of the manuscript and A, see p. 109 above). In short, Thorkelin's mistake seems to have come through the mediation of A's. On 160v12, both copied the scribe's correction of gumen to guman as gumean. The same mistake occurs in hreaþe for hraþe (171v20) and dæld for dæd (192v6). Perhaps not being able to see the point beneath the e, they both copied feorð for forð on 180r14, a forgiveable oversight in view of the scorching along the damaged edge of the manuscript.51

There are other shared mistakes worth noting. In many cases where both A and B make mistakes supported by the appearance of the manuscript, the evidence rather suggests that Thorkelin's error was influenced by A's. In these cases he seems to have followed A's rendering because the manuscript presented a confusing image. Some good examples come from 179 verso, the second page of the palimpsest. First, A and B tend to agree on the problem readings Zupitza used for his “freshening-up“ theory. Thus they both read fæs (originally væs in B) for fær in line 2; rihde for rende in line 10; innon for innan (originally innan in B) in line 14; fec for fea in line 16; and bealc (later emended to bealo in B) for beale in line 19. But they also agree on a number of misreadings where the manuscript does not so clearly support them: begent for begeat in line 2; hyððe for hydde in line 7; ðer (ðere in B) for ðær in line 9; eal for eall in line 12; forda (later forða in B) for worda and hold for heald in line 16; hruce (later hruse in B) for hruse in line 17; and hit for hyt in line 18. It is not hard to see how A made all these mistakes on this badly damaged page, but one would expect Thorkelin to avoid at least some of them, unless he was referring to A.

Shared mistakes vaguely supported by the manuscript are by no means restricted to the palimpsest. See, for example, a for o in weold (140v1); e for c in rinc (145v14); l for s in magas (152r6); b for þ in fæþme (172r3); c for k in kyning (181v19); e for a in hycgendra (186v21); æ for e in wælréc (189A[197]r5); and t for d in mid (196r20). Though once again we cannot be sure that Thorkelin's misreading stems from A's in such cases, once again it puts less of a strain on one's credulity to assume that he was using A in tandem with the manuscript. Despite the variety of support from the manuscript for these errors, it is unlikely that Thorkelin would have repeated all of them if his examination of the manuscript had been completely independent.52

Of course, not all shared mistakes suggest conflation. One error A and B often have in common is d for ð, but since Thorkelin commits this error 620 times (compared to A's 130 times) no special significance should be attached to it.53 On the other hand, not all cases should be discounted. B's ð for d in ádfære (194v17), for example, probably derives from A, who understandably interpreted the roaming accent mark over a as a cross-stroke for d. B first copied d, but later “corrected” it to ð. Thorkelin often presents minim-clusters in a way that suggests some kind of dependence on A. Coincidences in these cases are more compelling evidence of conflation than those involving d and ð because Thorkelin was notably careful to distinguish letter-forms in strings of minims. Both Thorkelin and his scribe wrote im for un in gefrunon (130v2), m for ni in wilnian (134r7), nn for mi in famiheals (134v12-13), un for im in brim (134v15), and m for ni in nioðor (189A[197]v16). Some of these shared mistakes may be conscious emendations on Thorkelin's part — brun (“bright”) for brim (“sea”), for example, or modor (“mother”) for nioðor (“lower down”). A's fragmentary “wym...” from the damaged edge of 189r18 (restored by modern editors to wynne) was probably the source of B's wyni, displaying once again Thorkelin's Icelandic roots as well as the condition of the manuscript.

Thorkelin sometimes changed the mistakes with minims he at first had in common with A. We have seen him make a good change with caines (132r15) and a bad one with side (157r5). At first he failed to recognize and underline the proper name deniga (“of the Danes”) when he studied A, doubtless because A had copied demga, writing m for ni. He repeated A's error in B, but later corrected both transcripts, underlining the reading in B. Where the manuscript is clear some mistakes in B are best viewed as emendations of readings in A. Thus on 141v7 he first wrote me for inc, an emendation evidently based on A's erroneous ine. Thorkelin emended both A and the manuscript in the same manner for two famous cruces, wundmi (160v4-5) and mwatide (179r19), and is mistakenly thought to have done it for hlitine (154v12) too. He first copied A's wuman (MS wunian 157v17), but then corrected m to ni. Perceiving they were wrong, he probably reanalyzed A's m's (all misreadings of the manuscript) in mge 155r16, glitiman 190r1, and m 195v9, to produce his own misreadings (nige, glitunan, and in). At any rate, inge, glitinian, and iu are all quite legible in the manuscript even now. Other probable emendations in B stemming from improperly joined minims in A are man for –nian in cunnian (161v4) and mowan for niowan (169r11).

Sometimes both A and B omit the same letter, the same parts of words, the same words, and once even the same manuscript line. The state of the manuscript probably accounts for both A and B overlooking letters in fre[m]man (132r9), bolge[n] (179r13), and s[w]ealg (198v5), as well as the entire words hwylc, gefeng (179r8) and meowle (198v1). Yet momentary conflation is surely the most plausible explanation when both A and B read on for what is clearly ond in the manuscript (143v12) and ð for what is clearly ðæt (188r21).

Figure 22

[Figure 22: (ðæt 188r21) (ð A76.3) (ð B119a)]

Likewise it most easily explains why both transcripts omit mearc on 132r11, and the same manuscript line, weard onfunde buon onbeorge bio wulfe, on 191v9.

A few times Thorkelin's corrections of errors he at first shared with A clearly suggest collation with the manuscript (for example, werhðo 143v1 from wehrðo and stow 160r17 from stor). However, there are far more indications of later proofreading against A than against the manuscript. In a thorough collation with the manuscript Thorkelin should have detected, but did not detect, his omission of an entire line at 191v9, a line that A had omitted. Collation with the manuscript is not easy to document. Even though Thorkelin's line-for-line format in B shows that he relied mainly on the manuscript for the bulk of his transcript, it does not show unequivocally whether he relied on A or the manuscript for corrections, not to mention those crucial readings that are now lost. If the B transcript indeed dates from the last months of his stay in England, it would help explain why Thorkelin collated B with A instead of with the manuscript. He may have left for Denmark before he had time for a thorough collation with the manuscript. Evidence of collation with A is manifest and ubiquitous in B, for Thorkelin frequently changed readings he had correctly copied to incorrect A readings. Later changes based on A, together with relatively meagre signs of proofreading against the manuscript, tend to reduce our confidence in all later additions to B, notably those along the damaged edges of the manuscript, where we most want to trust B.

To be sure, some features in B can only be explained by reference to A. Among other things, B's copy of 130 recto provides clear evidence of repeated collation with A in a bizarre series of errors Thorkelin made between lines 17 and 19. When he first copied the manuscript, Thorkelin made a major omission at line 17 by writing after gare most of line 18 (his wine magas georne hyrdon oððþet (with e-hook) instead of the rest of line 17 (here sped gyfen wiges weorð mynd. and crossed þ). He then compounded his error by making a long dittograph, recopying all of line 18 where it belonged — another indication that he wrote the first words of each line before copying the lines themselves. Later, after learning to expand crossed thorn as þæt, he erased the misplaced text and wrote the rest of line 17 where it belonged. Later still, he wrote seo gegod (the start of line 19) in the right margin of line 17, and then crossed out these words. He did not erase them, because he had already erased some earlier marginalia between lines 17 and 19, and this erasing had made a hole in the paper between seo and gegod.

