Printer Friendly

The Englishing of Juvenal: computational stylistics and translated texts (1).

Introduction

Comparisons between translations and their originals often shed light on cultural differences, large and small. At one level of meaning, such inter-language comparisons draw our attention to the elusive connotations and the ultimate untranslatability of conceptual words like anima and natura or, again, of words like amor, caritas, and amicitia. At another, we find small Saussurian mismatches like those between sword and gladius or between otiose and otium.

But comparisons among several English versions of any much translated text give a sharper focus to many suggestive points of style. Such intra-language comparisons are likely to be of special interest when the translators themselves are authors of distinction. Consider the opening lines of Juvenal's Tenth Satire, the poem upon which the present paper concentrates:
 Omnibus in terris, quac sunt a Gadibus usque
 Auroram et Gangen. pauci dinoscere possunt
 uera bona atque illis multum diuersa, remota
 erroris nebula. quid enim ratione timemus
 aut cupimus?

(1-5)

In his literal prose version of 1852, Lewis Evans renders these lines
as follows:
 In all the regions which extend from Gades even to the
 farthest east and Ganges, there are but few that can
 discriminate between real blessings and those that are
 widely different, all the mist of error being removed.
 For what is there that we either fear or wish for, as
 reason would direct?

(p. 102)

John Dryden's rendering (1693) is brief and vigorous:

 Look round the Habitable World, how few
 Know their own Good; or knowing it, pursue.
 How void of Reason are our Hopes and Fears!

(1-3)


The best known English version of the poem is Samuel Johnson's imitation, The Vaniy of Human Wishes (1749). It is only about two-thirds as long as Dryden's version, chiefly because Johnson reduces Juvenal's satirical illustrations to terse, ironic apophthegms. But, at the beginning, Johnson launches expansively and moves forward at a stately pace:
Let Observation with extensive View,
Survey Mankind, from China to Peru;
Remark each anxious Toil, each eager Strife,
And watch the busy Scenes of crouded Life;
Then say how Hope and Fear, Desire and Hate,
O'erspread with Snares the clouded Maze of Fate.
Where wav'ring Man, betray'd by vent'rous Pride,
To tread the dreary Paths without a Guide;
As treach' rous Phantoms in the Mist delude,
Shuns fancied Ills, or chases airy Good.
How rarely Reason guides the stubborn Choice,
Rules the bold Hand, or prompts the suppliant Voice [...]

(1-12)


The passage is less marked than most by the resonant Latinisms in which Johnson customarily takes advantage of every pertinent connotation in either language ("And restless Fire precipitates on Death" [20, emphasis mine]). But, beyond the typically Augustan compounds of epithet and noun or the formal balancing of abstracts, Johnson declares his own particular hand in phrase after phrase. If "the busy Scenes of crouded Life" are those of London, his other imitation of Juvenal, the prospect of "tread[ing] the dreary Paths without a Guide" takes us to the little elegy on Robert Levet. If "fancied Ills" and "airy Good" are at the heart of Rasselas, the somber view of life that underlies this passage is epitomized in the Soame Jenyns review and pervades the whole of Johnson's oeuvre. And whereas Reason, the sovereign faculty, should guide, rule, and prompt our thoughts and actions, we are easily deluded by the "treach'rous Phantoms" of the Imagination. While Johnson shares these concepts with many of his contempo raries, the metaphors that bear them are very much his own.

Any devoted reader of Style is well equipped to refine on these few observations. The same may be said of anyone who benefited from a youthful acquaintance with Brower's The Fields of Light (1951) and other notable works of a time when our purpose as readers, we were taught, was rather to submit ourselves to the text than to submit it to our predilections. Like Johnson himself, we were well aware that the precious self should not (and cannot) be entirely subjugated. But one might acknowledge that truth, so we were taught, without rushing to embrace its loathly absolute, the pernicious doctrine that only selfhood matters and that works of literature, like all other externals, are grist to the egoistic mill.

To set Dryden's brilliant elision, "or knowing it, pursue" beside the middle quatrain of Johnson's opening is to see, in miniature, the value of comparison in stylistic analysis. The question to be pursued in the present paper has to do with the value and the limitations of a different form of comparison, the sort where computational stylistics draws upon statistical analysis. The particular method to be employed is a measure of stylistic difference that I have been developing and testing over the last two years or so. More detailed accounts of the "Delta procedure" and a report on a series of trials with two hundred seventeenth-century poems can be seen in two forthcoming articles (Burrows "Authorship" and "Delta").

Stylistic Signatures

Let us begin with the matter of attribution, by far the most familiar use of computational stylistics. In the trials mentioned above, taking cases where the authorship of the texts was in no real doubt, the object was to pick out the true author of each poem from a field of twenty-five poets of the period. All told, the Delta procedure succeeded with ninety-four poems out of two hundred, even though a hundred of them were of fewer than five hundred words in length. It succeeded with thirty-six poems out of the forty whose length exceeded fifteen hundred words. Of the twenty whose length exceeded two thousand words, it succeeded with nineteen. Since the texts to be considered in this paper range from 2766 words to 6350, the Delta procedure might therefore be used here with great confidence if it could be assumed that poets stamp their stylistic signatures as firmly on a translation as on their original work. In reality, of course, that very question is at issue. The identifiability of the translator of a text that originates in a foreign language makes for subtle attributional problems in which different levels of stylistic versatility and different ideas of translation and imitation have a bearing.

But what, precisely, is this procedure? (2) Most of the methods employed in computational stylistics rest upon multivariate statistical comparisons between some characteristics of a given specimen and those of an appropriate set of norms. For an excellent recent specimen, see Holmes. The characteristics, which are used as statistical variables, comprise the relative frequencies of various simple phenomena such as alphabetic characters, strings of characters, whole words, or common grammatical forms. The advantage of working with whole words rests on their accessibility and their meaningfulness. They help us, in particular, to form close and fruitful inferences about the outcome of an inquiry. Whichever class of variables is chosen, it has become customary, in recent years, to allow the particular variables to "declare themselves," thus obviating, as far as possible, the danger of a pre-determined outcome. The words used, for example, might be the hundred most common in the database that provides the norms fo r a particular inquiry. In this sort of work on language, so our researches teach us, a wealth of variables, many of which may be weak discriminators, almost always offer more tenable results than a smaller number of strong ones.

The multivariate statistical instruments now most used in computational stylistics are designed to portray interrelationships of resemblance and difference across a whole set of specimens. The outcome makes it possible to form explanatory inferences bearing, for example, on the likely authorship of a given specimen. These methods, however, are unsuitable for the crude but useful task of ranking many candidates in a single, all-embracing hierarchy, thus singling out the statistically most eligible among them. Even a ranked series of aggregates or means would serve that purpose if a sound basis were available. But, because the scores for any given specimen on the chosen set of variables will always diverge in both directions from the norms for the database, an aggregate or mean divergence will comprise an arbitrary mixture of positives and negatives.

Now while the differences between positives and negatives--high scores, say, for the in this specimen but low ones for land me--are most instructive, they are not the heart of the matter. An expression of difference, pure difference, is what we need. If all the positive and negative divergences were rendered as absolute divergences, their overall aggregate or their mean might be of interest. A delta score is just such a mean divergence. (The term delta represents D for Difference.)

The first step in the Delta procedure is to establish a frequency hierarchy for the most common words in a large group of suitable texts. The texts are arranged in subsets representing the work of numerous authors appropriate to the particular task in hand. With texts of a bygone era, it is usual and desirable to standardize spelling and to expand contracted forms of expression in order to reduce the influence of trivial or accidental variations. It has also been our practice, in Newcastle, to tag some of the more common homographic forms in order to distinguish the different uses of words like so and that. When the word counts have been made, the frequencies are standardized as percentages of each authorial subset so that the larger subsets do not exert an undue influence on the composition or ranking of the hierarchy.

Working on these lines, we have formed a database of verse by twenty-five poets of the English Restoration period? This set of over half a million words yielded the frequency-hierarchies used for several recent studies of authorship. (The most recent example is Burrows and Craig.) The same database yielded the norms for the Delta project. For this project, however, I have added a further range of texts. Those to be considered in the present paper are fifteen English translations of Juvenal's Tenth Satire, listed below in "Works Cited." They range chronologically from Henry Vaughan (1646) to Jerome Mazzaro (1965) and Peter Green (1967). Those of Evans, Green, Ramsay, and Sheridan are couched in prose, the rest in verse of one kind or another. All but one are independent of the main database. The exception is Thomas Shadwell's version (1687), retained here in order to demonstrate the need for rigor on this very point.

