University of Virginia Library

Models for the Textual Transmission of Translation: The Case of John Trevisa [*]
by
D. C. Greetham
I

A great number of works written in Middle English are in fact translations, in one sense or another, of originals written in French, Latin, Italian, and so on. This has long been recognised by mediaevalists in the voluminous investigation—and frequent publication—of "sources and analogues", where the critic might be able to observe the structural, narrative, linguistic, and aesthetic changes wrought by the translating author. On the textual side, there has been much editorial concern for establishing specific readings (and justifying them by appeal to these sources) in works by major authors. But there has been comparatively little study of the theory of such textual relations; the attention of critics, literary and textual, has been on the particular manifestation of translation in a given author, a given work, or a given reading, and the discussion of translation has therefore tended to be embedded either in editions of Middle English works or in critical examinations of mediaeval authors. For example, although Vinaver attempted to draw general theoretical maps for charting the possible range of scribal errors that could occur in the act of copying,[1] when it came to the editing of his Works of Sir Thomas Malory he was content to produce stemmata that reflected the specific textual conditions which obtained between the Winchester manuscript, the Caxton edition, and the French sources.[2] There was no theoretical study of the possible routes of transmission across the translation gap, but only of the actual (and therefore inevitably limited) translation process as used by his author—and this is perhaps only right when an editor is concerned to defend the individual emendations introduced into his text. Where arguments over translation have occurred, as for example in David C. Fowler's attack on George Kane and E. Talbot Donaldson's smoothing of the B-Text of Piers Plowman to conform with the A, they have

usually concentrated on the validity of certain readings rather than on the theoretical attitude to translation.[3]

In fact, there are theoretical positions underlying the work of Vinaver on Malory, Kane-Donaldson and Fowler on Langland. For Vinaver, it might be appropriate to suggest that his models for translation all rest upon the notion of fidelity, so that where alternatives in the Winchester and Caxton versions occur, the test is an appeal to the French, on the assumption that this will yield Malory's likely reading.[4] For Kane-Donaldson, the argument

might be that internal harmony—metrical, alliterative, textual—should be the basis for editorial decisions, no matter what potential "interference" there may be from external authority (such as bibliography or translation), thereby assuming a prejudice on the editor's part that the author's work is motivated by a similar aesthetic.[5] For Fowler, in his response to Kane-Donaldson, the position would be that internal harmony is to be modified by textual standards available through such external authorities as translation. As Fowler says (p. 32), "One should not change a line just to make the author's words fit the text he is citing; but caution is surely advisable in emending if the MSS uniformly attest a reading which translates the Latin quoted."

What is currently missing from, or undeveloped in, the theory of editing in Middle English is a model or models which can be employed in evaluating the conditions to be found in the translated and translating languages and texts.[6] It may also be that similar models could be useful in other periods

and other fields, for as far as I am aware, the filiation paradigms constructed by such theorists as Maas, Pasquali, Quentin, Dearing, and Greg have been concerned only with establishing suitable methodologies within a given language, and have not normally devoted much attention to the problem of textual criticism between languages.[7] Thus, Dearing's "rings" are no more help than were Greg's "calculus" or Quentin's "positive critical apparatus" in charting the alternative structures and transmissions possible between, say, Latin and English. Stemmatic models have been generally responsive to textual features encountered in the transmission of texts in a single language, not of texts moving from one language to another—and this despite the fact that translation may frequently offer assistance in emendation which is not available to editors working on texts recoverable only from extant copies in one language.

I propose to begin this process of creating models for stemmatic analysis of translation by citing textual examples from the work of one of the most prolific and influential of Middle English translators, John Trevisa, a translator who moreover left considerable discussion of the nature and mechanics of translation.[8] I therefore select Trevisa in part because his known (and avowed) position and practice on procedures for translation make him an ideal author for such a "control" necessary for the preliminary charting I suggest, and in part because his actual work appears to be governed by what we may call the "clere and playne" doctrine, involving a self-effacement of the translator which reduces the potential creativity (or idiosyncracy) of the derived text to a manageable or discoverable level, despite the lack of holographs

for the author's work. My hope is that the models outlined in this discussion can perhaps be used in other periods and fields for the charting of similar problems.

I posit three theoretical classes, generating four ideal models between the translated and the translating languages:

II

Examples using MODEL A (with contexts):

                 

	Context	Suggested Emendation
a)	Eng.: þerefore erroure of naciouns and dyte of secular profytes (p. 791)[+]	*poetes
	Lat.: et gentilium error et secularum carmina poetarum
b)	Eng.: And more freliche þey ben þere yclepede Agareni (p. 745)	*vereliche
	Lat.: unde verius vocantur agareni
c)	Eng.: in þe eende of Siria (p. 743)	*hede
	Lat.: in capite Syrie
d)	Eng.: Mantua was ybounde of Mantus (p. 821)	*yfounde
	Lat.: mantua que a manto . . . est condita

                               

e)	Eng.: þe londe is playne with moche lese and pasture (p. 761)	*mareys
	Lat.: terra plana pascuosa palustris
f)	Eng.: and to Tracia in þe northe eeste	*southe eeste
	Lat.: ab euro autem tracie
g)	Eng.: stonden in occean aforne þe lefte syde (p. 762)	*lesse
	Lat.: in occeano sunt site contra leuam
h)	Eng.: Boeme is a partie of Messia toward þe eeste syde and weste by Germania (p. 742)	*nexte
	Lat.: bohemia pars est messie ad plagam orientalem iuxta germaniam posita
i)	Eng.: Withynne þees londes . . . estewarde ben Rodes (p. 749)	*ilondes
	Lat.: inter has insulas . . . ab oriente est rodus
j)	Eng.: An þe men of Asia ben nameliche disposed in þat (p. 753)	*meneliche
	Lat.: Homines vero asie . . . mediocriter in hoc se habent
k)	Eng.: The ryuer Pacco worshepede þis londe wiþ tokenes and brookes of golde (p. 778)	*torenes
	Lat.: quam fluuius pactolus extulit diuitiis torrentibus aureis
l)	Eng.: And erþe Germania is a riche londe and noble of strengþe (p. 732)	*eiþer
	Lat.: vtraque germania diues est terra et inclita