This curious sequence of blunders is elucidated at crucial points by reference to A, whose original omission of line 18 spawned Thorkelin's hapless marginalia. When he noticed the long dittograph he had created in lines 17 and 18, Thorkelin was apparently proofreading against A, not the manuscript. The long erasure in the margins of lines 17-19 cannot be read with certainty, but it appears to have been line 19. In other words, in addition to correcting his dittograph on the authority of A, Thorkelin continued copying A in the margin, not realizing at first that the discrepancy in their texts was caused by A's omission of line 18. As soon as he noticed his mistake he erased it. However, in the course of a second stage of proofreading with A, he began to make the same mistake over his erasure of the first. This time he got only as far as seo gegod before realizing his blunder, which he at once crossed out.

A study of Thorkelin's copy of folio 180 raises many pertinent questions about his use of the A transcript. On the recto, the readings in B along the right edge in the first six lines imply that the manuscript was almost as badly damaged when Thorkelin copied as it is now: in line 1 he records only “fe...”, and the e is almost certainly an ill-founded conjecture; in line 2 “dug..” was added later; in line 3 everything after helm was already gone; in line 4 everything but the first minim of mynd was added later; and in line 6 sio was probably added later. Except for the doubtful e in line one, these readings agree in every respect with those in A, so we have no way of knowing whether the later additions were made in collation with A or the manuscript. The same is true of the later corrections (se on an erasure and superscript e in hearda) in line 3. But there are two indications of conflation with A while copying (healfs for healfe 10, and feorð for forð 14), both from the damaged edges, and a great deal of evidence of collation with A on the rest of the page.

This kind of evidence normally becomes apparent only when A makes a mistake that in some way influences Thorkelin. On this page A makes three mistakes, two minor and one major, that led Thorkelin to alter his original transcription. In line 6, A copied MS herepád as herepað mistaking the accent mark for a as a cross-stroke for d. Thorkelin originally copied herepad, but later crossed the d  too; furthermore, he changed p to r, thus emending the word to hererað which he gamely translates in his edition as exercitus Dux (p. 169). The change of p to r has no manuscript support. It probably comes indirectly from A, for Thorkelin knew that A often wrote p for similar insular letters, including insular r. In line 11 A copied mæs for næs, an error perhaps induced by man in the line above. Thorkelin originally copied næs, the manuscript reading, but later changed the n to in in an obvious effort to make sense of A's m. The p in MS hearpan, the next word, is ironically changed to w, another apparent emendation of a correct A reading.54

A's major blunder affecting Thorkelin on this page was a haplograph (from æfter to æfter) in lines 8 and 9: æfter beorne nemæg byrnan hring. In collation with A, Thorkelin underlined A's omission in his own transcript, and later strangely followed A by omitting it himself in his edition. He may have somehow confused A's omission with one he made in lines 16 and 17: an æfter eallu(m) unbliðe hwear[f] dæges. Perhaps the word æfter triggered his confusion. In any event, though he later discovered his mistake by collating with A, he curiously “corrected” it by writing dages (in his late, shaky handwriting) on top of and in line 16 of B. It was not until preparing his edition that he restored his omission through A's testimony.

In his copy of the verso Thorkelin's scribe left three blank lines for 180v1-3, since they had been deliberately effaced in the manuscript. In B, Thorkelin first tried to copy what he could see, but in Denmark, confronted by the three blank lines in A and the fragments in B, he revised his transcript to agree with A. He placed a line of dots above line 1 and below line 3, and wrote in the margin, “Hic lacuna trium linearum sive 15 versuum incidit qui in autographo defuisse videntur et enim membrana, ex qua hoc apographum desumptum hic vacua est.”55 The note was presumably written in Denmark, since it more accurately describes the appearance of A than the manuscript. By the time he finished his edition, Thorkelin no longer understood the import of his note, which he printed as, “Hic lacuna incidit, quæ XV versibus respondet absentibus.”56

There are many cases where Thorkelin first made an emendation but later changed his erroneous reading to the correct manuscript reading. It is usually impossible to tell, however, whether the correction was made by reference to A or to the manuscript, since virtually all of the A readings in these instances faithfully represent the manuscript. Even his correction of the emendation siðan, the Icelandic spelling of Old English siððan (140v5), which A omitted, may derive from the restoration Thorkelin himself made in the A transcript before copying B.57 The correction of the emendation sceal to scealc, on the other hand, for manuscript scealca- (157r20) surely derives from A since it only partially restores the manuscript reading. Though normally we cannot be sure whether Thorkelin based his corrections of deliberate emendations on A or the manuscript, the odds would seem to favor later corrections from A over early ones from the manuscript, since Thorkelin might at first be reluctant to admit that his emendation was wrong-headed. He found A a welcome resort more than once as soon as he began translating.

In the following examples of extemporaneous emendations later corrected, all of the corrections could come as well from A as from the manuscript. The point here, however, is not to argue for Thorkelin's ultimate recourse to A, but rather to illustrate his tendency to make conscious emendations in the course of copying the manuscript. Long lists could be made of emendations with no deducible meanings (for example, volgen for bolgen on 167v5 or mæn for læn on 187v1) and of possible emendations that might only be scribal mistakes (ne for me, for instance, on 150v8, or mine for nime on 162v11). Even without this vast material, the variety of self-evident emendations is impressive. One could almost devise an entire paradigm from errors relating to the demonstrative se: on 174r12 he first copied mere (“pool”) for me se; conversely, he copied se for (“sea”) on 157r5; for a while he had swa se for swæse (170v18); in the genitive he first miscopied þæs for þæt (147A[131]v4), and both væs (before he changed v to w) and þær for MS þæs (167v13, 190r3, 196v13); in the dative he gave a variant spelling, þæm for þam (182v4), switched to another declension, ðare for ðam (174r3), and to another person in another declension, ðu for ðam (192r2); for the accusative þone he first wrote the phrase þe ne (158r8).

Thorkelin's emendations while copying, like his capitalizing proper names, are intentional changes not to be confused with ordinary scribal errors. If Thorkelin had copied for for rof only once or twice in his transcript the mistake might reasonably be seen as scribal metathesis; but when it appears five times between folios 134 and 145 it can only be seen as an emendation.58 Cases like these provide convincing evidence that many of Thorkelin's supposed mistakes in transcription should really be seen as emendations, even though they were later corrected. Some examples of words Thorkelin corrected that most likely began as emendations are: scip for scir (137r5), geheape for geheawe (145v6), rand for fand (148v19), scopen for scofen (149v20), geaf for gear (160r2), rand for the abbreviation for ond (later corrected to And, 167v4), sefa for selfa (170r13), sceal- for steal- (173r15), selfra for selra (178v7), lond for hond (185r15), beseah for geseah (corrected to Geseah, 190r9), sið for soð (192r6), heafode for heafde (194r8), ealde for ealdor (194v11), on for (196r3), on for of (196v21), and of for oft (196r9). Many of the corrections were made before Thorkelin mechanically changed v's to w's in his transcript: for example, svoru for sporu (151r20), stov for stop (161r3), vif for gif (186r1), vel for sel (186r11), and være for fære (194v17). Unfortunately, we do not know when he changed v's to w's, not even whether it was done in England or in Denmark. He certainly did not need the manuscript to make the change.

From what survives in the manuscript, however, we do know that a large number of later changes based on A are wrong, and Thorkelin would have known it as well if he had consulted the manuscript. Thus on 130v11 he later changed manuscript c to t in sinc, while reversing the error in etan and eteð (139v20, 140r5). He first copied MS reaf (138v20), but later decided that A's reof was right. Though he had accurately copied lind (135r14), he later deleted n to agree with A. Likewise, he changed MS gryrum (140v18) to A's grymmum and copied A's on for o in MS swogende (198r18). Examples stemming from A's early problems with insular letterforms that resemble wynn illustrate Thorkelin's manifest dependence on A here. Once again the later changes show Thorkelin making emendations, not deliberating over doubtful manuscript readings. Thus he had no manuscript support at all when he altered r to w in rica (138v18) or when he copied A's rine for MS wine above vme in B (139v6). He was also emending the manuscript when he altered þ to w in þege (132v6), and w to þ in wæs (135r10). Having miscopied wræce for MS wræc on 139r20, Thorkelin compounded his error by writing þrec above the line, a wrong alternative he took from A. Similarly he wrote A's wystrum in the margin beside þystrum, the manuscript reading (130v16).

