[Page numbers of the printed text appear at the right in bold. ]

John Trevisa and the English and Continental Traditions
of De proprietatibus rerum

West Virginia State College

   The first two Volumes of Trevisa's translation of Bartholomaeus Anglicus' encyclopedia used 2 Latin texts as controls for editing the Middle English text--MS Bodley 749 and the Vatican copy of a text printed in 1485 by Georg Husner of Strasbourg. The textual commentary uses some 18 Latin MSS to consider what Latin Trevisa may have actually had before him.1 Just how much, though, has been studied? The Latin text was not edited, but several studies of different portions were made: M. C. Seymour, the general editor, collated the first 5 folios of each book; Ralph Hanna studied Book V; Susan Clinton's work on Book X is separately available as her doctoral thesis, and Traugott Lawler did Book XI.

   The average folio of the English MSS provides for just under six pages in the printed text, so five folios means just under thirty pages. There are 1,396 pages; taking away Books V, X, and XI and the first thirty pages of each other Book leaves 715 pages not included in the survey of Latin MSS for the textual commentary. That accounts for just under half the text, then, but favors the first Volume: the three Books studied are all there, it contains thirteen Books to the second's six, and the unsurveyed material in the second Volume is nearly three quarters of all unsurveyed material.

Would results based mostly on the first half of the text be sufficiently representative of the whole?

   A quick check for comparison is to collate, edit, and study the results of the last half of the last Book; and this is what I have done--the sections on mathematics, measures, and music. This section is thrice the size of Clinton's Book X, about 3% of the whole encyclopedia and just over 8% of the unsurveyed part of Volume 2. The results are somewhat different from those published by Clinton and in the textual commentary. Clinton finds her

The textual commentary's results are rather similar: [IK] [(JLV)OTX₁] [(MNQ)SUX₂X₃Y] [PRW] with O possibly based on a MS like R and with S appearing "to change its character at the end of Book IV" (3:4).

   In the last part of the last book, however, one large family of MSS appears: JLNSTUVY, within which JS and NT are pairs and JLSVY is a group; but JLV is not as close as for Clinton or the textual commentary. Of the other MSS, PQR form a trio and WX₃ form a loose pair. I and O are free agents, both separately contaminated with NT. (K is wholly lacking, M is mostly missing, and Y lacks perhaps a quarter of the text.) Of the correcting hands, that for U evidently used a source not related to JLNSTVY.

   When M. C. Seymour surveyed "Some Medieval English Owners of De proprietatibus rerum" the year before the Trevisa translation appeared, he noted that, "Because of textual incompatibility (in some cases supported by a later date of writing or inaccessibility in a known location), none of [the MSS listed] could have been used by Trevisa as the copy-text for his translation" (157 n. 1). Trevisa finished his translation at Berkeley on 6 February 1398/99, as he himself tells us (1396); any MS he could have used must have been written before then and must have been available

to him for some time, if not in his personal possession then at least available for lengthy consultation somewhere he could be for long periods.

   One likely suspect, as Seymour noted, is the MS willed in 1368 by Simon Bredon (d. 1372) to Queen's Hall, Oxford, which Trevisa had entered in 1369.2 Trevisa might have had this in his hands for a time. On 6 January 1376, Thomas Carlisle, a northerner, was elected provost of Queen's; the southerners, among whom was counted Trevisa, objected to this. The dispute went on and, two years and four days later, the southerners took off with many items of college property, including the college seal. Repeated demands for return of the property, including books, led to an indenture drawn upon 13 May 1378 attesting the return of goods including the college seal and twenty-four books whose secundo folio incipits are recorded. De proprietatibus rerum is not on this list; but the list may have been incomplete, for as late as 20 October 1379 John Trevisa is cited among others who had not yet returned various goods, including unspecified books.3 Even if Bredon's copy were not one of the books taken, Trevisa may have had access to it earlier and long after, for he evidently rented rooms at Queens, whether or not he occupied them, in the 1380's and 1390's (Fowler "New Light" 305), not long before he finished this translation.

