University of Virginia Library

When George Kane published his edition of Piers Plowman A in 1960, he readily acknowledged the necessity—in certain situations—of allowing one's copy-text to stand unrevised by critical scrutiny (152-156). The cases in question were those where the ambiguity of the textual evidence, or the evenness with which it was balanced, or its paucity, prevented the attainment of a valid critical reading. Fortunately, however, relatively few textual problems in A seemed to him of such obscurity as to preclude editorial discrimination. By 1975, when Kane and E. T. Donaldson published their edition of the much longer B version, even fewer situations appeared to require a merely passive assent to copytext. Now the editors believed themselves able to correct the inferred B archetype by comparing it to parallel passages in A and C, the earlier and later versions of the poem.[1] Thus they hoped to restore not only many substantive readings garbled in the extant manuscripts of B but also the minutiae of the authorial style, which scribes had tended to blur through the process of repeated copying. This attempt at restoration led them to displace many strongly attested B manuscript readings.

Shortly after the appearance of the Kane-Donaldson edition of B, several reviews expressed skepticism about this feature of the editors' approach, as well as about what the reviewers took to be an undisciplined application of eclectic methodology in general (e.g., Turville-Petre, Pearsall, Fowler). Since that time, however, only Lee Patterson and Charlotte Brewer have examined the matter extensively. Patterson's discussion, though acute in noting some of the incompatible philosophical assumptions behind the Athlone project (71ff.), is broadly enthusiastic and sounds rather naive in its notion that Kane can somehow validate his editorial practice on an abstract plane where it remains mystically immune to particular falsifications (65, 68-69).[2] Eclecticism as an editorial approach may indeed be unimpeachable by any number of examples of its imprudent application, but surely the only yardstick for measuring the success of a particular editor's practices—and the only one that Kane himself will accept (1975, 220; 1989, 185)—is our estimate of the percentage of correct reconstructions in his text. By contrast, Brewer's assessment of Kane-Donaldson is both sharper in tone than Patterson's and more consistently insightful concerning the specifics of their editorial decisions but may miss the mark in its categorical rejection of Kane's editorial criteria and the means by which they were formulated (89; see Appendix 1 for a detailed review of Brewer's argument).

Meanwhile, the Athlone editions have established themselves as the definitive scholarly texts of Piers Plowman, with the first two volumes having been reissued in 1988 and the Russell C version imminent. In the course of its success, the Athlone project has managed to bequeath its brand of eclecticism to a number of later editors, some of whom have applied the method with a full measure of Kane's boldness but with little of his intuition and philological expertise. Yet it has also stimulated, among certain younger medievalists, an unfortunate distrust of all "eclectic" approaches to textual editing (e.g., Bowers).

Reacting against the sometimes feigned, sometimes exaggerated empiricism of the Athlone B version, yet also attuned to a stylish nihilism prevalent in recent literary criticism, these scholars have sensed an erosion of the epistemological confidence necessary for systematic discrimination of textual variants and have begun urging, in effect, a return to "best-text" editions of medieval works (cf. Kane's description of this movement: 1988, 195-196). Their efforts to act as go-betweens in contemporary theory's courtship of indeterminacy are sometimes inadvertently seconded by older scholars whose

skepticism about critical editions is more pragmatic than programmatic—grounded in plentiful, sad experience of the very real limitations circumscribing any attempt to recover authorial text from a complex scribal tradition. For such disillusioned veterans it is enough if the version in question can be claimed as historically credible, as an example of "what actually was read by the poet's near-contemporaries." The current penchant for studying scribal versions—and treating scribal errors as instances of medieval literary criticism (cf. Windeatt, passim, and Pearsall 1985, 103, with Kane 1988, 194)—attests to the same skepticism. Such factors also probably account, in part, for the extraordinary popularity of Jerome McGann's approach, which defines the production of literary meaning as a collective, public endeavor (with the author at the center of a complex matrix of influences) rather than as an isolated act of creative self-expression (47-48, 81). Where Piers Plowman is concerned, the most tangible manifestation of this textual relativism is seen in recent efforts implicitly questioning the established classification of the extant manuscripts into three discrete versions (Pearsall 1985, 99-100; Bowers).[3]

Much of this study may likewise seem—to those whose predilections are described above—aimed at undermining the possibility of any critical edition of Piers Plowman B. Nevertheless, I am generally in agreement with the Athlone editors in regarding the aforementioned trend as an unhealthy one (Kane 1989, 208). My own outlook is moderately eclecticist[4] and my only quarrel with Kane and Donaldson, though a very extensive one in its ramifications, concerns the level of evidence necessary to displace copy-text or archetype with a hypothetical reading.

Essentially, the problem with Athlone is this: while Kane allows in theory for the possibility that available textual evidence may be inadequate to support any critical judgment, in practice he and Donaldson rarely refrain from making such judgments and seldom impose on themselves exacting standards of evidence. Instead, they prefer to shift the burden of proof to the B arche-type,

whose authority they find it convenient to impugn. Though Kane admits that scribes did, in fact, frequently copy accurately (1960, 126), he and Donaldson also know that the extant copies of B point to an archetype that contained many undoubted errors and therefore feel entitled to treat these witnesses (even when they all agree among themselves) as guilty until proved innocent. Usually the B manuscripts can establish their veracity (in cases where the versions overlap) only by their happening to agree with the version of the text chosen by Kane in his edition of A or with the version reflected in their preferred C manuscripts (Huntington MS 143 and BL MS Add. 35157). Failing that, the editors feel completely free to choose between B variants without regard to attestation and on grounds that sometimes appear to be little more than ad hoc rationalizations (cf. 1975, 163, concerning the primacy of "discrimination as a determinant of originality . . . above that of attestation").

A more candid and disciplined approach would begin by acknowledging that a considerable number of the paradigmatic textual situations in Piers B are so fraught with ambiguity as to preclude isolating any single variant with recognizably superior claims to originality.[5] Oftentimes the Laud or Trinity reading seems altogether plausible, but other witnesses offer one or more similarly credible competitors. In some of these cases, a close inspection of all aspects of the problem will tip the scale; nevertheless, frequently the Laud or Trinity reading can be supplanted only by arbitrarily preferring one particular hypothesis and ignoring comparable ones that might be adduced in favor of the copy-text or the inferred B archetype. In a second class of situations, the archetype is obviously corrupt, but one can readily suggest three or four possibilities that would suit all of the requirements of the text for meter, alliteration, style, and sense. In either of these circumstances, attempting to rationalize a chosen reading purely on the basis of eclectic principles seems futile and tends to undermine the authority of the method in its proper sphere. In instances of the first sort, one should simply follow the copy-text, or correct it, where necessary, to conform to the archetypal reading. As for examples of the second kind, a patently defective archetypal reading certainly cannot be allowed to remain in our edited text; one of the more acceptable readings must be selected, but the reader should be informed—in the textual notes—that the choice was, within the range of possibilities available, arbitrary rather than reasoned. Then the other viable emendations may be listed and

evaluated.[6] To do otherwise tacitly invites the reader using our text to overlook difficulties rather than to confront them.