The simplest form of corrective paradigm is demonstrated in MODEL A (Fig. 1). This is the Perfect Linear Class of consistent source and consistent derived text, with, however, some identifiable 'problem' in the translation, noticeable, we assume, by context but not by collation.[9] Stemmatically, this relation can be represented by the straightforward genealogy of the first model, where S is the source, the wavy line the act of translation, O” the archetype of the extant MSS in the translating language (in this case English), and the parenthetical ABCDE the agreement of those MSS. Note that the model would assume the same shape whether there were such a 'problem' in the translation or no, and the validity of McKerrow's rule of contextual

identification[10] would have to be questioned before any editorial methodology for these examples could be invoked. Thus, arranging the readings given here to support MODEL A in a sequence dependent upon their likely contextual identification and upon no other consideration, possibly only erþe (in putative error for *eiþer) and maybe nameliche and tokenes would be contextually startling enough for the editor to be alerted to a problem in translation. Extra-textual information, derived, for example, from geography and history, would be necessary to alert the editor abiding by the contextual rule in the cases of ybounde/yfounde, moche/marys, northe eeste/ southe eeste, lefte/lesse, weste/nexte, londes/ilondes, but no purely contextual analysis would isolate profytes or freliche or even hede, all of which are eminently sensible scribal rationalisations, and all of which fit their contexts perfectly. If translation is to be considered as a textual paradigm at all, then McKerrow's strict contextual rule, at least in reference to the translating language, must be dropped from our editorial discipline in MODEL A. Note further that all the suggested restorations (marked with an asterisk) should ideally be palaeographically closer to their corrupt derived forms than would be possible in a similar Latin corrupted tradition. I say this bearing in mind Housman's strictures on the limits of the palaeographic rationale,[11] and would argue that the principle is demonstrably clear in, for example, the relations of eende/hede, ybounde/yfounde, northe eeste/ southe eeste, lefte/lesse, weste/nexte, londes/ilondes, nameliche/meneliche, tokenes /torenes, erþe/eiþer, and freliche/vereliche (with the latter assuming a scribal muting of an initial voiced SW v to f), and has high probability in all but the profytes/poetes pair, where an editor might have to decide whether palaeographic variation in the source could not also have produced a similar effect (i.e. Latin prophetarum/poetarum). To put it as simply as possible: for MODEL A to be the correct paradigm for this type of textual problem, it must be easy for the scribal eye in the translating version to create the extant corrupt form from the putative original and virtually impossible for the scribe of the translated version to have made a similar error which could have been rendered into English by the translator. If a Latin MS reading prophetarum were to be discovered, then our suspicions about the fitness of MODEL A for this particular case would be confirmed. In the light of the universal poetarum all we can say with confidence is that MODEL A seems more plausible than any other, based on the evidence to hand. For verius, capite, palustris, euro, leuam, iuxta, mediocriter, torrentibus, and utraque an English scribal corruption of an originally correct translation would take care of all the evidence and would give us a useful translating paradigm.

Examples using MODEL B (with contexts):

Context and Variants

• a) Eng.: þe walle was of brode/brande tiellen (p. 724) Lat.: murus de coctili bitumine compactus
• b) Eng.: Of þat toune/toure þe cite þat þere is ybilt (p. 744) Lat.: turrim a qua ciuitas ibi edificata
• c) Eng.: for þere þe lynage/langage was confounded of hem þat belded þe tour Babel (p. 738) Lat.: ibi lingua edificantium turrim babel est confusa
• d) Eng.: and bereþ plente of corne and wyne/whete/wynde/of wordes (p. 758) Lat.: fecunda frugibus et vinifera
• e) Eng.: Also Inde is moste fulsome/holsome in westren wynde (p. 769) Lat.: Est autem india fauonio vento saluberrima
• f) Eng.: And þis londe was in olde ryme/tyme yclepede Tynatria (p. 811) Lat.: antiquitus trinacria dicebatur
• g) Eng.: And he is sharpe of wordes/woodes (p. 779) Lat.: Est autem siluis aspera
• h) Eng.: The londe is clippede/cleppede alle about with þe see (p. 748) Lat.: terra est mari vndique clausa

A more complex corrective paradigm involves the problem of consistent source versus inconsistent translation, in the Imperfect Linear Class represented by MODEL B. Figure 2a is a straightforward example. In describing

Babilonia, Trevisa's text in DPR reads, "þe walle was of brode/brand tiellen" to translate the Latin coctili='burned'. This example reflects the same basic principle as the first model, that the divergence (or 'error') should be more plausible palaeographically in the translating than in the translated language. That is, an omitted nasal macron from a could very easily produce the rationalised scribal form brode, whereas a similar (non-extant) corruption of coctili to produce the same effect would be difficult to argue on palaeographic grounds. Emendation from a copytext brode to brand is therefore justified. The same principles hold for the alternative English forms listed in 2b-2h.

This model (as its stemmatic arrangement demonstrates) can be considered to follow the same transmission form as MODEL A, except that in this case, both 'corrupt' and 'pure' readings are retained in the extant MSS, whereas in MODEL A, either 'pure' or 'corrupt' (but not both) readings appear, and the linear form is unmodified. This ideal MODEL B is obviously the easiest to use in editing translated texts, since there is a wider range of evidence available—including, we assume, the auctorial form—and conjecture, palaeographic or contextual, has a consequently reduced prerogative. Unfortunately, this ideal may not always be as common as the modern editor might desire, and it may conceal sophistications which may not be immediately apparent.