   Regardless of date or location, however, the MSS used could be taken in varying degrees as representing their lost ancestors or other relatives which Trevisa might have consulted. Still, the lack of closeness between Trevisa's translation and what the Latin MSS offer is complicated not just by the sometimes bewildering variety of choices available among the variants but also by the fact that good English often requires grammatical and rhetorical constructions different from those of Bartholomew's Latin. For example, nam particularia. quorum quelibet sunt in se perfecta. perfecta sunt quando ad unitatem sunt redacta (4.15-5.1) is rendered "for alle particulere [th]inges, [th]e whiche euerich is parfite in hitsilf, be[th] parfite whan [th]ey be[th] redact into oon" (1355.14-16). Since Trevisa translated quelibet as "euerich," the verb sunt needs be singular--"is." This sort of matter does not help us discover the Latin he worked from. Similarly, his preference for native English active verbs in parataxis over Latin passive verbs

and participles in hypotaxis blurs our view of what Latin conjunctions, inflections, or other details lay before him when he wrote: when he renders Universitatis conditor Deus in circulo designatur (46.9-10) as "God is creatour and makere of alle and is designed in a cercle" (1369.24-25), we cannot tell if the first "is" occurs because Trevisa saw Universitatis est conditor Deus as in NOQT or because he added it himself to make better sense in English. To these must be added Trevisa's peculiarities, such as the habit of occasionally adding his own comments or examples,4 his penchant for English doublets (even occasional triplets) for Latin words or phrases,5 and his evident dislike for sive and vel, generally rendered as "and."6 In addition to the comments he added, often there are bits and pieces he left out, which may be due to their being lost in his Latin copy, to their loss in the tradition of copying Trevisa's text itself, or to his own omission of them.7

   As Lawler points out, Trevisa's vocabulary could reduce a variety of Latin words into a single English one (281-82), even though it vacillates between giving "an English technical vocabulary" and giving "the Latin term currency in English" (280). He modifies what Perry before him had pointed out (civ), showing how Trevisa earlier might "sacrifice accuracy for intelligibility" but in De proprietatibus rerum has "a firmer regard for accuracy" (282). This helps us discover the Latin behind Trevisa's Middle English. Where he reproduces the Latin term alone or in a doublet, we can be fairly certain of the reading in his Latin text; elsewhere, we cannot. Thus Dorica cytharum at 107.6--for Isidore's Dorica lingua kioapa (III.22.2)--with its bewildering array of variants among the Latin MSS, is in Trevisa simply what he saw: toricha citharum (1391.22).

   Nonetheless, Trevisa did nod at the end of this long work; e.g., he anglicizes reducuntur and redacta to "be[th] reduct" and "be[th] redact" (4.14 and 5.1,1355.12 and .16) where Fristedt would expect a form of "a[y]enlede."8 Similarly, he does not grasp that Dyatesseron enim et dyapente et diapason ab antecedentis numeri nominibus nuncupantur (115.1-2) means that the name dyatesseron is based on the earlier numerical term tessera and that dyapente is based on the prefix penta-, though we should excuse his not realizing diapason is based on the Greek for "stave"; instead of telling us that these terms were coined from preexisting numeric

terms, he confuses us by saying they "haue[th] [th]e names of [th]e nombres [th]at go[th] tofore in [th]e bigynnynge of [th]e names" (1393.29-30).

   But usually the translation is what produces uncertainties about what Trevisa saw. For example, in the beginning of the text, we read Ad dictas proprietates rebus insertas (1.3), which Trevisa gives as "To [th]e forsaide propretees of [th]inges" (1353.25); "forsaide" could represent either dictas as in most MSS or predictas as in WX, and "of [th]inges" does not let us know if his Latin copy-text read rebus insertas as in OPQRU, rerum insertas as in INT, or rebus insectas as in JLSVWXY. The lack of closer rendition may reflect Trevisa's confusion or unwillingness to deal more precisely with but a small phrase of little significance. And Trevisa often simplified, which further complicates attempts to find what he worked from: the nice bis senos scilicet xii becomes the simple and clear "twies sixe" (30.6 1363.23), the self-evident definition of a sphere--exterius evexa interius convexa--the natural "holow wi[th]inne" (44.13-14 1368.34), and the redundant Mulgarium est vas in quo pecorum lac mulgetur the direct "Mulgarium is a milkyng vessel" (69.4-5 1378.22).