There is, of course, no special authority residing in any single lection of the inferred archetype of a medieval poem with such a confused textual history as that which must lie behind Piers Plowman. Even so, a neutral base of this kind serves two modest but essential purposes: it fulfills the editor's practical need for a consistent source of merely viable readings (in the numerous instances where a more probable variant cannot be located) while also reminding editor and reader alike that the critical task is not primarily speculation but adjudication—and that adjudication can only operate in the presence of sufficient evidence. Hence the best feasible text of Piers Plowman B would be one that had been critically scrutinized throughout but whose synthetic readings[7] were confined to cases where significant differences of likelihood between variants may be discerned. When offered by the editor, a synthetic reading (whether purely conjectural or not) should establish its rights by its superior "fit" in the immediate or general context, or by its ability to account for the erroneous production of its competitors, or by its greater conformity to the accepted norms of contemporary literary usage (grammatical, lexical, or metrical) and, where possible, by its correspondence to undoubted samples (or, at least, statistical profiles) of our author's own corpus. Ideally, it should also be able to show a plausible pattern of attestation in its favor.[8] Such appeals must assume the burden of proof rather than shifting it, in the manner of a French prefect, to the documentary defendants.[9]

As long ago as 1960, Kane would have found this kind of editorial policy rather timid-something less than a true eclecticism. Thus he comments in the prolegomena to A that

Implicitly lumped together in this argument are two very different editorial policies, both of which are repudiated by Kane but without sufficient distinction having been drawn between them. The first is that of the editor who prints from a single manuscript (except where its contents are manifest nonsense); the second is that of the editor who scrupulously examines all of his textual evidence but who refrains from drawing quibbling distinctions between virtually equal variants and thus often must follow his archetype or copy-text faute de mieux.

Whatever Kane might have thought of the latter approach thirty years ago, his own well-known audacity and rigor were considerably tempered, during the editing of A, with prudence and flexibility. Yet by the time that he and Donaldson produced their edition of B some fifteen years later, the bolder side of his editorial persona had won out. It is only natural to wonder why this occurred. Finding an answer may gain us some valuable insights into the ruminations of perhaps the greatest modern editor of medieval vernacular texts, as well as an enhanced appreciation of the perils inherent in all eclectic editing. In seeking such an answer, we must begin with an understanding of the crucially different textual traditions embodied in the A and B manuscripts. Apparently, it was a failure, early on, to appreciate the full import of this difference that led Kane to develop, in the editing of A, habits of mind, specific editorial practices, and even particular textual commitments that would serve him poorly during the B project.[10]

Seventeen authoritative manuscripts of the A version survive, arranging themselves, roughly, into one family of six (T, R, U, D, Ch, H2), one family of four (E, A, M, H3), two pairs—(V, H) and (W, N)—and three unrelated

manuscripts, J, K, and L. This arrangement is complicated by that fact that W appears to become part of the AMH3 group in Passus 9-11 (after E has lapsed) and by the general tendency observable in all of these groupings for convergent variation (whether by inspired correction, contamination, or mere coincidence) to obscure truly genetic relationships. Faced with the task of editing these manuscripts by recension, Knott had sought to explain away the self-contradictory evidence of his A collations by positing a considerable number of changes of exemplar in the copying process (402-403, 412); but surely Kane is correct in objecting that the means by which this state of affairs came about is less important than its consequences: while the basic family relationships seem secure, they are contradicted so frequently that, lacking a gift of clairvoyance, editing by classical recensionist principles is simply impossible (1960, 84-85, 114).

The salient factors in this situation are, therefore, (1) that we have a rich variety of witnesses so arranged that they provide at least three or four independent sources of testimony for almost every point in the text of the poem, but (2) that those witnesses quite frequently disagree with each other over a fairly wide spectrum of textual possibilities (even within putative families), so that (3) the archetypal reading is often open to some doubt but, where inferable, generally sounds cogent, stylistically effective and, thus, very likely to be authorial. Furthermore, no single manuscript, even T (Kane's copytext), is sufficiently free from corruption to make a "best-text" edition plausible. The circumstances could scarcely be more favorable, or more challenging, for applying the methodology of "eclectic" discrimination.

Like the A version, Piers Plowman B is represented by well over a dozen manuscript witnesses. As in the case of A, most of these are from the fifteenth century though they tend, on the whole, to be datable earlier in the century than the A manuscripts. While two manuscripts of B and an authoritative printed text[11] date from the sixteenth century (G, S, and Cr respectively), between four and six may date to the last years of the fourteenth century (W, Bm, Bo, C, and perhaps L and R). By contrast, only MS V—and Z for those who reject its status as a separate version—among the A witnesses has a claim to comparable antiquity, though at least two others, the AC splices T and Ch, are thought by Doyle to have been written "in good anglicana formata of c. 1400" (46).

Though the number of authorities for the text of B is, then, roughly the same as for A—sixteen as opposed to seventeen—and of approximately the same vintage, the text to which they bear witness could scarcely be more different in its editorial status. For one thing, the family relationships among the B manuscripts are much more defined and predictable, though trivial

convergent variation is frequent (a persistent phenomenon in all three versions of Piers) and sometimes obscures them. Moreover, there are, in general, far fewer substantive variations between the text of B in any two of its manuscripts than is true of A.[12] A cursory glance through the bands of lemmata at the bottoms of pages in both Athlone texts will suffice to validate this point. This sounds very promising for the possibility of editing by recension; and in fact much of the archetype of B is recoverable (cf. Kane 1989, 212; 1948, 5) by a loose kind of recensional formula, viz., when mss W, L, and R agree in a reading—especially if supported by M, G, Cr, or Hm—one can usually be fairly certain that the reading thus attested was found in the common parent of all the extant B manuscripts.

Unhappily, it is precisely here that B begins to pose problems unparalleled in the A tradition. While A almost always provides three or more discrete witnesses to its readings, the hypothetical stemma of B offers—at best—only two lines of independent testimony regarding its point of origin. Of these, Schmidt's limb alpha is witnessed by a mere two manuscripts, one of which has suffered sporadic losses of text (R) while the other (F) is notably eccentric and heavily sophisticated. At times, therefore, our glimpse of the common B parent is limited to a single reliable viewpoint (that of beta = W, Hm, Cr, G, Y, O, C2, C, Bm, Bo, Cot, L, M, H), and the security of any inferences about its nature is always qualified by our never having more than two angles of vision.

Far more serious in practice, however, is a second difficulty. The inferential archetype of the extant B manuscripts, though definable with greater consistency and certainty than that of A, was obviously corrupt to a significant degree. This is especially clear in the second half of the poem (Passus 11-20), where comparisons with parallel passages of C reveal more than 50 failures of alliteration alone.[13] Moreover, this common ancestor of the surviving B copies tended to lapse occasionally into lumps of two or three lines of flat, prosy paraphrase. Of course obvious versification errors also occur with some frequency in the A tradition, but they are much more troublesome in B because of the narrowness of the stemma and the tendency for such faults to affect all witnesses.

