Can the Oxford candidacy be saved? A response to W. Ron Hess, "Shakespeare's dates: their effect on stylistic analysis".

There is a differency between a grub and a butterfly, yet your butterfly was a grub.

Coriolanus: 5:4.12

OLD-SCHOOL Oxfordians still fondly remember the trick that Oxfordian scholar Louis Benezet, Sr. liked to play on Stratfordian English Lit professors in the 1940's. He would give them a seventy-line mixture of passages from Shakespeare and Oxford, defy them to tell one author from the other, find they had great trouble in doing so, and conclude that his experiment showed their styles to be barely distinguishable. (1) Much has changed since those days. In 1980, Steven May (79-84) showed from external evidence (and over Charlton Ogburn, Jr.'s objections that "he is unconcerned with stylistic criteria" [396]) that some of the passages Benezet thought were Oxford's were in fact written by Robert Greene, and that five other poems confidently assigned to Oxford by J.T. Looney and other Oxfordian scholars (following A.B. Grosart), were not Oxford's work.

In 1987, our students in the Claremont Shakespeare Clinic, sponsored by the Sloan Foundation and the Shakespeare Authorship Roundtable, began what became a seven-year computer study of all testable Shakespeare claimants, to see which, if any, matched Shakespeare. For the first two years the tests were favorable to Oxford, attracting much interest among Oxfordians and warm invitations to present our results to Oxfordian audiences. But in 1989 we discovered what looked like serious flaws in our then-best test (Valenza 1990) and turned to six other tests that showed mismatch after mismatch between Shakespeare and twenty-seven testable poet claimants, including the front-runners Oxford, Bacon, and Marlowe. Oxford's poems flunked five of the six new tests and seemed particularly different from those of Shakespeare.

When the students reported these results to the Shakespeare Authorship Roundtable in 1990, they got worldwide media attention. A round of "refutations" ensued in Oxfordian publications and in private correspondence with Oxfordians. Some of these made worthwhile points; some did not. But the invitations to respond to them in Oxfordian publications and meetings stopped completely. We made a few revisions in our tests and published our general findings in mainstream journals, Computers and the Humanities (1991), and Notes and Queries (1991a). (2)

HOWEVER, it was not the end of our work. After a two-year break to build up and edit our text archive, the Clinic met again to test playwright claimants. This time the students validated fifty-one tests for plays and an additional eight tests for poems, besides the six we used on poems in 1990. As before, Shakespeare's core works seemed to have high internal consistency, and no claimant work, no poem or play from the "Shakespeare Apocrypha" (noncanonical works ascribed at one time or another to Shakespeare), came close to matching Shakespeare. The most discrepant core Shakespeare play had three Shakespeare rejections in fifty-one tests; the least discrepant claimant play had eleven. These results appeared in the April 1996 issue of Computers and the Humanities and were vigorously, but we think ineffectually, attacked by our former collaborator, Don Foster (his 1996a, our 1998). They were reprinted with slight revisions in our 1998. (3)

Key to Table 1:

HCW: Hyphenated Compound Words per 20,000 words

Rel. clauses: Relative clauses per 3,000-word block

BoB7, Modal distance, Open lines, Feminine endings, Enclitics, and Proclitics per 1,000 lines: see text. Shakespeare ranges are set in bold type.

Shakespeare Baseline:

A: fourteen 3,000-word blocks of Shakespeare's poems; all but "A Lover's Complaint" (Shakespeare authorship doubtful) and "Phoenix & Turtle" (too short);

B: fifty-six 3,000-word blocks of verse from selected Shakespeare plays;

C: twenty-eight 1,500-word blocks of Shakespeare's poems, minus "LC" and "Pht"; same as A, but with 1,500-word blocks.

Remarks:

g: results can be sensitive to differences of genre (poem verse v. play verse);

e: results can be sensitive to differences in editing, such as spelling and punctuation;

p: results can be sensitive to differences in prosody, that is, verse structure, meter, stanzaic structure, and rhyme schemes;

s/m: results can be sensitive to differences in subject matter;

t: results can be sensitive to differences in time of composition.

All ranges and results except those in the boxed area are based on comparisons between the entire Oxford poem corpus, per Steven May, 1980, and 3,000-word Shakespeare poem and/or play verse blocks. Ranges and results in the boxed area are based on comparisons with 1,500-word Shakespeare poem blocks, and compare only the 1,388 words of Oxford's poems that are in iambic pentameter (I-5) with like-sized I-5 Shakespeare poem blocks. See Shakespeare Baseline, above, for details. Shakespeare's most discrepant poem block, both 3,000-word and 1,500-word, is the first block of Venus and Adonis.