   There are quite a few instances where none of these sixteen Latin MSS can account for Trevisa's translation. Clinton noticed in her study of Book X that, although "W agrees with the translation most often, this may be because it is an early manuscript of high quality that preserves a text of the same calibre as Trevisa's reconstructed Latin source" (76). Where his copy agrees in correct readings, its affiliation cannot be identified. Quite often when Trevisa agrees in correct readings involving numbers and other matters, we cannot know if this is by independent correction or by his having had a good Latin copy.9

   Yet it must be remembered that Trevisa was frequently a bit careless in his intent to translate the ideas faithfully. For example, at 76.1 Bartholomew, listing some measures, finished up saying that a pertica is passus ii, id est pedes x: Trevisa has the correct x at the end, thus agreeing with IOQW (X has no reading and the rest have xx), but does not correct his error "elleuene" for ii id est in spite of having been reminded just a few words before that a passus is pedes v (1380.35). The error is not original with him, for it also appears in SUW, so it seems that for this whole passage he could have had a good text, like W, before him. Where Bartholomew

discusses how among numbers alius pariter par, alius pariter impar, alius est impariter par (20.14-21.1, Trevisa simply states "Som is par, and some is vnpar" (1360.18-19). Though here the Latin MSS are confused, they are basically clear when the terms are defined individually; but in Trevisa the distinction is wholly lost: pariter par becomes simply "par" (21.1 1360.19) and pariter impar, impariter par, and impariter impar all become "impar" (21.7 1360.24, 21.11 1360.28, and 22.3 1360.33).

   At times it seems we find Trevisa following two texts simultaneously. When giving references, he gives de materia ... materia as de materia vel natura ... natura where Q has natura both times at 53.4-5 (1372.12-13), his spelling "Phisolagia" is close to O's phisalogie and R's phisologie at 74.6 (1380.16), and he follows U² in adding Ezechielis to the xl at 76.5 while following X in adding capitulum d (1381.4-5).10 Where Bartholomew wrote centurias at 75.7, Trevisa first follows JLNOTU with "centories" and then adds the spelling of IQ as an alternative: "o[th]er centuries" (1380.27). When dealing with 86.12, Trevisa did not recall Matthew 10: 29 nonne duo passeres asse veneunt as witnessed by the confusion in the Middle English MSS over the last word, which is spelled out only in NT of the Latin MSS--IQU have v and JLOPRSVWX have ve. The Middle English MSS are split over variations upon veniunt evidently followed by some form of vyfe (1385.8). Here, his comparing MSS could explain the readings, if we assume he noted down ve from one MS, which he understood as meaning veneunt, and added v from another, then taken as the number five.

   On the whole, Trevisa's translation does not seem to follow any of the main textual groups, being more eclectic, as already seems evident from the examples posed. Simply counting up the support of Latin MSS where Trevisa errs in fairly certain agreement finds that his Latin copy resembled most, in order, Q in seventeen errors, R and X in thirteen, P in twelve, and I, O, and W in ten.11 Anything else--including JLNSTV and U, but ignoring the incomplete M and Y--agrees with Trevisa less than ten times. It is curious that IOPQRWX in agreement also tend to show readings original to Bartholomew (although they do not form a family). Why, again, JLNSTV is isolated as a group is not clear. But it is clear that in this part of De proprietatibus rerum Trevisa's Latin copy most resembles the trio PQR, Q most of all, then X, and then I, O, and W.

   Nonetheless, Trevisa's Latin copy does not resemble PQR as a group: that is, readings supported by the trio PQR which help us to identify these as a group are not readings that Trevisa seems to have seen in the Latin. Rather, his Latin copy bears some resemblance to what PQR came from. Earlier, we saw that X, though having some loose relationship to W, is generally independent of the MSS edited; here, it is as close to what Trevisa saw as are P and R. PQRX does not form a related group of MSS except insofar as the translation is concerned; it occurs only once in error, at 44.9, adding a superfluous scilicet after vero where Bartholomew discusses finding a "geometric" middle and gives the example that vi et xii multiplicata faciunt septuagesies dispondius, media vero viii et ix multiplicata tantum faciunt. Evidently not understanding this simple arithmetic, Trevisa omits the latter part, leaving only "sixe and twelue ymultiplyed make[th] two and seuenty" (1368.28-29).12