B, therefore, if approached conventionally and by itself, is deceptively easy and rather less glamorous to edit than A, affording fewer opportunities for clever reconstructions or discriminations of variants; but the resultant text will reflect some significant archetypal blemishes that cannot possibly be effaced unless one is willing to undertake an almost continuous collation of the text with those of the other two versions. Kane had been unwilling to follow such a method when editing A, and for a very good reason. As he had correctly perceived then, emending any version of an authorially revised

document on the basis of such collations depends on the editor's being able to ascertain whether the passage in question has been subjected to small, sporadic changes by the author. But accurately judging the ephemeral indecisions of a poet who died some six centuries ago is, in many instances, impossible. As Kane himself points out,

Yet when Kane and Donaldson came to edit the B version, they found themselves entangled in well over 1,000 "special circumstances," nearly 700 of which involved them in replacing an archetypal B-version reading with one from A and/or C. Kane realized that such a massive number of exceptions constituted a de facto change of editorial policy and admitted as much in a tersely worded note to the prolegomena of the Athlone B edition: "The editor of A now considers that he allowed insufficient weight to readings from other versions in his editing, and that his earlier view of the situation . . . was mistaken" (75, n. 15).

Was the inferential B archetype patently defective to a far greater degree than even the Athlone editors had initially suspected? Hardly. A four-year study of these myriads of "special circumstances" has convinced me that considerably fewer than half can safely be presumed to attest to archetypal B corruption. A startling number of the remainder exemplify tiny and trivial differences whose relative authenticity Langland himself would have been hard pressed to certify. How one chooses to handle cases of palpable and significant archetypal errors (comparison with A and C does reveal over 250 such instances in B) will invariably pose a number of difficulties for any editor of this multi-versional poem; had Kane and Donaldson simply failed to anticipate this problem, which is particularly acute in the B version, there would have been no need for this survey of their work. But such is not the case. Hundreds of the purported errors listed in their chapter on "The Archetypal B Manuscript" show signs of special pleading, of having been martialed in behalf of a viewpoint already formulated. Why would the Athlone editors have been drawn to exaggerate the defects of an admittedly less-than-perfect B archetype? To see why that course was tempting, we must review briefly the theoretical and practical commitments that Kane had made in the 1950s while editing A.

Faced with the impossibility of editing A by recension, but having at nearly every point a rich variety of witnesses to its text, Kane had opted to evaluate each set of A variants purely on its own merits. In doing this, he had

the advantage of certain heuristic tools of analysis developed by earlier eclectic editors from working with classical texts and the New Testament (Kenney 21-74), where stemmatic reconstruction usually lacked feasibility. According to one of these rules of thumb (the lectio brevior), when faced with an amorphous group of variants one should generally favor the briefer reading. The assumption here is that scribes tend to be more verbose and emphatic than authors. Similarly, one should prefer difficult or obscure readings (the lectio difficilior) to obvious or easy ones on the assumption that scribes are less likely than authors to employ hard words or constructions, whether the difficulty be lexical or grammatical. On the contrary, their tendency will be to smooth out authorial usage into predictable syntactic frames and commonplace phrases and perceptions (Kenney 43; Kane 1960, 130, n. 2, citing Griesbach). Thus, all else being equal, one should give greater weight to variants that reflect archaic or unusual linguistic usage than to their more obvious competitors.

In general, the assumption underlying the use of these tools is that spurious readings will tend to distinguish themselves from genuine ones by the ease with which they may be accounted for as having arisen from those genuine ones, either deliberately or subconsciously, by one of several processes familiar in the copying of all handwritten texts. Of course neither of the aforementioned principles has any compelling logical force in itself; but they seem sufficiently rooted in common sense and observation to provide an editor with a means of assessing probabilities in cases where little or no other evidence can be summoned (a frequent problem with both classical and biblical texts, in which early witnesses are rare and contamination is not).

In attempting to apply and extend these rules of thumb, Kane initially measured their usefulness against a controlled sample of textual situations in A where manuscript evidence overwhelmingly favored a particular variant and where, in consequence, there was little doubt about authorial intention. He was encouraged by the discovery that they did, in fact, appear to predict originality successfully in many instances (127). So accurate did they seem that he came to believe these principles could be used even to overrule the manuscripts in selected cases where the latter plainly supported an easier or more explicit reading (148-149, 158). In other words, he employed a limited base of textual evidence to formulate canons that would allow him to overrule majority attestation in other circumstances. As Kane summarizes the entire process, one must apply to textual cruces "the knowledge of characteristic scribal behaviour gained from passages where the direction of variation is scarcely in doubt, hoping by this means to establish presumption of originality among available readings" (146).

Brewer has argued that, in doing this, Kane involved himself in circular reasoning (75ff.).[14] Whatever hypothetical force such an objection may carry,

in practice Kane seems on fairly solid ground: many contemporary scientific theories are developed in the same way and for the same purpose, viz., to allow us to filter out anomalies from the very body of data which underlies our hypothesis. One might add that, in the case under discussion here, two (theoretically discrete) bodies of data are involved (readings from the A tradition—whose stemma is very broadly based—and readings from the B tradition—whose stemma is never based on more than two inferred witnesses). Thus Kane and Donaldson do not necessarily contradict themselves in using majority readings in A to formulate principles by which they can often reject majority-attested B lections.[15] A more serious theoretical objection is that Kane's reasoning perpetually flirts with the fallacy of converse accident.[16] He and Donaldson begin by inferring (validly, I believe) that principles undeniably illustrated in hundreds—if not thousands—of textual situations apply generally throughout the poem; however, they then proceed to apply these heuristic devices inflexibly and pervasively, as though the fact of their general relevance guarantees their applicability to particular cases, regardless of the uniqueness of those cases. Moreover, they frequently seem to opt for the first hypothesis that occurs to them and overlook the possibility that two or more competing eclectic principles of explanation may require weighing against each other. While Kane is quite correct in noting that textual editors always must settle for less than strict proof (1960, 148), his account of his procedures in editing A sometimes appears to imply that the mere act of explaining how a rejected reading might have arisen is tantamount to establishing that it did arise that way. Likewise, at certain points, the discussion leaves one with the impression that Kane believes in a form of the "coin-toss" fallacy of classical probability theory: that our being able to account for one set of problematic variants by means of some particular eclectic principle actually increases the likelihood that the same principle can be validly applied to the next, unrelated crux (e.g., 1960, 127-128, 145).

To be sure, Kane avers, near the end of his prolegomena to A, that each crux is unique and that he has tried to prevent any of his "various resources for determining originality . . . from becoming rules" (165); readers must judge for themselves how far either of the Athlone editions actually embodies this strategic concession. My own impression is that neither does very well but that the problem remains almost wholly theoretical in A, whereas its consequences are more tangible in the B edition. The reason for the difference resides in the fundamentally different textual situations typical of the two versions. In A we frequently encounter such a wide range of viable variants, variants showing little or no correlation with assumed stemmatic relationships, that one is reduced to the bare necessity of choosing between them on the basis of some rationally plausible and consistent principle, however slender. With B, on the other hand, where the range of variants is customarily narrow and the common parent often easily inferrable, more substantial and contextually specific evidence is necessary to support a non-archetypal lection since a rebuttable presumption of authenticity, however minimal in many cases, attaches to any reading shared by the putative progenitors of B's alpha and beta limbs.