Some of his collations with A of course helped Thorkelin improve his copy, allowing him to add readings later that he did not originally record. We cannot know why he omitted the readings in the first place. Perhaps he wanted more time to consider them, or perhaps they had crumbled away before he made his copy. Malone says,

In the B transcript one can distinguish two hands: an earlier and a later. The earlier hand is that characteristic of Thorkelin when he copied the text in the year 1787. The later hand is that characteristic of Thorkelin many years afterwards, when he was preparing his edition of the poem. It differs from the earlier hand in that it shows signs of age (it may be described as a shaky hand).... Nearly all the text is in the B hand but the Thorkelin hand appears now and then, especially in illegible or doubtful passages. (p. 22)

What should give us pause is that these illegible or doubtful passages all occur at the now lost edges; and while we are pausing we ought to remember that Thorkelin prepared his edition twice, once in 1815, and once in 1807, when all of his work was destroyed in the British bombardment of Copenhagen. It should by now be obvious that Thorkelin used more than “two hands,” more than three for that matter, in his copy of Beowulf.

Thorkelin's shaky hand betrays numerous readings as very late restorations, long after he had left England. Some obvious cases are þine (129r19), seo gegod (130r17), medo (130v1), fah stig (137r3), þr in þrymmum (137v1), -um and that hie (146r1), -lac (174r1), flet (174v4), g in gled (174v5), and dri in drihten (178v1, not from A). Yet we do not know when in the twenty-five years between the making of B and the first edition Thorkelin's hand grew unsteady. Perhaps the shaky hand should be associated with an illness while Thorkelin was still young (he was 39 years old when he returned from England, only 63 when he published his edition, and 77 when he died). We must presume that many other readings were also added on the basis of A, since B provides so little proof of collation with the manuscript, and so much of collation with A. The w in wedera (139v1) was added after the transcript was finished, when Thorkelin was no longer using v for wynn, for the w was never a v. Moreover, the restored w does not seem to be written on dots, though the ink is clearly different.

Malone leaves the impression that Thorkelin always left dots for letters he was unable to make out:

When B in his copying came upon places in the Cotton text which he was unable to make out, he would set dots for the illegible letters, roughly one dot for each letter. In collation he was often able to decipher parts or the whole of a passage previously illegible to him. In such cases he would write what he had deciphered on the appropriate dots, thus covering the dots up. It is usually possible to detect the former presence of the dots... (p. 23).

Thorkelin, however, used dots inconsistently, often to record a lacuna instead of illegible letters. It would be quite dangerous to rely on the number of dots to judge the number of missing letters. Moreover, when dots are written over, it may only mean that Thorkelin eventually agreed with a reading A preserved for him, or that Thorkelin later made a conjectural restoration of a reading lost before he copied the manuscript. A few examples of his use of dots will show how unreliable they can be.

At the end of 129r19 they appear in the palpably conjectural reading “...þine,” which as we have seen is a late addition based on an emendation of the A reading on 129v1. Here the three dots undoubtedly represent lost letters, not illegible ones, and Thorkelin cannot have fruitfully guessed how many there were, since the restoration was made when he no longer had access to the manuscript. His unreliability expands on the verso, where this restoration properly belongs. When he copied the manuscript Thorkelin apparently conflated the manuscript reading, “...rme þ hine,” by writing “. .þe nine,” which he subsequently altered to “..þæt hine.” Here his two dots correspond to three still legible letters as well as the estimated three letters, bea-, presumed lost at the edge. This is an extreme but not an isolated example. On 133v2 he later wrote du- on three dots that once preceded -guþe. On 143r3 he left three dots for supposedly illegible letters, but later wrote leoht above them. In an original blank space he later wrote “svy..s-” before -colde on 159v1. One would have expected six dots first if he was recording the number of illegible letters as he copied. Though we can still see ðy sel at the end of 180v3, Thorkelin made only two dots here. On the other hand, he made six dots for a reading he eventually restored as dri- on 178v1. On 147A(131)r6 he wrote three dots for what Zupitza accurately describes as “an erasure of some five letters.” Thorkelin was surely marking with dots what he thought were lacunas on 136v1 and 178r3. In the first case he wrote “...mas,” thinking that something was missing before mas, but later crossed out the dots. In the second he wrote “niða...” for niða, but did not delete the dots.

Most of his later additions can be identified only by changes of ink, not by dots underneath. Often there is no great difference in handwriting, and in most cases we have no way of knowing whether his restoration is based on a closer look at the manuscript or a later look at A. Some instances where the handwriting does not change and where the facsimile hides the change in the ink are to in sweoto (133r6), fr in frean (144v11), self (150r2), and (150v1), on woe(158r1, on woc in A), ha in hatan (161v1, ha- also added later to A, perhaps by Thorkelin), and fuslic (161v1). These later additions in B, all made without benefit of earlier dots, may have been made as a result of later scrutiny of the manuscript by Thorkelin. It is equally possible, though, that they are based on later scrutiny of A and later conjectures.

In many cases it is impossible to determine when Thorkelin made his later changes and additions. This is unfortunate because it would be useful to know the stages of his proofreading, particularly whether it was done with the manuscript in England, with A in England, with A in Denmark, or with neither the manuscript nor A, in his rooms in London, Edinburgh, or Dublin, or in his own home in Copenhagen. We must admit that we are usually in the dark when it comes to making these crucial distinctions. Malone remarks that B's copy of 146r “shows four inks, each used, of course, at a different time” (p. 31). Yet how do we confidently date the inks, not to mention the four hands? Malone discusses this page as if it were an isolated case, but his comments still help to illustrate how inadequate our answers to these questions must be:

Nearly all the page is copied in the ink of 1787. The last two words of the top line of the page, that hie, are in another ink, much lighter in color; the hand is still firm and smooth. The -um of fultum [preceding these words] is in a shaky hand; the ink differs but slightly from that of 1787. The reading ride at the end of the fourth line of the page is likewise in a shaky hand, except for the first stroke of r. In this hand ea at the end of line five and wolc at the end of line 16 were also written, or so I think. The changes of v to w on this page seem to me to be in a fourth ink, not very different in color from that used in writing that hie but a bit darker. The hand strikes me as shaky here too, but since we have only isolated strokes it is hard to be sure. The cross-marks that turned d into eth on this page agree in ink and hand, I think, with that hie.59

At face value, Malone seems to identify five inks, not four: the ink of 1787; the one used for that hie and the crossmarks; the one used for -um in fultum; the one used for ride, ea, and wolc; and the one used to change v's to w's. When were they used? Malone implies that that hie was written when Thorkelin still had access to the manuscript, but if so why did he write th for þ, an unparalleled error best attributed to years away from the manuscript. He also implies that -um, preceding that hie, was restored after these words in a shaky hand he elsewhere identifies with Thorkelin's old age (p. 22). Yet, as he says, the ink “differs but slightly from that of 1787.” What, then, is the evidential value of differences of ink and hand?