Our 1996/1998 play results have little bearing on the Oxford candidacy because none of Oxford's plays have survived, but the new 1996/1998 poem tests do permit a significant updating of the Oxford findings we published in 1991. Highlights appear in Table 1 to the left:

Our testing methodology

Our methodology can be summed up in three short phrases: "clean baseline," "block and profile," and "silver bullets." "Clean baseline" means that we tried to test from a pure Shakespeare baseline, from which anything thought authored or co-authored by someone else was excluded. In Oxford's case we also tried to use a clean comparison sample: the poems Steven May assigned to Oxford in 1980, not his "possibly Oxford" poems, nor the A Hundreth Sundrie Flowres poems claimed for Oxford by some Oxfordians. (4) "Block and profile" means that, when we calculated a Shakespeare profile for a given trait--say, hyphenated compound words--we aimed to compare like-sized blocks with each other. Large blocks and large numbers normally tell you more than small blocks with small numbers because they average out more variance. With giant, play-sized blocks, 20,000 words or so in size, we could easily validate fifty-one tests. But for poems the blocks have to be smaller. Oxford and Bacon wrote only 3,000 words or so of testable poetry; Marlowe, two 3,000-word blocks, not counting his translations. These, and corresponding 3,000-word blocks of Shakespeare, should have more "noise" and wider test profiles than 20,000-word play blocks, and most of them did. We could only validate fourteen tests for 3,000-word and 1,500-word Shakespeare blocks, not fifty-one. 500-word blocks are so short and noisy that few of our tests are usable.

Besides size, one should also try to match for other variables: genre (whether a work is play verse, prose, or poem), time of composition, subject matter, editorial conventions (spelling and punctuation), and prosody (meter, stanzaic structure, etc.). Many of these appear in the "Remarks" column of Table 1 and should figure prominently in any discussion of whether Louis Benezet was right in treating Oxford like a mature butterfly, properly comparable to Shakespeare, and not like a juvenile caterpillar or "grub," unsuitable for comparison to Shakespeare's mature work. It is seldom possible to match perfectly for all of these at once, but there are often ways to try one combination against the other and see how much difference it makes. In Oxford's case, we have matched our 3,000-word blocks for genre (poem v. poem, or poem v. play verse) and spelling (Riverside Shakespeare spelling, including hyphenation), but not for prosody or time of composition--unless, as Ron Hess and other Oxfordians argue, we are wrong about our Shakespeare dates. Ninety-nine percent of Shakespeare's verse, but less than half of Oxford's, is iambic pentameter, and most of Oxford's poems far predate Shakespeare's plays, as conventionally dated. For verse-tests, which are considered sensitive to prosodic variations (boxed in Table 1), we used 1,500-word blocks matched for genre (poem v. poem), spelling, and meter (iambic pentameter v. iambic pentameter), but, again, not for time of composition by conventional reckoning.

Finally, there is our preference for "silver-bullet" tests, which attempt to disprove common authorship by showing lack of resemblance, rather than "smoking-gun" or "thumbprint" tests that attempt to prove common authorship by showing supposedly unique resemblances. (5) Part of this preference may have come from our assignment by the Shakespeare Authorship Roundtable, which was to use computers to shorten the list of credible claimants. But most of it stems from the fact that, with imperfect tests, a "couldn't-be" finding is much more telling than a "could-be" finding. In thirteen years of looking, we have heard many claims of "unique quirks" shared by one claimant or another with Shakespeare, but none has ever been shown, like thumbprints or smoking guns, to be free of false negatives or positives. Absent such a showing of perfection, fitting a size-five slipper does not prove that you are Cinderella, only that you are a could-be. You could also be Little Miss Muffet or Tiny Tim. But not fitting the tiny slipper is strong proof that you are not Cinderella.

Our test outcomes updated

That means that the seven strong, validated Shakespeare tests that the Oxford poem sample passed (listed in our 1996 and 1998, Appendix Five) are much less interesting than the seven that it flunked. The former are nothing more than could-be's. Only the latter (plus hyphenated compound words, a close-call Oxford pass, and open lines, a time-sensitive verse-test passed by Oxford's iambic-pentameter poems and not available in our earlier work for I-5 only) are listed in Table 1. Table 1 compares Oxford's poems with Shakespeare's least typical core poem block, the first 1,500 words--or the first 3,000 words--of Venus and Adonis. In every case, the most discrepant Shakespeare block fits (albeit sometimes barely) within the Shakespeare profile we used, while, in every case but two (HCW's and open lines), the matched Oxford block does not. Let us look at the Oxford outcomes.

Grade Level: Shakespeare's poems have much longer sentences and/or longer words than Oxford's, testing no lower than the tenth-grade level. Oxford's poems test at the seventh-grade level (our 1996, 1998 Appendix Five). This test, which compares Oxford's lightly-modernized punctuation with that of the Riverside Shakespeare, is sensitive to editorial preference, but comparing original-punctuation Oxford with original-punctuation Shakespeare would make the gap even wider. It seems to us a clear rejection.

HCW's: Oxford's poems have fewer hyphenated compound words per block than any like-sized Shakespeare poem block, and fewer HCW's than ninety-seven percent of Shakespeare's like-sized play-verse blocks. But, to be cautious, we re-edited Oxford's poems to mark every arguable Riverside hyphenation; we expanded our Shakespeare verse baseline to include plays, as well as poems; and we broadened our Shakespeare profile to include the highest …