In formulating eclectic principles for editing A, therefore, Kane skirts serious theoretical difficulties; but it is hard to see what other general approach would have worked better, if one were to do more than merely publish a facsimile of MS T. In truth, no editor who aims at recreating an authorial text can avoid repeatedly using some set of heuristic principles. The greater danger is entirely practical: recognizing the specific contexts where these tools are likely to be helpful and what their limits are.

Although they are sometimes spoken of as if they were predictive and prescriptive rules resembling the laws of Newtonian physics, the usual principles of eclectic editing are, in their actual application, retrospective and descriptive, like the laws that explain historical phonological changes. When

we are confronted with a lection that we suspect as scribal, these rules offer us some typical patterns of scribal behavior that may account for the unauthentic production of that lection. But they can seldom give us more than marginal assurances that the word or phrase in question did in fact arise in any particular way, and they can never predict in advance exactly what will happen when a given scribe confronts a designated piece of text. Instead, they provide a probability-based confirmation of suspicions that we have already entertained on other grounds, whether stylistic, grammatical, or historical. Ideally, they should act as a check on editorial intuitions of originality by causing us to question whether our favored reading could plausibly have stimulated scribes to generate its rejected competitors; but all too often they are in fact dragooned into rationalizing such intuitions. This happens because these rules tend—especially in hard cases—to be highly context-specific and easily reversible; that is, they often represent opposite sides of the same editorial coin.

Thus, if we have ill-advisedly chosen a very precise and vivid phrase as authentic, we will be strongly tempted to dismiss another plausible variant as scribal for its tendency toward blandness or flatness (Kane 1960, 134) or even censorship (138). Conversely, if we have been taken in by an actual case of scribal censorship or euphemism, we will easily defend our choice by labelling its more striking competitor "overemphatic" and "lacking in subtlety." Consider, for example, the famous crux in B Passus 7 concerning Piers's resolve to change his manner of life: does he swear to be less busy about his bilyue (as Kane and Donaldson emend on the basis of two A manuscripts) or about his bely ioye (as all B manuscripts uniformly attest)? Leaving aside the question of authorial revision (which can hardly be ruled out in this case), one way of addressing the issue is to ask whether bely ioye seems overemphatic to us (hence the work of a vulgarizing scribe) or whether bilyue seems somehow euphemistic and vague (produced by the vulgarizing scribe's brother—the timid, flaccid scribe).[17]

As with the issue of comparative emphasis, deciding which of two readings is the more difficult can frequently become perplexing. In a given set of variants, one word may be much rarer lexically, but its competitor may, as part of the phrase before us, reveal a subtler or richer meaning or a less

typical syntax.[18] Though recognizing that "hardness" may be of various kinds, Kane and Donaldson occasionally underestimate the potential complexity of such situations and tend to prefer lexical "hardness" above all (see Appendix Two). In other instances, we find ourselves in situations where there is little doubt as to which of several choices constitutes the harder reading but where it is often problematic—as Kane himself notes—to distinguish a harder reading from a variant "that make[s] nonsense or present[s] 'some perverse and distorted grammatical complication'" (1960, 159).[19] Thus the judgments we must make are in fact controlled by a three-value logic, not a bipolar one. Harder readings must be distinguished not only from easier ones but from impossibly hard ones. In sum, the categories of common scribal error help only to authenticate our decisions; they do not make them for us, except in the easy cases where we would not have needed them anyway.[20]

For the most part, Kane's edition of A qualifies as a consistent and rigorous example of critical reasoning in employing these standard tools of eclecticism, but it is by no means clear that he ever articulated to himself the exact nature of the practical dangers that his intuition was usually avoiding. This may partially explain why he and Donaldson are less fortunate in employing these tools in their editing of B.

Another of the snares encountered by the Athlone editors in working with B is also foreshadowed in Kane's A version: reductivism. Kane tended to approach A's cruces monistically, viewing virtually all textual anomalies as consisting of a single correct reading to be discerned plus a host of erroneous ones to be eradicated. It is probably true that the entire Greg-Bowers tradition of eclecticism vaguely encourages such an unrealistic expectation, but Kane and Donaldson did little to resist it.

The consequences are twofold. First, since in the majority of ordinary cases a single correct lection does indeed lie hidden behind the garbled readings of the various witnesses, editorial monism lures one into believing that

that original lection (however nugatory the difference between it and its competitors) can nearly always be ascertained and that one's duty lies in doing so. And no sooner has one decided that monosyllabic adverb X is slightly likelier to be authorial than its rival, Y, than one also notices that Y's occurrence may be accounted for in terms of some familiar category of scribal error. So great is our relief at having successfully drawn such a fine distinction that we are disinclined to ask whether our favored choice might likewise be dissolvable in the same critical acid—much less to ask whether the total mass of textual evidence indicates that our author was indifferent to the tenuous distinction that we have just made and might, in fact, have used the two monosyllabic adverbs interchangeably. Thus, where editorial modesty would dictate simply following some base manuscript for indeterminable cases of this kind, our text instead becomes hypercritical and claims more for itself than is feasible.[21]

This is the only objection that I would register, for example, against Kane's and Donaldson's editing of the last two passus of B, where they use a number of readings from C manuscripts to repair the B archetype. Precisely because the C manuscripts reveal that there has been no substantive change in these passus—and thus authorial revision can be no factor—Kane and Donaldon were entitled to infer that C sometimes preserves a better state of this section of the B text than that reflected in the B manuscripts. In a large majority of cases involving Passus 19 and 20, then, their inferences seem astute and certainly carry a preponderance of likelihood if not inevitability.

But even here common sense tells me that they have occasionally gone too far, as when they imagine themselves able to decide between to (C21.164) and for to (B19.164) in the b-verse "Iesu to / for to seke." At least one of the C manuscripts, Camb. Ff.5.25, agrees on this reading with B and the difference is far too tiny to attempt to resolve by classical eclectic analysis anyway.[22]

The same exact variants are discriminated by Kane and Donaldson at B19. 360/C21.360 in the phrase "Conscience to / for to teche." Similar cases, approximately a dozen in all, are decided with equal confidence, including such spidery fine dilemmas as preye versus And praye (B19.357/C21.357), to þe versus to (B19.197 / C21.197), and or versus or ellis (B20.228 / C22.228).