Table 1 represents a small Microsoft Excel worksheet. It offers a simplified version of the procedure, bringing the top twenty words of the database of Restoration verse to bear on a plain question. Can we demonstrate, by this means, that our main sample of Dryden's verse differs less from his translation of Juvenal's Tenth Satire than from The Vanity of Human Wishes? Columns A and B list the twenty most common words in descending order of their frequency in the main database. Column C shows their mean frequencies, all represented as percentages of that set while Column D shows the corresponding standard deviations. Columns E, G, and J show the scores for our sample of Dryden's verse, for his translation of Juvenal X and for The Vanity of Human Wishes respectively while Columns F, H, and K give z-scores representing their divergences from the means of the main set.

The z-scores are used to obtain cognate figures for all the words in a hierarchy where the original frequencies fall away sharply from top to bottom.(4) The object is to treat all of these words as markers of potentially equal power in highlighting the differences between one style and another. Columns I and L, respectively, show the absolute differences between the z-scores for Dryden and his Juvenal X, and those for Dryden and The Vanity of Human Wishes.

By translating positive and negative measures of difference into absolute differences, as shown in Columns I and L, we obscure some useful stylistic information. But we are now able to derive meaningful totals and means for the whole range of differences. These are shown in I3 and L3 and in I4 and L4. A "delta score," as I term entries like those in I4 and L4, can be defined as "the mean of the absolute differences between the z-scores for a set of word-variables in a given text-group and the z-scores for the same set of word-variables in a target text."

The respective delta scores of 1.041 and 1.756 show, as one would hope, that Dryden's Juvenal X is less unlike Dryden's other verse than is The Vanity of Human Wishes. And since an average difference of 0.7 between small numbers like 1 and 1.7 is a sharp one, it is fair to say "much less unlike." Twenty words, of course, are too few for confident analysis, especially when several of them are pronouns of volatile frequency. But the larger differences shown (like those for the and with where Johnson has very high z-scores, and those for but and not, where he runs very low) might well repay a closer examination of the texts themselves. And even twenty words are enough to show why the differences we wish to add up and average out must be derived from z-scores and not from the original text percentages. The text percentages fall away so rapidly as the list extends downward that even sharp differences among lower-order words would be obliterated, in the total, by those from higher in the order.

While this is a satisfactory outcome, the task of distinguishing Dryden from Johnson is not difficult. The Delta procedure really begins to come into its own when it is used to test such questions as how Dryden's Juvenal X compares with our twenty-five authorial sets. Table 2 shows how the full multi-author version of the procedure is used for this purpose.

If Columns A-C, where the output is recorded, are passed over for the moment, Table 2 begins like Table I. Columns D-G show the upper range of the descending hierarchy of common words, standardized means for the frequency of each word in our main database, and the corresponding standard deviations. In its complete form, much too extensive to print, the worksheet from which Table 2 is drawn includes the hundred and fifty most common words, ranging down to those that occur about once in every thousand in the main database. The words tagged so as to distinguish homographic forms are accompanied by parenthetic abbreviations: (i) for infinitive, (p) for preposition, (rp) for relative pronoun, and (c) for conjunction. Columns H-I now provide a site for pasting in the scores for any chosen test piece and for displaying the z-scores derived by setting those scores against the means and standard deviations given in Columns F and G. In Columns J-L and M-O respectively, we have the entries for the first two of our twen ty-five authorial sets. (In its complete form, the table continues until it includes them all. The vast arithmetical power of spreadsheets like Microsoft Excel also allows room for other sets to be added as desired.)

Columns J-L (and each of the corresponding trios that follow) give the standardized score for each word in a particular authorial set, the corresponding z-score, and the absolute difference between each z-score and that of the test-piece. Cells L2-L5 and O2-O5 (and those corresponding to them in the pages not shown) sum up the columns beneath, giving a count of the number of entries and the sum, mean, and standard deviation of those entries. As in Table 1 these means are our delta scores. But, refining on Table 1, they are auto-copied across to Column B where each is listed beside the name of the appropriate author. (The entries in Rows 35-37, towards the foot of Columns A-C, allow for comparison with other authors or for the testing of control-sets.)

Cell B3 shows the minimum entry in Column B. Glancing down, we see, in Cell B17, that this is not Dryden's delta score but Thomas D'Urfey's. On this test, then, Dryden's Juvenal X is least unlike D'Urfey. Dryden himself ranks second out of twenty-five. The results shown in Cells B16 and B17 are assessed in Cells C16 and C17, the two lowest of a fresh set of "delta z-scores" derived from the delta scores in Column B. At-l.454 and -1.172 respectively, these "delta z-scores" for D'Urfey and Dryden do not diverge far from the other twenty-three. Certainly the divergence is not sharp enough to suggest that, if the identity of this particular translator were actually in doubt, they would be the only feasible candidates.

By repeatedly entering new sets of scores in Column H, we can apply the Delta procedure to as many test-pieces as we wish, each test being independent of the rest. Treated in this fashion, the present set of Juvenal translations yields the mixed outcome exhibited in Table 3. The outcome for Dryden is as we have seen above. Thomas Shadwell's Juvenal X (1687) shows a strong but spurious affinity for his authorial set when it is allowed to remain part of that set (against which it is being tested) but ceases to do so when it is treated as an independent entity. (The effects of this difference in treatment can be seen in the delta scores shown in Cells D31 and D37 and in the corresponding delta z-scores shown in Cells E31 and E37.) The strength of this difference is easily explained. The fairly small body of verse that can definitely be attributed to Shadwell chiefly comprises the occasional poems of his period as Poet Laureate. That being so, the inclusion or exclusion of his Juvenal X alters the stylistic colo ring of his whole authorial set.

In Rows 38 and 39, Table 3 also takes account of two versions of Juvenal X whose translators are not members of our set of twenty-five but whose other verse makes it possible to undertake comparative analyses. The scores in Cells F38 and G38 show that Henry Vaughan's Juvenal X, a most uncharacteristic early performance, is quite unlike his other secular work, let alone his pious ejaculations.

But when The Vanity of Human Wishes is compared with Johnson's minor verse (London excluded), it proves unmistakably Johnsonian. It is, after all, the highly idiosyncratic work of a poet who was writing well over half a century later than most members of our main set. It differs less from Milton than from any other of our twenty-five poets. But the scores given in H39 and I39 show its true affinity.

Similar tests of three English translations of Juvenal's Third Satire also yield mixed results. Dryden's version has him ranking only ninth out of twenty-five. John Oldham's version shows a clear affinity for Oldham. And Johnson's London is even more strikingly his own than was The Vanity of Human Wishes. Oldham ranks first out of the twenty-five and Johnson's score lies well above any of them. With other translations again, Dryden ranks only seventh out of twenty-five as a candidate for the authorship of his translation of Ovid's Ars Amoris and fifth out of twenty-five for his Aeneid II. And Sir John Denham ranks only fifth out of twenty-five for The Destruction of Troy, his version of Aeneid II.

At this point, it would appear that our earlier question yields an uncertain but intriguing answer. To conclude, as these results allow, that some translators can be identified but others not would amount to accepting that the Delta test is of no real value in a case where the identity of the translator is doubtful or unknown. And, since this position is seriously at odds with the results obtained in the tests of "true authorship" mentioned earlier, it would follow that some translators are so sensitive to their task that their own stylistic signatures completely disappear behind their image of the foreign author whose work they are representing.

Table 4, however, offers an altered perspective. It embraces all fifteen versions of Juvenal X instead of being confined to the four that were produced by poets whose other work allows us to undertake direct authorial comparisons. And it is focussed on a different question. Table 3 was designed to show which author, out of twenty-five, differs least from a given version. But Table 4 is designed to show which version, out of fifteen, differs least from a given author. Each of the five entries in the first column picks out the lowest score of that row of fifteen. The highlighted figures, in later columns, show where each of those scores lies. The versions by Vaughan and Johnson yield the lowest scores in the last two rows. Shadwell's version yields the lowest score in each of the appropriate rows--the invalid one where his version is included in his authorial set and the valid one where it is excluded. And even Dryden, quite the most elusive translator of our earlier analyses, scarcely eludes identification. Hi s obscure contemporary, Henry Higden, scores 1.195 to Dryden's 1.196. Such a difference, of course, is too slight to bear any serious weight.

We are much too sophisticated, nowadays, to suppose that these fifteen translators have each undertaken the same task. It seems fair to maintain, nevertheless, that, as translators and/or imitators of the same original, their several tasks are decidedly more like each other than is usual in literary composition. Now when their work is tested, three of those four where the appropriate comparison is possible betray their identity while the fourth remains only flimsily concealed. In those four cases, the question posed by the test is this: from which of these fifteen versions of Juvenal X does our main set of Dryden (Shadwell, Vaughan, or Johnson) differ least? Since each test is independent of the rest, each of the three successful results (for Shadwell, Vaughan, and Johnson) is obtained against odds of one in fifteen ("with replacement," as the statisticians say). The fourth result (where Dryden ranks second, albeit so narrowly) is obtained against odds of two in fifteen. When all of these odds are compounded (1/15 x 1/15 x 1/15 x 2/15), the likelihood that this result could come about by chance is around 25,000 to 1. Although statistical analysis never deals in certainties, such odds as these do not encourage doubt.