These are not trivial concerns. To depend on Thorkelin's copy of Beowulf as a source for lost readings we must know when and where his restorations were made. Differences in ink cannot be precisely dated: one that is “slightly” different from the one used in 1787 might just as well come from 1815 as 1789. Nor can differences in hand be reduced to two stages, an early one used by Thorkelin as a young man in England, and a late one used by him as an old man in Denmark. Nor can we always tell whether Thorkelin's collation was with the manuscript or with A, nor consistently detect authentic manuscript readings among conjectural emendations and A readings where the manuscript no longer survives. Only occasionally are there some helpful clues. Folio 191, as we have seen, furnishes some clear information about Thorkelin's proofreading. Since he omitted the same line as A but never noticed the omission it is probable that Thorkelin never proofread his transcript of this folio with the manuscript. The superscript w over the first p in speop v19, moreover, shows emendation without the manuscript at an advanced stage, since the w was never a v, the letter Thorkelin otherwise used for wynn when he originally copied the manuscript.

In addition to proofreading and conflating with A, Thorkelin indulged in conjectural restorations and in a wide variety of both conscious and unconscious emendations. An undisguised example of the former comes from a composite case on 132r20-v1. In line 20 he first emended the scribe's “7”'s (abbreviations for ond) to g's, thereby creating two new words to alliterate with gi- at the end of the line: gylfe gorcneas swylce gi-. Later he erased the imagined g's, and correctly wrote and above each erasure. He next inserted -gantas after gi- at the end of the line, making this conjectural restoration in two stages: most likely -ntas first, from A's reading on the verso, and then -ga- whenever the editorial spirit moved him, not minding that -gantas actually belonged on the next page. This misplacement of text is best explained as direct copying from A. By the time Thorkelin copied the manuscript all that he could see at the start of the verso was “þa”, whereas A could still see “...ntes”, which either he or Thorkelin altered to -ntas. It was undoubtedly Thorkelin who in A made the guesses before these letters, starting with an o (gio, “formerly, of old”) but ending with -ga, an excellent restoration even if it was conjectural.

Though conjectural restorations in B are ordinarily not susceptible to either proof or disproof, Thorkelin gives us reason to be cautious about readings he provides that are not confirmed by A. The various signs of editing, as opposed to straightforward copying, throughout his transcript warn us to be wary of later additions in B along the lost edges of the manuscript. Just as he made emendations as he copied, so Thorkelin may have also made conjectural restorations as he copied. There can be no doubt that Thorkelin often anticipated (both correctly and incorrectly) the meaning of what he was copying. That this tendency could lead to extemporaneous conjectural emendations seems self-evident, but it can be documented as well.

We have reason to believe that the upper left corner of 143v1-3 had crumbled away by the time Thorkelin copied the manuscript. In line 1 he added helle later.60 In line 2 he recorded “...nigt,” a copying error based on the manuscript as it appears now, not as it appeared to A when the manuscript was intact.61 Thorkelin misread ni for the right side of d combined with u, and he misread the scribe's makeshift e (he started to make a u) as a t. The start of line 3 was already gone, too, for Thorkelin has þeoves here where A has lafes. In this case Thorkelin obviously restored a lost reading by anticipating the meaning. Unfortunately he assumed that the proper name he began in the previous line was Ecgþeow, Beowulf's father, instead of Ecglaf, Hunferð's.62

A crucial feature of Thorkelin's transcript that has gone all but unnoticed is the preponderance of emendations. They occur on virtually every page of his transcript. The following discussion of misreadings that seem to be deliberate emendations only begins to treat the subject. All of Thorkelin's mistakes need to be carefully examined to determine whether they are more likely to be emendations or scribal errors. One clear emendation, his change of le to Ic on 135r10, inspires some confidence.63 Less worthy treatment of pronouns occurs in his changes of hie to he (137r6), hi to he (165v12, 193r19), þe to he (134v20), he to þe (147A[131]v12), his ellen to he sellen (198v16), and possibly even him to Danish hun (175v15).

A favorite emendation was changing the Old English spelling burg to the Scottish spelling burgh, another sign that Thorkelin made his transcript after returning from Scotland in 1788.64 When such a form persistently recurs, there can be no real doubt that the change is a deliberate emendation. He misunderstood the comparative form selra and regularly emended it and its oblique forms to selfa, selfe, and selfan;65 once he emended selran to stelran (170v1). On one folio he repeatedly expanded the abbreviated form of the adverb or conjunction þonne as the demonstrative adjective þone.66 He noticed that the Beowulf scribe used two spellings for the name Ohtere (beside Ohthere), and emended the variants (193r15, r18). Repetitive changes like these prove that Thorkelin was purposefully emending the text as he copied, not just making random scribal errors. With this assurance it becomes necessary to scrutinize other changes as possible emendations too. For example, he appears to be making deliberate changes of inflection when he copies laþed for laþes (148r13), alwealdum for alwealdan (150r11), ænige for æigre (150v9), fagre for fagne (154v11), þæt for þas (165v6), and acigdre for acigde (196v21). And he is beginning the editorial task of normalizing spellings when he copies wutun as wutan (188v15) and scufun as scufon (198r7).

His interpretation of word-division in the manuscript, another editorial function, probably leads to other emendations. He did not recognize Ongenþeoes burgum (173r11) and changed -gen- to gan, perhaps thinking that on gan þeoes burgum meant something like “he went into the servant's town.” He changed wig beget (“war befell”) to Wigbege, presumably someone who was nealles folc cyning (192r13). He wrote vigheton for we geheton (188v2), possibly because the edge of the manuscript had fallen away by the time he copied. Another spurious vig had surfaced earlier as an emendation for wic (187r21). He analyzed MS þemec (189v3, i.e., þe mec) as þam ec (“that too”?) by changing the first e to æ. On 167v22 he emended healdanne (“possessing”) to heal donne (“the hall then”?), the latter a frequent spelling of his for ðonne.67

Many errors, it is true, can equally well be common scribal errors rather than deliberate emendations: o for a in cealdost (142v3), metathesis of r and l in ealra (167v19), c for t (and v for w) in wat (170r5), m for ni in niode (176v10), f for s in est (177v11), b for þ in þær (183r1), b for h in heortan (185v11), o for u in lungre (189v10), and n for r in bordum (198r13). But in the last analysis there are so many forms that cannot be cogently explained as mere mistakes that it seems dangerous to give Thorkelin the benefit of the doubt if we want to feel secure about the text. He has ne for nu on 155v20 and ne for no on 181r20, weard for geard on 157r6, fæste for ræste on 157r17, þe for te on 157v14, leoþ for leod on 148r2, se for ge on 149r10. He had swylces for hwylces on 179r17 (and later pointed both s and w for deletion); norge for burge 192v3 (and later “corrected” the reading to borge); and wiða was vida, but never MS niða on 172v10. He inserted spurious letters when he wrote gespring for geswing 148r20, stearwum for searwum 152v6, oft for of 171r20, dreahte for ðreate 183v2, and guðe for uðe 192r14; he strategically omitted a letter in is for wis (196v2), and meaningfully expanded an abbreviation incorrectly in cu(nne) for cu(m) (176r3). In cases like these it seems fairly certain that Thorkelin was thinking more about his incipient edition than about the accuracy of his transcript.