The second consequence of editorial monism is that it predisposed Kane to discount the possibility that some variants in the extant A manuscripts may be authorial variants.[23] Granted our awareness that Langland repeatedly revised his poem over a period of many years, what more likely than that some of the better pairs of variants in surviving copies of A reflect ambiguities in the author's own mind and copy? Yet Kane effectually rejects this hypothesis. Why? Because it creates untidy problems to think about. Such a utilitarian attitude is patent in his offhanded and grudging acknowledgment that

This description makes it sound as though we are obliged to opt for a known fact (scribal error) over an undemonstrable hypothesis (authorial revision), but the truth is that, where Langland is concerned, both factors seem equally certain in general yet sometimes difficult to distinguish in particular cases. What we cannot assume a priori (cf. Kane 1988, 184) is that we know which of these factors is likelier to be at work in producing a specific set of variants. We can only guess at that as we come to have a feel for the unique features of each successive textual situation.

In essence, authorial variation as an important factor in the determination of A readings appears to have been ruled out categorically rather than case by case.[24] To have done otherwise would have risked exposing the fragile

contingency of the usual tools of critical analysis—designed, as they are, to ferret out no more than a single authorial reading uniquely embodied in distinctive language.

By the time work began on the editing of B, therefore, Kane and Donaldson were already committed to a system and a text. First, they were committed to testing variants by means of some double-edged but venerable eclectic principles that had proved quite valuable in editing A. These techniques tend to yield more ambiguous results with B, where one has not only to speculate on the usual scribal motives for departing from copy but also on possible authorial motives for fiddling with the phrasing of A. If the variant in question is relatively inconsequential and isolated, determining its authenticity can be troublesome. Secondly, Kane and Donaldson were committed to a pair of procedures that were quickly to prove incompatible when applied to B (that is, a rigorous analysis of all relevant textual evidence versus a cautious refusal to employ evidence from other versions of the poem). The problem here was that their collation of B was revealing, as already noted above, a sizeable number of corruptions in the archetype of the extant B copies. Hence

the temptation to invoke readings from the other versions, even if it made a mess out of logic and their prior methodology. Finally, they were saddled with a whole host of specific textual choices already made in editing A—choices that (so they seem to have believed) would have to be defended in B, however poorly attested there, since identical or virtually identical lines often occurred in that version. Though these difficulties must perforce be described separately, they actually constituted three facets of the same editorial challenge: how to evaluate and deal with minute, sporadic authorial revision.

None of these problems appears to have been anticipated, and the failure to do so probably has a fairly obvious explanation: when Kane was editing the A version during the 1950s, there seemed to be no need for a close and continuous monitoring of the lectional evidence from parallel passages in the other two versions. In many cases he appears to have had only vague ideas of the attestational array or the various lectional possibilities in B and C. Nevertheless, not sloth but serious editorial scruples caused this state of affairs. To have done otherwise would have been to tamper with the jury of his own mind by introducing inadmissible evidence.[25] After all, he had already decided concerning these parallel passages that it would be "impossible to say of many lines whether their various versions originated with scribe or author" (147). Hence, "the readings of the B-version and . . . those of C, can never be treated as primary authorities in determining originality for A" (157).

One can easily imagine Kane's and Donaldson's discomfort, therefore, when they began to examine their B collations. The evidence they found there must often have seemed concocted by a malicious fate to expose the inadequacies of the reasoning adopted in Professor Kane's edition of A. Wherever there were parallel passages, the problem kept cropping up: a lection endorsed as authorial in A would turn out to be only weakly supported in B. In fact, in far too many instances it was altogether unsupported. Instead, presenting itself with unanimous or near-unanimous witness, would be a completely different reading—sandwiched into an alliterative line otherwise indistinguishable from one in A.

So what was the difficulty? Cannot authors fluctuate in revising their work, even where relatively polished phrasing is concerned? Yes, to be sure, but, as the Athlone editors must have realized, in this matter degree is everything. Hence they were willing to concede that a limited amount of trivial authorial variation does occur between the versions of Piers. Not to have done so would have seemed too self-serving. They would have had to rewrite hundreds of tiny examples like that in B7.171, which reads, "That Ioseph was Iustice Egipte to loke." Its analogue in the A-version reads identically except that the last stave word—unanimously supported by the A manuscripts—is kepe. In C we find the very same line, but now its final stave word is saue.

Hardly a major issue of style or substance, but in this example Langland exhibits the same self-doubts and fussiness, or inattention to detail, that afflicts the rest of us at times.[26]

A similar case occurs in Passus 10 of B, but here Kane and Donaldson felt obliged to intervene. Will's friendly reception at the home of Clergy is being described and the Dreamer tells us that there "Was neuere gome vpon þis ground, siþ god made þe worlde / Fairer vnderfongen" (230-231). An unexceptionable set of verses? So one might imagine; but in the A version Langland appears to have written "siþ god makid heuene" (173). In these circumstances the editors emended B to conform to A and tried to convince themselves that B's choice for its final stave word constitutes a stale scribal substitution whereas A's phrase is so much fresher as to preclude the possibility of Langland's having written both.

Thus although occasional examples of minute versional variation could be passed over with a judicious silence, Kane and Donaldson could not acknowledge the authenticity of a sizeable number of such petty changes—changes that frequently seemed to result in no measurable improvement of the line and sometimes appeared to weaken it. If they had conceded the reality of this state of affairs, they would have been undercutting the mystique of a specially privileged authorial sensibility (cf. Bowers' recent assessment of Hoccleve's many inconsequential revisions). And this kind of mystique was essential in guaranteeing the monistic inevitability of the process by which Kane had discriminated scribal variants in A.[27]

But sometimes the problem was even more acute. A considerable group of these troubling but strongly attested B lines that differed in small ways from the equivalent lines in A turned out to differ specifically in that they embodied readings that had already been rejected in the editing of A, readings that then had seemed to Kane either not quite so good or less broadly attested than the readings that had been adopted. The unique features of many dozens of these lines are clearly archetypal in B, and in a majority of these cases no one would ever have suspected these B readings of having scribal origins. Nevertheless, Kane and Donaldson presumably feared that to allow them into the Athlone B version would have invited mockery of the eclectic

principles by which these same readings had been rejected in A, where their occurrence had almost always been attributable to some known category of scribal error.[28]

This is, in fact, exactly why Kane and Donaldson could not leave the second of my two aforementioned examples as they found it in the B manuscripts, for they knew that one of the A manuscripts, Ch, also reads "seþ god made þe world" rather than "siþ god makid heuene" at this point. Consequently, denial of this lection's authenticity in B was almost a foregone conclusion: to allow it to pass there would be to concede that revising authors and erring scribes can sometimes change texts in ways that are either attributable to the same causes or, at the least, indistinguishable in their results. [29]

Identical motives appear at work in their treatment of a tiny variant in B3.284. The archetype of the b-verse of this line read: "for kynde wit me it tauʒte." Kane and Donaldson choose to excise the direct object, it, on grounds that the phrase is awkward and excessively explicit. They prefer Kane's choice from the A version: "for kynde wit me tauʒte." Most readers would fail to see how the editors can justify a change of this sort on such slender grounds. Would Langland himself have noticed such a change? But once more, what is really at stake is more subtle than the perennial editorial myopia that causes each of us to imagine himself able to distinguish fly specks invisible to others. In this case the problem is that a whole set of A manuscripts (T, R, U, D, Ch, H2) had read "it me tauʒte" and Kane had already disallowed that possibility as stylistically inferior. Unfortunately, because he lacked a computerized record of his choices, Kane's prodigious memory here failed him; hence on several later occasions in the poem—notably, passages where A offers no reading and thus no potential embarrassment—he and Donaldson were quite willing to allow that the authorial usus scribendi might encompass

such structures as "me it + transitive verb." In fact, at Passus 18.351 they allow the very phrase, "me it tauʒte," as authorial.