The upshot is twofold. The Delta procedure shows a capacity for making very subtle differentiations when sufficiently subtle comparisons are undertaken. And the notion that, though they can often be elusive, our "stylistic signatures" are a real presence in our writings is upheld.

"Durfeyism"

Considered from yet another perspective, our main set of results reveals yet another strong and illuminating pattern. Table 3, in which four versions of Juvenal X were matched against our full array of authors, showed that Dryden ranked next after Thomas D'Urfey as least unlike Dryden's own translation. In the other cases, D'Urfey ranked third out of the main twenty-five for Shadwell's version, first of all for Vaughan's, and seventh for Johnson's.

Table 5 is a summary of such authorial rankings for the full set of fifteen versions of Juvenal X. For each version in turn, it shows the top and bottom ends of the authorial rank-order. (Save for the four versions already mentioned, there is no ranking for the actual translator because we have none of his other work as a basis for comparison.) Table 5 shows that, for fifteen versions down the three centuries from Vaughan and Stapleton to Mazzaro and Peter Green, D'Urfey ranks first out of twenty-five on five occasions, second or third on another eight, and fifth and seventh on the remaining two. An even stronger corollary appears at the foot of the fifteen sets of rankings, where Katherine Phillips ("Orinda"), Robert Brome, or both lie at the rear of the field in every case but one.

Such a striking outcome might arise from what is known as a statistical artefact whereby D'Urfey's scores acted as a lowest common denominator for the main database. But that possibility is discountenanced by the fact that it appears to be confined to these Juvenal translations. D'Urfey lies well down the list of twenty-five candidates when Dryden's Ovid, Congreve's Ovid, Dryden's Virgil, or Denham's Virgil are subjected to the same analysis. The first two, in fact, show a strong affinity for Congreve, the third for Nahum Tate, and the fourth a weaker one for Congreve. To my mind, these affinities (especially the first) are so apt as to suggest that D'Urfey's work may reflect something that diverse English translators, who may never have heard of D'Urfey, have in mind when they think of Juvenal. If this were true, it would suggest, in turn, that translators see rather similar features in a given target text or target author.

The initial hypothesis, that, without even knowing it, English translators tend to "see" D'Urfey in Juvenal, can be tested either by an appeal to the reader's ear or by recourse to the numbers from which it derives. With only an ode or two as clear exceptions, the twenty thousand words of D'Urfey's verse in our main database are mostly couched in a brisk, vernacular form of seventeenth-century English. His prologues and epilogues hector the audience or the critics. His songs, many of them published in Pills to Purge Melancholy (1719-20), are mostly low-life bawdry. And his satires are much like this:
"A satyr upon London"
Who in old Sodom would live a day,
Grow deaf with rattling of coaches;
Where folly and noise is called brisk and gay
And wit lies in studying debauches?
With stinks, which smoke and rank fogs display,
Who'd be offending their noses,
That in the sweet shades of the country may
Sit cool under bushes of roses?
Town fops in riot consume every day,
The cit will cheat his own brother,
And the ladies haunt the park and the play
To laugh and rail at each other.
Our funds are wanting, our credit decays,
The French are publicly arming;
And for all the daily noise is of peace
It never conies to confirming.

(1-16)


Among the most Durfey-like of our fifteen versions of Juvenal X was that of Henry Higden (1687). A closer analysis of the word counts supports that finding, and its opening lines say nothing to the contrary
Survey Mankind, muster the Herd
From smoothest Chin to deepest Beard;
Search every Climate, view each Nation,
From lowest to the highest Station;
From Eastern to the Western Indies,
From frozen Poles to the Line that singes:
Scarce will you find one Mortal Wight
Knows Good from Ill, or Wrong from Right,
Because clouds of Lust and Passion blind
And bribe with Interests our Mind;
And while they Combat in our heart,
Our Fondness crowns the conquering part.
What is the thing under the Sun,
That we with Reason seek or shun?

(1-14)


Turning from these jaunty measures to measures of the arithmetic sort, one can readily construct a crude statistical model of the "Durfeyan Juvenal" by uniting and then averaging the word counts for our fifteen versions. For the word the, with scores ranging from 3.525% of the word tokens in Mazzaro's version right up to 7.918% in Johnson's, the mean is 5.27%. For and, with scores ranging from 2.205% in the anonymous Dublin version of 1675 to 4.483% in Johnson's, the mean is 3.262%. And so on downward, for the full list of 150 words. When the Delta procedure is applied to this whole set of average scores, the set, not surprisingly, shows most affinity for D'Urfey.

How successfully can the model be used to identify those words on which the affinity rests--those words where both D'Urfey and the artificial model stand apart from the other twenty-four members of the main database, rising above or falling below the mean frequencies in that larger set? And can it be said that these words are "Juvenalian"?

The words that best satisfy the first stipulation are the personal pronouns and the definite and indefinite articles. The articles, especially the, are more frequent in both D'Urfey and the model than in the main database. Whenever they are abundant, a text will obviously be heavily populated with common nouns, as names either for things or for persons not named by their own names. The main personal pronouns also betoken the broad "terms of reference" of a text. They observe the same overall frequency pattern in D'Urfey and in the model. In rendering Juvenal's tales of vice and folly, he, him, and his are prominent. In rendering the narrative cast of his work, the first-person pronouns (whether singular or plural) and the second-person pronouns (whether modern or archaic) have little part to play. The polar opposition in such "terms of reference" between work like D'Urfey's and the personally oriented work of poets like Brome and Katherine Phillips is highlighted, and it does much to explain their relative po sitions in the rankings shown in Table 5.

The "thingishness" indicated by a high incidence of the is marked again, in both D'Urfey and the Juvenal model, by an abundance of such major prepositions as of, by, for, in, to, on, and at. (The preposition like is common in D'Urfey but not in the Juvenal model because the translators follow Juvenal in making very little use of similes.) The consequent wealth of prepositional phrases, often adjective-studded, gives a great density of texture.

There is, on the other hand, a dearth of many common conjunctions. D'Urfey runs comparatively high on and, if, whilst, how, and the loose conjunction so. The Juvenal model runs low on all of these and also on but, as, when, where, since, though, and the relative pronoun which. (Who and whom lie near their normal frequencies.) As will appear when we return to the particular case of The Vanity of Human Wishes, a dearth of most connectives is not necessarily a sign of loose stylistic organization. But it is usually so. In striving for a Juvenalian effect, the more formal of our translators seem to have slackened their customary syntactic discipline. In the cases of Dryden and Shadwell, where we have a body of other verse to bring into comparison, this explanation can be given direct support. Their versions of Juvenal fall far short of their other verse in the frequency of most connectives.

But, potent as they are in distinguishing a rhetorical framework, these articles, pronouns and connectives do not give off the Durfeyan (or Juvenalian) flavor that marks most of the translations. The sorts of common words whose abundance usually marks the vernacular do not stand out at all. It seems that the statistical model has yielded a picture strong in its main lines but stripped of idiosyncrasy, like an Identikit picture of one person based on the differing impressions of many witnesses.

Particularities

No single version of the poem can quite match the model because all statistical models rest upon a useful fiction. Take the "average household," in which 1.8 children, 0.4 clogs, and 1.3 motor vehicles take their unimaginable places and in which 0.41 of the spouses will divorce their partners. The real value of such models emerges when they are used as the basis for close comparisons--comparisons, in our case, among the fifteen translations from which it derives. In order to highlight die most striking resemblances and differences among them, the Delta procedure is used in a slightly different way. The main database of verse by twenty-five poets is set aside in favor of our model. As can be seen in Table 6, the scores for the model (labelled "Juv10 means") are entered in two places, once as "Base" and once as "Model." The fact that they are identical is the reason for the column of z-scores valued at zero. The ensuing fifteen sets of scores, of which only Dryden's and Johnson's are displayed in the page from Table 6 that is printed here, are thus measured as particular divergences from a zero-rated base.

Just as in Table 2, the delta scores are autocopied into Column 2 and the delta z-scores derived from them are given in Column 3. On this occasion, space allows a further refinement to be shown. The delta scores and delta z-scores for each author are re-copied at the foot of the first three columns, where they are sorted in order of their increasing difference from the base.

In this hierarchy of difference, Shadwell's version is least unlike the model while Johnson's is by far the most unlike. Keeping in mind that the model is derived directly from these fifteen translations, one might have supposed that changes in the language over the centuries would have left Green and Mazzaro furthest from the rest or that the prose versions would stand free of the verse translations. But, possibly because they are drawn together by fidelity to the original, three of the prose versions join Shadwell and Dryden as the five least unlike the model. And, at the other extremity, Johnson's idiosyncrasy prevails over Mazzaro and Vaughan--the most and the least recent members of the set.