For example, the English MSS variants (Figure 3) Assyria/Affrica—which,

• 3. Assyria (Assyria)/Affrica (p. 738)
• Frisia/ffrigia (p. 788)
• for gredynes of germaynes/gyauntes (Lat.: germani)

of course, with long s become even more similar: Assyria/Affrica—could be evaluated by reference to the consistent Latin Assyria. But since the proper names would not in fact be subject to 'translation', but rather to (identical) 'transliteration', there are no pure palaeographic grounds for assuming that the divergence in English authority does not reflect either a non-extant divergence in the Latin (i.e. contamination across the language boundary—see MODEL D2 below), or the preservation in Trevisa's source of the 'incorrect' Affrica which, through later scribal corruption, returns by chance or determined variation to its original Assyria (i.e., along a two-way street). Now, none of these latter conditions can be demonstrated through the extant evidence, but neither can they be totally disproven. They may be unlikely, and therefore emendation to Assyria justifiable, but they should at least be considered as qualifications to our ideal model. The Frisia/ffrigia example illustrates the same relation, as do many such occurrences in the Trevisa text of DPR. Another example should make this clearer. Trevisa's text reads, "for gredynes of germayns/gyauntes" as a translation of the consistent Latin germani (Figure 3). There is no doubt that historically, it was the Germans rather than the Giants who were responsible for the particular military aggression described, but of course, the textual critic is not necessarily concerned with historical truth but with textual accuracy.[12] The Latin germani certainly provides us with the authority for the germayns, but a putative gigantes is arguably very little further removed palaeographically from germani than is gyauntes from germayns, so that the same conditions as Assyria/Affrica could possibly apply, even though translation, not transliteration is now involved. An editor would probably be justified, under these circumstances, in using the act of translation as defence for an emendation of copytext gyauntes to germayns, but should search for any evidence of parallel corruption in the Latin source.

These ambiguous cases, where the simple model still serves as the most plausible explanation of textual history, but where other possible routes have to be considered, gradually merge into a totally different model where a consistent Latin reading is of virtually no help in establishing the correct English form. We may still keep MODEL B before us as a paradigm, but its validity will begin to become questionable in the next few examples.

• 4. a) raþer/first/fyrst/furst etc. (Lat.: quondam)
• b) ful/file/foul/full//fowle axen (Lat.: vilior . . . cinerem)

For instance, the variants (Figure 4a) raþer/first/fyrst/furst and so on are to a large degree 'accidental' divergencies which could depend upon scribal habit or whim. The problem is the form raþer, which OED, citing the meaning of "of earlier times", quotes as a Trevisa neologism (in this particular sense) and as obsolete, rare". There can be no doubt that the first form, with all its variants, is the most comfortable to the modern reader/ editor, but the doctrine of lectio difficilior probior est might be invoked here, since the dubiousness of raþer, preserved in only one manuscript and far outweighed by the other readings in a distributional analysis, would explain a scribal substitution, but not vice versa. Editorial retention of raþer (as a translation of Latin quondam), supported but not confirmed by the Latin MSS, would be justified here, especially if (as so happens in this case), the raþer variant occurred in a manuscript bearing many other signs of 'authority.'

The use of a Latin corrective gets even more difficult in Figure 4b, where we are confronted with the series ful/file/foul/full/fowle axen as a translation of Latin vilior . . . cinerem. The problem here is not what the Latin means, but whether these forms can all be regarded as 'accidental' scribal/ auctorial variants, with the same semantic core, or whether some should be seen as substantive variants (perhaps by accretion), so that gradual scribal corruption has, in effect, produced a 'different' word. For example, is ful to be considered as a form of full and therefore as an 'error', or is it a form of

foul and therefore 'correct'? Does file possibly indicate an erroneous scribal correction or smoothing—voiced to unvoiced—of initial SW v in vile (much closer to the Latin form), or is it simply a decayed remnant of an original ful(e) with minim error? By a series of eminently clear palaeographic transmissions, we could postulate at least two separate routes to explain the extant English forms: vilior translated by the SW Trevisa as vile, corrupted scribally to file in an attempt to smooth out the SW dialect, then further corrupted into ful, foul etc. OR: vilior translated as foul, with a series of 'accidental' scribal variants to follow (ful, file, etc.).

• 5a. Mesopotania hat/hyghte/is cleped/is calde/hap ethimologia of grew (p. 757) mesopotania grecam ethymologiam possidet
• 5b. x (Lat. form) [is called] y (Eng. form) [from] z (language/nation) + gloss
• 5c.
  

A similarly ambivalent structure could be suggested for a slightly different problem, where Trevisa's text (Figure 5a) reads, "Mesopotania hat/ hyghte/is cleped/is calde/hap ethimologia of grew" as a translation of "mesopotania grecam ethymologiam possidet". The forms is cleped and is calde are very common scribal variants of a translation of vocatus, nominata etc., but of course, in this case they misrepresent the Latin sense, as does the palaeographically unrelated hyghte. For the moment then, we can either assume that these forms are indicative of 'bad' MSS (and that the translation was originally correct), or that the forms are 'good' (and the translation is in error). Another, perhaps corrupting, influence may be mentioned here. In this section of DPR, the most pervasive rhetorical formula for the opening statement about each country described is the pattern shown in Figure 5b: the Latin name of a country (e.g. Gallia) is given its English version (France), explained as deriving from its inhabitants or their language, which is then given a suitable gloss. The formula is developed from the pressure of Isidore's transcendental etymological explicatio for the nature of all matter, but the textual significance here is that the formula could very well have forced the apparently similar rhetorical shape of this particular sentence into a pre-ordained mould which is, in fact, totally unsuitable for its semantic content. The problem is to decide whether scribe or author were responsible for the moulding. Here we can turn to the other two readings: hat/haþ. Were it not that t/th/þ variance can be so whimsical as to be virtually meaningless in establishing textual primacy (especially in 15C MSS), we could assume that haþ is correct (=possidet) and that hat is either correct—if it is to be considered as a form of the verb have) or incorrect—if it is a form of hatte=to be called. But hat's ambiguity could be shared by haþ, so we are not necessarily any closer to the evaluation of either the translation or the extant forms. We can summarise our uncertainty in Figure 5c: where ethymologiam possidet is either translated correctly as hat (meaning has)

ethimologiea, but rationalised scribally—first by the normalising hap, then by the formulaic hyghte, which in turn is levelled to the individual scribal usage (is cleped etc.). OR, the Latin is mistranslated by the author as hyghte or hatte (both meaning is called and therefore generating scribal variants of this, one of which is hat, scribally altered to haþ).