Similar considerations may explain the extraordinary respect shown by Kane and Donaldson for the most eccentric of the B manuscripts, Oxford Corpus Christi MS 201 (known as F). Since nearly 100 of their emendations in B depend on this manuscript alone, the editors are obliged to opine that F's scribe had access to a B manuscript superior to the archetype of the extant copies and that he occasionally corrected himself from that source. In reality, there is no doubt that the F scribe was sophisticating, that he collated and at times conflated several copies with his own exemplar. One such copy may well have been an exceptional B manuscript, but the evidence adduced by Kane and Donaldson seems easier to account for on a more mundane supposition, one already noticed by Schmidt: i.e., F agrees many times with the Kane A version against the other B manuscripts because at least one of its conflational sources was an A manuscript (275).[30] Significantly, Kane and Donaldson invoke the authority of F in 67 different instances where an A version reading agrees with it, but in cases where there is no corresponding A line (that is, over nearly 70% of B's text!), they use it only 31 times to displace the common B reading. In only 17 cases out of the 98 where they substitute F for Bx do they employ it with support exclusively drawn from C manuscripts (the other 14 cases concern passages not found in either A or C), and in only 8 of these examples is the F reading clearly identical with the C archetype. The real proof of the pudding is seen in how often Kane and Donaldson cite F as the basis for their emendation but then actually choose a reading from Kane A that is similar to but not identical with F.[31] It is also worth noting that, in 32 out of the 67 aforementioned cases where the editors invoke an isolated F reading because of its support for the text of Kane A, one or more A-version manuscripts supports the generic B reading that Kane and Donaldson are seeking to efface!

Were Kane and Donaldson, therefore, simply deceived in imagining that the A and C versions could be used to purge archetypal corruption from B? Should they have held steadfastly to the versional impartiality and agnosticism announced in the A version? Not altogether. In some cases they must surely be correct. There are, for example, about a dozen occasions where, when B is defective in alliteration, a stave word found in both the other versions may confidently be supplied. Out of a total of some 150 instances in the Athlone edition where A and C agree against B and are used to emend it, approximately half are necessary or probable. Some of the other half are highly improbable and many fall into the category of "possible, but far from necessary and having no more intrinsic probability than the B archetype's alternative."

An example of a close call occurs in Passus 10.5a, where Kane and Donaldson emend B to conform to A and do so with some plausibility. Here the archetypal B manuscript must have read, "'Wel artow wis,' quod she to wit." By contrast, the A version archetype read, "'Wel artow wis, wit,' quod she." No great matter is involved here, but it seems somewhat unlikely that, having once composed a syntactically tight, effective half-line like A's, Langland would later have written the looser, more awkward phrase of B. Moreover, it is easy to explain how the process of copying could have started from the reading in A and ended up producing the version seen in the B archetype. In fact, it is already apparent among the A manuscripts, no less than three of which agree with the B archetype. Four other A manuscripts read the verse, "Wel artow wis, quod she" and altogether omit the vocative "wit" from the verse. Apparently, the omission resulted from eyeskip, with the phrase "wis, wit" causing "wit" to be dropped. Then other scribes, sensing the need to clarify the person being addressed in this line, restored the reference by means of a following prepositional phrase, "quod she to wit." As Greg observed long ago, the easier it is to account for a mistake, the easier it is to suppose that the mistake may have happened more than once (1927, 20, n. 1). Thus Kane must have felt fairly secure in invoking the principle of coincidental convergence to explain archetypal B's agreement here with a rejected A-version reading.

Even this example, however, is less clear than it first appears because it looks as though the C archetype also may have read "quod she to wit." When faced with this possibility we can, of course, continue to invoke the principle of coincidence. If we do, we will say, "A and B scribes made such an error, so why not the archetypal C scribe as well?" In an unknown number of cases, such a line of defense must be correct, and it will always protect our hypothesis from direct falsification. However, when summoned in the face of every attestational embarrassment, it has the correlative effect of progressively eviscerating our theory's probability. On the other hand, if we forego this defense of the proposed emendation, we are left contemplating two distasteful possibilities: (1) that some scribe in the A tradition inadvertently made the line better than it was originally; or (2) that after scribal corruption had occurred Langland either failed to notice or was content to accept the inferior reading in

B and C and thus conferred upon it his implicit blessing (cf. Russell 1962, 42-43). The second of these alternatives looks somewhat more probable, but it forces us to ask ourselves whether our editorial duty includes rescuing our Wordsworth from apparent lapses of attention and taste,[32] or whether our duty is to the work he produced, however inferior it may have been (in isolated instances) when compared to his own earlier effort.[33] In posing the issue of our editorial allegiance, I am not aiming to beg the question at stake with Kane, viz., "to what degree does the inferential archetype of the extant B manuscripts adequately represent Langland's original of that version?" But our judgment about this matter cannot be held hostage to programmatic notions about our author's customary excellence; instead, it must be continuously reformulated in the light of each textual situation. We will, to be sure, eventually arrive at a general view of this relationship, but it can never be allowed, as Kane and Donaldson appear to have done on many occasions, veto power over the unique specifics of the particular variant array.

There are, of course, dozens of cases where the evidence for interversional emendation is stronger than in the aforementioned example.[34] But B 10.5a

still represents, with all of its deceptive simplicity, a fairly typical locus for editorial intervention. Unfortunately, many of the targets for this activity in the Athlone edition of B seem much less warranted. In situations where only A and B share a line, my calculations suggest that nearly 80% of the Kane-Donaldson emendations from A are unnecessary and show little or no superior probability.

It might seem that, wherever both of the other versions agree against B in reading a shared line, we would be fairly safe in emending; but things are seldom so simple. Even if we put aside the intrinsic likelihood that Langland must sometimes have changed his mind in revising from A to B, only to change it back again while reworking B to C,[35] we are left with many cases that are hardly transparent or compelling. Sometimes the agreement of A and C is itself the questionable product of a tenuous chain of inference. That is, similar arrays of variants may exist in all three versions but with strikingly different proportions of attestation. Having chosen one of these in editing A, the editors are inclined to choose the same variant in C, however weakly attested; then, in spite of their aversion to stemmas, they vote these two inferential archtypes against a nearly unanimous B reading.[36] Convergent variation thus transmutes the most intransigent lemmata into clarity. Above all, Kane and Donaldson are unwilling to allow that Langland might occasionally have lapsed into revisions of merely scribal quality in B, only to repent and rediscover the merits of his more felicitous original in C (cf. Kane 1989, 210-211, concerning the "postulate of excellence"). And yet our own experience as writers demonstrates every day the probability of just such an occurrence.