   As in our conclusions about the MSS' relationships, Clinton's conclusions and mine about what Trevisa's Latin copy most resembled again differ but for the possibility that Trevisa compared texts. In the latter part of Book XIX, Trevisa's copy was something resembling both the ancestor of PQR and that of X; in Book X "Trevisa's primary exemplar seems to have been a manuscript related to the group NPQRSUWXY. It is safe to rule out P, R, andY as possibilities," though "he sometimes followed readings exclusive to IJLMOTV" (76). Ruling out P, R, and Y leaves NQSUWX; so the only agreement we have is that W and X look at all like what Trevisa saw. The two being generally unrelated late in Book XIX and not particularly close in Book X leads to the conclusion that Trevisa conflated copies or owned a copy that itself had conflated different traditions. Seymour concludes that Trevisa's copy-text was related to a subgroup JLOTVX, "a single Latin manuscript, no longer extant, primarily of an L-type text which had been sporadically conflated with an I-type manuscript" (3: 9). His evidence is based on selected readings throughout the work, though favoring the first thousand pages, and is admittedly provisionary. Now, of course, the three of us obviously disagree, but I would say that this apparent contradiction is due to the evidence surveyed, to its presentation, and to the history of how Bartholomew's text was copied on the continent.

   In his 1974 article, Seymour included the British Library Sloane MSS 471 and 511, written by English hands, among those of English origin; but for the Trevisa project not only were the fragmentary and abridged MSS left out of consideration but so also were these two, for they were considered as being "of the French rather than of the English tradition" (Clinton 12). In all, Seymour traced 23 substantial MSS (157), 15 with extracts including one in Welsh (158), 5 in bindings (156-57 n. 4), 36 references to ownership (159-63), and 7 uses by other writers in English (164). International travel and commerce were well developed and the English "nation" at the University of Paris had already had a long history, its seal first appearing on an act of 1252 and some of its masters founding St. Andrews University in Scotland as late as 1413 (Toulouse 16, 28). Bartholomew himself had taught at Paris 13 and left long before Trevisa began his translation.14 Even during Trevisa's lifetime the Hundred Years War did little to lessen the English nation's lists of suppositi, those over whom the nation had jurisdiction (Boyce 31, 28 n.4).

   Trevisa himself travelled abroad. In his translation of Higden's Polychronicon, which he says he finished 8 April 1387, he noted that he had seen the hot baths "at Akene in Almayne" and used those "at Egges in Sauoy" (2:61), that he changed money at "Brisak uppon [th]e Ryne" (6: 259), and that Higden is wrong about there being only one kind of French, for "[th]ere is as many dyuers manere Frensche in [th]e reem of Fraunce as is dyuerse manere Englische in [th]e reem of Engelond" (2:161). Clearly, he must have crossed the Channel, passed through Belgium or Holland to Aachen just inside Germany, then probably to Cologne for a boat trip along the Rhine, stopping at Breisach (just west of Freiburg) not far from the Swiss border, and on, either inland through Aix-les-Bains between Geneva and Grenoble or elsewhere, maybe stopping at Aix-les-Bains on the return trip, perhaps a land route across France, though when he speaks of "money of turoneis" he merely speculates about practice in Tours (6:259). If he went to Italy on this trip, he did not reach Rome, for his comment that the Coliseum "was [th]e place of [th]e ymages of provinces and of londes" (6: 337) betrays his ignorance and simple repetition of what Higden wrote before (1: 216-19) and his comment that the "Alpes bee[th] hi[y]e hilles in Lumbardie side" (4:343) shows no particular visit beyond the Alps.

   This was not his only trip abroad, for the Close Rolls show that on 29 January 1387/8 Angelus Christofre, a Lombard banker living in London, received a license to make a letter of exchange for twenty marks payable to Trevisa, probably for a trip to Italy (Richard II 5: 524). And the Treaty Rolls in the Public Record Office show he gave power of attorney to Robert of Hodersale and William of Faryngton on 5 November 1390 when he was about to go abroad--qui de licencia Regis ad partes transmarinas profecturus est (Perry lxviii). Evidently, by the time he finished this translation in 1398 he had already been on the continent at least three times. Since he could have obtained a copy of De proprietatibus rerum on any of his trips, he could well have used a MS not of English origin. Further work, then, should include study of continental MSS, especially those in England.