Now, there is little doubt that an editor's acceptance of the translator's general competence would favour the first alternative and that a reading of hat (i.e. has) would represent this belief most clearly. But the pressure of the rhetorical mould and the existence of a high degree of scribal ingenuity can qualify this position. One final possibility is that possidet was omitted in the Latin manuscript from which Trevisa made his translation and that the more expected hyghte forms are therefore a justifiable paraphrase of an apparently meaningless Latin sentence and therefore represent auctorial intention.

In each of the cases in this section, despite the specific difficulties deliberately introduced into the discussion, we have been following basically the same paradigm—a consistent Latin authority to be used as a potential control for inconsistent English witnesses. The stemmatic MODEL B remains clear behind the clouds of possible corruption. Both MODELS A and B assume no prima facie evidence for corruption in S, an assumption which, strictly speaking, should depend upon a critical edition of S itself, and even of the sources of S. This latter sophistication is perhaps less of a difficulty in the generally tight plagiarism (or embedding of authority) found in the textual continuum of so many mediaeval works, particularly the encyclopaedias, but a brief illustration will demonstrate the inherent snare that the editor should be aware of.

• Eng.: The pleyne parties þerof þat beren corne ben byclyppede about . . . (p. 739)
• Lat.: pane/pone . . . circumdatur

Trevisa's source, the De Proprietatibus Rerum of Bartholomaeus Anglicus, is itself very highly dependent on Isidore's Etymologiae. Isidore has been critically edited, but Bartholomaeus has not.[13] In describing Bactria, Trevisa's text reads (Figure 6, giving MODEL C), "The pleyne parties þerof þat beren corne ben byclyppede about . . ." which should not be subject to emendation by contextual evidence, for the sentence seems to make perfect sense and also appears to be an adequate if not completely accurate translation of Bartholomaeus' "pane . . . circumdatur". However, the critical edition of Isidore makes it clear that the 'correct' reading for this sentence should be pone=behind, in the back, not pane, and that Trevisa's translation is therefore in error, although the extant MSS of Bartholomaeus of English provenance all read pane and should therefore support the Trevisa reading. Since the palaeographic discrimination between the two Latin forms pone/pane is so slight, without the actual MS from which the translation was made, this numerical weight is not totally reliable, but should probably be accepted as confirming the 'accuracy' if not the 'correctness' of the translation. Emendation of "þat beren corne" to "behind" etc. is therefore unjustified. The ambiguity of the textual relationships in this second type of Imperfect Linear transmission (i.e. with the imperfection in the translated, not the translating, language) would give us either of the MODEL Cs, (figs. 6 or 7) dependent upon bibliographical evidence for variance at an earlier or later stage in the transmission of the original from which the translation was made.

Examples of MODEL C:

     

	Latin	English
a)	fontibus/montibus	*founteyns/mounteyns
b)	filia/filio	douƷter (p. 727)

                 

c)	aurea/antea	toforehonde (p. 778)
d)	sinus/filius solem	alone þe sone (p. 810)
e)	solis ardoribus/solis arboribus	bare ooneliche of treen (p. 737)
f)	*fecunditatem/securitatem	plente (p. 771)
g)	*canibus/captibus/raptibus/capitibus	with houndes (p. 732)
h)	*infertilitatis/fertilitatis	bareyne (p. 810)
i)	*bouis/bonis	and catalle (p. 763)
j)	*anno/auro	in on Ʒere (p. 763)
k)	apri	beers or beers boars (p. 742)

This general pattern—a consistent English text but an inconsistent Latin —has no textual (or rather editorial) significance as long as one of the Latin variants can be seen as the source of the English uniformity. For example, in Figure 7, a fontibus/montibus variance in Latin and a uniform English mounteyns would not suggest emendation, but rather that the Latin source must surely have read montibus not fontibus. There is a possible palaeographic quibble to be argued here, for while there is no extant English reading founteyns, this putative form is no more orthographically distant from mounteyns than is fontibus from montibus (see the Assyria/Affrica, germayns/gyauntes example above). But without evidence for an English corruption parallel to the Latin—a condition to be discussed in MODEL D— the error must be confined to the Latin textual tradition.

The readings of the Figure 7 type can, of course, be very useful in identifying MSS which could not have been used as a source by the translator (and may thus reinforce the bibliographical arguments based on provenance), but do not involve any immediate editorial prerogative. For example, the Latin variants filia/filio versus the English douƷter demonstrate that the English reading must derive from the filia group, since it is palaeographically much less easy to proceed from son to douƷter than from filio to filia. 7c-d illustrate the same situation. As a further extension of this general principle, we may postulate a (lost) Latin reading to explain the consistent English form. For example, the English plente may descend from a lost fecunditatem in error for the extant MSS reading securitatem, which would, with long s, be very similar to the lost form; and the English with houndes (Latin captibus, raptibus, capitibus) presumably derives from a lost canibus. Similar conditions govern the infertilitatis/fertilitatis, bouis/bonis and auro/ anno variants. The last case listed in fig. 7 (apri/beers etc.) offers a slightly different problem and interpretation. The English text omits any reference to the apri (i.e. boars) and it might therefore be assumed that Trevisa's source similarly omitted the apri—i.e. a simple MODEL C. However, since the previous word in the English translation, beers, is palaeographically very similar to boars, the omission might have resulted from scribal homeoteleuton in the English copying (i.e. MODEL B). Otherwise translation is in these cases being used as a means of plotting the textual history of the translated language rather than the translating.