Another curious feature of the Athlone methodology is how little scope it appears to grant to contamination as a factor in explaining some of the suspicious situations where a variant strongly attested in one version will also appear in two or three isolated copies of another. Of course Kane was well aware of gross lateral transmission between the versions and documented dozens of instances (1960, 30-31) in which whole groups of authentic B or C lines are somehow injected into A-version copies. When this happens, an editor's obligation to his version is clear: he must excise the alien material. However, neither Kane nor Donaldson ever really addressed the problems posed by the same phenomenon when it occurs on a smaller scale than that of the line or paragraph.[37] Potential cases of micro-contamination were usually

interpreted by Kane as instances of coincidental convergent error, with the preferred variant (i.e., that chosen in his edition of A) being original to all versions and the rejected one having arisen spontaneously in each version through independent scribal error.[38] Donaldson, by contrast, once seems to have accepted the possibility of considerable memorial, if not collational, micro-contamination (1955, 183)—and, by implication, the idea that more than one variant in a given set might be authorial—but provided no guidance for detecting the phenomenon or dealing with it.

Yet the likelihood is that such micro-contamination did happen—repeatedly but unsystematically. The early copyists betray no realization that the poem exists in three different, autonomous versions.[39] The most that can be

said for them is that they were aware that some of the circulating copies were longer or more complete than others and that some seemed to contain better readings than others. The survival of numerous interversionally spliced copies attests both to this scribal ignorance and to their employers' demands for the fullest, most accurate text of the poem available. These two factors alone make it probable that some of the more complex sets of shared A/B/C variants result from individual lections (both authorial and scribal ones) having passed between manuscripts of different versions (cf. Kane 1960, 55, n. 1, where Housman's claim for sporadic lectional contaminations in Juvenal manuscripts is cited).

How often it happened is less clear. We know that it took place in the sixteenth century, and there is every reason to believe that it occurred in the fifteenth as well. Among the A manuscripts, MS Harley 6041 (H2), a composite AC manuscript, shows numerous (usually trivial) marginal and interlinear "corrections" throughout in a different, later hand and ink. These interpolations all come from an unknown B manuscript. The extraordinarily high percentage of isolated, archetypal B readings in MS Society of Antiquaries no. 687 (M, another A-version copy) suggests that something similar happened to one of its ancestors. The same may be true of the now famous MS Bodley 851 (Z).[40] I have already mentioned that a process of this kind probably influenced MS F of the B tradition. The combination of many inconsequential agreements—and some highly significant ones—between early parts of B MS G and the A tradition (against the evidence of all other pure B manuscripts) invites the hypothesis that a similar process of sporadic microcontamination affected one of G's ancestors (viz., its exemplar for Prologue-Passus 7).[41]

As for the most blatant of these cases, the evidence of his ink suggests that the sixteenth-century corrector of H2 set himself the task of marking this manuscript for medial pauses (by reference to another copy already so punctuated) and scribbled interlinear corrections only when his eye happened to notice interesting discrepancies between H2 and his checking copy. Presumably the corrector used a manuscript of B for this task not because he had a deep interest in the subtleties of Piers Plowman textual variation but because it was the copy most readily available to him. From our perspective it would seem worse than counterproductive to proof a copy of one version of a poem against a copy of a different version, but such cross-checking of the text of Piers Plowman versions can be done more easily than might be imagined. The trick lies in the macaronic character of the poem. As a rule, the hundreds of Latin tags are set off by rubricated or stylized lettering, making them readily useable signposts to specific parts of the poem's three recensions, in which many passages are closely parallel.

We can observe the scribe of MS Hm 114 (or one of his predecessors) doing this very thing in the copying of Passus 3 of B. The gist of his text leaves no doubt that his primary source here was of the second version of the poem. Moreover, at this early stage in his work he has not begun the wholesale mixing of batches of lines from each version that characterizes the later passus of his copy. But several times we catch him consulting parallel passages of an A and a C manuscript and adding to his copy a Latin scriptural citation not found in the B archetype.[42] Doing this leads him to notice small differences

of phrasing in the English text, and occasionally to contaminate his exemplar, as when he alters "Yiftes or yeresyeues bycause of hire Office(s)" (B 3.100) to "Gyftes or presentes in ʒougþe or in elde" (cf. A 3.89). Although MS K of the A version parallels this line in Hm 114 perfectly, it seems likely that the substitution in these two manuscripts of "presentes" for "yeresyeues" (which is archetypal in both A and B) is the product of mere coincidental convergence, an illustration of the process whereby scribes sometimes create the illusion of textual affiliation by choosing identical easier synonyms in eliminating hard words. On the other hand, the occurrence of "in ʒougþe or in elde" in Hm 114 requires a different explanation. Here there is no doubt that the phrase in question is archetypal in A but illegitimate in B, which reads "bycause of hire Office(s)." Someone in the Hm 114 tradition found this very distinctive—though bland—reading in an A copy and, for whatever reason, thought it more authoritative or appealing than the B reading it replaces. Perhaps he was dubious of the metricality of the B archetypal line (= aaa/ xy).[43] His motive may have been as simple as an illegible patch of copy in his exemplar, but similar small examples in this and other manuscripts sometimes seem explainable only in terms of the uneven whimsy of a poetaster.[44] Ironically, the same process appears to have happened the other way round with MS W of the A tradition. Someone in the transmission history of this manuscript chose to evict the native A reading ("in ʒouþe or in elde") in favor of the B reading ("bycause of hire Offices").

Thus the existence of the same pair of alternative readings in two versions of Piers does not automatically prove the presence of any error at all, nor the accidental convergence of scribal inadvertencies; rather both versions will, on the odd occasion, have originally contained a separate authorial reading and each in turn will have been deliberately "corrected" (i.e., contaminated) from the other tradition. This same state of affairs occurs at times on a larger scale. That is, an A manuscript (or several of them) will have acquired the B version of a particular passage, while one or more B manuscripts will reproduce the A version of the same passage. For example, at B 3.51-62, manuscript pair R/F omit the B version and read instead three lines that correspond to the A version of this passage. Conversely, MS N of the A tradition witnesses the much longer passage original to B/C! Various explanations are conceivable for this example, including mere coincidence.[45]

But with some sets of interversionally shared variants, one cannot rule out the possibility that, at an early stage of transmission, a heavily revised authorial foul copy (with ambiguously marked interlinear and marginal afterthoughts) was repeatedly used, both as exemplar and checking copy.[46] If, in fact, this took place, it would follow that some of the contamination is authorial in origin.

Under such circumstances, especially where versional exchanges of individual words or brief phrases are concerned, the bipolar logic of the Athlone method breaks down; so it comes as no surprise to see that the editors will have none of this. Adducing lateral transmission is all very well when one is explaining the impossibility of genetic recension within a given version (Kane 1960, 53-114 passim), but the three versions are conceived of as hermetically sealed off from each other—at least at the level of the individual lection. After all, only such a theory of independent vertical transmission will allow tacit re-introduction of stemmatics at a later stage so as to emend B by mechanically voting A and C against it.