Both Dryden and his old enemy, Shadwell, offer brilliant specimens of high Restoration satire. Gifford's version was the standard translation for a century and more. A close study of Evans, Higden, Barten Holyday Esq., or the anonymous Fellow of Trinity College, Dublin, would also be possible but might seem less rewarding. Let us, instead, return to the beginning of this paper and take The Vanity of Human Wishes as our principal example of what the word counts have to tell us about the language of these texts.

Johnson's idiosyncrasy begins to declare itself, as was noted earlier, in the astonishing frequency of the, with a rate of use much more than double that of Mazzaro and a z-score of 1.921. Why is this so? It does not spring from the well recognized fact that Johnson, like his contemporaries, is much given to generality. He often represents abstract concepts, article-free, as abstract nouns or clothes them in personifications: he exceeds all our other translators--whether his predecessors or his successors--in his recourse to life, power, heart, name, and many others:
Yet hope not Life from Grief or Danger free,
Nor think the Doom of Man revers'd for thee:

(155-56)


His comparatively sparse use of the two indefinite articles and of most prepositions indicates that he does not often generalize through the use of post-modified nouns ("a youth of scholarly inclination" or "scholars in the making").

His habit of generalizing by translating a typical case into real or virtual specimens of synecdoche and metonymy is what turns his use of the into a mannerism. There are seven examples in the opening twelve lines of the poem, as quoted earlier. Elsewhere in the poem, they cluster even more closely:
When first the College Rolls receive his Name,
The young Enthusiast quits his Ease for Fame;
Resistless burns the Fever of Renown,
Caught from the strong Contagion of the Gown.

(135-38)


The typical distribution of the main personal pronouns in these texts has already been considered. Here, too, Johnson establishes patterns of his own. With some striking exceptions, his personal pronouns fall below, often far below the norms of our fifteen versions. The exceptions are her, his, and the three archaic forms of the second person. These last are mostly addressed to "the young Enthusiast," to the spirit of Democritus, or directly to the reader. Many of the instances of his are associated by Johnson, as by Juvenal himself and the other translators, with historical personages taken as ominous examples. In Johnson's version, Wolsey, Charles XII, and Xerxes attract over twenty instances. Elsewhere, however, where he reverts to his habit of singling out types, the needy Traveller, the young Enthusiast, the Hind, the Dotard, and others like them attract as many instances again. The teeming Mother attracts one instance of her. All the rest are associated with Britain and Austria or with a legion of perso nifications--Beauty, Peace, and Virtue; Chance and Misfortune; Science, Reason, and Learning. I dwell on these uses of personal pronouns because they are as personal as it gets. All the other second- and third-person pronouns run low in Johnson's poem. Of the first-person pronouns, there are single instances each of my, we, and our, this last, "Our supple Tribes" (95), being a terse reflection on the venality of the British voter of the day. The other first-person pronouns do not appear at all.

Johnson's heavy reliance on and is made obvious by its comparatively high frequency. But his recourse to this form of co-ordination (and, in less degree, to or and nor) is made remarkable by the subordinates they displace. The translators, as we saw, make much less than the normal use of these words. But Johnson runs lower still-lower than all but one of the other translators in his use of if and lowest of all for as, when, that, and since. He alone makes no use of since and he is among those who make no use of whilst or the conjunctive form of so. He runs low for while, though, and how. He runs well below the average in his use of relative pronouns, partly no doubt because of his propensity, noted earlier, of linking epithet and noun. He makes little use of but. Even his and, naturally enough, is often used to link words or phrases rather than clauses. Johnson, in short, is less accustomed than others to connect his ideas through the ordinary syntactical hierarchy than through marshalling them in parallels, antitheses, climaxes, and even larger architectonic forms.

As his delta scores and delta z-scores in Tables 3 and 4 imply, all this is as characteristic of Johnson's other verse as of this poem. We have seen that, in his translations, Dryden is able to conceal his hand. This is one facet of the stylistic versatility that runs throughout the vast range of his oeuvre. But Johnson, like Milton, strikes his own note and holds it. At that point, we are left to form our preferences without benefit of stylistics, whether traditional or computational.

If both forms of stylistics balk at this last hurdle, they are also at one in their care for detail and their use of comparison. Such parallels are accompanied by an underlying complementarity in difference. The close reader sees things in a text--single moments and large amorphous movements--to which computer programs give no easy access. The computer, on the other hand, reveals hidden patterns and enables us to marshal hosts of instances too numerous for our unassisted powers. Even in the common case where we do not have fifteen versions of one original to bring into comparison, these principles hold good.
Table 1

Specimen of Delta Procedure

 A B C D E F

 1 Test-piece
 2 COUNT 20
 3
 4
 5
 6
 7 Main set Dryden
 8 Word Mean SD Score z-score
 9 1 the 4.242 0.630 4.935 1.100
10 2 and 3.770 0.501 3.498 -0.542
11 3 of 1.821 0.315 2.012 0.607
12 4 a 1.601 0.430 1.656 0.128
13 5 to(i) 1.419 0.272 1.535 0.428
14 6 in(p) 1.358 0.189 1.464 0.561
15 7 his 1.154 0.323 1.733 1.792
16 8 with 1.022 0.208 0.910 -0.536
17 9 to(p) 1.014 0.131 0.877 -1.047
18 10 is 0.938 0.312 0.768 -0.546
19 11 but 0.923 0.195 1.025 0.525
20 12 he 0.803 0.241 0.707 -0.396
21 13 all 0.781 0.193 0.576 -1.066
22 14 I 0.766 0.391 0.219 -1.399
23 15 it 0.766 0.239 0.647 -0.497
24 16 as 0.710 0.224 0.614 -0.428
25 17 their 0.641 0.237 0.872 0.975
26 18 her 0.623 0.336 0.406 -0.645
27 19 not 0.616 0.174 0.642 0.147
28 20 be 0.586 0.167 0.411 -1.050

 A B G H I

 1 Dryden, Juvenal 10
 2 COUNT 20
 3 SUM 20.824
 4 mean (= "delta") 1.041
 5 STDEV 0.800
 6
 7
 8 Word Score z-score Abs. diff.
 9 1 the 5.513 2.018 0.918
10 2 and 3.822 0.105 0.646
11 3 of 2.502 2.164 1.556
12 4 a 2.733 2.632 2.504
13 5 to(i) 1.529 0.404 0.024
14 6 in(p) 1.830 2.500 1.939
15 7 his 2.479 4.100 2.309
16 8 with 1.042 0.100 0.636
17 9 to(p) 0.857 -1.202 0.154
18 10 is 0.927 -0.037 0.509
19 11 but 0.880 -0.220 0.745
20 12 he 1.181 1.573 1.969
21 13 all 0.486 -1.529 0.463
22 14 I 0.278 -1.249 0.150
23 15 it 0.255 -2.140 1.643
24 16 as 0.162 -2.448 2.020
25 17 their 0.556 -0.357 1.333
26 18 her 0.371 -0.750 0.104
27 19 not 0.486 -0.742 0.889
28 20 be 0.463 -0.737 0.313

 A B J K L

 1 Johnson, Vanity
 2 COUNT 20
 3 SUM 35.111
 4 mean (= "delta") 1.756
 5 STDEV 1.077
 6
 7
 8 Word Score z-score Abs. diff.
 9 1 the 7.918 5.835 4.735
10 2 and 4.483 1.423 1.965
11 3 of 2.242 1.336 0.729
12 4 a 1.229 -0.863 0.992
13 5 to(i) 0.940 -1.765 2.192
14 6 in(p) 1.229 -0.683 1.244
15 7 his 1.988 2.583 0.792
16 8 with 1.482 2.215 2.751
17 9 to(p) 1.012 -0.016 1.032
18 10 is 0.181 -2.426 1.880
19 11 but 0.470 -2.328 2.853
20 12 he 0.578 -0.932 0.535
21 13 all 0.217 -2.926 1.860
22 14 I 0.000 -1.959 0.561
23 15 it 0.036 -3.056 2.559
24 16 as 0.145 -2.527 2.098
25 17 their 0.470 -0.720 1.695
26 18 her 0.470 -0.454 0.191
27 19 not 0.145 -2.701 2.848
28 20 be 0.145 -2.651 1.601

Table 2

First Page of 150-word Delta Worksheet (Dryden's Juvenal 10 as A
Test-Piece)

 A B C

 1
 2 MAX 1.695
 3 MIN 1.157
 4 MEAN 1.355
 5 STDEV 0.136
 6 OUTPUT
 7 delta scores delta z-scores
 8 Behn 1.456 0.747
 9 Brome 1.688 2.459
10 Butler 1.283 -0.529
11 Congreve 1.311 -0.323
12 Cotton 1.330 -0.179
13 Cowley 1.366 0.082
14 Denham 1.413 0.428
15 Dorset 1.314 -0.302
16 Dryden 1.196 -1.172
17 Durfey 1.157 -1.454
18 Gould 1.360 0.041
19 Marvell 1.381 0.191
20 Milton 1.276 -0.582
21 Oldham 1.249 -0.777
22 Phillips 1.695 2.511
23 Prior 1.270 -0.620
24 Radcliffe 1.429 0.548
25 Rochester 1.383 0.207
26 Sedley 1.348 -0.045
27 Settle 1.251 -0.766
28 Shadwell 1.262 -0.684
29 Swift 1.216 -1.019
30 Tate 1.235 0.883
31 Walter 1.472 0.865
32 Wharton 1.525 1.255
33
34
35 Shad sans Juv 1.486 0.968
36 Vaughan 1.635 2.071
37 Johnson 1.534 1.325
38
39
40
41
42