   There are eleven complete manuscripts of continental origin now in England, and six incomplete ones.15 In Europe and the US, there are well over one hundred more.16 Only a few were available to me, and to satisfy curiosity just a little I looked at them in a few places where certain MS groups had unique readings and where the general picture was of interest.17

   For example, among both the English and the continental MSS sometimes there is glossing, sometimes there is not. At 11.9-10, saying Aristotle praised God by the number three, LQUVWX have the gloss Nota de frigiditate virili, though it does not make much sense. This gloss is given also in twelve continental MSS,18 it is omitted from two others (El Esc o and Ste-Gen, just as in NT), and thirteen more MSS lack glosses at large, just like IJOPRSY.19 One other MS reads instead Nota de firmitate virili, and four more agree in this reading but place it by 12.5-7, which says the quadrangle signifies the stability of the Church and of the faithful soul.20 Evidently it was misplaced in the majority. Much later, at 94.1-4 on subtle voices, the gloss in LNTWX and probably also in QUV (where it may be in the gutter) reads Nota contra juvenes doctores; here fourteen continental MSS agree, two omit it, a leaf is missing in yet another, the same thirteen and IJOPRSY have no glosses, and two MSS show variants: Nota contra detractores juvenes and Nota juvenes doctores.21 Similarly, where some chapter divisions were not made by one or more of the English MSS, those chapter divisions were also not made in some continental ones and may even have been corrected.22

   Where readings unique to known groups were checked, a curious event occurred: readings peculiar to INT and to JLNSTV were almost without exception not attested by any continental MS.23 Eight unique errors of PQR surveyed among the continental MSS produced an interesting find:24 BN lat 346A agrees throughout, two more agree but when they lack portions of the text, and at least six of the eight errors are shared by six other MSS. Moreover, BN lat 16099 would agree with PQR seven times, having its own error once, but for corrections in all but two of the readings chosen. Sixteen of the MSS never agree with PQR. All of those sixteen but one agree with WX at least six of nine times, with an ambiguous abbreviation the only thing forbidding complete agreement among WX and two of them, which never agree with PQR.25

   Further, where Trevisa's Latin copy evidently read Exechielis xl capitulum d (76.5, 1381.4-5), this is the reading only in the English MS X and in twelve of the continental MSS.26 Though this would make it seem Trevisa used a MS related to WX and the continental cousins, it is not necessarily so, for later his version toricha (1391.22) seems based on torica (107.6) in MRV and Arras, Autun, BN lat 346A, 16098, and 16099, Brugge, Brux 7568, El Esc o, Hunter 391, Schullian, Sloane 511, Ste-Gen, and Upp, which was later compared to some other copy (like T or C-F, Eins, Kob, Praha, Reims, W 2337, or Zürich) and expuncted to show an eyeskip that left out this word. Here, his Latin copy evidently would not resemble a MS like WX and its relatives.

   Moreover, we know that--on the continent--De pro prietatibus rerum was copied in the University of Paris pecia system. In the university list of 1275 that defines the prices of books the stationers should make available to the scholars, the work is included as being in 102 pecie for rent at four sous apiece; in the list of 1304, it is in 100 pecie at six sous apiece.27 The complexities of what copying by pecia means are too much to go into here, but I will note two things. First, if we estimate how much text each pecia would take, we can divide up our text into presumed pecia texts and see if MSS change affiliations from one such portion to another; in the section I have studied, some did while others did not. So far as I know, no copy of De proprietatibus rerum with marks showing pecia divisions has come to light. But effects of the pecia system can be felt in a general way.

   Although the MSS ranged widely in number of folios, from U's 183 to N's 343, the average was 236; when the stationers provided De proprietatibus rerum in 100 pecie, this meant one pecia would produce 2.36 leaves, and when in 102 pecia 2.31 leaves. If we rely on an average 101 pecie, the result is 2.34 leaves. The result for each individual MS would be different, of course, but at present it is the average that concerns us, since it is the fairly constant size of the text that is, in effect, the common denominator of all the inconstant sizes of the MSS. My portion of Book XIX averages about 9.37 leaves, ranging from barely over six in W to nearly twenty in X. So if a pecia would result in about 2.34 leaves, then the average 9.37 leaves in my portion should be the product of a shade over four pecie. Although this does not discover the actual demarcations (particularly since the final pecia probably also included the list of authors, not part of the text edited), it does reveal that--if the MSS vary affiliation because of different pecie in their traditions--such variations could be discoverable by considering the text one fourth at a time.28

   An interesting thing happened with the trio PQR: although it agreed 13 times in the first quarter and 12 times in the last, it agreed only 6 times in the second and 4 times in the third. Most notably, the pair WX varied greatly in closeness. The two almost never agree in the first and last quarters--only 4 and 3 times--but are much closer in the middle, agreeing in the second 9 times and in the third 8 times (though three of these as IWX). Unless a threefold increase of coincidental agreement can account for this, it seems evident that the background of these two MSS implies changes of exemplars.