The next degree of textual divergence—an inconsistent source and a parallel inconsistent translation—is perhaps the most difficult to draw a model for. If the inconsistencies seem independent, then we could simply combine MODELS B and C, but if they appear to be in parallel, then we might have to argue either for separate but equal translations (MODEL D1) —in general an unlikely event—or for a contamination model (D2) which crosses the language boundary—perhaps just as unlikely, since it assumes that a scribe was dissatisfied with his English exemplar and sought to correct it not from another English MS but from a Latin text different from that used by the original translator.[14]

Neither of these models is necessarily satisfactory to explain the actual conditions of textual transmission through translation, however, and we might instead turn to a specific, and highly complex, example (Figures 10-11) to demonstrate the subtleties which the theoretical models could obscure.

In a series of variants on the name of the river Moldau (or Vltava), the Latin MSS of DPR present a bewildering array of choices. These include: Waldani/multa/vltava/moldani/mulda/multe aque etc. Similarly, with all of the expected enthusiasm for proper name variation (a scribal characteristic which has led some theorists to banish proper names from the pantheon of critically emendable variants, on the ground that they are too interesting to scribes), the English MSS provide us with fulta/vulta/multa/many þinges/many ryuers. How are we to arrange this seeming confusion into any logical pattern which will tell us anything valuable about the Latin text, the English text, or the translation from one to the other? Ignoring the double-translation model as bibliographically implausible for this particular work (and, one suspects, for most others),[15] we could create what might look like

a workable contamination model (fig. 10) which is essentially a specific working out of a D2 type.

In this model, an original Moldau-like form (or its Slavic equivalent Vltava—a discrimination which might lead to an investigation of the linguistic sympathies of certain scribes) could have undergone a series of permutations within the Latin transmission, to produce e.g. multa from mulda (itself a form of Moldau), and multe aque as a Latin scribal rationalisation of an apparently noun-less adjective. The translator would then have read multa in his source, but instead of rationalising as the Latin scribe had done, would regard the word as a neuter plural and render it simply as many þinges. In the context, however (a discussion of the rivers of Bohemia), this would seem at best weak, and at worst, nonsense. Two alert scribes would then seek to turn this supposed nonsense into something more appropriate for the context—one either by 'inventing' the apparently missing word ryuers (in a manner similar to that of the Latin scribe who wrote multe aque) or perhaps by consulting the Latin MS of this scribe; the other by going to a different Latin MS bearing the corrupt (or original) Slavic Vltava, and rendering this as fulta/vulta in his own version. This explanation takes account of all the variants and plots them in a transmission system which is not impossible to demonstrate, but it is clumsy and puts far too much responsibility on the shoulders of too many scribes. Its deficiencies are that, because of a superficial parallelism of form (Vltava and vulta versus mulda and multa), it immediately assumes that contamination is the only way such parallels could have been created. We should note that it is not quite as ingenuous as that theory which would simply select the extant English variant which happens to be morphologically closest to either (or both) the modern English received form or that of the Latin originals. By this approach, one would merely look through an atlas until one came across the river in question, and then make a textual selection of, e.g. vulta as the 'obvious' English rendering of Vltava, assuming that it must therefore be auctorial—a promotion of 'correctness' over 'accuracy'.

In place of these two schemes, let us attempt a model (fig. 11) which relies neither upon the atlas nor the editorial research of too many scribes.

In this model, we assume that the Latin source for the original English translation read multa (just as in the previous model), but that instead of regarding this as a common noun and 'translating' it, Trevisa simply saw it as a possible variant for the proper noun=Moldau (which, of course, it is) and transcribed it so. Alternatively, he could have felt uncomfortable with the context if he had regarded it as a common noun, and therefore left it in

its Latin form rather than risk a translation, in effect creating a neologism by simple duplication of form. In either case, his fair copy would have read multa, and our arguments are guilty inevitably of the intentional fallacy if we pursue the discriminations further than this likelihood.

Now the scribes enter the scene. Dissatisfied with multa as either a neologism or a proper noun—and we will see later why this might occur—one of them 'translates' the word as many þinges which, of course, makes nonsense of the text, and is therefore rationalised by the artful scribe of many ryuers. On the other hand, through a series of very simple minim confusions, the original multa becomes first * uulta, then vulta, and then, via normalisation or smoothing of an apparent SW initial voiced v, fulta. This model has the virtue of embodying the 'simplification' theory, and preserves the independence of the single act of translation, which should be one of our basic assumptions unless there is consistent or developed bibliographical, historical, and linguistic evidence for multiple translation or correction.

Another example of a type D redrawn as B and C occurs in figure 12, where apparently parallel variants in the Latin and English (numedia/in media, the latter with an English 'translation') can be explained by minim error (in Latin and English) and scribal rationalisation (in English only).

The four models constructed so far hae all assumed a fundamental auctorial fidelity to the source; the problems have been in identifying and describing the likely transmission of this fidelity. We should now turn to a different sort of problem: conscious or determined auctorial variation. An immediate objection, of course, is that if the standard of fidelity is removed, then how are we to measure the degrees of scribal, as opposed to auctorial, interference with the text, which now lacks a consistent paradigm? There is no easy answer to this objection, except perhaps that a close study of translating habits as observable through the three 'simple' models might be sufficient

to establish, at the very least, some degree of probability in each textual variance. In fact, with Trevisa the case may be a little more straight-forward than with other translators—and justifying yet again the use of his texts as a standard to measure divergencies. First, his views on the art of translation are known (though obviously we should always be careful of assuming that theory and practice are identical); second, if we divide these conscious departures into two basic types (the gloss-type, or divergence from the norm of the source, usually in the form of an addition; and the neologistic-type, or divergence from the norm of the current medium, usually in the form of a substitution), we shall see that Trevisa leaves a number of careful signposts to the identification of auctorial method.

