London · a lost play at court in 1613, a bookseller's entry of 1653, an adaptation staged in 1727
Does the play Lewis Theobald staged in 1727 as Shakespeare's descend from the Cardenio that Shakespeare's company played at court in 1613, and if so whose writing is still in it?
What happened. In the winter of 1612 to 1613 the King's Men, the acting company that Shakespeare wrote for and part-owned, played twenty times at the court of King James I, and the royal household's accounts list one of the plays as "Cardenno"; a second payment covers "Cardenna", played on 8 June 1613, and neither record names an author. The title comes from Don Quixote, the novel by Miguel de Cervantes (1605; in English 1612), whose Cardenio loses his bride Luscinda to his friend Fernando and runs mad in the mountains; the play was never printed and is not in the First Folio, the collected edition of Shakespeare's plays put together in 1623, and the next trace is forty years on, when on 9 September 1653 Humphrey Moseley, a London bookseller, entered "The History of Cardenio, by Mr Fletcher & Shakespeare" in the Stationers' Register, the book trade's record of who held the right to print what, John Fletcher being the playwright who succeeded Shakespeare as the King's Men's chief writer and wrote Henry VIII and The Two Noble Kinsmen with him; Moseley never printed it, and no manuscript of it is known. On 13 December 1727 Lewis Theobald, a London editor and playwright who had made his name the year before by correcting the poet Alexander Pope's edition of Shakespeare, staged Double Falshood; or, The Distrest Lovers (Double Falsehood in modern spelling) at the Drury Lane theatre as "Written Originally by W. Shakespeare", its plot Cervantes's Cardenio story with the names changed, and claimed in his preface to own three manuscripts of the play, one in the hand of John Downes, prompter to one of the two London theatre companies from 1662, and once owned by the actor Thomas Betterton. Nobody else ever recorded seeing them; a notice at a revival in 1770 placed the "original Manuscript" in the museum of the Covent Garden theatre, which burned in 1808, and no account of the fire mentions it.
Why it is a mystery. Nothing ties the 1613 play or the 1653 entry to Theobald's text except his own word, and whatever he had has been rewritten at least once and probably twice, and rewriting destroys most of what tells one author from another, so four positions have been held since the play appeared. One side holds that Theobald wrote it himself and invented the manuscripts (Pope mocked one of its lines in 1728; George Steevens and Edmond Malone, the leading Shakespeare editors of their day, doubted it in 1780; Tiffany Stern, a historian of the early modern theatre, argued the case in full in 2011); a second holds that it is an old play by someone other than Shakespeare (Richard Farmer, a Cambridge scholar, proposed James Shirley, the leading playwright of the 1630s, in 1767; Malone, and in 2017 Giuliano Pascucci, proposed Philip Massinger, who succeeded Fletcher as the King's Men's chief writer in 1625). A third side holds that Fletcher wrote it: every count of the verse since the American critic Gamaliel Bradford's in 1910 finds a break at scene 3.3 (act 3, scene 3) with Fletcher's habits after it; Stephan Kukowski (1991), Jonathan Hope (1994) and MacDonald P. Jackson (2012), all specialists in who wrote what in this period, give scenes 3.3 to 5.2 to Fletcher; and Naseem Alotaibi's doctoral thesis of 2016 holds that Fletcher is the only author the text will yield. The fourth side, taken by the play's modern editors, Brean Hammond (his Arden Shakespeare edition, 2010), Gary Taylor (2016) and the New Oxford Shakespeare (2017), whose Authorship Companion volume sets out who wrote what, is that Fletcher and Shakespeare wrote it together and that the text is in three layers: a Jacobean original (Jacobean meaning from the reign of James I, 1603 to 1625), a rewriting of the 1660s for the company of William Davenant, who ran one of the two London theatres after the Restoration of the monarchy in 1660 and rewrote several of Shakespeare's plays, and Theobald's revision of 1727.