Finally, there are many cases where Thorkelin originally had the correct reading, but later changed it, unequivocally showing emendation rather than scribal error. For this reason, Scyld's body is carried to the beach, waroðe, but not to the water, faroðe (129v6). Hroðgar's men see the results of Grendel's war-craft during the night, on nihtan, instead of at dawn, on uhtan (132v13). Wulfgar is advised to hurry home (Beo ðu on ofeste ham in gan), rather than to order (hat) Beowulf to enter Heorot (138v6). Hroðgar's hall-troop (flet-werod 140v12) is relegated to the status of foot-soldiers (fet-werod) while a Welsh troop goes forth (eode Vealh þeod forð) instead of Hroðgar's queen, Wealhþeow (144r5). Heremod was apparently only left behind (forlaten) with his enemies instead of betrayed (forlacen) into their hands (149v7). Not a boar (swin) but hard work (swinc) appears over the Danes' helmets (158r20). At one point victory (sige), not the sun (sigel), shines from the south (173r9). Viking sea-kings — brimcingas — replace mere ships (brentingas, 190v21). While some emendations (like specan to sprecan 192r6) are easy to grasp, some (like hearpan to hearwan 180r11) defy explanation, without recourse to Thorkelin's Latin translation, which must have gone through numerous metamorphoses. Beowulf is described as ginfæste (long-winded?) in his last words (gingæste word, 191r9). These examples of later changes show even more clearly than the emendations made while copying that Thorkelin deliberately read into the copy what he wanted or expected the text to say.68

In a few critical places, however, Thorkelin seems unusually careful. The first copyist deliberately left blank spaces in his transcript when he was confronted by severely damaged folios — such as 179, 180v1-3, and 198 — no doubt because he knew he was not qualified to make educated guesses about letters he could not clearly read, and because he had no opportunity to finish proofreading his work later. Although Thorkelin himself had no such compunctions or restrictions, B is still at its best where A does not provide readings for Thorkelin to conflate with the manuscript. On folio 179 Thorkelin's enhanced reliability unfortunately produced no additional readings, but on 180 verso, where A deliberately omitted the first three lines, Thorkelin alone preserves for us five letters (bearn) at the end of line 1, and three letters (win-) at the end of line 3. On 198 recto he alone pieced together syfone from around the hole burned in the vellum in line 1, perhaps aided by the context, eahta sum, in line 2. On the last page of the Beowulf manuscript, 198v, only Thorkelin saved the bi- in bið v17, and showed that there was an abbreviation mark over the g in freog v18, even if he did expand it to -en instead of e.

Since B remains generally unreliable, however, I think we must, with due caution and due exceptions, depend far more heavily on A. By relying solely on A we can restore 1898 letters that are now gone at the edges. Thorkelin B adds 72 additional letters, but it is certain that some of these letters are conjectural restorations. To be sure, the conjectures are often persuasive, such as gi[ga]ntas 132v1 and [dri]hten 178v1, which were clearly made when Thorkelin no longer had access to the manuscript. Similarly, plausible readings like ða com 169v1 and genam 175r1 were not the readings Thorkelin originally recorded. We should not close our minds to other possibilities in such cases. On the other hand, we should not doubt the authenticity of Thorkelin's heard where the manuscript now has hea... 139r2, since A merely omitted the line by an oversight.

Elsewhere, however, Thorkelin may well have supplied intelligent but not necessarily authoritative restorations in his copy. Anyone with a rudimentary knowledge of Old English could have supplied the missing letters in wed(e)ra 139v1, (þ)a 140v6, (hw)ær 147A(131)v1, (h)is 147A(131)v3, hi(l)de 163r3, col(1)en 169v4, (s)iððan 173v1, (on) ende 174v3, guð(um) 178r12, (o)n 178v8, (b)am 178v11, byrnend(e) 180r19, (win)trum 180v3, on (þære) westenne 181r2-3, meowla(n) 193r17, wundu(r) 195r14, and hæfd(e) 198r19. As a matter of fact it is more than likely that he made some errors in these cases. For example, he probably expanded an abbreviation mark by writing guðum for guðu(m) 178r12; we can be pretty sure that he wrote i for y in dryhten 178v1, since the second scribe always writes the word with a y; though his conjectural s in siððan 173v1 cannot be doubted, he wrote y for i here (cf. A and MS); perhaps he wrote hwær for ðær 147A(131)v1; surely he interpolated þære for þam 181r2, since westen, a neuter noun, always occurs with þam in the dative elsewhere in the poetic records.

It is notable that many of the letters Thorkelin alone was able to record are parts of proper names, which he would have been able to restore with or without the manuscript. Without B, we too can restore the w in ecgþeo(w)es 160v8, the u in beow(u)lf 167r17, and even lac in hige(lac) 174r1 and are in hroðg(are) 177r2. Thorkelin had a tendency to anticipate proper names, to write them out without carefully considering the manuscript. We have already seen that he once miscopied Ecgþeow for Ecglaf; on the same line he silently corrected manuscript gredel 143v3 to grendel. In one place where A does not provide a reading, Thorkelin first wrote beo for what is now beal, perhaps because he saw bearn ecgþeowes in the next line and so began to write beowulf (178r11-12). He seems to have been thinking of (H)roðgares on 160r20, where he first copied roderas as rodgeras. Wealhþeow was briefly on his mind when he changed f to W in fealh 156v6, and wrote and erased þ as a superscript after it.

In many cases, independent of B, our easiest corrections of A's mistakes will coincide with readings in B, though we will not know whether Thorkelin made obvious corrections of A or copied the manuscript itself correctly. In other cases, by depending solely on A, we will lose unique readings from B we formerly considered authoritative. But the chances are strong that many of these readings are merely conjectural restorations, which we with dictionaries, concordances, and many editions, including Thorkelin B and his De Danorum rebus gestis, are in probably a better position to make today than Thorkelin was in the eighteenth century.


The exact date of Thorkelin's transcript of Beowulf is in itself a matter of little consequence. Yet the date and reliability of the transcripts are central to our understanding of the transmission and foundation of the text of Beowulf in modern times. A general chronology for the Thorkelin transcripts — the sequence of copying, as well as the sequence of proofreading and of otherwise altering the original copies — is therefore a matter of the first importance, one that Thorkelin left in great confusion.

We must deduce a time-sequence, moreover, from often conflicting and always inadequate records. What records we do have tend to support the 1787 date Thorkelin gives for transcript A. He was actively trying to employ a scribe at the British Museum, apparently for the first time, in April of that year. James Matthews, the staff-member he seems to have hired, died before the year was out. On the other hand, we have compelling reasons to doubt the 1787 date Thorkelin gives for his own transcript. They show that he was busy studying and copying other manuscripts in the first half of the year and busy in Scotland in the second half. The extensive records of his use of manuscripts at the British Museum inspire confidence until April 1790, when they abruptly cease listing his name. It was at this time that Thorkelin was offered a major post in the British Museum, and although he declined it, perhaps he had earned the privilege of working with some “lowly” manuscripts without following protocol. The Keeper of Manuscripts was an old friend of his by then.

With the aid of a more plausible sequence of copying than the one Thorkelin implies, we can study the progressive deterioration of the manuscript and the relative value of the transcripts in a clearer light. A page-by-page account of letters now lost to the manucript, but restored by means of the transcripts, reveals the full extent of the damage to the manuscript from the time of the transcripts to the present. This list of lost letters first shows how much of the text can be restored from A alone, and then how many additional readings can be restored from B. But the list of restorations, by itself, can be misleading. Were it not for the accompanying collations we might even conclude from this evidence that the manuscript was in better condition when B copied than it was for A, that B was copied first, and that B was a more reliable witness than A. A page-by-page account of the collations not only makes it easy to see the specific mistakes and idiosyncrasies of each transcript and the relation of A to B on any given page, but also discloses the frequency and distribution of these features and the overall relationship between A and B throughout the manuscript.

The collations of A with the manuscript display the predictive shortcomings of a good facsimile in the days before photography. A's ignorance of Old English was fortunately an irrelevent factor, for rather like a camera he confined his efforts to an objective, letter-by-letter copy of the text. Human fallibility led him to copy some of the strange, insular letterforms as if they were familiar, 18th-century ones; to abandon especially illegible passages on purpose without recording anything; and to leave out a number of other passages by mistake. These uncameralike characteristics notwithstanding, Thorkelin's scribe was a skillful and conscientious facsimilist, whose mistakes as a result were virtually restricted to a confusion of letters resembling one another, and who was notably alert to the task before him.