 A D E F G H I

 1 Test-piece
 2 MAX 150
 3 MIN
 4 MEAN
 5 STDEV Dryden Juvenal 10
 6 DERIVED FROM DATABASE INPUT
 7 Word Mean SD Score z-score
 8 Behn 1 the 4.242 0.630 5.513 2.018
 9 Brome 2 and 3.770 0.501 3.822 0.105
10 Butler 3 of 1.821 0.315 2.502 2.164
11 Congreve 4 a 1.601 0.430 2.733 2.632
12 Cotton 5 to(i) 1.419 0.272 1.529 0.404
13 Cowley 6 in(p) 1.358 0.189 1.830 2.500
14 Denham 7 his 1.154 0.323 2.479 4.100
15 Dorset 8 with 1.022 0.208 1.042 0.100
16 Dryden 9 to(p) 1.014 0.131 0.857 -1.202
17 Durfey 10 is 0.938 0.312 0.927 -0.037
18 Gould 11 but 0.923 0.195 0.880 -0.220
19 Marvell 12 he 0.803 0.241 1.181 1.573
20 Milton 13 all 0.781 0.193 0.486 -1.529
21 Oldham 14 I 0.766 0.391 0.278 -1.249
22 Phillips 15 it 0.766 0.239 0.255 -2.140
23 Prior 16 as 0.710 0.224 0.162 -2.448
24 Radcliffe 17 their 0.641 0.237 0.556 -0.357
25 Rochester 18 her 0.623 0.336 0.371 -0.750
26 Sedley 19 not 0.616 0.174 0.486 -0.742
27 Settle 20 be 0.586 0.167 0.463 -0.737
28 Shadwell 21 you 0.580 0.252 0.116 -1.841
29 Swift 22 they 0.564 0.234 0.278 -1.222
30 Tate 23 for(p) 0.559 0.114 0.741 1.603
31 Walter 24 by(p) 0.555 0.100 0.533 -0.211
32 Wharton 25 my 0.512 0.370 0.069 -1.195
33 26 we 0.510 0.275 0.278 -0.844
34 27 from 0.500 0.127 0.371 -1.017
35 Shad sans Juv 28 that(rp) 0.476 0.228 0.208 -1.173
36 Vaughan 29 or 0.471 0.165 0.371 -0.610
37 Johnson 30 our 0.460 0.268 0.394 -0.247
38 31 thy 0.451 0.247 0.232 -0.887
39 32 was 0.437 0.140 0.232 -1.468
40 33 this 0.426 0.095 0.347 -0.828
41 34 when 0.426 0.105 0.371 -0.528
42 35 are 0.413 0.134 0.486 0.550

 A J K L M

 1 Behn
 2 MAX COUNT 150
 3 MIN SUM 218.385
 4 MEAN MEAN 1.456
 5 STDEV STDEV 1.196
 6
 7 Score z-score Abs. diff. Score
 8 Behn 4.202 -0.064 2.082 3.883
 9 Brome 3.925 0.311 0.206 4.695
10 Butler 1.783 -0.121 2.285 1.229
11 Congreve 1.479 -0.283 2.915 1.750
12 Cotton 1.331 -0.323 0.727 1.666
13 Cowley 1.120 -1.264 3.764 1.198
14 Denham 0.912 -0.747 4.848 0.978
15 Dorset 0.944 -0.371 0.471 0.812
16 Dryden 0.986 -0.217 0.985 1.026
17 Durfey 0.797 -0.452 0.415 1.642
18 Gould 0.797 -0.648 0.427 1.222
19 Marvell 0.792 -0.043 1.616 0.897
20 Milton 1.179 2.063 3.592 0.840
21 Oldham 1.382 1.574 2.823 1.093
22 Phillips 0.733 -0.138 2.002 1.290
23 Prior 0.673 -0.167 2.282 0.765
24 Radcliffe 0.355 -1.206 0.849 0.711
25 Rochester 0.299 -0.961 0.212 0.200
26 Sedley 0.539 -0.441 0.301 0.989
27 Settle 0.617 0.188 0.925 1.016
28 Shadwell 1.133 2.196 4.036 0.620
29 Swift 0.272 -1.249 0.026 1.049
30 Tate 0.475 -0.739 2.342 0.758
31 Walter 0.479 -0.717 0.506 0.569
32 Wharton 1.221 1.914 3.108 0.339
33 0.290 -0.800 0.045 1.226
34 0.396 -0.816 0.201 0.318
35 Shad sans Juv 0.636 0.699 1.872 0.968
36 Vaughan 0.442 -0.175 0.434 0.660
37 Johnson 0.290 -0.634 0.387 0.951
38 0.769 1.290 2.177 0.213
39 0.507 0.500 1.967 0.389
40 0.304 -1.283 0.455 0.552
41 0.544 1.117 1.644 0.345
42 0.382 0.229 0.779 0.691

 A N O

 1 Brome
 2 MAX 150
 3 MIN 253.200
 4 MEAN 1.688
 5 STDEV 1.251
 6
 7 z-score Abs. diff.
 8 Behn -0.570 2.588
 9 Brome 1.847 1.742
10 Butler -1.883 4.047
11 Congreve 0.347 0.285
12 Cotton 0.908 0.504
13 Cowley -0.847 3.347
14 Denham -0.543 4.643
15 Dorset -1.006 1.105
16 Dryden 0.087 1.289
17 Durfey 2.254 2.291
18 Gould 1.536 1.756
19 Marvell 0.392 1.181
20 Milton 0.301 1.830
21 Oldham 0.836 2.085
22 Phillips 2.196 4.336
23 Prior 0.246 2.695
24 Radcliffe 0.296 0.654
25 Rochester -1.258 0.508
26 Sedley 2.135 2.877
27 Settle 2.579 3.317
28 Shadwell 0.157 1.998
29 Swift 2.071 3.293
30 Tate 1.753 0.150
31 Walter 0.128 0.339
32 Wharton -0.469 0.726
33 2.603 3.447
34 -1.430 0.412
35 Shad sans Juv 2.155 3.328
36 Vaughan 1.145 1.755
37 Johnson 1.835 2.082
38 -0.961 0.074
39 -0.340 1.127
40 1.316 2.144
41 -0.768 0.241
42 2.078 1.528

Table 3

Results of Delta Tests for Four Versions of Juvenal 10

 A B C D E
 1 Dryden (1693) Shadwell (1687)
 2 4317 words 3979 words
 3 SUM 33,865 33,438
 4 MEAN 1.355 1.338
 5 STDEV 0.136 0.131
 6 MEDIAN 1.330 1.336
 7 MINIMA 1.157 0.958
 8 MINIMA -1.454 -2.896
 9 Delta Delta
10 score z-score score z-score
11 Behn 1.456 0.747 1.380 0.324
12 Brome 1.688 2.459 1.620 2.158
13 Butler 1.283 -0.529 1.222 -0.882
14 Congreve 1.311 -0.323 1.256 -0.623
15 Cotton 1.330 -0.179 1.342 0.033
16 Cowley 1.366 0.082 1.301 -0.277
17 Denham 1.413 0.428 1.313 -0.184
18 Dorset 1.314 -0.302 1.309 -0.219
19 Dryden 1.196 -1.172 1.332 -0.039
20 Durfey 1.157 -1.454 1.176 -1.233
21 Gould 1.360 0.041 1.352 0.108
22 Marvell 1.381 0.191 1.430 0.704
23 Milton 1.276 -0.582 1.392 0.416
24 Oldham 1.249 -0.777 1.166 -1.306
25 Phillips 1.695 2.511 1.575 1.808
26 Prior 1.270 -0.620 1.276 -0.472
27 Radcliffe 1.429 0.548 1.374 0.281
28 Rochester 1.383 0.207 1.402 0.489
29 Sedley 1.348 -0.045 1.408 0.537
30 Settle 1.251 -0.766 1.336 -0.009
31 Shadwell 1.262 -0.684 0.958 -2.896
32 Swift 1.216 -1.019 1.300 -0.287
33 Tale 1.235 -0.883 1.323 -0.108
34 Waller 1.472 0.865 1.433 0.731
35 Wharton 1.525 1.255 1.462 0.946
36
37 Shad, sans Juv. 1.486 0.968 1.319 -0.143
38 Vaughan 1.635 2.071 1.633 2.251
39 Johnson 1.534 1.325 1.667 2.512