   While the numbers themselves are not particularly significant, the changing frequency is, for it is that which indicates subtle shifts of affiliation from part to part and thus implies potential changes of pecia exemplars. This may best be seen in U's increasing agreement with JLNSTV: without I and O, U agrees with JLNSTV only 3 times in the first quarter and 5 in the second, but this grows to 12 times in the third and 16 in the last. So what is U doing in the first two? In the second, nothing in particular, for it never agrees with anything else more than twice. In the first, it agrees with P and with X 3 times apiece and with I 4 times; this may mean very little, for IU occurs 5 times in the third and 3 times in the fourth quarters, and both I and U agree with JLNSTV 4 times and

3 times in the last two quarters respectively. Only in the third quarter does JLSUV occur, and that only 3 times. So it appears that the ancestry of U may well lie in the Paris university pecia system, and that U's readings increasingly come from the common ancestor of JLNSTV

   Second, the fact of copying by pecie means that results of MS study for some portions of the whole text may not agree with results of study of other portions. Some MSS--English as well as continental--may descend from within the pecia system, others from without, and yet others may be mixed. Only a complete study can show which ones descend how. These three different surveys produce similar yet divergent results because of the nature of what was surveyed. The transmission of De proprietatibus rerum both within the Paris university pecia system and without it has produced a tradition where the MSS may or may not shift their allegiance from one section to another, where they are at times unwavering and at times seemingly arbitrary in deciding with whom, with what other MS, they will side.

   There is yet room for one final example: in Book XVIII,29 Wien 2287 reads Si quis fumigaverit domos formicarum, flores et germina et frondes corrodendo violant--if anyone fumigates the ants' homes, they ravage the flowers, buds, and foliage by gnawing away at them. It is not that the ants are really this vengeful, but that the MS has lost about a half a column, comprising cum sulphere velorigani ... ad summitates arborum scandunt. The continental MS Schullian agrees in this omission, as do Eins and the English MS Y. Now, even though the missing text is added in the bottom margins of BN lat 346A, BN lat 16099, and the English MS Q, a most remarkable thing happens: BN lat 346A simply adds the text, Q glosses it, and BN lat 16099 places those glosses inside the addition and marks them by underlining.30 However, we cannot tell whether Trevisa saw a copy with the text where it belonged or in the margins, just that he did not have a copy that lacked that portion of the text.

   This conspicuous instance of textual error and correction shows both the necessity and the difficulty of studying all the available MSS and throughout the whole of the text. The relationships of the Latin MSS are mutable and fickle, depending on which portions of the text are read; settling their relationships is difficult enough, but clearly it cannot be done by surveying bits and pieces

as we all have done to date. Discovering Trevisa's Latin relies heavily on a certain identification of the textual relationships, yet even so striking an instance as the last can have little import in the context of identifying his Latin source.

   Either Trevisa used two or more unrelated MSS or he had a single MS conflated of two (or more) traditions represented in England by different branches of a textual tradition originating on the continent. In this, all three of our studies agree. It is undeniable that the tradition of copying De proprietatibus rerum in England is an integral part of its being copied in France, in Germany, Spain, Italy, and elsewhere. The different results shown by the textual commentary to the Trevisa edition, by Clinton, and here should not be seen as contradictory or inconclusive; rather, they should indicate the reality of a complex textual tradition underlying the history and development of Bartholomew's text and the fragile state of our understanding of Trevisa's work at translating it. To return to the opening question, we can answer that each study is sufficient to the material it surveys. Results summarized above are as precise as possible for the small portion of the whole encyclopedia studied; results given elsewhere are accurate representations of what can be known given what had been investigated. What remains is to begin with the discoveries of all and, acknowledging the strengths and apparent disparities, to reach forwards towards a more full understanding and appreciation based not on isolated sections or on hopeful surveys, but on thorough study of the whole encyclopedia and its history.