A's alertness is often paradoxically manifested in what might be called “useful errors,” ones that highlight unique phenomena in the manuscript. Collating A with a printed transcript of the manuscript is therefore quite deceptive, obscuring the most valuable aspects of A's work, the facsimile features. Thus, when he commits an unwonted mistake it is invariably prudent to seek its source in the manuscript or the facsimile. Where the manuscript is now destroyed, we have good reason to trust A as an instructive recourse for describing its condition in 1787. In short, what at first glance seem to be A's most egregious mistakes are in fact his most useful contributions to textual scholarship.

The indications are that Thorkelin made his transcript long after that of A and this circumstance helps explain the particular features of Thorkelin's work. He must have thoroughly studied A before deciding that it was not an adequate source for an edition. His use of A at this early stage is reflected in the A transcript itself, where Thorkelin on the opening pages underlined what he thought were proper names, linked what he thought were faulty word-divisions, and indeed supplied some passages A had inadvertently omitted. In light of this early work in A, Thorkelin's heroic decision to make another full copy of the poem, instead of completing the far less arduous collation he began in A, means that he intended to improve significantly on A. By then, alas, the manuscript was in worse condition than it was when A copied. As a result, Thorkelin's admirable intentions almost inevitably led to conflation and collation with A, and emendation and conjectural restoration of the manuscript.

The collations of B with the manuscript display the predictive shortcomings of an edition-in-progress in the days before rigorous methods were established for editing Old English texts. Often Thorkelin's misreadings in B are not mistakes, as such, but editorial emendations. The most obvious examples are seemingly trivial changes in spelling, capitalization, punctuation, and the like. His rudimentary knowledge of Old English along with his recognition of some connections between Old English and his native Icelandic account for a multitude of other deliberate departures from the manuscript. Unlike A, Thorkelin strove to understand what he was copying, and his most serious mistakes come about by erroneously anticipating the meaning of a passage and accordingly emending it to meet his expectations.

In the context of a damaged manuscript, even trivial editorial changes become consequential, for they are not easy to correct without recourse to the manuscript. In A's facsimile, the mistakes are almost always objective, based on A's perception of the manuscript; but in B's incipient edition, they are normally subjective, producing a bewildering variety of mistakes not susceptible to simple, objective explanations or solutions. We cannot really fault Thorkelin for beginning his edition in B, of course, nor for adhering to 18th-century editorial practice in the process; but at the same time we cannot safely rely on his readings for the lost parts of the text.

Moreover, there is a complicated and largely irrecoverable sequence of copying and revising in B that further undermines its reliability as a textual witness. Two main stages are easy to distinguish: the earliest one, when Thorkelin copied from the manuscript, and a late one, when with a shaky hand he made unsupported changes and additions in Denmark. In between, however, it is impossible to determine when and where Thorkelin made his many alterations, for time and place of origin cannot be deduced from mere differences of ink. Compounding and proliferating these uncertainties, Thorkelin fully prepared his edition twice. It is more than likely that he made many of his changes in Copenhagen before 1807, and many more between 1807 and 1815, none of which could have been based on the manuscript. To be able to trust Thorkelin's copy of Beowulf as a source for lost readings we must know when and where his restorations were made.

The nature and extent of Thorkelin's unreliability in B emerge most clearly in a full collation of A and B. This new collation does not merely confirm, folio by folio, that A contains fewer mistakes than B. It also reveals, folio by folio, how much Thorkelin depended on A, both in the course of copying the manuscript and in the course of later proofreading of B. He evidently used A in tandem with the manuscript to help him supply readings that had crumbled away since A made his copy. He certainly used A in place of the manuscript when repeatedly proofreading his transcript in Denmark. Agreement between A and B accordingly loses its textual significance. This collation, in other words, drastically undermines the evidential value of the B transcript, for we cannot confidently detect authentic manuscript readings among conjectural restorations and A readings where the manuscript is now destroyed.

The inaccurate date and the general unreliability of Thorkelin's transcript of Beowulf, however, must be kept in perspective. In neither case is there cause to think that Thorkelin planned to deceive posterity. He dated his transcript in the same way he dated his research trip, from its conception rather than its birth. After all, Thorkelin B was his copy of the manuscript, not ours. From his point of view his legacy to us was the edition he began in his transcript and labored on for over a quarter of a century. Our enormous, unpayable debt to him is his timely discovery of the poem and his foresight to commission a professional transcript that successfully rescued Beowulf from the ravages of fire.

Lexington, 1 November 1983

Notes to Part Three

1. The tally of “Lost Letters” in Part Two shows that, with A alone, we can confidently restore 1898 letters (not counting 10 punctuation marks, or “points,” and counting ampersands and abbreviation marks as single letters). B provides a total of 72 letters not found in A.

2. “The Text of Beowulf,” Proceedings of the American Philosophical Society 93 (1949), pp. 239-240.

3. See N.F.S. Grundtvig, Bjowulfs Drape (Copenhagen, 1820). Grundtvig had the best guidance imaginable in Rasmus Rask, who not only helped him with Old English, but copied and corrected the transcripts for him (p. xxxiii).

4. See, for example, bold- for blod- (148r19, 150r16), calm- for clam- (151r1), hint- for hnit- (159r20), and half- for hlaf- (190r19).

5. With the notable exception of nine times writing ig for ing (137v12, 141v19, 172v3, 182r3, 183v18, 184r15, 192v4, 193r13, 194r4).

6. Editors of Old English texts ultimately abandoned the use of Junian fonts because undergraduates were baffled by the “3's” (g's), the s's and r's, and the ubiquitous “p's” (mainly wynns and thorns).

7. In the first 918 lines he wrote p for w 73 times, and mistakenly “corrected” p to w 6 times; and he wrote d for eth 20 times and eth for d 5 times (2 of which were formerly d).

8. My tallies (done by computer) often differ from Malone's. Here many of the cases Malone cites (for example, æghwylcne 621, cwen 641, swealh 743) are thorn-like p's or p-like thorns, and so are not included among mistakes of p for wynn; others accurately represent the form in the MS (cwom 2073, geswac 2681, getruwode 1533, hwile 2159, sceawode 2793); one is gone from the MS and should not be counted (wit 589); and some are even corrected (wæs 223, 467, weold 702). My collations list 73 cases of p for w in the first 918 lines, and 7 cases in the second scribe's part of the MS.

9. Many of Malone's cases are thorns with round bows (for example, guþ 678, þa 488, 620, þær 493, þam 425, þe 506, þeah, þelu 486, þeoden 417, þeos 484, þu 506), and one was actually corrected to thorn (þa 720). There are only 7 cases of p for þ listed in the collations.

10. See þanc 928, A30.2; þæm 1868, A55.16; and þin 2048, A60.14.

11. Malone lists only 14 cases of thorn for wynn. He felt that A also wrote wynn for thorn 35 times (15 from the first 918 lines), but here the data are problematic, as are the 39 additional cases of thorn-like wynn and the 56 cases of wynn-like thorn.

12. See also geþyldum 1705, Hroþgares 717, synscaþa 707, þæm 1868, þe 441, 1759, þeaw 1246, þeodne 2709, þin 2048, þingian 156, þioden 2336, þohton 541, þrag 2883, þu 1175. The collations show that A wrote w for thorn only 6 times before line 918 and 6 times after, though it must be admitted that personal judgment must decide whether or not an ascender is long enough for thorn. Many of Malone's 35 cases of wynn for thorn, however, do have distinct ascenders.

13. There are 27 cases of r for “p” left uncorrected in the first 918 lines: 4 of r for p, 12 of r for w, and 11 of r for þ. There are only 8 cases of “p” for r: 2 of p for r, 4 of w (formerly p) for r, and 2 of þ (formerly p) for r. The latter mistakes illustrate A's mechanical corrections of p to wynn and thorn by tracing over the original bows. He also changed one correct r to w in this way, and two correct r's to þ's.