How it was judged. The documents (two court payment records of 1613, Moseley's entry of 1653, Theobald's licence, dedication, preface and one later statement, and the notices of 1770 and 1808) establish only that the King's Men had a Cardenio in 1613 and that a bookseller assigned it to Fletcher and Shakespeare forty years later, so the text was compared with 264 plays and other works of the period, about 5.2 million words, most of them from the Text Creation Partnership, a programme that keyboards early printed books into searchable text. The measurements were distances based on the 300 commonest words, a statistical classifier, the impostors test (how often a candidate author beats a crowd of other writers; above 0.8 counts as verified) and counts of rare phrases and of small words such as "hath" and "doth" that had gone out of use by 1727; each was tried first on Henry VIII and The Two Noble Kinsmen, where scholars agree which scenes Fletcher wrote, and on the corpus's 45 rewritings of older plays made after the Restoration of 1660 and in the 18th century, each measured for how much of the original wording it keeps, with the 39 whose originals are plays by Shakespeare or Fletcher alone scored for how much of an author survives revision, and only then on Double Falsehood scene by scene, with Theobald's own 93,000 words put through the same tests as a control. What tipped the balance is the text: the second half, from scene 3.3, carries Fletcher's habits of grammar and rhythm as strongly as scenes he is known to have written, at a level none of Theobald's own writing reaches, so a Jacobean play by Fletcher lies under it, and the first half keeps an older layer under heavy rewriting in which Shakespeare fits best of the candidates tested and Massinger and Shirley do not fit at all, though no single scene can be given to Shakespeare with confidence. Put together with the court record of 1613 and the Register entry of 1653, that makes a Fletcher and Shakespeare original the most likely source, at about 64 in 100 (the evidence supports a range of 50 to 75), and Theobald's own invention at about 5; discounting the 1653 entry entirely would lower the first figure to about 50, and proof that the play follows Thomas Shelton's 1612 English translation of Don Quixote, which a phrase count could neither confirm nor refute, would raise it to about 75.
Fletcher wrote the second half; Shakespeare probably wrote in the first half, though that is not proven; Theobald did not make the play up. About 64 percent that the play descends from a Fletcher and Shakespeare original.
The 64 is a judgment with its range stated beside it. Fletcher's presence is measured; Shakespeare's is a reading the models cannot verify at scene length, and nobody has yet rerun the computation.
Weighing the documents and the text (the weights are stated in the report's section "Where the percentages come from"): about 64 percent that the play descends from a Fletcher and Shakespeare original, 24 that it descends from a Fletcher play written alone or with another Jacobean partner, 4 that it descends from a Jacobean play without Fletcher, 5 that Theobald wrote it himself, 3 something else. These are judgments rather than a model's output, and nobody has yet rerun the computation. The range the evidence supports for the first figure is 50 to 75: about 50 if the 1653 Register entry is discounted entirely, about 55 if the Shakespeare readings in 1.2, 1.3 and 3.2 are treated as noise, about 75 if the argument made by Taylor with Steven Wagschal that the play follows Shelton's 1612 translation is accepted as proven (a phrase count could neither confirm nor refute it).
Scenes 3.3, 4.1 and 5.2 show John Fletcher's habits as strongly as his own scenes in Henry VIII and The Two Noble Kinsmen: on the impostors test, where 0.8 counts as verified, they score 0.82, 0.84 and 0.79 (the last just under the line, with every other method agreeing), and the 125 chunks of Theobald's own plays, translations and verse average 0.19 on the same test, with a maximum of 0.63 (two chunks reach 0.60). Scenes 4.2 and 5.1 are probably Fletcher's too, on the classifier and the phrase counts rather than on this test. Scenes 1.1 to 3.2 are not Fletcher's; under heavy later rewriting they keep an older layer (thirteen "hath" and "doth" in 14,900 words against four in Theobald's own 93,000; Jacobean turns of phrase at about twice the rate of his writing). Three passages read as Shakespeare on every model tried (the second half of 1.2, scene 3.2, the opening of 5.2), but the same models call John Webster, another playwright of the period, and Massinger "Shakespeare" often enough that one scene's label cannot be trusted on its own. Shakespeare fits best of the candidates tested; Massinger (Malone 1780, Pascucci 2017) and Shirley (Farmer 1767) do not fit at all. Act 2 was rewritten most.
The strongest case against is Stern's (2011): the manuscripts are unwitnessed, Theobald's statements about them inconsistent, his career one of imitation and adaptation, and the "Shakespearean" parallels the product of his immersion in Shakespeare. The first three points hold, and they are why the documents alone give no more than about 60 percent. The fourth is answered by measurement: Theobald's immersion in Shakespeare did not lead him to write "hath" or "doth" more than four times in 93,000 words, to reach Jacobean phrase density at the median, or to produce Fletcher's fingerprint at the strength of the verified scenes anywhere in his own writing. Her article was not read in full; its argument comes through Alotaibi (2016) and Taylor (2016).
From the weighing set out in the report's section "Where the percentages come from". The 24 percent for a Fletcher play with another partner is a judgment: the plays of Beaumont and Field, two other dramatists of the period, and Davenant's plays from before the theatres closed in 1642 were too small to test as partners. Discounting the 1653 Register entry entirely lowers the first share to about 50; accepting the Shelton-1612 argument as proven raises it to about 75.
The documents are few: two court payment records of 1613 naming a play called Cardenio, the London bookseller Humphrey Moseley's entry of 1653 assigning it to Fletcher and Shakespeare, Lewis Theobald's licence, dedication, preface and one later statement, and the notices of 1770 and 1808. All were read in transcription or scan; the Rawlinson manuscript of the 1613 payments in the Bodleian Library and the Stationers' Register were not examined. Tiffany Stern's article of 2011, which argues that Theobald wrote the play himself, and MacDonald P. Jackson's chapter of 2012 were not read in full; their arguments come through Naseem Alotaibi's thesis of 2016 and Gary Taylor (2016), both read in full.