For example, to deal with the gloss-type first: where he finds peculiarities in his Latin original which seem to need explanatory material, Trevisa will usually mark his addition with his own name. Thus, in the chapter on Vitria, he inserts: "Treuisa: here lacketh for ne mencioun is ymade whider-warde þis ilonde shulde be. Some men wolde wene þat þis ilond is WyƷt, but WyƷt comuneliche is yclepede insula vecta and þis ile is here yclepede Vitria" (DPR, xv. 172, p. 823).

This marking is another manifestation of the translator's concern with fidelity: the reader must know what is genuinely found in the original and what is not. For the most part, this fidelity to source does not conflict with concern for the reader's understanding, although Trevisa had remarked that on occasion he must "set a resoun for a word" (see Appendix).

At its simplest level, the gloss for the sake of the reading public may be no more than the tendency to form doublet expressions, which is strictly speaking not an 'addition' to the text—for it is often highly debatable which is the 'core' of the doublet and which the explanatory expansion. The genuine gloss itself can be observed at three levels of complexity, illustrated in

Figure 13. The first, "bloo mennes londe" is a simple etymological translation

13. GLOSS-TYPE:

         

Latin	English
Ethiopia	Ethiopia, bloo mennes londe (p. 754)
antipodes	pe antipodes, men þat hauen here feet aƷenst oure feet (p. 754)
castores	castores, beasts þat leuen both in water and londe and gelden hemself whan þey been yhuntede (p. 785)
solis vicinitas	for þe sunne is nyƷe and rosteþ and tosteþ hem (p. 754)

of the Latin (or properly Greek) and can be regarded as a 'perfect' gloss. The second, "þe antipodes . . ." etc. is still etymologically motivated, but offers a slight vernacular amplification, almost syntactic rather than semantic; and the third, "castores . . ." etc. becomes not only an etymological gloss and amplification, but a description of function too. In fact, the etymology—leading to castration—is obviously taken for granted. The first two would probably not alert an editor to a search for possible Latin source-equivalents for the gloss, but the third properly should. As it turns out, all three seem, on bibliographical evidence, to be genuine translator's glosses.

All of this is straightforward enough. But the introductory section of Ethiopia continues, "for þe sonne is nyƷe and rosteþ and tosteþ hem", as a translation of solis vicinitas. In other words, there is no trace in the Latin for the semantic bud from which the humorous "rosteþ and tosteþ" might have bloomed as a gloss. Are we then to assume that the principle of auctorial fidelity must put this fascinating little doublet down to scribal fancy, for it is no longer bound by even loose rules of translation, nor marked as a 'Trevisa' insertion? Stylistically the "rosteþ and tosteþ" is typical Trevisa, though, as we have seen, used elsewhere as a technique for enlarging in English the semantic range of a Latin word, which now seems to be lacking in the source. Should we perhaps assume that the Latin is at fault and that a word meaning "to toast" or "to roast" or both has dropped out of the extant MSS? Certainly, any editor with a sense of humour would be loath to sacrifice the marvelously idiosyncratic comment, which is present in all English MSS; but not only is rhyme reinforcement very unusual in the standard Trevisa doublet, so is this humour we are so anxious to preserve. Such an indefinable personal quality is impossible to demonstrate on technical grounds, and it would be a glum job indeed to testify to the relative dourness of John Trevisa and Bartholomaeus Anglicus. Textual criticism must become aesthetic criticism at points like this, and the theoretical models will retreat.

Moving on to a more complex model for the gloss, we shall see that there may, however, be occasions when technical assistance will help. In describing the Propontes, Trevisa's Latin source reads as in Figure 14a, with a very wide range of English variants, much simplified in the versions given here, for almost every MS has its own idiosyncratic way of interpreting the Latin,

14. NEOLOGISM-TYPE:

• 14a. Eng.: þe same water spredeþ and maketh {þere þe costes} of þe see
• {þat same coste}
• þat is yclepede Propontides {þat is to seye þe broode þerof conteneþ}
• {sone þerof conceyued/conteyned }
• {but lii bi lix /fifti} paces (p. 757)
• Lat.: eadem/autem aqua et facit propontidem qui mox in quinquaginta/l passibus coartatur
• 14b. Eng.: þe gumme þerof turneth into þeffecte of/in to effecte of/in to white/in to fecte/in/in to glas (p. 764)
• Lat.: cuius gummi infectum vitorum reddit

pointing to a good deal of scribal confusion. Certain problems can be tested by one or other of our earlier models: for example, same with eadem/autem, sone with mox, and the passage contains far too many detailed problems of this type to be dealt with in full here. Our major concern is with the coartatur: conteneþ/conceyued/conteyned variants, and with the apparent gloss, "þat is to seye". This latter phrase most often occurs in Trevisa when explaining a difficult Latin etymology. Given that we accept our translator as competent, we should suppose that the phrase, if it is an introduction to such a gloss, would conclude with a correct explanation of the etymology. But Propontides does not mean "the breadth contains fifty paces" and so we must either propose that our translator nodded or that the phrase is a scribal rationalisation of what might have seemed an impossibly corrupt sentence. Pursuing this line of investigation, and moving into the second type of conscious variation—the neologism—we should now look at the coartatur problem and see what help that may give. The word has the basic sense of "confining" or "drawing together", but none of the English MSS does full justice to this sense. Again, we might suppose that our translator had erred, and that a form of 'contained' was his incorrect translation of the Latin. However, invoking the lectio difficilior principle, another solution appears more likely.