Much of the discussion hitherto merely elaborates in a theoretical context points made by various reviewers more than a decade ago. My final point, however, has never before been addressed. As soon as we accept random lateral transmission as a significant (though far from exclusive) cause of the shared variants mentioned above,[47] another methodological difficulty comes into clear focus: Athlone B's massive application of interversional stemmatics must partially—if unconsciously—hinge on an obscure and unprovable hypothesis concerning the three versions' order of publication. The Kane-Donaldson operational assumption was that publication order corresponded to compositional order (which, since Skeat's work, has been confidently established as A>B>C). Yet only George Russell has ever articulated anything resembling a detailed description of this hypothesis—perhaps because it seemed so obvious and commonsensical to his colleagues.[48] Having overestimated

(in my view) the role of coincidental convergence in producing parallel sets of variants among the three versions, Kane and Donaldson seem to have attached little importance to the question of publication order. All that one needed to know for editing a poem whose versions were genealogically discrete in vertical descent was the order in which the author had composed them.

But if intermittent, often trivial, microscopic contamination did characterize the early scribal history of the versions, collating A and C variants against B to find B's archetypal errors (though far from valueless or impossible) would require more sophisticated and cautious methods than those actually applied.[49] Similarly, if the versions were not released in the same order in which they were composed, scribal perceptions of the relative authority of their divergent readings might well have differed from ours, and the direction of influence (i.e., for memorial as well as collational contamination) might not coincide with what we would expect from compositional order.

Of course any discussion of this kind must remain speculative, for—barring the discovery of fresh manuscripts—relevant evidence concerning the actual order of publication of Piers A, B, and C is slim and ambiguous. One clue—dialect distribution of the extant copies—seems compatible with the Athlone hypothesis but can hardly be said to strengthen it.[50] Several others,

equally slim, raise doubts about it. Thus Doyle's recent paleographical study of the manuscripts concludes that the oldest extant manuscripts are of the B and C versions, not of A (36 and passim; cf. Parkes's estimate of Trinity College, Dublin, MS 212 cited in Kane 1985, 912). In both cases we possess a significant number of manuscripts apparently produced before the end of the fourteenth century. Surviving copies of A, on the other hand, seem noticeably later than we might have expected for a "first edition": all but Vernon and Bodley 851 clearly consignable to the fifteenth century and many to its middle or latter half.[51] Similarly intriguing is the fact that A copies were, on at least four separate occasions, completed by splicing them to C conclusions but never, so far as we now know, by using B (cf. Kane 1988, 181, 186 for the view that this circumstance is merely accidental). And yet B appears to have circulated more widely than C. Many extant B copies show a multilayered complexity of dialects typical of London production (Samuels 240-241) but also indicative of a wider circulation (and more extensive recopying) than that achieved by the C version, whose pattern of dialect distribution is tightly focused in the southwest Midlands (cf. Russell's opinion, n. 48 above). Concerning C's range, Samuels observes: "That such a distribution could have been the work of a colony of western immigrant scribes in London seems unlikely in the extreme. . . . Such a concentration of C-MSS in dialects of the south-west Midlands is unlikely to have come about without an authorial presence in the area, and Skeat's view that the author returned to Malvern in later life is thus shown to be highly probable" (240).

What, then, are we to make of the anomaly that puzzled Russell, viz., that B, which has always been the most popular version of the poem, reveals a remarkably homogenous—though uneven—text and a tightly bifid stemma (perhaps suggesting a tradition that never consisted of many manuscripts)? The answer that seems best to account for these facts is that B was not only the earliest version of the poem released (indeed the only one sanctioned by the author himself) but that the reproduction of its text was, in the beginning, rigidly though unintelligently controlled by some presumably non-authorial agency (cf. Kane-Donaldson 42, on the production of Bm/Bo/Cot).

Despite the obvious deficiencies of that text, B quickly achieved something like a canonical status (modern parallels easily suggest themselves). This status allowed its unique readings—the bad as well as the good—in later decades to infect, in both subtle and blatant ways, various manuscripts of the other versions—especially A copies (since C remained largely confined to a small and remote region). Nevertheless, it survives in slightly fewer copies today than the less influential A and C versions because of the greater volatility in London literary fashions, which must have made it look much more outdated (and controversial?), by the middle of the fifteenth century, than its cognates would have seemed in Norfolk or Herefordshire.[52]

To conclude, though one cannot place much weight on the slender clues provided by paleography and dialectology, they do at least suggest an alternative hypothesis concerning publication, one that seems fully as likely as the one taken for granted in the Athlone project. Namely, that A (though representing an earlier state of the poem than B) was, because of its fragmentary nature, virtually unknown (may not even have circulated among a coterie) and passed with Langland's other papers and the incomplete C version into the hands of a literary executor.[53] Kane and Donaldson are themselves inclined to believe in the existence of such an executor (127). If such a set of circumstances occurred, and the transmission histories of A and C were in the beginning briefly intertwined[54] (the only A manuscript with numerous Worcestershire dialect features is Vernon, the oldest), then one of the reasons why surviving A and C manuscripts often seem textually closer

to each other than to B would have nothing to do with their "superior" quality. Another problem that this might partially account for is the commonly occurring situation where only A and B parallel each other and where a reading of the B archetype is reflected in one or more of the A manuscripts. As noted above, the Athlone methodology assigns the usual cause of this phenomenon as "coincidental convergent variation" and archetypal B is therefore convicted of corruption. A likelier explanation in some of these cases is that scribes already familiar with the earliest published version (i.e., B), reacted against different readings they found in A and replaced them from memory or by deliberate collation with B readings.[55] Such an explanation is far from all-encompassing and should not be construed as a monolithic replacement for coincidental convergence.[56] Rather, it is grounded in the realization that similarly suspicious textual situations can arise from a variety of causes, some of which do not diminish an archetype's credibility. Such an approach allows us to acknowledge that, in most cases, Kane was probably right about the correct reading of the A version, but it also affords to B the customary assumption of innocence denied it by the Athlone editors.

That is the essence of my methodological reservation about the entire Athlone project. Certainly one would be remiss to aim at merely reconstructing the archetype. In essence, Skeat did that. And Professor Kane is assuredly right when he reminds us that perfection is not a viable alternative. Whatever method we adopt, and however flexible we are in adapting it to unanticipated situations in our text, we will sometimes err. But in editing as in law it seems essential that we decide in which direction we will err—either by convicting some of the innocent or by letting some of the guilty go free. I think far less offense is done to the integrity of our author's text and to our own daily experience of the complexities of composition by the latter policy than by the former. I would not be understood to be invoking the analogy of an American criminal trial and to be insinuating that the archetype must be guilty beyond all reasonable doubt before emendation occurs. Civil procedure's requirement of a preponderance of the evidence is all I would ask. But the burden of proving should be on the editor, not the archetype.