The text was compared with 264 plays and other works, about 5.2 million words, most of them from the Text Creation Partnership, a programme that keyboards early printed books. Four kinds of measurement were made: distances between texts based on how often they use the 300 commonest words; a support-vector machine, a statistical classifier trained to tell the authors apart; the impostors test, which asks how often a candidate author beats a crowd of other writers and counts a score above 0.8 as verified; and counts of rare phrases and of small words such as "hath" and "doth" that had gone out of use by 1727.
Each method was tried first on Henry VIII and The Two Noble Kinsmen, where scholars agree which scenes are Fletcher's, and then on the corpus's 45 rewritings of older plays made between 1661 and 1772, each measured for how much of the original wording it keeps; the 39 whose originals are plays by Shakespeare or Fletcher alone gave survival curves, which say how much of an author's signal outlasts revision. Only then was Double Falsehood scored, scene by scene and in 400-word windows, with 93,000 words of Theobald's own plays, translations and verse put through the same tests as a control.
Figure 1 shows Double Falsehood scene by scene: the probabilities the classifier gives each of its four groups of writing, Fletcher, Shakespeare, Theobald and 18th-century plays, the impostors scores, and the density of phrases found only in Fletcher's plays. From scene 3.3 Fletcher's figures reach the level of his own scenes; nothing from 1.1 to 3.2 is his. Figure 2 shows the three-class model (Shakespeare, Fletcher, 18th-century) on Henry VIII and The Two Noble Kinsmen: it recovers the accepted hand in 35 of 41 scenes. Figure 3 shows how much of an original author's signal survives rewriting: his probability is 0.14 when an adaptation keeps under 5 percent of his wording and 0.90 above 70 percent.
Fletcher's habits of grammar and rhythm run through the second half at a level none of Theobald's own writing reaches, so a Fletcher play lies under it. In the first half Shakespeare fits best of the candidates tested, but no single scene can be given to him with confidence. With the documents of 1613 and 1653, that gives about 64 percent for a Fletcher and Shakespeare original and about 5 for Theobald's own invention. The 64 is a judgment: it falls to about 50 if the 1653 entry is discounted and rises to about 75 if the play is shown to follow Thomas Shelton's 1612 translation of Don Quixote.
Already established by others. The break at 3.3 and Fletcher after it were established by Bradford (1910), Walter Graham (1916), who later edited the play, Oliphant (1927), Kukowski (1991), Hope (1994), Hammond (2010), Jackson (2012) and Tarlinskaja (2017); the thinness of surviving Shakespeare by Jackson (2012) and Robert D. Hume (2016); Theobald's and the Restoration hands in Act 2 by Oliphant, Richard Proudfoot, the Arden series' general editor, Taylor, Tarlinskaja and Pascucci; the documentary chain as John Freehafer (1969), who traced the play's history, and Hammond (2010) set it out. The three-layer view, a Jacobean original by Fletcher and Shakespeare under a Restoration adaptation and Theobald's revision, is Hammond's (2010) and the New Oxford Shakespeare Authorship Companion's (2017).
What historians have said. Every count made since Gamaliel Bradford's in 1910 finds a break in the verse at scene 3.3, with Fletcher's habits after it; Kukowski (1991), Hope (1994), Hammond's Arden edition (2010) and Jackson (2012) all put Fletcher in 3.3 to 5.2. Hammond's edition and the New Oxford Shakespeare Authorship Companion (2017) hold that the text is in three layers, a Jacobean original by Fletcher and Shakespeare, a Restoration adaptation of the 1660s and Theobald's revision of 1727, with the later layers thickest in Act 2 and in 3.1 to 3.2. The case for Shakespeare has always rested mainly on the documents. Against the standard view, Stern (2011) argued that Theobald, a practised imitator, could have written the whole thing; Alotaibi (2016) holds that Fletcher is the only author who can be recovered; Malone (1780) proposed Massinger and Pascucci (2017) found him in 2.2 to 2.4; Farmer (1767) suggested Shirley.
Nothing here overturns what historians already think. The measurements agree with the standard view on Fletcher, on the placement of the later layers, and on the Julio and Leonora passage of 1.2 (Julio and Leonora being the play's names for Cardenio and Luscinda) and the opening of 5.2. They read the evidence for Shakespeare differently: the Shakespeare-unique phrase links used by Jackson (2012) and Taylor (2016) turn out not to tell Shakespeare from other Jacobean dramatists, since Fletcher's and Massinger's scenes carry them as densely as Shakespeare's own, while Fletcher-unique links do. Stern's case (2011) that Theobald wrote the whole play is answered on the text at about 5 percent; her article was not read in full. Massinger and Shirley are tested at scene length against equal corpora for the first time, and neither appears.