14. He wrote r for p only 4 times after line 918, and p for r only once; similarly, he wrote r for þ only 2 times, and never þ for r.

15. Malone does not count -ðed for -deð (A51.18, 168r1), citing it instead as a case of metathesis.

16. The cross-stroke is slightly torn at the right, but the letter as a whole is certain when bright light it held behind it. See “The state of the Beowulf manuscript 1882-1983.”

17. Malone cites three additional cases where A originally had d, which he later changed to eth, but these changes look like Thorkelin's work too (eard 2493, A72.13; gerad 2898, A83.3; and reced 2719, A78.7).

18. Similarly, he conceivably interpreted the first occurrence of the x-shaped rune for eðel (141v18) as a false start, and so omitted it in his transcript. Malone lists 129 cases of d for ð, but many should not be included because they accurately represent the appearance of the manuscript (for instance, freoðe 167r20, -sweorð 175v2, and wearð 179v10), or because they are conjectural readings from the lost edges (see soðe 142r1, ðe 174r8, ðam 176r9, and wið 187r1). In the latter cases, which are very frequent, Malone has not followed his own principle of determining A's mistakes by collation with that part of the manuscript that remains intact. Even without these exceptions, however, the collations still can list 130 cases of d for ð.

19. Malone lists 51 cases of c for t, overlooking only wit (145v8, A23.11). It should be noted, however, that two cases (utan 2297 and geata 2658) come from the lost edges of the MS (181r1 and 189A[197]r2). Malone lists 22 cases of t for c, overlooking 8 additional cases: scipes 129v11, wot 130r11, sceawedon 134v2, scip 135r12, cyðanne 135v4, micel 135v18, lic 142v6, and sceaða 180r19.

20. One case Malone cites of t for c, -weorc (166v19, A50.10), actually has an original c that was later mistakenly crossed. Some of the cases he cites of c corrected to t are in reality cases of A's early form of t hypercorrected to t (e.g., stapa A4.10, selest and tid A6.3, gast A7.4), and there are probably many more cases now hidden by heavy crossmarks. Most likely Thorkelin is responsible for these “corrections”. But all of the cases Malone cites from the second scribe's part of the MS are c's A immediately changed to t's while copying, and need not be counted in a list of errors.

21. The second scribe drew the top of his t with a tilde-like stroke, compared to the first scribe's flat top.

22. See hreþrinc A54.19 and -rine A55.1. Errors are frequent between his catchwords, no doubt because A mechanically rewrote them for their functional value. We can tell that they were sometimes added later by the form of the letters (note the shape of the bow in þearf A19.20 long before A learned to distinguish wynn and thorn from p).

23. Malone missed 10 additional cases: swege 156v18 (A38.7, formerly s), sæce 172v7 (A57.15), beowulfes 173r14 (A58.11), sæcce 173v11 (A59.2), -ðioes 183r4 (A70.1), - 183r12 (A70.5), syððan 189A(197)v19, 193v18 (A77.18, A84.14), seft 189v15 (A79.3), and scepen 193r4 (A83.11).

24. Malone overlooked f for s in swege 156v18, and does not mention that f was later corrected to insular s in fres 153r16. Note that the s in slæpendne 147A(131)r1 is hybrid, a low caroline s.

25. Malone finds only 4 cases of s for f, missing cræft 146r2 (caroline f) in the first scribe's part of the MS, and nefa 173r4, beowulfes 173r14 (both insular s's), and efnde 194v15 (caroline) in the second scribe's part.

26. See dæg 176v10, hlæw 198v13, ðær 179v9, wæg- 198v6, wæter- 179v13.

27. Actually, three more illustrate the same difficulty arising from eg and et ligatures (egesa 147r2, þegnas 157r10, and mette 147A[131]r10).

28. A's naive “corrections” of low s to f (and of eth to d and t to c, if one sees them from this angle) are relatively innocuous compared to the heady editorial changes Thorkelin makes in the course of his transcribing.

29. Some examples chosen at random are m for in in inge- 155r16 (A36.12); m for ni in cyning 134r18 (A8.3); nn for mi in fa/mi- 134v12-13 (A8.15); 4 ligatured mimins for mi in micel 130r19 (A3.4); mi for nu in nu 135r19 (A10.4); mu for nui in nu ic 162v6 (A45.1); and im for un in wun- 163r16 (A45.18). A is no more at fault here than is the Beowulf scribe.

30. Wrenn, for example, posits a “disastrous chemical reagent” to explain the condition of pages 179r and 198v, even though he realized that “in Thorkelin's day they seem to have been little better than they are now.” He does not discuss the condition of 179v. Beowulf, with the Finnesburg Fragment, 3rd ed. W.F. Bolton (New York: St. Martin's Press, 1973), p. 12. It is unlikely that anyone would have been treating the MS with chemicals before the Thorkelin transcripts were made. [I now believe that Wrenn was correct. See “The nathwylc Scribe and the nathwylc Text of Beowulf,” Poetry, Place and Gender: Studies in Medieval Culture in Honor of Helen Damico, ed. Catherine E. Karkov (Medieval Institute Publications, Western Michigan University, Kalamazoo, MI, 2009), pp. 98-131; see also the text of fol. 179r and v in the third edition of Electronic Beowulf (2011).]

31. “Thorkelin's Transcripts of BeowulfStudia Neophilologica 14 (1941-42), p. 28. This article deals mainly with the representation of MS pointing in the two transcripts. It is revealing that Malone found that “In three passages... amounting to 26 lines of the poem, transcript B has no less than 27 points not to be found in the MS text, or an average addition of about one point a line” (p. 28).

32. See “The Text of Beowulf,” p. 240, and note 5, where he accounts for his earlier “erroneous conclusion” by saying that it “was based on a study of the photostats only.”

33. See especially 175v1, 177r1, 177r3, 177v1, 178v2, 179r15, 179v9, 179v17, 180r1, 180r2, 180r16, 181r3, 182r1, 189A(197)r20, 194v3, 194v4, 196v3, and 198r21.

34. He crossed thorn 3 times on the first page (129r8, 9, and 12), but thereafter carefully distinguished thorn from the abbreviation for þæt.

35. Sometimes Thorkelin skimmed through his transcript changing v's to w's so quickly that the wet ink left “chicken tracks” on the opposite page (e.g., p. 88a).

36. For example, 159r20, 162v15, 163v16 and 164v12. Other Icelandicisms appear in wræk (for wræc 132r15), hun (for him 175v15 and hu 176r9), mægi (a misinterpretation of the scribal correction from mæg to mæl 163v13), nefni (later nefu, for what we restore as nefne 186r13), suni (for sunu 137v5), þeim (later corrected to þam 195v15), and wyni (for what we restore as wynne 189r18).

37. Thorkelin already knew the genealogical matter in Wanley's first excerpt from the opening lines, and as he told Bülow in the letter of 3 November 1786, he felt the rest of the poem would considerably expand on the information in Saxo. See Part One (p. 17, and note 49). The fact that he begins with Fitt I, rather than translating from the start of the poem, reminds us that he first learned about the poem from Wanley's Catalogus, whose main extract of Beowulf, from Fitt I, may well have inspired Thorkelin to begin his translation here, where it really begins to increase “the good Saxo...not a little.” To be sure, Thorkelin relied on Wanley's excerpt when he restored medo in the margin of A.

38. There can be little doubt that Thorkelin at first intended to improve A rather than make another entire transcript. Some early tinkering can probably be seen in his changes of p to wynn and c and c-like t to t.