 A F G H I
 1 Vaughan (1646) Johnson (1749)
 2 4374 words 2766 words
 3 34,423 48,750
 4 1.377 1.950
 5 0.078 0.206
 6 1.373 1.969
 7 MINIMA 1.267 1.597
 8 MINIMA -1.402 -1.715
 9 Delta Delta
10 score z-score score z-score
11 Behn 1.482 1.344 1.969 0.093
12 Brome 1.497 1.529 2.360 1.991
13 Butler 1.271 -1.345 2.114 0.796
14 Congreve 1.339 -0.485 1.622 -1.597
15 Cotton 1.373 -0.055 2.120 0.827
16 Cowley 1.301 -0.965 1.803 -0.714
17 Denham 1.381 0.058 1.972 0.106
18 Dorset 1.382 0.062 2.048 0.474
19 Dryden 1.376 -0.007 1.805 -0.706
20 Durfey 1.267 -1.402 1.813 -0.668
21 Gould 1.314 -0.803 2.024 0.359
22 Marvell 1.323 -0.691 1.896 -0.261
23 Milton 1.377 0.007 1.597 -1.715
24 Oldham 1.315 -0.793 1.813 -0.666
25 Phillips 1.559 2.324 2.366 2.022
26 Prior 1.344 -0.424 1.632 -1.546
27 Radcliffe 1.425 0.609 2.212 1.276
28 Rochester 1.493 1.485 2.066 0.563
29 Sedley 1.373 -0.044 2.054 0.506
30 Settle 1.291 -1.096 1.942 -0.039
31 Shadwell 1.347 -0.383 1.921 -0.139
32 Swift 1.377 0.001 2.000 0.243
33 Tale 1.290 -1.110 1.653 -1.444
34 Waller 1.453 0.973 1.992 0.202
35 Wharton 1.472 1.211 1.957 0.036
36
37 Shad, sans Juv. 1.479 1.298 1.995 0.218
38 Vaughan 1.426 0.622 2.001 0.247
39 Johnson 1.788 5.237 1.297 -3.175

Table 4

Delta Scores for Fifteen Versions of Juvenal 10

 Dryden Shadwell Vaughan Johnson
Delta Minima (1693) (1687) (1646) (1749)

 Delta Delta Delta Delta
 Behn 1.456 1.380 1.482 1.969
 Brome 1.688 1.620 1.497 2.360
 Butler 1.283 1.222 1.271 2.114
 Congreve 1.311 1.256 1.399 1.622
 Cotton 1.330 1.342 1.373 2.120
 Cowley 1.366 1.301 1.301 1.803
 Denham 1.413 1.313 1.381 1.972
 Dorset 1.314 1.309 1.382 2.048
1.195 Dryden 1.196 1.332 1.376 1.805
 Durfey 1.157 1.176 1.267 1.813
 Gould 1.360 1.352 1.314 2.024
 Marvell 1.381 1.430 1.323 1.896
 Milton 1.276 1.392 1.377 1.597
 Oldham 1.249 1.166 1.315 1.813
 Phillips 1.695 1.575 1.529 2.366
 Prior 1.270 1.276 1.344 1.632
 Radcliffe 1.429 1.374 1.425 2.212
 Rochester 1.383 1.402 1.493 2.066
 Sedley 1.348 1.408 1.373 2.054
 Settle 1.251 1.336 1.291 1.942
0.958 Shadwell 1.262 0.958 1.347 1.921
 Swift 1.216 1.300 1.377 2.000
 Tate 1.235 1.323 1.290 1.653
 Walter 1.472 1.433 1.453 1.993
 Wharton 1.525 1.462 1.472 1.957

1.319 Shad, sans Juv. 1.486 1.319 1.479 1.995
1.426 Vaughan 1.635 1.633 1.426 2.001
1.297 Johnson 1.534 1.667 1.788 1.297

 Stapleton Holyday Anon. J.H.
Delta Minima (1647) (1673) (1675) (1683)

 Delta Delta Delta Delta
 Behn 1.432 1.599 1.382 1.499
 Brome 1.571 1.668 1.359 1.467
 Butler 1.305 1.407 1.167 1.222
 Congreve 1.329 1.472 1.467 1.319
 Cotton 1.379 1.486 1.257 1.217
 Cowley 1.358 1.318 1.265 1.287
 Denham 1.266 1.397 1.263 1.299
 Dorset 1.469 1.494 1.279 1.288
1.195 Dryden 1.336 1.430 1.394 1.298
 Durfey 1.271 1.332 1.227 1.150
 Gould 1.440 1.516 1.083 1.271
 Marvell 1.491 1.437 1.409 1.427
 Milton 1.352 1.529 1.641 1.479
 Oldham 1.328 1.428 1.240 1.221
 Phillips 1.593 1.569 1.348 1.533
 Prior 1.295 1.487 1.461 1.349
 Radcliffe 1.431 1.491 1.251 1.251
 Rochester 1.439 1.529 1.314 1.269
 Sedley 1.412 1.427 1.294 1.324
 Settle 1.361 1.344 1.285 1.280
0.958 Shadwell 1.248 1.275 1.254 1.241
 Swift 1.412 1.439 1.284 1.267
 Tate 1.344 1.461 1.425 1.342
 Walter 1.525 1.559 1.406 1.478
 Wharton 1.546 1.544 1.477 1.443

1.319 Shad, sans Juv. 1.477 1.462 1.445 1.468
1.426 Vaughan 1.446 1.737 1.645 1.556
1.297 Johnson 1.876 1.827 1.962 1.793

 Higden Sheridan Gifford Evans Ramsay
Delta Minima (1687) (1739) (1802) (1852) (1918)

 Delta Delta Delta Delta Delta
 Behn 1.383 1.687 1.574 1.477 1.503
 Brome 1.548 1.826 1.825 1.616 1.617
 Butler 1.314 1.346 1.469 1.324 1.249
 Congreve 1.105 1.573 1.325 1.395 1.501
 Cotton 1.435 1.516 1.604 1.401 1.340
 Cowley 1.207 1.524 1.374 1.517 1.512
 Denham 1.252 1.594 1.486 1.489 1.500
 Dorset 1.282 1.555 1.563 1.387 1.347
1.195 Dryden 1.195 1.538 1.309 1.486 1.459
 Durfey 1.112 1.373 1.274 1.265 1.344
 Gould 1.299 1.577 1.462 1.379 1.431
 Marvell 1.331 1.584 1.400 1.489 1.535
 Milton 1.151 1.492 1.300 1.353 1.499
 Oldham 1.180 1.497 1.408 1.350 1.336
 Phillips 1.650 1.932 1.804 1.719 1.712
 Prior 1.075 1.501 1.229 1.369 1.461
 Radcliffe 1.415 1.568 1.571 1.479 1.420
 Rochester 1.331 1.613 1.581 1.484 1.509
 Sedley 1.321 1.669 1.561 1.593 1.487
 Settle 1.209 1.420 1.341 1.340 1.450
0.958 Shadwell 1.148 1.434 1.338 1.358 1.337
 Swift 1.167 1.472 1.390 1.384 1.353
 Tate 1.130 1.487 1.256 1.363 1.492
 Walter 1.335 1.629 1.576 1.494 1.605
 Wharton 1.452 1.686 1.577 1.475 1.485

1.319 Shad, sans Juv. 1.371 1.634 1.547 1.568 1.544
1.426 Vaughan 1.447 1.814 1.670 1.668 1.684
1.297 Johnson 1.352 1.819 1.317 1.749 1.915

 Mazzaro Green
Delta Minima (1965) (1967)

 Delta Delta
 Behn 1.710 1.574
 Brome 1.696 1.490
 Butler 1.536 1.319
 Congreve 1.493 1.487
 Cotton 1.568 1.343
 Cowley 1.530 1.394
 Denham 1.637 1.466
 Dorset 1.483 1.337
1.195 Dryden 1.496 1.381
 Durfey 1.358 1.264
 Gould 1.556 1.386
 Marvell 1.607 1.427
 Milton 1.633 1.478
 Oldham 1.471 1.325
 Phillips 1.729 1.727
 Prior 1.483 1.383
 Radcliffe 1.550 1.343
 Rochester 1.544 1.452
 Sedley 1.515 1.443
 Settle 1.415 1.328
0.958 Shadwell 1.513 1.345
 Swift 1.410 1.261
 Tate 1.546 1.458
 Walter 1.634 1.541
 Wharton 1.633 1.589

1.319 Shad, sans Juv. 1.712 1.529
1.426 Vaughan 1.693 1.665
1.297 Johnson 1.769 1.746

Table 5

Summary of Authorial Rankings for Fifteen Versions of Juvenal 10

 Dryden (1693) Shadwell (1687)
 4317 words 3979 words
 2 1 (invalid)
 Durfey 1 Durfey 3