The basic problem with any neologism is that it is new, and therefore will presumably not be immediately recognised by the scribe copying a text containing many such coinages. Since Trevisa, in his role as translator, was one of the greatest neologisers of his period, we might expect such scribal confusions to be frequent. Postulating for the moment, by an admittedly conjecturalist leap, a 'better' English translation ("constrained"), we should suppose that a scribe would seek to normalise the offending new word and reduce it to acceptable late Middle English (a form of "contain"). The evidence of OED in fact confirms that "constrain" is a Trevisa neologism and that such a scribal interference, thereby changing the whole meaning of the latter part of the sentence, is quite plausible. It might also be noted that a different passage in DPR also confirms this guess at "constrain", where this same word is indeed used to translate coartatur, but this time without scribal

interference, possibly because the context would not admit a substitution and nonsense seemed preferable to wholesale slaughter. Returning again to the "þat is to seye" phrase, we can, after having resolved the "constrained" tangle, presume that "þat is to seye" is not a true auctorial gloss at all, but a scribal intervention necessitated by the displacement of Trevisa's neologism and the consequent bungling of the sentence. An intelligent but harmful scribe (apparently the most dangerous sort) thought he might save the passage by conceiving of the latter part as a gloss—but in error.

More typically, the neologism is in Trevisa's case almost the opposite of a gloss: that is, the retention of the original form of the Latin or the substitution of a slightly anglicised equivalent (see the multa example above). The coartatur/constrain problem is a genuine translation, but many of Trevisa's neologisms are virtually transliterations. The technique is to drop the Latin word whole into the syntax of the English sentence and then make it behave as if it were English, rather than to form semantic equivalents through the juxtaposition of pre-extant English morphemes (as in the Old English translation of trinitas as þriness). The problem for the editor, and the modern reader, is that sometimes this technique works, and the word catches on, but that sometimes the neologism becomes a virtual nonce-word, and therefore unrecognisable outside the context of translation and extremely liable to scribal interference.

For example, the Latin cuius gummi infectum vitorum reddit of Figure 14b is rendered as "þe gumme þerof turneth into þeffecte of/in to effecte of/ in to white/in to fecte/in/or in to glas" in the various MSS. The scribes were obviously at their wits' end again. What could the author mean, and how are we, the modern readers, to discover what he actually said before this overlay of scribal ingenuity took place?

Again, assuming the translator to be competent, we can postulate that he knew that infectum meant 'imperfect'. But none of the English forms comes very close to such a word. Therefore, the next step is to suppose that the word he used was unknown to the scribes, who therewith buried it by rationalisation. In fact, traces of the direction of this rationalisation can still be seen in the forms þeffecte, fecte, effect, all of which could be the decayed remains of an original anglicised neologism infect, to mean 'imperfect' and to translate infectum. Note that contextually there would be no grounds for arguing that the first of the variants listed "turneth into þeffecte of glas" was in error, so complete is the scribal reworking. Furthermore, it is only the apparent nonsense of fecte and white (the latter of which showing a scribe having thrown up his hands in dismay at his author's obtuseness), which would alert us to the traces of an earlier, neologistic form. But none of this need have occurred if the neologism had been immediately accepted. Since all the English MSS were copied some time after the author's death, a successful neologism would theoretically have become established by the time the scribes were at work, and they (and we) would have had no problems with infect meaning 'imperfect', provided, that is, the troublesome

word had not been removed even earlier in the stemma. Again, OED and MED confirm that infect is a Trevisa neologism, and one which failed to enter the language permanently.

The last textual problem to be considered—that of such confirmation of an editorial emendation based on translation—has already been hinted at, but some general cautions are probably useful here. As mentioned earlier, the postulated emendation constrained as a translation of coartatur was confirmed not only by a similar passage elsewhere in DPR, but also by the evidence of OED. But ideally, we should have worked out, on critical and textual grounds, a likely auctorial version before the discovery of supporting material, otherwise there will obviously be the tendency to make the desired reading fit the context. In this sense, such apparent aids as concordances and lexicons might actually be hindrances to solid editorial emendation, as might the evidence of historical dictionaries, especially by the confirmatory 'short circuit' where the only citation for the problematical word or usage is in the very context which the editor is attempting to restore. When trying to recapture the textual processes whereby a particular reading became 'corrupt', it is clearly better to have worked through the chronology of transmission and to have satisfactorily explained each stage, than to have leapt to the conjecture whose sole virtue is its reinforcing the aesthetic predispositions of the editor. Such was the danger in the Vltava/Moldau case cited earlier, where a facile acceptance of an apparently appropriate word would have had to ignore, or sacrifice, any consideration of the textual routes to the establishment of a reading.

III

Since all the citations used to support or illustrate the four basic models and their extensions are drawn from the work of a single author (in fact from a limited section of only one translation), I obviously cannot expect that MODELS A-D will prove capable of accommodating all textual conditions where translation is involved in establishing or charting the transmission of a text. Indeed, as has been shown, there are in Trevisa no genuine examples of either MODEL D1 or D2, both of which can probably be broken down into a combination of MODELS B and C. However, the fact that there is a theoretical need for a fourth class, and that this class would result in models looking something like D1 or D2, does confirm that the focus of this discussion is more on the construction of usable models than on representing the work of one author through stemmatic diagrams (as happens in the Vinaver/Malory example mentioned earlier).

The area most likely to need elaboration is doubtless the gloss/neologism alternative, especially in the degree of variance from the norms of context and source. It is here that Trevisa's characteristics as translator are most influential over the paradigms suggested in this article, for his neologising tendencies (a result of his sense of the English language as needing semantic enlargement) and his "clere and playne" doctrine combine to establish fairly

trustworthy limits on the translator's prerogative. But one can certainly imagine circumstances where this prerogative can be so enlarged as to reduce the validity of the gloss/neologism alternative as a means of describing the liberties undertaken.

When, for example, Chaucer is charged, in the Prologue of The Legend of Good Women, that "thou hast translatid the romauns of the rose" (l. 329 [F version], 255 [G version]), are we to suppose that the same act is invoked as when Johnson insists that "Poetry cannot be translated, and therefore it is the poets that preserve languages" (Boswell, Life of Johnson, 11 April 1776)? Whatever Johnson meant by that remark, should we assume that Chaucer had proved him wrong by tackling a poem like the Romaunt? Johnson's suggestion would deny exact equivalence (of the sort we have been observing in Trevisa) in any other medium—even syntactic—a notion disputed by Locke's claim that this function of equivalency as opposed to definition is the only proper responsibility of translation: "This is to translate, and not to define, when we change two words of the same signification one for another" (Essay on Human Understanding, 3. 4. 9). The problem, of course, is in the slippery term "signification", which (as we have seen) to Trevisa frequently meant enlargement or expansion. But could it then be argued that LaƷamon's Brut, with its more than doubling of the 15,000 lines of Wace, is in any appropriate sense a translation of his French "source", or is the Roman de Brut itself a translation of the Latin of Geoffrey of Monmouth? The game, played with some wit by Valerie Flint in her suggestion that Geoffrey's Historia is a parody,[16] could continue back to Geoffrey himself with his fanciful claim that he was only "translating" the little Welsh book given to him by Walter of Oxford.

Translation, through its emphasis on continuity and tradition rather than individual creativity or idiosyncracy, becomes a mediaeval staple for the reticent poet, and is exploited with considerable skill by the humble Chaucerian narrator of, for example, Troilus and Criseyde, where, despite the claimed subservient role of the English poet, the use of MODELS A-D would hardly occur in editorial practice.[17] As a device for defending one's work against the supreme charge of "fiction", translation may therefore become one of the rhetorical artifices of fiction itself, and the ironic stature and methods of the translator could deliberately obscure any models to be found in charting this new construct. For most authors in Middle English, however, one can probably proceed downwards from a conventional source study, establishing the putative degree of fidelity, to a genuinely textual consideration

of the value of these models in describing and resuscitating the translated text. The more prominent the models, the less original is the translator and the more purely textual the criticism of the translation. As the models recede (as arguably they do in Piers Plowman but not in Malory), then the larger the editorial prerogative to reconstruct without deference to a source reading.

APPENDIX: TREVISA THE TRANSLATOR

The purpose of this brief appendix is to describe, for readers who may be unfamiliar with Trevisa's work, his general principles for translation as discussed in his theoretical writing on the subject. The texts covered here will be well-known to most mediaevalists, who should not therefore expect a re-evaluation of Trevisa's translation techniques.[18] For non-mediaevalists, however, the materials may provide an illuminating background to the specific cases cited in this study.

In the Dialogue on Translation [19] the problem is stated in characteristically plain terms: that people of different nations "vnderstondeth others speche no more than gaglinge of gees. For jangle that one neuer so fast, that other is neuer the wyser though he shrewe hym instede of good morrow. This is a grete meschyef that foloweth now mankynde." The remedy for this cultural and social problem is "that some man lerneth and knoweth many dyuerse speches. And so bytwene strange men of the whiche neyther vnderstondeth others speche suche a man may be mene and telle eyther what other wole mene."

The translator is thus a "mene" between oppositions created by the difference of language, and it is important not to undervalue this directly communicative (as opposed to individually creative) theory of translation, since it not only determines the linguistic motivation for translation but also characterizes stylistic and rhetorical effects—often to the dissatisfaction of the modern critics. Witness, for example, A. J. Perry's unease: "He wished to be understood—this resulted in wordiness,"[20] although the comment seems eventually to be only a reflection of the conventional (and convenient) Trevisa use of the doublet as a reinforcement of the semantic "core" of the Latin original; the problem is not so much Trevisa's wordiness (in any rhetorical

sense) but the density of the Latin vocabulary and the translator's need to provoke late Middle English into extending its lexicon (by the means suggested in the discussion above).

In the shadow-boxing of the Dialogue, it is the Lord who is given all of those arguments which Trevisa is to exemplify in the translations themselves: that a "skylfull" translation is one that "myght be knowe and vnderstonden," rather than one demonstrating the creative power of the translator; that "prose is moore clere than ryme, more easy and more playne to knowe and vnderstonde" and should therefore be employed wherever possible. In fact, all of Trevisa's major translations are of Latin prose works, and when verse is occasionally embedded therein, he will usually render it as prose, sometimes with a suspiciously anti-poetic comment like "God woot what þis is to mene". The doggerel verses at the head of DPR (if they are by Trevisa), merely reinforce the implication that verse was a foreign and uncomfortable medium to the translator.[21]

There is no need to be concerned here with the cultural or literary or even theological defence of translation in the Dialogue, except to note that, in citing Jerome, Alfred, Caedmon, and Bede (and the act of "prechyng" as "very translacioun"), Trevisa is determined to establish and affirm the continuity not of literary merit (for, as he remarks, the Hebrew which Jerome translated was already "good and fayr and ywrytte by inspyracioun of the holy gost"—i.e. Jerome did not improve the original by turning it into another language), but rather the continuity of the subservience of the translator to the work and its author, and the duty and necessity of following the "clere and playne" doctrine.

How did Trevisa imagine that this theory would turn out in practice? A partial answer to this question (beyond the translated texts themselves) is contained in the Epistle to Lord Berkeley: "For travell wol I not spare comfort . . . to make this translation clere and playne, to be knowe and understondyn. In some place I shall set word for word and actiffe for actiffe and passife for passife arowe right as it standeth without changinge the ordre of words. But in some places I must change the order of words and set actiffe for passife and aƷenward. In some places I must set a resoun for a word, and tell what it meaneth; but for all such changing the meaning shall stand and not be chaunged."

This practical description of technique presents us again with an espousal of fidelity, but a fidelity which may modify its form if not its nature or intention. Thus, the concept of subservience may be rather ambiguous: is the translator being more faithful by setting "passife for passife arowe right as it standeth," and perhaps thereby producing an unintelligible sentence in English, or by giving "a resoun for a word" (e.g. in glosses or neologisms), and destroying therefore a verbatim parallel in the hope of achieving a greater clarity, and thus semantic rather than morphological or syntactic "fidelity" to the source? As is obvious here, Trevisa is at least conscious of the problem and its alternative resolutions, and as I hope is evident from the body of this article, he could arrive at his goal of fidelity through one of several practices.

Notes