39. In his edition, Thorkelin was overly cautious about proper names, so that he failed to identify many persons. Grundtvig complained that he overlooked “Sigmund Volsung, Kong Hrædel med to of hans Sønner: Herebald og Hædkyn, Higelaks Søn Hardred, og Friser-Høvdingen Hengest [sic].” See Bjowulfs Drape, pp. xxxi-xxxii.

40. “det halve deraf er nu færdigt i hensigt til Versens rette Afdeeling, og en fierden deel er oversat”; “Arbeidet finder Jeg ellers tungere end Jeg havde ventet. AngelSax. prosa forstaaer Jeg. Men sie förer, som det Islandske, et forskelligt Sprog, Vendinger og Tankens gang.” Sorø 86.A.78.

41. Most of them are transmogrified, for better and for worse, in the edition he produced a quarter of a century later. There, for example, his Gif stol appears as gif Stol with the Latin translation si Ad thronem (p. 15); his wig Weorþunga appears (with a new proper name) as Wig weorþunga, translated Martem supplicibus (pp. 15-16); and ofer swan rade is accurately enough rendered as Ultra cygnorum vias (p. 17).

42. The g's in nægling look especially odd. At the ends of 175r1-4 in A, the g's in ge-, grim, geong, and higes are odd-looking, too, and have been tampered with (black ink over brown).

43. See 134v6, 139v5, 139v8, 155v17, 161v10, 163r15, 164r22, 166v15, 187v16, 189A[197]v17, 189r13, 191r1, 192v15, 196r16, 198r15.

44. See 132v5, 134v17, 137v13, 137v16, 141r15, 146v11, 155r12, 185v13.

45. The usual restoration, f[yrd], is not well founded for other reasons, too. The existence of f might well be doubted in view of missing readings in B on the recto; and the final letter is l (or a stroke resembling l), which still survives under the paper frames. See “The state of the Beowulf manuscript 1882-1983.”

46. In his 1817 collation of the MS with Thorkelin's 1815 first edition of the poem, J.J. Conybeare made dots beneath letters he could no longer see in the MS. By this testimony well over 900 letters Thorkelin records were gone from the edges by the time of the collation. However, the collation is not entirely reliable, for Conybeare failed to make dots beneath letters that are palpably conjectural restorations in Thorkelin's edition. For example, both transcripts show that the bottom corner of 195r was gone when they were made (A has laðm and B has laðn for r20, both have gry for r21); yet in his edition Thorkelin reads laðne and gryre, and Conybeare passes over both conjectural restorations without marking them. A more accurate collation made by Sir Frederic Madden in 1824, recently discovered by J. R. Hall in the Houghton Library at Harvard, shows far more deterioration than Conybeare noted. I would like to thank Whitney F. Bolton for his generous, long-term loan of Conybeare's copy of Thorkelin's 1815 edition, and J. R. Hall, for informing me of the existence of Madden's collation soon after he found it.

47. “De tre fölgende Blade paa urigtig Plads.” In 1983 Mrs. Kirsten Weber, Assistant Keeper of Manuscripts, added another note that reads “NB: De tre blade sad i rigtig rækkefølge ved restaureringen i 1983” (the three leaves were put in the correct order by the restoration in 1983).

48. The only exceptions to the two-sheet construction of his gatherings are a three-sheet gathering for fols. 133-138 and a four-sheet gathering for fols. 139-146. Three of the two-sheet gatherings, however, now have a leaf missing: the first, for fols. 129-132, is lacking the last leaf, suggesting that fol. 147A(131) was formerly misplaced here (corroborated by Thorkelin's original pagination and his use of the catchword fyrst on the last page); the second, for fols. 150-152, is lacking the third leaf, suggesting that Thorkelin made a mess of the leaf and removed it; and the third, for the misplaced fols. 165-167, is lacking its last leaf. I am grateful to Mrs. Mette Jakobsen, the binder from the Arnamægnean Institute in Copenhagen, who rebound Thorkelin B in 1983, for confirming my account of the gatherings.

49. Even p for o in hors is vindicated by the current state of the MS.

50. In fact, the second line on both pages supports this hypothesis. Thorkelin added blode at the end of 165r2 later, and could only see geato, not geatolic (A's reading), at the end of 161r2. Moreover, the letters he erased after geato are not -lic, as one might expect, but -lde, perhaps because he was trying to make some sense of the combination of geato (161r2) and blode (165r2): hence -tolde.

51. One place where Thorkelin was surely following the MS, not A, occurs where he copied mægi for mæl on 163v13. The Beowulf scribe corrected his mistake clearly enough by drawing a line through the g and writing a rather short l (which Thorkelin misread as i) above it.

52. It is worth mentioning in this context that shared mistakes not well supported by the MS are comparatively rare, since A was not prone to make unfounded errors. For a few cases where it does happen, see e for c in sinc on 133v7 and wicge on 135r5; o for a in rond (158v3); a for 1st e in hemninges on 172v8; and u for a in hyrsta on 198v10.

53. Malone's tallies are too low in both cases. He says in “The Text of Beowulf” that in B “I have counted 599 cases of this error — I am sorry I could not make it an even 600” (p. 241; see also his “Introduction,” p. 23). He lists 129 cases from A (p. 11), not always correctly.

54. What Thorkelin eventually made of these misreadings in his edition is be heals sines / Hear wan (remarkably translated, per cervicis nervos / Crine lividis, p. 169) for MS be healfe / næs hear-pan (180r10-11).

55. See Beowulf and the Beowulf Manuscript, p. 246.

56. The note made no sense whatever to Conybeare when he made his collation, for Thorkelin mistakenly placed it after the lines he was describing as a lacuna. Conybeare amusingly registered his confusion in the copy he used for his collation by changing “XV versibus” to “III litteris,” referring to the last few letters on 180v3 (p. 170).

57. The spelling siðan appears to reflect Thorkelin's tendency to give Old English a “Dano-Icelandic” flavor (cf. Danish siden and Icelandic siðan). He makes a comparable “mistake” in siþan (146v1, 145v9, 157v11), which he corrected only once.

58. On fols. 137v18, 143v20, and 144r11 the emendation was later corrected; on fols. 134v1 and 145v7 it was not.

59. Italics added and verse numbers deleted for Old English words (pp. 31-32).

60. Fred Robinson is perhaps right that A miscopied helle for MS healle and that B copied A instead of the MS. See “Elements of the marvelous in the characterization of Beowulf: a reconsideration of the textual evidence,” Old English Studies in Honour of John C. Pope, ed. Robert B. Burlin and Edward B. Irving (Toronto University Press, 1974), pp. 129-130.

61. A correctly copied wit duge but someone, most likely Thorkelin, erased A's e later.

62. For a similar case of anticipation where he is probably right, see 160v8 (B -wes vs. A -æs).

63. Note that he is another possibility, though. Actually the emendation now preferred is [Ic hwi]le, not Ic for le.

64. See 145v16, 149v15, 153r17, 172r13, 196v6; cf. beargh for bearh 164r6. Harvey Wood points out that he inspired John Jamieson to write the Etymological Dictionary of the Scottish Language (Edinburgh, 1808), pp. 215-219, 408-418.

65. See 156v3, 162r18, 168v7, 170r13, 178v13.

66. See 184r6, r17, r18, and v3.

67. Compare the emendation, donne for dome 149r20.

68. Many of these mistakes survive in his edition, where, for example, To brimes waroþe is translated Ad maris litus (p. 5), forlaten is relictus (p. 70), sige is cursu (p. 148), and bremcingas is maris Reges (p. 208); sometimes they are corrected (e.g. swin, porcos, p. 98), presumably by reference to A, but usually Thorkelin's translation attempts to make sense of unresolved errors and corrections alike.