 1 -1.454 Durfey -2.896 Shadewell
 2 -1.172 Dryden -1.306 Oldham
 3 -1.019 Swift -1.233 Durfey
 4 -0.883 Tate -0.882 Buttler
 5 -0.777 Oldham -0.623 Congreve
 6 -0.766 Settle -0.472 Prior
 7 -0.684 Shadwell -0.287 Swift
 8 -0.620 Prior -0.277 Cowley
**
24 2.459 Brome 1.808 Phillips
25 2.511 Phillips 2.158 Brome

 Vaughan (1646) Johnson (1749)
 4374 words 2766 words
 circa 19 (above 1)
 Durfey 1 Durfey 7

 1 -1.402 Durfey -1.715 Milton
 2 -1.345 Buttler -1.597 Congreve
 3 -1.110 Tate -1.546 Prior
 4 -1.096 Settle -1.444 Tate
 5 -0.965 Cowley -0.714 Cowley
 6 -0.803 Gould -0.706 Dryden
 7 -0.793 Oldham -0.668 Durfey
 8 -0.691 Marvell -0.666 Oldham
**
24 1.529 Brome 1.991 Brome
25 2.324 Phillips 2.022 Phillips


 Stapleton (1647)
 3433 words
 Durfey 3

 1 -1.553 Shaewell
 2 -1.361 Denham
 3 -1.313 Durfey
 4 -1.059 Prior
 5 -0.961 Butler
 6 -0.717 Oldham
 7 -0.708 Congreve
 8 -0.636 Dryden
**
24 1.802 Brome
25 2.031 Phillips

 Holiday (1673) Anon (1675)
 3308 words 6350 words
 Durfey 3 Durfey 3

 1 -2.077 Shawell -2.121 Gould
 2 -1.613 Cowley -1.399 Butler
 3 -1.456 Durfey -0.883 Durfey
 4 -1.324 Settle -0.768 Oldham
 5 -0.747 Denham -0.669 Radcliffe
 6 -0.634 Butler -0.651 Shadwell
 7 -0.417 Sedley -0.623 Cotton
 8 -0.406 Oldham -0.572 Denham
**
24 1.458 Behn 1.272 Wharton
25 2.204 Brome 2.680 Milton

 J.H. (1683) Higden (1687) heridan (1739)
 3956 words 4906 words 4121 words
 Durfey 1 Durfey 3 Durfey 2

 1 -1.731 Durfey -1.423 Prior -1.672 Butler
 2 -1.083 Cotton -1.219 Congreve -1.463 Durfey
 3 -1.038 Oldham -1.172 Durfey -1.102 Settle
 4 -1.033 Butler -1.045 Tate -0.994 Shadwell
 5 -0.852 Shadwell -0.918 Shadwell -0.703 Swift
 6 -0.749 Radclliffe -0.902 Milton -0.590 Tate
 7 -0.602 Swift -0.789 Swift -0.549 Milton
 8 -0.578 Rochester -0.700 Oldham -0.511 Oldham
**
24 1.646 Behn 1.844 Brome 2.009 Brome
25 1.966 Phillips 2.554 Phillips 2.828 Phillips

 Gifford (1802) Evans (1852) Ramsay (1918)
 4087 words 4226 words 4031 words
 Durfey 3 Durfey 1 Durfey 5

 1 -1.481 Prior -1.694 Durfey -0.028 Buttler
 2 -1.312 Tate -1.122 Butler -1.187 Oldham
 3 -1.194 Durfey -0.964 Settle -1.180 Shadwell
 4 -1.032 Milton -0.870 Oldham -1.146 Cotton
 5 -0.978 Dryden -0.839 Milton -1.112 Durfey
 6 -0.875 Congreve -0.788 Shadwell -1.084 Dorset
 7 -0.792 Shadwell -0.745 Tate -1.026 Swift
 8 -0.776 Settle -0.687 Prior -0.384 Radclife
**
24 2.142 Phillips 1.714 Brome 1.520 Brome
25 2.276 Brome 2.712 Phillips 2.430 Phillips

 Mazzaro (1965) Green (1967)
 2894 words 3767 words
 Durfey 1 Durfey 2

 1 -2.050 Durfey -1.470 Swift
 2 -1.494 Swift -1.447 Durfey
 3 -1.440 Settle -0.939 Butler
 4 -0.840 Oldham -0.884 Oldham
 5 -0.715 Prior -0.862 Settle
 6 -0.713 Dorset -0.779 Dorset
 7 -0.608 Congreve -0.717 Cotton
 8 -0.572 Dryden -0.716 Radcliffe
**
24 1.708 Behn 1.533 Wharton
25 1.916 Phillips 2.802 Phillips

Table 6

First Page of 150-Word Worksheet (Juvenal 10: Fifteen Versions vs.
Model)

 Base
MAX 1.324
MIN 0.628
MEAN 0.773
DTDEV 0.171


 Delta Juv 10 means
 scores z-scores Word Mean SD

Dryden 0.674 -0.576 1 the 5.270 1.379
Johnson 1.324 3.217 2 and 3.262 0.758
Shadwell 0.628 -0.846 3 of 2.324 0.871
Vaughan 0.861 0.512 4 a 1.914 0.509
Stapleton 0.700 -0.427 5 to(i) 1.075 0.265
Holyday 0.805 0.187 6 in(p) 1.539 0.239
Anon 0.786 0.074 7 his 1.918 0.396
JH 0.713 -0.352 8 with 1.008 0.284
Higden 0.722 -0.297 9 to(p) 1.107 0.180
Shendan 0.685 -0.512 10 is 0.955 0.319
Gifford 0.713 -0.348 11 but 0.726 0.244
Evans 0.652 -0.709 12 he 1.335 0.323
Ramsay 0.666 -0.626 13 all 0.632 0.267
Mazzaro 0.908 0.787 14 I 0.224 0.087
Green 0.759 -0.083 15 it 0.452 0.238
 16 as 0.449 0.189
 17 their 0.651 0.193
1 Shadwell 0.628 -0.846 18 her 0.426 0.144
2 Evans 0.652 -0.709 19 not 0.411 0.213
3 Ramsay 0.666 -0.626 20 be 0.524 0.206
4 Dryden 0.674 -0.576 21 you 0.459 0.379
5 Sheridan 0.685 -0.512 22 they 0.415 0.244
6 Stapleton 0.700 -0.427 23 for(p) 0.781 0.188
7 JH 0.713 -0.352 24 by(p) 0.688 0.187
8 Gifford 0.713 -0.348 25 my 0.102 0.067
9 Higden 0.722 -0.297 26 we 0.289 0.138
10 Green 0.759 -0.083 27 from 0.441 0.137
11 Anon 0.786 0.074 28 that(rp) 0.454 0.277
12 Holyday 0.805 0.187 29 or 0.570 0.166
13 Vaughan 0.861 0.512 30 our 0.297 0.137
14 Mazzaro 0.908 0.787 31 thy 0.149 0.137
15 Johnson 1.324 3.217 32 was 0.392 0.175
 33 this 0.380 0.163
 34 when 0.343 0.077
 35 are 0.425 0.151

 Model Dryden (1693)
MAX 150 COUNT 150
MIN SUM 101.17
MEAN MEAN 0.674
DTDEV STDEV 0.443


 Juv 10 means
 Mean z-score Score z-score Abs. diff.

Dryden 5.270 0 5.513 0.177 0.177
Johnson 3.262 0 3.822 0.738 0.738
Shadwell 2.324 0 2.502 0.204 0.204
Vaughan 1.914 0 2.733 1.609 1.609
Stapleton 1.075 0 1.529 1.714 1.714
Holyday 1.539 0 1.830 1.219 1.219
Anon 1.918 0 2.479 1.417 1.417
JH 1.008 0 1.042 0.122 0.122
Higden 1.107 0 0.857 -1.386 1.386
Shendan 0.955 0 0.927 -0.091 0.091
Gifford 0.726 0 0.880 0.632 0.632
Evans 1.335 0 1.181 -0.475 0.475
Ramsay 0.632 0 0.486 -0.545 0.545
Mazzaro 0.224 0 0.278 0.629 0.629
Green 0.452 0 0.255 -0.829 0.829
 0.449 0 0.162 -1.515 1.515
 0.651 0 0.556 -0.494 0.494
1 Shadwell 0.426 0 0.371 -0.380 0.380
2 Evans 0.411 0 0.486 0.356 0.356
3 Ramsay 0.524 0 0.463 -0.293 0.293
4 Dryden 0.459 0 0.116 -0.906 0.906
5 Sheridan 0.415 0 0.278 -0.563 0.563
6 Stapleton 0.781 0 0.741 -0.214 0.214
7 JH 0.688 0 0.533 -0.831 0.831
8 Gifford 0.102 0 0.069 -0.486 0.486
9 Higden 0.289 0 0.278 -0.081 0.081
10 Green 0.441 0 0.371 -0.515 0.515
11 Anon 0.454 0 0.208 -0.888 0.888
12 Holyday 0.570 0 0.371 -1.206 1.206
13 Vaughan 0.297 0 0.394 0.708 0.708
14 Mazzaro 0.149 0 0.232 0.603 0.603
15 Johnson 0.392 0 0.232 -0.914 0.914
 0.380 0 0.347 -0.199 0.199
 0.343 0 0.371 0.365 0.365
 0.425 0 0.486 0.410 0.410

 Johnson (1749)
MAX 150
MIN 198.53
MEAN 1.324
DTDEV 0.740



 Score z-score Abs. diff

Dryden 7.918 1.921 1.921
Johnson 4.483 1.610 1.610
Shadwell 2.242 -0.095 0.095
Vaughan 1.229 -1.345 1.345
Stapleton 0.940 -0.509 0.509
Holyday 1.229 -1.296 1.296
Anon 1.988 0.178 0.178
JH 1.482 1.674 1.674
Higden 1.012 -0.525 0.525
Shendan 0.181 -2.429 2.429
Gifford 0.470 -1.051 1.051
Evans 0.578 -2.342 2.342
Ramsay 0.217 -1.555 1.555
Mazzaro 0.000 -2.584 2.584
Green 0.036 -1.747 1.747
 0.145 -1.608 1.608
 0.470 -0.938 0.938
1 Shadwell 0.470 0.308 0.308
2 Evans 0.145 -1.252 1.252
3 Ramsay 0.145 -1.841 1.841
4 Dryden 0.000 -1.212 1.212
5 Sheridan 0.181 -0.961 0.961
6 Stapleton 0.615 -0.888 0.888
7 JH 0.398 -1.552 1.552
8 Gifford 0.036 -0.983 0.983
9 Higden 0.036 -1.832 1.832
10 Green 0.651 1.529 1.529
11 Anon 0.253 -0.727 0.727
12 Holyday 0.759 1.140 1.140
13 Vaughan 0.036 -1.912 1.912
14 Mazzaro 0.325 1.287 1.287
15 Johnson 0.181 -1.205 1.205
 0.072 -1.888 1.888
 0.217 -1.636 1.636
 0.253 -1.139 1.139


Notes

(1.) The research on which this paper is based has been generously supported by the Australian Research Council and the University of Newcastle. I am also indebted to Harold Love of Monash University and to my colleagues in the Centre for Literary and Linguistic Computing at Newcastle.

(2.) The description of the procedure in the next few paragraphs follows that given in Burrows ("Delta").

(3.) The present corpus of 540,244 words ranges widely across the work of the following twenty-five poets: Aphra Behn (1640-89) 21,705 words; Alexander Brome (1620-66) 29,539; Samuel Butler (1612-80) 30,932; William Congreve (1670-1729) 30,917; Charles Cotton (1630-87) 12,625; Abraham Cowley (1618-67) 19,272; Sir John Denham (1615-69) 30,092; Charles Sackville, Earl of Dorset (1638-1706) 9,586; John Dryden (1631-1700) 18,238; Thomas D'Urfey (1653-1723), 18,757; Robert Gould (1660?-1709?) 29,110; Andrew Marvell (1621-78) 23,282; John Milton (1608-74) 18,924; John Oldham (1653-83) 32,462; Katherine Phillips (1631-64) 29,004 ; Matthew Prior (1664-1721) 32,000; Alexander Radcliffe (floruit 1669-96) 11,889; John Wilmot, Earl of Rochester (1648-80) 12,725; Sir Charles Sedley (1639?-1701) 10,304; Elkanah Settle (1648-1724) 24,080; Thomas Shadwell (1642?-92) 14,540; Jonathan Swift (1667-1745) 30,974; Nahum Tate (1652-1715) 20,333; Edmund Waller (1606-87) 16,443; Anne Wharton (1659-85) 12,511. Most of the corpus was p repared by John Burrows and Harold Love, assisted by Alexis Antonia and Meredith Sherlock. The Marvell subset was contributed by Christopher Wortham.

(4.) An outline of the calculation and use of z-scores can be found in introductory manuals of statistics. But readers in need of such help may be best served by the lucid plain-language account in Kenny's The Computation of Style, 57-58.

Works Cited

Brower, Reuben Arthur. The Fields of Light: An Experiment in Critical Reading. London: Oxford UP, 1951.

Burrows, John. "'Delta': A Measure of Stylistic Difference and a Guide to Likely Authorship." Forthcoming in Literary and Linguistic Computing.

_____. "Questions of Authorship: Attribution and Beyond." Originally presented as the Roberto Busa Award Lecture for 2001 at the ACH-ALLC Conference at New York University. Forthcoming in Computers and the Humanities 37 (2003).

Burrows, John, and Hugh Craig. "Lucy Hutchinson and the Authorship of Two Seventeenth-Century Poems: A Computational Approach." The Seventeenth Century 16 (2001): 259-82.

Denham, John. The Destruction of Troy. London: 1671.

Dryden, John. The First Book of Ovid's Art of Love. 1709. The Poems of John Dryden. Ed. John Sargeaunt. London: Oxford UP, 1910.

_____. The Second Book of Virgil's Aeneid. 1697. Harvard Classics. vol. 13. New York: Collier, 1910.

_____. The Third Satire. Translations from Juvenal. 1693. The Poems of John Dryden. Ed. John Sargeaunt. London: Oxford UP, 1910.

_____. Translations from Juvenal. 1693. The Poems of John Dryden. Ed. John Sargeaunt. London: Oxford UP, 1910.

Evans, Lewis. The Satires of luvenal ... Literally Translated into English Prose. London: Bohn, 1852.

Gifford, William. The Satires of Decimus Junius Juvenalis: With the Satires of Persius. 1803. London: Dent, 1954.

Green, Peter. Juvenal: The Sixteen Satires. London: Penguin, 1967.

Higden, Henry. A Modern Essay on the Tenth Satyr of Juvenal. London, 1687. Early English Books 1641-1700. University Microfilms. Ann Arbor, University of Michigan: 1983.

Holmes, David. "A Widow and Her Soldier: The Case of the Pickett Letters." Literary and Linguistic Computing 16 (2001): 403-20.

Holyday, Barten. Decimus Junius ,Juvenalis and Aulus Persius Flaccus. Oxford, 1673.

Iuvenalis, D. Iuni. Satura X. <www.thelatinlibrary.com>

J. H. Esq. The Tenth Satyr of Juvenal Done into English Verse. London, 1683. Early English Books 1641-1700. University Microfilms. Ann Arbor, University of Michigan: 1983.

Johnson, Samuel. London. 1738. Johnson: Prose and Poetry. Sel. Mona Wilson. London: Hart-Davis, 1963.

_____. The Vanity of Human Wishes. 1749. Johnson: Prose and Poetry. Sel. Mona Wilson. London: Hart-Davis, 1963.

Kenny, Anthony. The Computation of Style: An Introduction to Statistics for Students of Literature and Humanities. Oxford: Pergamon, 1982.

Mazzaro, Jerome. Juvenal: Satires. Ann Arbor: U of Michigan P, 1965.

Oldham, John. A Satyr, in Imitation of the Third of Juvenal. 1682. The Poems of John Oldham. Ed. Harold F. Brooks and Raman Selden. Oxford: Clarendon, 1987.

Ramsay, G. G. Juvenal and Persius. London: Heinemann, 1918.

Shadwell, Thomas. The Tenth Satyr of Juvenal. 1687. The Complete Works of Thomas Shadwell. Ed. Montague Summers. London: Fortune, 1927.

Sheridan, Thomas. The Satires of Juvenal Translated. 1739. New York: AMS P, 1978.

Stapleton, Robert. Juvenal's Sixteen Satyrs. London, 1647. Early English Books 1641-1700. University Microfilms. Ann Arbor, University of Michigan: 1983.

Vaughan, Henry. Juvenal's Tenth Satire Translated. 1646. The Complete Poems. Ed. Alan Rudrum. London: Penguin, 1976.

The Wish, being the Tenth Satyr of Juvenal, periphrastically rendered ... by a ... sometimes Fellow of Trin. Coll. Dublin. Dublin: Tooke, 1675.

John Burrows (john.burrows@netcentral.com.au) is an emeritus professor of English in the University of Newcastle, Australia. His many publications in computational stylistics over the last twenty years include Computation into Criticism (Oxford: Clarendon, 1987). In 2001, he became the second recipient of the Roberto Busa Award for Computing in the Humanities.
COPYRIGHT 2002 Northern Illinois University
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2002 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Burrows, John
Publication:Style
Date:Dec 22, 2002
Words:12722
Previous Article:Reading literature/culture: a translation of "Reading as a Cultural Practice".
Next Article:Meaning, language, and mind: an interview with Mark Turner.
Topics:


Related Articles
Sara Laviosa and Valerie Cleverton. Learning by Translating. A Course in Translation: English to Italian & Italian to English.
Preface.
Paul Simpson. Stylistics: A Resource Book for Students.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters