Picture of Gabriel Egan G a b r i e l   E g a n  .  com

"Shakespeare: Editions and Textual Studies in 2019" Not the Year's Work in English Studies

Only one major scholarly edition of Shakespeare appeared this year: All's Well That Ends Well edited by Suzanne Gossett and Helen Wilcox for the Arden Shakespeare Third Series. My copy's imprint page says that this edition was first published in 2019 but the British Library catalogue and the Arden publisher's website both say 2018. The edition's Preface indicates that Gossett wrote the Introduction and Commentary and Wilcox established the text and wrote the textual notes, but this review will treat them as equally responsible for all matters in their book.

    Gossett and Wilcox's Introduction (pp. 1-119) reports that it was Frederick Boas in 1896 who first christened All's Well that Ends Well a 'problem play'. There is no stage history before the eighteenth century. Because the performance history is bound up with the critical responses, there is no separate performance history in this Introduction. The play's title promises a happy ending but it is not clear that the audience actually gets one. The 'problem play' category is the critical response to this -- and to other plays with similar complexities such as Measure for Measure, Troilus and Cressida, and even Hamlet -- but Gossett and Wilcox think it a useless label. All's Well that Ends Well is the last comedy that Shakespeare wrote, and Gossett and Wilcox detail the jarring shifts in mode and tone across the script.

    The source for the main plot is Giovanni Boccaccio's Decameron via William Painter's The Palace of Pleasure. Gossett and Wilcox relate the source plot in detail and then describe the differences in Shakespeare's telling of the tale. Because there are bed tricks in All's Well that Ends Well and Measure for Measure and because the one in All's Well that Ends Well comes from the play's source, it has long been assumed that All's Well that Ends Well must have been written first. The logic here is that it is inherently unlikely that Shakespeare first invented a bed trick for Measure for Measure and only then wrote a play, All's Well that Ends Well, for which the source happens to have a bed trick.

    In the New Oxford Shakespeare Authorship Companion (of which the present reviewer is co-editor), Gary Taylor argued that duplicated stage directions, act divisions, and calls for cornetts in the Folio text of All's Well that Ends Well all suggest a manuscript annotated for use in the theatre, and moreover -- on account of the act divisions and cornetts -- that this annotation was done after the King's men got possession of the Blackfriars theatre in 1608. But this does not tell us when the play was originally composed. Maybe the Loue labours wonne mentioned by Francis Meres was another name for All's Well that Ends Well, in which case it was written in the 1590s.

    The fact that the opening of All's Well that Ends Well parallels the opening of Hamlet could be used to argue that they were written around the same time, or that All's Well that Ends Well was much later if we think that no one recycles their most recent work. But then in examining what can go wrong after marriage -- a subject not normally explored in comedies -- All's Well that Ends Well is more like the late plays The Winter's Tale and Cymbeline. Gossett and Wilcox detect a possible allusion to the Gunpowder Plot of 1605 in lines 4.3.18-24 ("Now, God delay our rebellion . . . o'erflows himself") but consider it pretty faint and cryptic (pp. 17-18).

    Gossett and Wilcox briefly survey the evidence in the 1986-87 Oxford Complete Works and in Douglas Bruster and Genevi ve Smith's "A New Chronology for Shakespeare's Plays" (reviewed in YWES for 2014) that narrows the date to 1604-8. MacDonald P. Jackson argued (in an article reviewed in YWES for 2001) that because there is a character called Spurio in Thomas Middleton's The Revenger's Tragedy (first performed early 1606) and because Paroles in All's Well that Ends Well mentions a man called Spurio, All's Well that Ends Well must be no earlier than 1606. But Gossett and Wilcox point out that since Middleton's play was put on by the King's Men, Shakespeare might easily have read it in manuscript a little before its first performance.

    That the Folio text of All's Well that Ends Well makes abundant use of the word 'God' puts its composition before the May 1606 act against stage profanity. Weighing a few other minor studies too, Gossett and Wilcox observe that a date of 1605 is consistent with the majority of the evidence if the play is sole-authored, but Laurie Maguire and Emma Smith have made a case for Middleton's hand in it, examined in detail in Appendix 2 (considered below) along with the attempted refutation of it by Brian Vickers and Marcus Dahl. Next Gossett and Wilcox survey the evidence adduced in the New Oxford Shakespeare Authorship Companion for the play being revived in the late 1610s or early 1620s with a layer of Middletonian annotations. This layering would account for some of the inconformities of the play.

    All the play's children have lost a father, and all the old people have lost a spouse. Gossett and Wilcox explain the British system of wardship and the Court of Wards to contextualize this. The script is inconsistent on how long Helen has been supported by the Count and Countess: her father died only six months earlier (1.2.71) but Bertram says her "breeding" was "at my father's charge" (2.3.113-114), which implies a much longer period of support. Directors often give Helen and Bertram a childhood-companions backstory in dumbshow. This raises a suggestion of incest in their marrying, and incest is also raised in the Countess's opening remarks about Bertram being a second husband to her, and is also raised in the Countess's repeatedly saying that she considers Helen as a daughter to her. The virtue of the countess and of Helen have been much debated by critics. 'Helena' is the minority form of her name: overwhelmingly she is Helen. In Troilus and Cressida, Helen of Troy is once called Helena, suggesting that these are the same name, as does the alternation between these two forms in A Midsummer Night's Dream, where another Helen/Helena pursues a man.

    Helen says that she is making a pilgrimage from Roussillon to the shrine of Saint Jacques (= Saint James), which is in Santiago de Compostela in north-west Spain (about 700 miles due west of Roussillon). So how does she end up going east into Italy? This seems to be Shakespeare's mistake, since he also has the Widow of Florence say that pilgrims going to the shrine of Saint Jacques often stay at her house on the way (3.5.93-95). When Helen is presented to the King of France some editors choose to have her disguised (since Lafeu seems not to know her) but then she immediately says who she is ("Gerard de Narbonne was my father") so a disguise would be pointless.

    The first recorded performance of the play was in 1741, but for a long time thereafter the sexual material (such as the bed-trick) was removed or toned down in performance. Gossett and Wilcox count 52 bed-tricks in 44 Stuart plays (p. 57). Whether Bertram and Helen are actually married by the hand-fasting that the King of France makes them do (to which Bertram does not verbally consent), and whether consummation is necessary for a marriage to be valid, were contestable by seventeenth-century English legal standards. The slippage in the terms of the conditions, from "show me a child" when Bertram states them to "are by me with child" when Helen says she has met them, means that even at the end it is uncertain that the conditions have been met. But the onstage audience seems to take the pregnancy as meeting the condition set.

    Bertram has no soliloquy so our knowing what motivates him is difficult. Gossett and Wilcox explore the parallels between Bertram's circumstances and the lives of Henry Wriothesley the Third Earl of Southampton and William Herbert the Third Earl of Pembroke (pp. 68-70). Both the real men were wards, opposed marriages made for them, sought military glory, and were known to Shakespeare. (Southampton was dedicatee of Venus and Adonis and The Rape of Lucrece and perhaps the young man advised to marry in the first 17 Sonnets, and Herbert was a dedicatee of the Folio).

    Gossett and Wilcox explore Paroles's malign influence as the explanation for Bertram's bad behaviour. Rory Loughnane has argued that the second scene of Paroles insulting his colleagues was added for posthumous revival before 1622. Gossett points out that if this is correct, it is evidence that the Paroles scenes were a major source of the play's attraction from the beginning. This leads Gossett and Wilcox to a discussion of the nature of foolery ('natural' versus 'artificial') in early modern thought, and how Paroles ends up like Lavatch in status. They finish their Introduction with an examination of how some productions have handled the difficult ending of the play.

    And so to the text of the play as presented by Gossett and Wilcox. As usual, this review does not offer an exhaustive account of their choices regarding emendation but rather a selection of the most interesting and disputable ones. Where in the Folio the Countess tells Helen to say "no more least it be rather thought you affect a sorrow, then to haue --" and Helen replies "I doe affect a sorrow indeed, but I haue it too" (1.1.50-52), Gossett and Wilcox choose to retain the Folio's dash, which gives the impression that Helen interrupts the Countess. But in a note, Gossett and Wilcox explain that Helen in her first speech rudely interrupting the Countess would be unlikely and suggest that Helen perhaps speaks her first line as an aside. Many editors treat the dash as a printing error and replace it with "it." to properly end the Countess's speech.

    At 1.1.55-58, the Folio has "Mo. If the liuing be enemie to the greefe, the excesse makes it soone mortall. | Ros. Maddam I desire your holie wishes, | Laf. How vnderstand we that? | Mo. Be thou blest Bertrame . . .". It is unclear what Lafeu (given the speech prefix "Laf.") is commenting on in this exchange and some editors reverse the order of his speech and Bertram's (prefixed "Ros.") so that Lafeu responds immediately to the Countess's speech (prefixed "Mo.") and then the Countess responds immediately to Bertram. Alternatively, Lafeu's speech makes sense if it is appended to his previous speech before this exchange, making it his first response to Helen's enigmatic "I doe affect a sorrow indeed, but I haue it too", which response then continues with "Moderate lamentation is the right of the dead, excessiue greefe the enemie to the liuing". Gossett and Wilcox leave the order of speeches as it is in the Folio and explain that ". . . Bertram may simply interrupt a discussion of Helen that he finds boring by asking his mother's blessing". That is, Lafeu ignores Bertram's rude interruption of the discussion and reflects on the last thing the Countess said.

    Where the Folio has Helen say that virtue's steely bones "Lookes bleake i'th cold wind: withall, full ofte we see | Cold wisedome waighting on superfluous follie" (1.1.104-105), Susan Snyder's Oxford Shakespeare single-play edition omitted the first occurrence of "cold". Gossett and Wilcox follow the Folio's reading, but observe that Snyder's reading improves the meter and wonder if in fact the problem is the second 'cold' being an inadvertent repetition of the first. This example and the two before it show the editors sticking with the Folio's reading while providing in their notes plentiful evidence that the Folio reading is in fact wrong. Given the strength of their arguments, it is unclear why they do not emend.

    At 1.1.146, the Folio has Paroles say of the loss of virginity that "within ten yeare it will make it selfe two" and Gossett and Wilcox follow G. Blakemore Evans in emending the "ten" to "t'one" (meaning 'the one'), on the assumption that the manuscript had 't'on'. This makes Paroles refer to the human nine-months gestation period so that one person becomes two within a year. But Gossett and Wilcox also footnote the possibility that Paroles is continuing his monetary-inflation metaphor, noting that a starting 'principal' will indeed double in 10 years if compound interest is paid at the rate of 10% per year (the maximum allowed in England at the time). Mathematically, though, the principal will double within 8 years at that rate (since 1.18 = 2.14) and in any case, as Gossett and Wilcox observe, Paroles's point requires a rather more rapid inflation than that. Supposing that the correct word is 'the' instead of 'ten' makes for easier sense but as Gossett and Wilcox note this is harder to explain as a misreading.

    The Folio has Paroles end his speech on virginity with a question, followed by Helen's response: "Will you any thing with it? | Hel. Not my virginity yet: | There shall your Master haue a thousand loues" (1.1.162-163). The problem is that the second line of Helen's response seems suitable to follow one in which the Parisian court (the referent of "There") has been mentioned, but it does not. One possibility is that a line or part-line before "There . . ." has dropped out and should be so marked as missing in a modern edition or even invented by the editor (as Thomas Hanmer did with "You're for the Court"). Helen could perhaps gesture to the "There" being the place Bertram is going. Gossett and Wilcox footnote here and at 1.1.110 the possibility that the entire exchange about virginity is a posthumous insertion by Middleton and that 1.1.100 originally read not "Are you meditating on virginitie?" as it does now but "Are you meditating on the court?" and was immediately followed by Helen's "There shall your Master haue a thousand loues" (1.1.164), as argued by Loughnane and Taylor in the New Oxford Shakespeare Authorship Companion.

    At 1.3.109, Gossett and Wilcox add "Dian no" before "queen of virgins" (where the Folio has nothing), to match "Fortune . . . no goddess" and "Love no god" in the preceding lines. Lewis Theobald was the first to correct this omission of two words, although he chose the longer form "Diana no".  Where the Folio has Helen refer to her hope as "this captious, and intemible Siue" (1.3.199) into which she pours the water of her love, Gossett and Wilcox follow the Second Folio in emending "intemible" to "intenible" and then modernize to "intenable" meaning incapable of retaining.

    The Folio has "L. Laf. Pardon my Lord for mee and for my tidings. | King. Ile see thee to stand vp. | L. Laf. Then heres a man stands that has brought his pardon" (2.1.159-161). Theobald emended "see" to "fee" (meaning to pay or be paid) and "brought" to "bought", and Gossett and Wilcox adopt these changes. The second is necessary because Lafeu can hardly ask for a pardon that he says he has brought with him, and with this change accepted the first change turns an awkward way of saying "Let me see you stand up" into a jokey initiation of the economic metaphor that Lafeu extends.

    Gossett and Wilcox print "great floods have flown | From simple sources, and great seas have dried | When miracles have by the greatest been denied. | Oft expectation fails, and most oft there | Where most it promises; and oft it hits | Where hope is coldest, and despair most fits" (2.1.137-142). The problem is that "Oft expectation fails, and most oft there" has no line to rhyme with, even though the preceding 12 lines are rhyming couplets and so are the next 68. Samuel Johnson thought that "When miracles have by the greatest been denied" lacked "import or connection" with its surroundings, and proposed that a line was lost. But Gossett and Wilcox see it as qualifying the preceding four words so that the denying of miracles by the great happens when "great seas have dried". This they explain as a reference to Pharaoh denying Moses's miracles (including drying the Red Sea, Exodus 14.21-22) and more generally the scepticism about miracles shown by great people. The final word in the passage, "fits", is Theobald's conjectured emendation of F's "shifts", which is hard to find sense for but easily explained as a misreading of "ffitts" in which a "f"' was misread as a long 's'.

    The King confirms his promise to give Helen whatever she demands if she cures him with "Ay, by my sceptre and my hopes of heaven" (2.1.190). This is Gossett and Wilcox's adoption of Theobald's emendation (suggested by Styan Thirlby) of the Folio's "helpe" to "heaven", and it keeps the rhyming couplets going and fits the nearby religious language. At 2.3.1-40, Gossett and Wilcox follow the distribution and attribution of speeches in the Folio that C. J. Sisson thought badly corrupted. In the exchanges here, Paroles gets to speak quite eruditely and knowledgeably, and Sisson argued that the clever things he says actually belong to Lafeu, with whom Paroles can merely agree by claiming that he has the same thoughts. Bertram speaks only once, saying just three words, until he reacts to being given to Helen at 2.3.106, and, according to Sisson, these three words are properly Paroles's and Bertram does not enter until 2.3.57 with the other young lords. For Sisson, the point of the speech is Paroles's interaction with Lafeu, but Gossett and Wilcox believe that in fact ". . . Lafeu is trying to speak to Bertram but is repeatedly interrupted by Paroles . . ." (2.3.9n), which plays comically.

    The Folio has two successive speeches by Lavatch: "Clo. Did you finde me in your selfe sir, or were you taught to finde me? | Clo. The search sir was profitable . . ." (2.4.32-35). Gossett and Wilcox adopt the solution from the New Shakespeare edited by Arthur Quiller-Couch and John Dover Wilson of inserting between these two speeches the reply from Paroles "In myself". They footnote Taylor's observation that Paroles might have said "In myself, knave" as a characteristic insult (since four lines later he calls Lavatch "A good knave") but Gossett and Wilcox decide that when inventing text "it seems best to keep it as brief as possible". Also against Taylor's emendation they cite the fact that "elsewhere Paroles calls Lavatch a fool" (indeed he does three lines earlier), but I would have thought that would be a reason to invent "In myself, fool" as Loughnane does in the New Oxford Shakespeare, where he supports the emendation by pointing out that Lavatch's reply engages the same insult: "much fool you may find in you".

    At 2.5.45-46, the Folio has Lafeu say to Paroles "I haue spoken better of you, then you haue or will to deserue" and rather than trying to make sense of the impossibly awkward "have or will to" Gossett and Wilcox adopt the emendation conjectured by Samuel W. Singer of adding "wit" to make "have wit or will to deserve". When Paroles calls Lafeu an idle lord, Gossett and Wilcox have Bertram reply "I think not so" although the Folio has "I thinke so" (2.5.50). As they point out, Paroles's surprise at Bertram's answer and Bertram's subsequent explanation of his high opinion of Lafeu make the Folio reading impossible. 

    A famous crux concerns Helen's hope that the bullets Bertram faces in war will whizz by him harmlessly. The Folio reads "Fly with false ayme, moue the still-peering aire; | That sings with piercing, do not touch my Lord" (3.2.111-12). The chief problem is making sense of the air being "still-peering", and Gossett and Wilcox adopt George Steevens's emendation of this to "still-piecing", which they gloss as meaning that "the air that continually (still) mends or join itself together (OED piece v. 1, 2) as the bullets fly through it". Discussing the plan to gull Paroles, Lord G. says in the Folio that they shall thereby see "what mettle this counterfeyt lump of ours will be melted" (3.6.36) and editors since Theobald have emended "ours" to "ore". (If the manuscript spelling were "oure" then the mistake would be particularly easy to make.) Gossett and Wilcox footnote David Scott Kastan's observation that the Folio reading "ours" has the merit of "unhappily acknowledging that Paroles is one of 'theirs', that is, the French".

    At 3.6.104-109, the Folio appears to become confused about just what Lord E. and Lord G. do in the gulling of Paroles and the seduction of Diana. Lord E. (in the Folio called "Cap. E.") says "I must go looke to my twigges", which means that he intends to set the trap for Paroles, leaving Lord G. to accompany Bertram in visiting Diana. But then Lord G. announces that he is leaving (which sounds like he is going off to set the trap) and Lord E. discusses the visit to Diana that he is about to make with Bertram. Later it is Lord E. who knows about the seduction of Diana (4.3.13-17) yet it is also Lord E. who leads the ambush of Paroles in 4.1. Gossett and Wilcox follow Susan Snyder's solution which is to have Bertram switch the lords' intended actions. After Lord E. says "I must go looke to my twigges", Bertram responds "Your brother he shall go along with me", which sounds like Bertram agreeing that Lord E. will set the trap and Lord G. will accompany Bertram in visiting Diana. But Gossett and Wilcox add a stage direction to Bertram's line to produce "BERTRAM [to Lord G.] Your brother, he shall go along with me" so that Bertram is saying "No, Lord E. shall accompany me in visiting Diana" and hence Lord G. goes off to set the trap. This saves having to change the speech prefixes in the remainder of this scene, but at the cost of some dramatic awkwardness. After all, why would Bertram announce this intended switch by telling Lord G. that Lord E. shall accompany him in visiting Diana when he might more easily say this to Lord E. himself, who is standing right there? And the intervention here requires that all the speech prefixes for Lord E. in scene 4.1, the gulling of Paroles, to be changed to speech prefixes for Lord G. There is no easy solution to the problem raised by the apparent confusion of these two characters.

    The Folio has Helen say that the bed-trick plan "Is wicked meaning in a lawfull deede; | And lawfull meaning in a lawfull act" (3.4.45-46). Editors since Theobald have thought that a typically Shakespearian chiasmus is intended and hence emend the last "lawfull" to "wicked" to produce one; the error of mistakenly repeating "lawfull" from the line above could easily be made in transcription or typesetting. Gossett and Wilcox make this emendation. At 4.1.39-42, the Folio has Paroles say, in regretting his boasts: "Tongue, I must put you into a Butter-womans mouth, and buy myselfe another of Baiazeths Mule, if you prattle mee into these perilles". Butter-women (that is, dairy-workers) were stereotypically talkative and Paroles's point is that he needs to speak less in future to keep out of trouble. It is unclear what a Bajazeth's mule might be, and Gossett and Wilcox adopt Hanmer's emendation of "mule" to "mute".

    In one of the most famous cruces in the canon, the Folio has Diana say to Bertram "I see that men make rope's in such a scarre, | That wee'l forsake our selues" (4.2.38-39), which makes no sense. Proposed emendations have ranged from minor (such as "scarre" to "snare") to major (such as Taylor's "I see that men make toys e'en such a surance, | That we'll forsake ourselves"). Gossett and Wilcox adopt the emendation that they attribute to P. A. Daniel of "make" to "may" and "scarre" to "snare" with "rope's" understood as "rope us", to give "I see that men may rope 's in such a snare". An objection to this, made by Taylor, is that the order of events is wrong: getting snared and then abandoning one's principles (forsaking oneself) seems the wrong way round, since the forsaking is the cause and should come first. In a note on this crux (reviewed in YWES for 2014), Karen Britland found in Psalms 140 an appropriate linking of snares and ropes ("cords" in the King James version) and also an image of turning-the-tables on an oppressor that suits what Diana is up to in demanding Bertram's ring, setting him up for the justice of the final scene. But this does not address Taylor's point about the order of events being wrong. (Britland is sceptical that Daniel invented the emendation attributed to him since she cannot find it in his Notes and Conjectural Emendations of Certain Doubtful Passages in Shakespeare's Plays of 1870, which Gossett and Wilcox give as the source. Nor can I.)

    There is in the Folio an exchange between Lafeu and Lavatch that seems to obscurely distinguish different kinds of herbs and pun on the near homophone 'grace'/'grass': "Laf. Twas a good Lady, 'twas a good Lady. Wee may picke a thousand sallets ere wee light on such another hearbe. | Clo. Indeed sir she was the sweete Margerom of the sallet, or rather the hearbe of grace. | Laf. They are not hearbes you knaue, they are nose-hearbes. | Clowne. I am no great Nabuchadnezar sir, I haue not much skill in grace" (4.5.13-20). Nicholas Rowe emended Lafeu's "not hearbes" to "not salad-herbs" and Lavatch's "skill in grace" to "skill in grass", the latter presumably because Nebuchadnezzar famously ate grass (Daniel 4.32-33). This makes explicit the grace/grass pun but without clarifying it. Frank S. Hook in 1951 proposed that Lafeu should actually say "They are not grass, you knave", which makes the whole exchange intelligible. Lavatch likens Helen to a herb of grace (perhaps pronounced to sound like 'grass') and Lafeu picks up the sense of 'grass' to deny it, and then Lavatch extends the pun by saying 'grace' but invoking Nebuchadnezzar who was known for eating grass. Gossett and Wilcox adopt only the second of Rowe's emendations, so that the exchange follows the Folio except that it ends with Lavatch saying "I am no great Nebuchadnezzar, sir; I have not much skill in grass".

    At 5.1.6, the Folio has "Enter a gentle Astringer" and Gossett and Wilcox retain this and simply modernize to Austringer (a keeper of hunting hawks). Other editors have emended to "a gentleman a stranger" (as F3 has), on the grounds that i) this is a new character not one of the previously seen gentlemen, and ii) nothing about his occupation matters for the scene so it would have been gratuitous to dress up an actor and/or give him a prop to convey this occupation. Perhaps Shakespeare planned to invoke the Austringer's occupation in the scene but as he developed it he found he did not need to.

    Gosset and Wilcox's first appendix is on "The Text of All's Well That Ends Well" (pp. 331-358). All's Well that Ends Well was one of the 16 plays entered into the Stationers' Register on 8 November 1623 as "not formerly entered to other men". Gossett and Wilcox describe in accurate detail the setting by formes of All's Well that Ends Well as determined by Charlton Hinman. The only problem arises when they refer to "the left-hand pages of the formes and . . . the right-hand ones" (p. 333), which is ambiguous. They might mean (and I think they do) the left and right halves of one side of the sheet printed from the forme of type (so left = verso and right = recto) but the same phrasing could refer to the left and right sides of the metal block of type that comprised the forme from which that sheet were printed (so left = recto and right = verso), which is a mirror image of the impression it leaves on the paper. Strictly speaking the 'forme' is the type not the paper printed from it. It would have been better if Gossett and Wilcox had avoided the terms 'left' and 'right' and stuck with the unambiguous labels 'recto' and 'verso' regarding to the printed sheet.

    This point of terminology matters when Gossett and Wilcox discuss which half of each forme Compositors C and D set when they started on All's Well that Ends Well. Gossett and Wilcox claim that ". . . with these two formes [V3v:V4r and V3r:V4v], they reversed their roles, with D setting the left-hand page and C the right-hand one" (p. 333). Even if we agree that here "left-hand page" means verso and "right-hand one" means recto", there is a problem in that Gossett and Wilcox do not cite a specific authority for these attributions of forme-halves to Compositors C and D, and there has not been unanimity.

    In his groundbreaking book The Printing and Proof-Reading of the First Folio of Shakespeare, Hinman gives pages V3v and V3r to Compositor A and V4r and V4v to Compositor C (Hinman 2:457). But in 1973 Trevor Howard-Hill ("The Compositors of Shakespeare's First Folio", Studies in Bibliography 26: 61-106, p. 75) reassigned Compositor A's two pages to Compositor D. Gossett and Wilcox appear to accept this reassignment since they do not mention Compositor A having any role in setting All's Well that Ends Well, but neither Hinman's nor Howard-Hill's assignments agree with Gossett and Wilcox claim about a swapping of sides by the two compositors, expressed as ". . . with these two formes [V3v:V4r and V3r:V4v], they reversed their roles, with D setting the left-hand page and C the right-hand one". Rather, Hinman's and Howard-Hill's assignments have one man (A or D) setting V3v and V3r -- one left-hand and one right-hand page, no matter which way round Gossett and Wilcox are thinking of handedness -- and Compositor C setting V4v and V4r (also one left-hand and one right-hand page).

    Investigation of this matter is made unusually difficult by an unfortunate error in the printing of the 1996 second edition of the Norton Facsimile of the First Folio, which contains Peter W. M. Blayney's summary of compositor attribution scholarship up to that date. In the midst of the rows of data for All's Well that Ends Well is one for page "Vv", which is a way of writing signatures that Blayney uses nowhere else in the table. For all pages he otherwise always gives the leaf number 1 through 6 and  uses a "v" for "verso". Since there is no row of page V4r the likeliest explanation is that "Vv" is a misprint for "V4r".

    Gossett and Wilcox go on to detail the setting of the rest of All's Well that Ends Well by Compositor B, paying attention to the order in which formes were set and the two substantial breaks in the work as he contributed to other typesetting tasks. Peculiarly, Gossett and Wilcox do not put to work on the textual problems this detailed knowledge about the order in which the pages of All's Well that Ends Well were typeset. Fredson Bowers showed that some of the mysterious features of the script, such as the variant forms of certain names, make more sense if we consider exactly when each compositor first encountered each character, which (because the pages were not set in reading order) will not be when the reader first encounters the character.

    For most of the twentieth century it was agreed that the manuscript copy for All's Well that Ends Well was authorial foul papers, on account of the printed book having inconsistent speech prefixes, ghosts, literary and imprecise stage directions, and authorial spellings and punctuation. Since the work of Paul Werstine in particular, it has become more common to accept that perhaps a scribal copy intervened between the authorial papers and the printed book because the marks of authorial papers would not necessarily be erased in scribal copying. That whatever the compositors had was annotated for theatrical use seems indicated by the presence of music cues for cornetts and sensible act breaks (both used from 1609). But three chief problems with understanding the nature of the printer's copy have remained: some particularly challenging cruces, uncertainty about characters' names, and unusual stage directions.

    Bowers showed that the variations in certain character names are explicable as the result of the three compositors failing to agree on forms and of Shakespeare altering his intentions as he wrote. The two gentlemen who enter in 3.2 may not be Lord G. and Lord E, the brothers Dumaine, who get their own appendix discussed below. Gossett and Wilcox remark on some odd wording that suggests that when Shakespeare broke off writing he left reminders of what to dramatise next, as in "Parolles and Lafew stay behind, commenting of this wedding", which they then do. Other oddities are best explained as the results of Shakespeare changing his mind during composition. Gossett and Wilcox draw attention to parts of pages where the compositors are noticeably trying to waste or save space on account of small misjudgements in casting off the manuscript copy. They think that the generally successful achievement of evenness across the play suggests that the copy was more likely a scribal transcript than authorial papers. If Middleton did make the additions to the play that the New Oxford Shakespeare proposes, they conclude, there is no sign in the printed book of disruption in the manuscript copy at those points.

    The second appendix is devoted to "The Authorship Debate" (pp. 359-368). Maguire and Smith started the debate with their suggestion in the Times Literary Supplement that Middleton had a hand in All's Well that Ends Well, for which Gossett and Wilcox summarize the evidence, and Brian Vickers and Marcus Dahl provided a sceptical response to it. This exchange is reviewed in YWES for 2012. Gossett and Wilcox then turn to the evidence presented in the New Oxford Shakespeare Authorship Companion of 2017 that Middleton added particular speeches and episodes to All's Well that Ends Well for a revival after Shakespeare's death. This form of co-authorship necessarily results in a multi-layered script in which typically Shakespearian (and unMiddletonian) features sit alongside typically Middletonian (and unShakespearian) ones, as we find in this play.

    In five essays totalling more than 50,000 words in the New Oxford Shakespeare Authorship Companion, Loughnane, Terri Bourus, Farah Karim-Cooper, John V. Nance, and Taylor presented substantial and interlocking evidence for Middleton's adaptation of All's Well that Ends Well, and Gossett and Wilcox summarize it here. The lines that Gossett and Wilcox number as 4.3.242-283 form a coherent whole that could be taken out without damaging the surrounding dialogue and action, and hence might have been dropped in by Middleton. (Being only 375 words, the series of speeches is rather hard to test by quantitative methods.) Other parts of the play that the New Oxford Shakespeare determined to be Middleton's are the King's speech at 2.3.117-144 and Helen's exchange with Paroles about virginity at 1.1.110-163. Gossett and Wilcox find this last claim particularly persuasive since it accords with longstanding critical opinion that the exchange is uncharacteristically vulgar for Helen.

    The last appendix is on "Casting All's Well that Ends Well" (pp. 369-74). Gossett and Wilcox find that there are the "usual four female parts" and "Mariana can be doubled" (p. 369). The two gentlemen who enter with Helen at the Countess's house in 3.2 may or may not be the same Lord E. and Lord G. elsewhere in the play; possibly they were originally a separate pair but got merged into Lord E. and Lord G. to reduce the cast size. Gossett and Wilcox's casting chart assigns all the roles to 14 adults and 4 or 5 boys. The fifth boy is needed if the small part of Mariana is played by an inexperienced boy, but it "might also have been taken by an adult actor" (p. 371). But in fact we know that adult actors did not play female roles.

    There were no monographs on our topic this year. A complete special double-issue (forming issues 1 and 2 of volume 31) of the journal Critical Survey was guest-edited by Terri Bourus on the topic of "Canonizing Q1 Hamlet" and most of its essays are relevant to this review. The first article is "Shakespeare's Early Gothic Hamlet" (pp. 4-25) by Gary Taylor, and begins with Catherine Belsey's remark that an important source for the ghost in Hamlet is the fireside tales Shakespeare would have heard as a child. In Q1 Hamlet, Gertred is unequivocally innocent, whereas in Q2/F she might not be and Hamlet doubts her. Q1 has an additional scene, Scene 14, just for Gertred to actively join with Hamlet and Horatio against her husband; this makes Gertred more sympathetic, more properly motherly.

    There are 271 words in Scene 14, which Taylor reckons makes for "2,419 possible word sequences of two, three or four consecutive words" (p. 9). I make it 270+269+268 = 798 possible sequences. Taylor did his micro-attribution test using the 346 plays in Literature Online (LION) first performed in 1564-1614. He found 33 such n-grams that are unique to Scene 14 of Q1 Hamlet and one other play, and of these 33 other plays 12 are by Shakespeare (two of them having two such unique matches), 3 are by Ben Jonson, 2 are by George Chapman and all other dramatists have no more than one each. Thomas Kyd has none. The Shakespeare links are predominantly to his early plays.

    Next Taylor tested the same 271 words of Scene 14 against the 15,109 books in the Early English Books Online Text Creation Partnership (EEBO-TCP) dataset that were published in 1564-1614, and finds 37 unique matches. 5 are to Shakespeare, 2 are to Robert Greene, and 2 are to Edward Grimeston and all other authors have only one each. Among the 15,109 books in EEBO-TCP for this period, Greene's canon is actually bigger (more works, more words) than Shakespeare's. Taylor then combined the results of the two tests (one on the LION dataset and one on the EEBO-TCP dataset), eliminating results that were not unique across both, and that left 12 n-grams of which 6 occur in the Shakespeare canon (two in Richard 2) and the other 6 n-grams falling one each in 6 other dramatists' canons. Thus Scene 14 of Q1-Hamlet is by Shakespeare. But readers will find it unlike other things Shakespeare wrote in 1599-1602, and the reason is that it is early Shakespeare.

    As Taylor points out, the ending of Titus Andronicus (written 1589) is like the ending of Q1 Hamlet in having several characters die in rapid succession, and the two endings are equally lacking in dramatic sophistication. In Titus Andronicus, Marcus refers to Lavinia's hands beautifully playing the lute (and the piteousness of their loss by amputation), which may be a metatheatrical reference to the same boy actor in the part of Ofelia in the Q1 version of Hamlet, "playing a lute". Hamlet and Titus Andronicus both end with invasions by Goths (since Norwegians and Danes were considered Goths), "led by a son in defence of his father" (p. 16), and both are about "the second marriage of a northern European widow-queen with adult male offspring from her first marriage, and a great father-warrior whose violent achievements haunt a younger generation" (p. 17). The Ghost in Hamlet is essentially Gothic, since its function (unlike say the framing Ghost in Kyd's The Spanish Tragedy) is to frighten other characters. In an appendix, Taylor lists all the hits his searching of LION and EEBO-TCP found.

    In "The Hybrid Hamlet: Player Tested, Shakespeare Approved" (pp. 26-42), Christopher Marino offers his reflections as a director putting on a student production of Q1 Hamlet. The Q1 version of Ofelia's first speech to Hamlet in the Nunnery Scene, about giving back his gifts, sounds like Corambis/Polonius's language, not the heartfelt language she has in Q2/F, so perhaps she is reciting a speech her father taught her and hence is letting Hamlet know that the encounter is a set up.

    The next essay is "Ofelia's Interrution of Ophelia in Hamlet" (pp. 43-57) by Michael M. Wagoner. In Q1, Ofelia interrupts Hamlet in the Nunnery Scene and is his equal, whereas the Ophelia in Q2/F is subordinate. Thus Q1 presents a stronger and more-equal-to-Hamlet heroine. In Q2, Hamlet gives Ophelia her cue "Farewell" in the middle of his speech "If thou dost marry . . . and quickly too. Farewell" as well as at the end of it, causing her to interrupt him and then get spoken over by him; this is really a case of him interrupting her since she is not allowed to finish. In Q1, Hamlet twice gives Ofelia her cue "a nunnery go" before actually using this phrase at the end of his speech "O, thou shouldst not ha' believed me . . . To a nunnery, go". Wagoner points out that Hamlet in Q1 keeps telling Ofelia to "go" (to a nunnery) and she stays put. If Bourus is right about the dating of the script underlying Q1, Hamlet and Ofelia are teenagers in that version and hence are more equal than in the Q2/F version where Burbage is the older actor playing with and dominating an apprentice.

    Douglas Bruster's contribution to the collection is called "Beautified Q1 Hamlet" (pp. 58-71). The mis-setting of prose as verse in Q1 Hamlet might have been an attempt to "confer prestige upon the text" (p. 58), since verse was what Shakespeare was known for. Ann Thompson and Neil Taylor's Arden3 modern-spelling Q1 Hamlet represents a lot of what Q1 sets at verse as prose, but when Bruster tries to illustrate this he does not reproduce the straight right edge that Thompson and Taylor use for prose so the effect is not easily appreciated. Bruster thinks that Thompson and Taylor should not have relined as prose the prose that Q1 mislines as verse. This leads Bruster to consider how Q1 presented itself and the fact that very few of its lines are actually set to look like prose (less than 1%).

    The few lines that are clearly set as prose in Q1 Hamlet tend to fall on the 4r pages of formes and turn-overs and turn-unders are especially common on 4r pages, leading Bruster to speculate that these features cluster there because of inaccurate casting-off of the manuscript copy for the purpose of setting Q1 by formes. In setting-by-formes, page 4r is "typically set last in each forme" (p. 63) and hence is the one where corrections of casting-off errors had to be made. That is, the use of prose and of turned lines was necessitated by errors in casting off. Bruster provides the data for these claims in his Table 1, but the assumptions and data used here are unsound.

    When setting by formes the pages can be set in any order, although it makes sense to do the four pages that complete a forme (inner or outer) before starting on the other four pages of the sheet. But there is no requirement to set the outer forme before the inner (containing the 4r page), as Bruster assumes. Establishing the order of setting of formes and pages requires more detailed analysis than Q1 Hamlet has yet been subject to, including such matters as the locations where distinctly damaged pieces of type reappear across the book, which can be used to show which pages cannot have been standing in type at the same time (since a single piece of type cannot be in two places at once). Furthermore, Bruster's Table 1 shows that the turned lines in Q1 are no more common on 4r pages than on 3r pages (6 a piece) and the 4r pages have the predominance of prose only after Bruster has discounted the 8 lines of prose on page I1r and the 4 lines of prose on page I1v because they "seem exceptional" (p. 62). Without this deduction, 1r pages across the book have as many prose lines as 4r pages do (11 a piece) and 1v pages come in second at 9 lines of prose in all.

    For estimating the length of the Q1 Hamlet manuscript copy, Bruster makes a supposition: "If, like other manuscripts of the time, it featured approximately twenty-five lines of dialogue per page . . ." (p. 64). It is not clear why he thinks 25 lines is any kind of standard: surviving play manuscripts often have two or three times as many lines per page. Bruster uses inappropriate terminology for how writers laid out words on the manuscript page, such as claiming that the dialogue was "justified along an imaginary line indented from the left side with enough room for speech prefixes" (pp. 64-65). The term 'justification' applies only to typesetting, and is not (as it seems to be used here) a synonym for alignment.

    Bruster attempts to deduce the width (measured in letters of the alphabet) of Q1's manuscript copy pages from the use of turn-overs and turn-unders in the printed book: "Judging from the length of Q1 s twenty-six turned lines -- that is, the total length of lines turned both above and below the main line, by means of open parens -- we can estimate that the manuscript copy and penmanship afforded room for the scribe to record approximately sixty-nine alphabetic characters (including spaces) per line of dialogue" (p. 65). The number is, Bruster writes, "derived by calculating the total number of spaces required for the twenty-six turned lines (that is, the 'turned' portions, minus their open parens) added to the lines they complete" (p. 65). Bruster believes that the manuscript had its prose set out like verse but in excessively long lines and the compositor was trying to preserve the lineation of that manuscript. It was Shakespeare's verse-heavy plays that sold best, so perhaps Q1 Hamlet was made to look like it was almost entirely in verse to help it sell. Bruster notes that the opening pages of The True Tragedy of Richard 3 (1594) and The Famous Victories of Henry 5 (1598) also set their prose as verse, which he thinks was an attempt to look sophisticated

    Terri Bourus begins "The Good Enough Quarto: Hamlet as a Material Object" (pp. 72-86) with the observation that it was A. W. Pollard in Shakespeare's Folios and Quartos (1909) who first identified the unnamed printer of Q1 Hamlet as Valentine Simmes, on the basis of its printer's ornament and typeface. The title page reports only that Q1 was "printed for N. L. and Iohn Trundell" but it has Nicholas Ling's device so we know that "N. L." is him. Pollard added Q1 Hamlet to the collection of books made by Simmes and Ling. The paper used for Q1 Hamlet was pretty good and there is little show-through; in that sense it is a good quarto.

    Zachary Lesser and Peter Stallybrass pointed out in an essay reviewed in YWES 2008 that Q1 Hamlet is literary in as much as it has commonplacing marks in it, and scholarly in that its title page says it was performed in the universities of Cambridge and Oxford. And the Q2 Hamlet title page, Lesser went on to claim in a monograph reviewed in YWES for 2015, imitates the Q1 title page and Q1 had not sold out when Q2 was published. Lesser thinks that the fact that Ling's device (his initials around a fish called a ling) is on the Q1 title page and the Q2 title page is explained by Q2 imitating Q1, but, as Bourus points out, Ling put this device on virtually all the books he published. The Q1 and Q2 Hamlet title pages are not quite identical: the alignment of the name of the author differs and in Q1 the first word "THE" is roman and in Q2 it is italic. Also, Q2 has the word "Hamlet" in much bigger type. Bourus notes other minor differences and thinks overall the title page differences are enough to render implausible Lesser's claim that a casual browser might not notice them in the bookshop and might end up buying a Q1 thinking he was getting a Q2.

    Bourus shows that other books published by Ling had successive editions whose title pages are at least as similar as those of Q1 and Q2 Hamlet and concludes that "Such material similarities arise naturally from the routines of book production" (p. 81). It would be hard to mistake Q1 and Q2 in the bookshop, as the latter was twice as long and hence twice as heavy. Bourus thinks there is no reason to suppose that stocks of Q1 remained unsold when Q2 was published. Secondary evidence that readers owned and used Q1 in the seventeenth and eighteenth centuries suggests that Q1 was not reviled as soon as Q2 was available, that Q2 was not thought of as the definitive edition. Q1 sold out within 12-18 months (to judge from Q2 being published) but Q2 did not sell out until 1611 (the publication date of Q3). It was Q1 Hamlet that was the big hit in the bookshops, not Q2.

    Dennis McCarthy's contribution to the collection is "Harvey's 1593 'To Be and Not To Be': The Authorship and Date of the First Quarto of Hamlet" (pp. 87-100). Jonson alludes to Shakespeare's acquisition of a coat of arms and the motto "Not without right" by having Sogliardo in Every Man Out of His Humour purchase a crest with the motto "Not without mustard". McCarthy thinks this alludes specifically to the scene in The Taming of the Shrew where the hungry Katherine is teasingly offered mustard without beef by Grumio. McCarthy also thinks that Sogliardo's "I'll give coats, that's my humour: but I lack a Cullisen" (Every Man Out of His Humour 1.2.146-147) alludes to the Hamlet Q1-only line "My coat wants a cullison!" (9.34), claiming that 'coat' near 'cullison' appears nowhere else in EEBO (meaning the searchable EEBO-TCP). I agree. The speech in Q1 using this line is, according to McCarthy, an allusion to the clown Richard Tarlton, who did have a jest about a cullisance on his sleeve, and since Tarlton died in 1588 the point makes more sense if delivered closer to 1588 than 1603, being a speech about topical extemporizing.

    McCarthy adds to this what seems to me a less persuasive claim of allusion. In Cynthia's Revels, Jonson has a praeludium in which one of the boys says "the Vmbrae, or Ghosts, of some three or foure Playes, departed a dozen yeares since, haue been seene walking on your Stage here" (STC 14773, sig. A4v). McCarthy notes that a dozen years before Cynthia's Revels would be about the date for the late 1580s Hamlet and it has a ghost that walks the stage, and Jonson appears to be alluding to its revival around 1600.

    The allusion to a Hamlet play in Thomas Nashe's preface to Greene's Menaphon in 1589 ("he will afford you whole Hamlets, I should say handfuls of tragical speeches") is well known, but after Nashe apologized for attacking Gabriel Harvey and others in his foreword to his Christ's Tears Over Jerusalem (1593), Harvey replied with his pamphlet A New Letter of Notable Contents (1593) with an attack that likens Nashe to vacillating Hamlet and makes specific allusions to Shakespeare's play. Most of these seem tenuous to me, but "teach a prince to be and not to be religious" (STC 12902, sig. B2v-Br3), in which Harvey's use of italics seems to denote quotation, suggests that Hamlet's famous speech was in existence by 1593.

    But how rare is the phrase "to be or not to be"? In fact, EEBO-TCP has 15 examples up to 1603 and 13 of "to be and not to be", but the collocation of these with "prince" genuinely is rare and EEBO-TCP has no examples. The trouble is, the word "prince" does not collocate with "to be or not to be" in Shakespeare's play. Rather, it is said by a prince, so we have to subjectively weigh this supposed link. In his response to Harvey, in a second edition of Christ's Tears Over Jerusalem, Nashe called Harvey a grave-digger and called his vainglory "his gentlewoman" whom "he hath painted over an inch thick" (STC 18367, sig **r), which echoes Hamlet's plea to Yorick's skull to tell his lady "let her paint an inch thick". McCarthy claims that "paint" occurs within 10 words of "inch thick" only in this work of Harvey's and Q1 Hamlet and then some later works that might be copying these, and I agree. But might Shakespeare have got the phrase from Harvey? Shakespeare around 1600 reaching back to Harvey's book of 1594 seems a lot less likely to McCarthy than Harvey alluding to a popular stage play. Since playing within the City of London was banned from 1594, the Q1 Hamlet title-page reference to a London performance must be to one before 1594. (Paul Menzer showed -- in an article reviewed in YWES for 2006 -- that all other play title pages referring to playing in London are earlier than 1594.) Thus, concludes McCarthy, Bourus is right: Q1 is the early Hamlet play by Shakespeare.

    In his contribution, Saul Frampton finds that the extra 12 lines that the "to be or not to be" speech has in Q2 Hamlet over Q1's version were inspired by Michel de Montaigne's essays published in John Florio's translation of 1603 ("'To Be, or Not To Be' Hamlet Q1, Q2 and Montaigne", pp. 101-112). In particular, Montaigne's "By Diverse Means Men Come Unto a Like End" has the same discussion of activity versus passivity that Q2's opening lines of "To be" has and that Q1's version lacks. Montaigne's essay also has a reference to the Black Prince, which might have given Shakespeare the idea for the "inky cloak . . . solemn black" (1.2.77-78) mentioned in Q2 but not in Q1. Frampton details some minor verbal matches between Montaigne's phrasing and the Q2 lines of "To be" not matched in Q1, including such things as whipping, drowning, outrage, and contumely.

    Frampton sees in the Q1 speech a standard Elizabethan articulation of Stoicism and he finds the Q2 version complicating and questioning this standard model, and he argues that the complication comes from Montaigne's Essays. Moreover the process is self-reflexive: the Q2 version focusses on our complicating second thoughts about every situation and is itself a complicating second thought about the Q1 version. Frampton traces some other more tenuous links between Montaigne's Essays and material in Q2 but not Q1 Hamlet. The speech in Q2 about Gertrude leaving "this faire mountaine" to feed on a "Moore" might be invoking Montaigne's name and acknowledging his contribution to the expansion of the play, since at the start of Florio's translation of Essays this very pun is made in Matthew Gwinne's dedicatory poem: "Montaigne, no cragg'd Mountaine, but faire plaine" (sig. A7r). If this is a connexion, then Shakespeare expanded on Q1 Hamlet using the 1603 printed edition of Essays, not a manuscript (which would be unlikely to have the dedicatory poem).

    In "Shakespeare, Virgil and the First Hamlet" (pp. 113-129), John V. Nance points out that if Martin Wiggins is right that Q1 Hamlet draws on the first (1594) edition of Christopher Marlowe and Nashe's Dido Queen of Carthage then it cannot be a play from the late 1580s. Nance reminds us that Bourus thinks that Q1 reflects the first version Shakespeare wrote, and that then in 1602 he made the version that ended up in the Folio, and that then in 1603-4 he revised it again to make the version represented in Q2. But if Wiggins is right that Q1 depends on Shakespeare reading the printed books of Kyd's The Spanish Tragedy (1592), Nashe and Marlowe's Dido Queen of Carthage (1594), and Jonson's Every Man In His Humour  (1601) then Bourus cannot be right. The big obstacle is the idea that Shakespeare's Troy speech (= Aeneas's Tale to Dido) depends on the play Dido Queen of Carthage. Eighteenth-century readers did not see Shakespeare being especially indebted to Dido Queen of Carthage and focussed rather more on the problem that Hamlet says that the play he is thinking of was naturalistic but then the extract we hear by the Player is bombastic.

    The connection between Hamlet and Dido Queen of Carthage arose when Alexander Dyce in his 1850 edition of Marlowe accepted John Payne Collier's emendation of the second line of Dido Queen of Carthage's "Which he disdaining whiskt his sword about, | And with the wound thereof the King fell downe" (STC  17441 sig. C1v) to "And with the wind thereof the King fell downe", which Collier did precisely in order to make it match "but with the whiffe and winde | Of his fell sword, th'unnerued father falles" (Q1 Hamlet E4r). Thus the link between Hamlet and Dido Queen of Carthage was made by Collier's emendation, which has been adopted by virtually every editor since. Nance makes a literary-critical comparison of the killing of Priam by Pyrrhus in Dido Queen of Carthage and Hamlet and finds the latter more sophisticated even in the Q1 version, including Pyrrhus's transformation from black (melancholic) to red (choleric). The only possible debt to Dido Queen of Carthage in Hamlet is the frozen pause of Pyrrhus before giving Priam his death blow, and this happens only in Q2 not Q1 Hamlet. Other Q2-only material -- the lines about Cyclops's hammer forging eternally strong armour -- seems to come directly from Virgil's The Aeneid.

    Conversely, Laurie Johnson focusses on lines that are in Q1 Hamlet but not Q2 or the Folio ("Unique Lines and the Ambient Heart of Q1 Hamlet", pp. 130-143), which pose a problem for the theory that Q1 was made by note-taking in the theatre. Johnson finds recurrence in Q1's unique lines of 'heart' imagery associated with Corambis. Johnson must of course first clarify just what he means by lines being substantially different or identical, and he goes for a definition in which no possible likeness of phrasing or sentiment or intention can be found. Johnson counts 272 such unique lines in Q1 (12.25% of the play). The first 7 of these unique lines have 4 occurrences of the word 'heart' and the unusually high prevalence of this word continues across the unique lines, and disproportionately these heart lines refer to Corambis or his children. The connexion is that 'Corambis' means two-hearted. William Cecil Lord Burghley had the motto "cor unum via un" (Latin for "one heart, one way") and maybe the Burghley family saw the play in its Q1 version with the character Corambis and all this heart talk and took it to be a pointed allusion to him and forced Shakespeare to change the name and cut the heart talk, resulting in the name Polonius and the loss of the heart talk in Q2.

    Johnson reckons that it is impossible that in a memorial reconstruction of the version underlying Q2 Hamlet the reconstructor patched in all the heart lines and changed the name Polonius to Corambis. (I wonder, is that especially unlikely? Once you have chosen the new name the idea of writing heart lines for him is not impossible.) Maybe one line in Q2/F actually alludes to the excised heart material that these editions lack: "I will wear him | In my heart's core, ay, in my heart of heart". This occurs in Hamlet's speech to Horatio "about the utility of a topical play to trigger the guilt of the powerful" (p. 141). This line is not in Q1. Why would someone remove it from the play while putting in the heart lines and the name Corambis, as the Q1-derives-from-Q2 theory would have it? Johnson reckons that this line was added in place of the heart-lines-and-Corambis material that was taken out of the Q1 version to make the Q2 version. In this speech, 16 lines were added to Q1 to make Q2 and perhaps this is alluded to in Hamlet's question to the players about maybe inserting "some dozen or sixteen lines" (present in Q1, Q2, and F). That is, the question about inserting 16 lines was present in the Q1 version but then Shakespeare actually did add 16 lines to revise the Q1 version to make the Q2 version.

    Tommaso Continisio's "'Brief Let Me Be': Telescoped Action and Characters in Q1 and Q2 Hamlet" is a literary-critical account of the Q1/Q2 differences, especially regarding Ophelia and Gertrude, and it will not be further noticed here. Charles Adams Kelly and Dayna Leigh Plehn's "Q1 Hamlet: The Sequence of Creation and Implications for the 'Allowed Booke'" (pp. 153-167) starts by rehearsing the standard arguments against Q1 being a memorial reconstruction. Kelly and Plehn think that Hamlet's line in Fratricide Punished "Ay ay, King, just send me off to Portugal, that I may never come back again, that is the best plan", after the King has said that he is sending Hamlet to England, alludes to "a disastrous military expedition to Portugal in 1589" (p. 156), so the German play perhaps comes from a 1589 version of the English Hamlet when this topical allusion would make sense.

    G. I. Duthie found in Fratricide Punished 21 plot elements that are also found in Q1 Hamlet but not Q2 and 36 that are found in Q2 but not in Q1, and since 21:36 is also approximately "the ratio of the total line count of Q1 vs. Q2" (p. 158), Kelly and Plehn think that this shows that Fratricide Punished "either derived from both Q1 or Q2 or that it predates both" (p. 158) but it could not derive from one and not the other nor could it predate one and not the other. There are plot elements in Fratricide Punished that are not in Q1 or Q2, and Kelly and Plehn think that this makes it impossible that Fratricide Punished is merely an abridgement of Shakespeare's play (why add them?), so it must be a source for Shakespeare's play. And since Fratricide Punished has a character called Corambus this leads to the line of descent being Fratricide Punished to Q1 Hamlet to Q2 Hamlet.

    Kelly and Plehn then turn to a statistical analysis of the fact that there are 209 lines that appear only in Q2 Hamlet (not in the Folio) that are absent from Q1, asking themselves if it is likely that someone remembering the script underlying Q2 would manage to forget all of these 209 lines. (As we saw above in relation to Laurie Johnson's essay, deciding just what it means for lines to be the same or different in different editions is not a trivial problem: we need some agreed measure of distinctiveness.) Calculating "The remoteness of the probability that none of the 209 Q2 lines would be remembered (if Q1 is a reported text)", Kelly and Plehn make a comparison with coin-tossing that is manifestly inapt.

    Kelly and Plehn write that "If you toss a coin two hundred times . . . .  It is statistically certain that you would not get zero heads in two hundred tosses" (p. 159, emphasis in original). They mean that it is extremely unlikely that you would get zero heads in such an experiment. It is trivial to quantify the probability: it is 0.5 to the power of 200, which is about one chance in 16 with 59 noughts after it. This is small, but not zero. Kelly and Plehn neglect to emphasize that they are referring to the lost manuscript versions of the play underlying Q1 and Q2, not the printed editions themselves, leading them to the absurd assertion that "It is a statistical certainty that Q1 is not derived from Q2 as printed" (p. 159, emphasis in original). Indeed, but more simply this is also an historical certainty, since Q2 had not been printed when Q1 was made. 

    The analogy with coin throwing is more fundamentally open to the objection that the omission in the Q1 version of lines from the Q2 version might have been intentional -- to abridge the script -- rather than mere forgetfulness. Kelly and Plehn counter this by asserting that unless we can find a reason, an "identifiable rationale" (p. 159) for such an intentional cutting, we should discount the possibility. By a similar series of steps whose logic this reviewer cannot follow, Kelly and Plehn conclude that Q1 cannot be based on a version that was intermediate between the Q2 version (assumed to be based on the author's unperformed first draft) and the Folio (assumed to be based on a script that had been rehearsed and theatricalized). But if Q1 is based on a version that preceded the versions in Q2 and Folio, how come many of Marcellus's lines in Q1 are similar to his lines in Q2, which is not true for other characters? Kelly and Plehn's hypothesis is that after producing the draft that underlines Q1, Shakespeare went back and applied second thoughts to the play, making it like Q2, but only in the first half.

    To explain why in one part of Q1 Hamlet there is a run of lines that end in full stops where commas are what the grammar needs, Kelly and Plehn give an inaccurate account of typesetting. They imagine that "Each letter or symbol was placed in a groove in the composing stick" (p. 162), but there was no such groove, and that "As each line of type was completed, it was placed with other lines of type, in an iron frame (or chase) corresponding to a press sheet, and a new line of type was composed" (p. 162). On the contrary, several lines were kept in the stick at once (each new line resting on the previous one), which was the point of having the stick, and the chase was placed around the type only once all the pages for one forme had been set. Because of this imaginary process of moving each line to the chase on its own, Kelly and Plehn suppose that the single line of type was the "unit of process" (p. 162) and hence if the manuscript was difficult to read -- as it might be if Q1 was set from authorial papers -- then the compositor might decide to punctuate the line without even trying to decipher what the next line reads. This supposition is far from the reality of typesetting in Shakespeare's time

    Kelly and Plehn explain one apparent aural error in Q1 ('honor' where 'owner' needed) as likely a mishearing during typesetting by oral dictation. This is not impossible, but to invoke such an hypothesis to account for just one reading seems indulgent. At this point Kelly and Plehn offer a definition of their threshold for deciding when two versions of a line are identical: when the difference is more than "a compositor might cause" in typesetting (p. 163). They offer counts and proportions for lines identical in Q1 and Q2 (49% of Q1), lines unique to Q1 (24%  of Q1), and lines "identifiably concordant but altered more than might be caused by the personal choices or careless work of a compositor" (27% of Q1) (p. 163). Kelly and Plehn end by describing how their students are going to use Q1 in creating a text for performance.

    The final essay in this special issue of Critical Survey is "What Doesn't Happen in Hamlet" (pp. 168-186) by Rory Loughnane. It responds to John Dover Wilson's classic account of the problem of the dumb show: if it depicts the murder he committed, why does Claudius not react to it but then go on to react badly to The Murder of Gonzago that follows it? Unlike the play it precedes, which Hamlet relates at length, telling the onstage audience things they could not possibly otherwise know (the characters, their relationships and motivations), the dumb show is unexplained and inexplicable.

    Depending on how you count them, there are about 150-186 dumb shows across 73-98 plays in the period. Loughnane surveys some references to dumb shows and uncovers an association with incompleteness, misdirection, and hypocrisy. Quite often an interpreter has to tell the theatre audience (and the play's reader) what the dumb show actually depicted or meant. In Hamlet, the dumb show tells the onstage audience nothing but since we the wider audience know the Ghost's account of the murder we can make sense of it.

    Loughnane prints and compares the dumb shows as presented in Q1, Q2, and Folio Hamlet. Q1's dumb show gives no explanation for the murder, which is conveyed in Q2 and the Folio by the poisoner kissing the crown. In Q1 the Queen goes away with the murderer without first showing a reluctance that he overcomes with gifts; thus Q2 and the Folio (with the reluctance and gifts) make plain that she was not in on the murder. In Q1 the King's body is present when the Queen goes off with the murderer, whereas in Q2 and the Folio the body is first removed. Thus Q2 and the Folio absolve the Queen. Loughnane interprets Hamlet's remark that the players will "tell all" (present in all three versions) as an expression of annoyance that the dumb show has given the game away. If the dumb show is so awkward, why did Shakespeare include it?

    Loughnane's answer is that Shakespeare knew that he was going to have The Murder of Gonzago be interrupted by Claudius rising, so the preceding dumb show is the only way he has of telling us, the wider audience, just how much more pertinent matter The Murder of Gonzago would have presented to the onstage audience if its performance had been allowed to continue. And that more is about Gertrude's behaviour after her husband's death, so what matters is whether Gertrude sees the dumb show. In having much less information overall, Q1's version of the dumb show gives us less of a sense of what Hamlet has in mind for shaming his mother.

    In an "Editorial Afterword" (pp. 187-195) to this special issue of Critical Survey, Graham Holderness and Bryan Loughrey revisit their early 1990s Shakespearean Originals book series and its Q1 Hamlet edition. They sketch Brian Vickers's negative review in the Times Literary Supplement and the subsequent exchange of letters which showed that many scholars did not agree with Vickers about the memorial reconstruction theory. Then they consider Janette Dillon's critique of the series, finding logical inconsistency in her objections.

    A large number of articles relevant to our topic were published this year, and most were on matters of authorship-attribution methodology and its application to a small number of disputed cases. These will be taken together after we first consider the articles on other matters, including editorial theory and practice. The most important article this year was a persuasive demonstration that the 1594 quarto of The Contention of York and Lancaster and the 1595 octavo of Richard Duke of York cannot have been made by memorial reconstruction of the plays underlying the editions of 2 Henry 6 and 3 Henry 6 in the 1623 Shakespeare Folio (Heejin Kim 'The Memorial Reconstruction Theory and Chronicles: The Henry VI Plays', Shakespeare 15[2019] 356-78).

    Edmond Malone made the point that if the scripts underlying The Contention of York and Lancaster and Richard Duke of York were made by memorial reconstruction  of the scripts underlying Folio 2 Henry 6 and Folio 3 Henry 6 then it is hard to explain why the 1590s editions contain whole speeches not in the Folio. It does not seem likely that someone trying to reconstruct something would just make up whole sections. (Or does it? I can imagine memorial reconstructors resorting to that expedient when their recollections failed them.) Peter Alexander in 1928 and Madeleine Doran in 1929 popularized the theory of The Contention of York and Lancaster and Richard Duke of York being bad quartos made by memorial reconstruction. Their key evidence was these versions' distortions of historical facts.

    Kim summarizes the turn away from the memorial reconstruction theory since the 1980s, and how editors have nonetheless persisted with some form of it, such as memorial reconstruction plus revision explaining the Quarto/Folio differences. Attribution scholarship has tended to work on the Folio Henry 6 plays. Kim's position is that The Contention of York and Lancaster and Richard Duke of York are not derived from the plays underlying Folio 2 Henry 6 and Folio 3 Henry 6 but from "older plays written prior to a process of revision" (p. 358). Alexander influentially argued that an author would consult the chronicles and get the historical details right, while actors just use their memories of lines they learnt. Kim sets to show that material unique to The Contention of York and Lancaster and Richard Duke of York is historically accurate. Moreover, the 1590s plays rely on  Edward Hall and John Stow's chronicles more than on Holinshed's, and the Folio plays do the opposite; this is evidence against textual dependence. 

    Before the Battle of Wakefield, Richard Duke of York and 3 Henry 6 number the Yorkists at 5,000, but Richard Duke of York numbers the Lancastrians at 30,000 (and names Westmoreland and Northumberland as leaders) while 3 Henry 6 gives 20,000 (and omits Westmoreland and Northumberland). This is consistent with Richard Duke of York relying on Stows's Annals (for the 30,000 men and the presence of Westmoreland and Northumberland) and 3 Henry 6 relying on Hall and Holinshed (for the 20,000 and no mention of Westmoreland and Northumberland). Kim traces other verbal details in which The Contention of York and Lancaster and Richard Duke of York seem dependent on Hall and 2 Henry 6 and 3 Henry 6 seem revised away from this source, perhaps by consultation of Holinshed. When Richard wants to incite Warwick to action by recalling his fallen kin, Richard Duke of York has him describe the death of Warwick's father while 3 Henry 6 has him describe the death of Warwick's brother. Linking Warwick's thirst for revenge with his father's death occurs only in Hall, however.

    When Somerset is killed under the sign for an ale-house called The Castle, The Contention of York and Lancaster follows closely the wording in Hall ("vnder the signe of the Castle", "before was warned", and "Castles") and 2 Henry 6 does not, which again shows an authorial hand not an actor's corruption. Likewise, Jack Cade's killing of Lord Saye and Sir James Cromer in The Contention of York and Lancaster closely follows the wording of Hall, and refers to the London location Mile End (found only in Hall), while 2 Henry 6 is more distant from this source. Thus it is impossible that the quarto is based on merely a poor recollection of the Folio version. Also, in revision things tend to drift away from sources not towards them, making revision of the 1590s version to make the Folio version more likely than revision of the Folio version to make the 1590s version.

    Roger Warren claimed in "'The Quarto and Folio Texts of 2 Henry VI" (reviewed in YWES for 2000) that The Contention of York and Lancaster is based on a report of a lost version of the play that is different from the Folio version. Kim considers particular pieces of textual evidence that Warren used and finds them insufficient to settle the case. A bit of internal verbal repetition is, as Laurie Maguire showed, not something exclusive to memorially reconstructed texts: we find it in good texts too. And the phrase "what wouldst thou haue me do" being in The Contention of York and Lancaster and Marlowe's Doctor Faustus is not necessary evidence of a poor memory being patched with matter from another play but might simply be the use of a common expression: Kim finds in EEBO-TCP many identical or similar phrases. And in any case, maybe Marlowe was the borrower. Kim points out that "Hieronimo beware, go by, go by" from Kyd's The Spanish Tragedy was echoed in nine later plays. Or it may be Marlowe just using in The Contention of York and Lancaster a phrase he used in Doctor Faustus. Likewise the mention of "the wild O'Neill" in The Contention of York and Lancaster and Marlowe's Edward 2 but not Folio 2 Henry 6 could be Marlowe repeating himself, or Marlowe in Edward 2 could be echoing the performed script underlying The Contention of York and Lancaster.

    Randall Martin in "The True Tragedy of Richard Duke of York and 3 Henry VI: Report and Revision" (reviewed in YWES for 2002) argued that the historical name Lord Bonville (as in Folio 3 Henry 6) becoming the unhistorical Lord Bonfield in Richard Duke of York is another sign of memorial reconstruction. Kim points out that Martin Wiggins observed that the historical Somerville is called Summerfield in Richard Duke of York. These just seem to be alternative forms of the name, and Bonville/Bonfield might be too.

    Martin thought that Richard Duke of York calling the historical Edward Brooke "Edmund Brooke" (where 3 Henry 6 just calls him Brooke) is not extra historical information (which would be evidence of conscious revision of the play) but random error. Kim points out that Stow's Annals actually does give his name as "Edmond Brooke", so this really was believed to be extra historical information and hence its appearance in Richard Duke of York is evidence of revision. Martin also misused Maguire's work on internal repetition as evidence of memorial reconstruction: she looked at the same evidence and concluded that memorial reconstruction was not shown by it. Likewise with anticipated phrasing as evidence of memorial reconstruction.

    In Folio 3 Henry 6 the Queen advises Henry 6 to fly a lost battle by referring to Edward and Richard and how "vengeance comes along with them". But Richard Duke of York lacks the reference to Edward and Richard and says, apparently meaninglessly, that "vengance comes along with him" with no obvious antecedent for "him". Martin gave this as an instance of memorial reconstruction garbling the sense, but Kim argues "him" is perhaps just an error for "them" and that this refers to the general threat to the king on a stage busy with violent action, which imperils the king. Also adduced as evidence for memorial reconstruction is the presence of aural errors, but Kim counters that these occur in good texts too and the faulty memory of a compositor as he temporarily keeps phrases in his head can also cause them.

    The strongest evidence for memorial reconstruction, Kim admits, is York's bungled genealogy. But maybe, Kim argues, just a few bits of the manuscript holding the version used in the 1590s were corrupt and had to be patched from memory. Wrecked verse is, according to Kim, a reasonable sign of textual corruption, and he looks at examples that are convincing as examples of corruption but not easily explained as memorial reconstruction. From the ragged meter around York's bungled genealogy in The Contention of York and Lancaster, Kim speculates that ". . . untidy, damaged, illegible, or lost parts of the manuscript might have introduced corruptions and disruptions" (p. 374) and shows that memorial reconstruction is less plausible since around the same place Warwick also has high-quality but unique-to-Q lines.

    In another important article this year, John Jowett argues that Thomas Heywood wrote part of Q1 Hamlet ('Whose Hamlet Mocks the Warm Clown?', PBSA 113[2019] 341-70). His central claim about Hamlet is that ". . . one block of text that is unique to Q1 belongs to 1602 rather than the 1580s, and . . . it was not written by Shakespeare" (p. 342). Two substantial passages in Q1 are unique to that edition: one is Scene 14 between the Queen and Horatio and the other is the 11-line, 113-word speech in Scene 9 from "And then you haue some agen" to "Maisters tell of it" (Q1 F2r-F2v), which Jowett calls the Warm Clown speech. Around late 1602 to early 1603 Heywood was writing the play The Blind Eats Many a Fly, which title is similar to the phrase "as the blind man catcheth a hare" in Q1 Hamlet, and Heywood wrote similar dialogue for and about clowns elsewhere.

    Jowett describes how he went looking in EEBO-TCP for words and word n-grams from the Q1 Hamlet Warm Clown speech that appear in the works of Heywood but not Shakespeare and those that appear in the works of Shakespeare but not Heywood. Heywood's overall canon is bigger than Shakespeare's, but that is on account of a lot of non-dramatic prose works. Jowett used all of Shakespeare's sole-authored plays and Heywood's plays plus enough of Heywood's non-dramatic works to make the two corpora of roughly equal size. Jowett searched for phrases used in the Q1 Hamlet speech and either Shakespeare or Heywood's corpus. Only exact phrasing matches were allowed, but ProQuest's EEBO-TCP engine was used with its spelling-variants feature switched on. Where a smaller n-gram match occurred within a larger one, Jowett separated out the two kinds in his lists.

    Jowett describes two validation runs: one using a piece of undisputed Shakespeare writing (the 113 words of Scene 3.2 as in Q2 Hamlet, "Speak the speech, I pray you, as I pronounced it") and one using a piece of undisputed Heywood writing (113 words from the clown in the first edition of The Golden Age). Tabulating the results from the first validation shows that twice as many (24 against 12) of the n-grams found in "Speak the speech" also turn up in the Shakespeare canon as in the Heywood canon, and that of the n-grams from the passage from The Golden Age 19 turn up in Shakespeare's canon and 29 in Heywood's. If instead of counting token-wise as here we count type-wise (so that the multiple occurrences of the same phrase count only once) then the figures are 12 Shakespeare and 7 Heywood for "Speak the speech" and 11 Shakespeare and 23 Heywood for the passage from The Golden Age. Jowett considers that his validation runs "yield a significant discrimination" (p. 351).

    When the Warm Clown speech from Q1 Hamlet is tested the same way, it comes out as Heywoodian not Shakespearian: 6:12 in favour of Heywood by types and 8:14 in favour of Heywood by tokens. Jowett discusses various ways of counting the data (such as including or excluding sub-n-grams that appear within larger n-grams) and the results are essentially the same: the passage tests more like Heywood than Shakespeare. Next Jowett considers some looser verbal connexions, and sensibly he does not count them since an element of subjectivity enters the evidence (that is, how loose is loose allowed to be?) They point the same way: against Shakespeare's authorship of the Warm Clown speech.

    What about the word "cinkapase" (= cinquepace) in the Warm Clown passage, which occurs thrice in Shakespeare plays (twice in Much Ado About Nothing and once in Twelfth Night) and nowhere in Heywood's corpus? Jowett explores the connotations of this term (generally negative) and notes that outside of Q1 Hamlet the Shakespearian uses cluster in 1598 (Much Ado About Nothing) to 1600 (Hamlet), with none at all in his many plays from earlier in the 1590s. In Microcynicon (published in 1599), Middleton uses the word "pace" for a dance in the context of writing down what is being performed, and that is exactly the context in the Warm Clown speech too. Jowett suggests that Q1 Hamlet got the idea and the word from Microcynicon rather than vice versa (p. 360).

    Another distinctive phrase in the Warm Clown speech is "as the blind man", but it is used there as part of a known proverb ("a blind man may catch [or kill] a hare"), so is not a marker of authorship. Jowett tracks some other unusual words and ideas in the Warm Clown speech that seem more likely to have been written after 1600. The first is the idea of writing down in tables what a clown says, which occurs in Samuel Rowlands's Greene's Ghost-Hunting Cony-Catchers (published 1602), and not before, and the reference to a coat lacking a cullison, which seems to come from Jonson's Every Man Out of His Humour (performed 1599, published 1600).

    Every Man Out of His Humour also contains a reference to eating porridge that seems to match the one in the Warm Clown speech. The Warm Clown speech's idea of a clown's single suit of jests being like a man's single suit of apparel seems to match the same idea in Thomas Dekker's Satiromastix (first performed 1601). Thus "The author compiling the Q1 clown lines is demonstrably picking up lines familiar from texts of 1596-1602, specifically, plays of 1598-1602, and, most especially of all, passages in them that were theatrically topical at the time of the 'Poetomachia' and the rise of the comedy of humours as reflected in the clown's own surliness" (p. 365).

    Jowett's conclusion is that the Warm Clown speech is more like Heywood than Shakespeare and more like writing of the early 1600s than writing of the late 1580s. The fact that the Warm Clown speech itself refers to performers' lines being written down is "a particular and unnervingly close echo of the textual technology most plausibly used for the compilation of the Q1 text" (p. 367), that is the recording of a performance by note-takers, as suggested by Tiffany Stern (in an article reviewed in YWES for 2013). To Jowett, this suggests that the self-referential Warm Clown speech was "written after the rest of the play had undergone disruptive transmission to reach its form as in other respects given in Q1" (p. 370).

    Two articles on the practice of Shakespearian editing appeared this year. In a study of the language used by scholars of Shakespearian textual matters, Laurie Maguire and Emma Smith note that the present reviewer began his account of the history of Shakespearian authorship studies in the Authorship Companion to the New Oxford Shakespeare with the story of a miscarriage of justice regarding the murder of Dorothy Woods in 1996, and near the end he referred to the miscarriage of justice suffered by Sally Clark, wrongly convicted in 1999 of murdering her two children ('On Editing', Shakespeare 15[2019] 293-309). The point of these two topical anecdotes was that stylometry can resemble forensic pseudo-science. For Maguire and Smith this focus on women and crime is troubling, since ". . . there is a sense that the underlying premise of editorial investigation is cherchez la femme" and in these stories women seem to be "passive proxies for the text itself" (p. 296).

    It takes special framing by Maguire and Smith to give this impression, since they have to omit the males in the stories. In the Dorothy Woods case that I described the supposed perpetrator (and victim of the miscarriage of justice) was male and the deceased was female, while in the Sally Clark case the supposed perpetrator (and the victim of the miscarriage of justice) was female and the deceased were male. The stories could not be more evenly balanced in respect of gender. Nowhere in the essay -- only in Maguire and Smith's minds -- were the "women figured as damaged text" (p. 296). The ungendered claim being advanced was that scientific evidence needs special handling if we are not to be misled by it.

    Maguire and Smith consider the gender balance of leading Shakespeare editions, noting that the 1986-87 Oxford Complete Works team was all male and that "The single-volume plays published by Oxford University Press in the World s Classics series admit only two women editors across the entire canon" (p. 296). The word 'admit' here is loaded, since it suggests that suitably qualified women applied to edit plays for the series and were excluded. That is not how book-series commissions are awarded (as Smith and Maguire must know): the invitations are in the gift of series editors and publishers. These invitations may of course be shaped by sex- and race-discrimination and by other biases such as personal rivalries and jealousy. But Maguire and Smith's analysis is logically invalid in looking at the outcome of a selection process and diagnosing bias solely because the demographic make-up of the resulting pool of people does not match the demographic balance of the wider human population.

    To see why this is illogical, consider the fact that a quarter of the winners of Nobel Prizes have been Jews. By Maguire and Smith's logic we would work backwards from this outcome to conclude that the over-representation of this ethnic group can be explained only by the committee in Stockholm being biased in favour of Jews. Sadly, this fallacious logic is in evidence everywhere these days. It lies behind a number of political trends within academia, including the RaceB4Race collective, which diagnoses systemic bias as the cause of the over-representation of white people in premodern literary, historical, and cultural studies. This makes no more sense than concluding from the fact that seven of the world's top ten computer scientists are Asians that non-Asians are being discriminated against in this field. Maguire and Smith pursue this fallacy to its illogical conclusion regarding the racial as well as sexual identities of various Shakespeare editors, apparently without ever noticing the irrationality of what they are claiming.

    The essay returns to reason when Maguire and Smith consider the increasing attention paid to matters of performance in Shakespeare editions, and trace the increasing willingness of editors -- sometimes going too far, they imply -- to tell readers exactly what they think is occurring on the stage. What happens, though, when a play is edited by one scholar, as with E. A. J. Honigmann's 1999 Arden3 Othello, but is then reissued with a new introduction (fronting the same play text) written by a different scholar, in this case Ayanna Thompson, to form the 2016 Arden3b edition, as Maguire and Smith call it? The result, they diagnose, is internal contradiction.

    Honigmann departed from his Folio copy text to reassign the line "This Lodovico is a proper man" from Desdemona to Emilia, and Thompson had to retain this reassignment since her task was only to write a new introduction, not to re-edit the text. But she was free nonetheless to explain that she thought the reassignment was mistaken. Maguire and Smith write that Thompson's "new introduction here reveals its limitations: to edit a text is to undertake a series of acts of critical interpretation which are always limited by the perspective of the editor" (p. 299). For Smith, at least, this has since become not merely a hypothetical problem. She has been appointed as General Editor of the Oxford World Classics series of Shakespeare works based on the New Oxford Shakespeare editions. For the most part, the editors who created the New Oxford Shakespeare editions are not being asked to write the introductions for these new editions and it will be interesting to see if further cases of conflict between text and introduction emerge from this process.

    Next Maguire and Smith turn to authorship attribution, and their particular contribution regarding Middleton's co-authorship of All's Well that Ends Well (reviewed in YWES for 2012). They observe that the literary-critical part of their argument was either not engaged with by other scholars or was dismissed using what Maguire and Smith consider gendered language: they were said to be "unsympathetic towards Shakespeare s heroines" and "indelicate" (p. 301). Maguire and Smith believe that although quantitative studies are essential, "literary criticism has a role to play in attribution studies", and they find it striking that "before anyone had ever thought of associating Thomas Middleton with Timon of Athens" (p. 301), John Bayley made the case in his Shakespeare and Tragedy of 1981.

    This chronology is incorrect. William Wells made the case for Middleton co-writing Timon of Athens in Notes and Queries in 1920 (112: 226-229) and Dugdale Sykes supported it in his book Sidelights on Elizebethan Drama in 1924. In 1975, Cambridge University Press published David J. Lake's The Canon of Thomas Middleton's Plays: Internal Evidence for the Major Problems of Authorship, which provided new evidence for the claim from colloquialisms and contractions. MacDonald P. Jackson devoted a chapter of his book Studies in Attribution: Middleton and Shakespeare (1979) to providing further support for the claim. That Bayley was comprehensively preempted does not invalidate Maguire and Smith's point, though: they are right that literary criticism has a role to play in attribution. The present reviewer would caution that it should not be a leading role.

    Maguire and Smith return to the matter of the credentials for editing and the degree to which current political fashions should shape the allocation of tasks. "Can a non-Jew edit Merchant of Venice?", they ask, and "Can a man edit The Taming of the Shrew?". Their answer is "Probably, but not at this juncture . . ." (p. 302). By couching this as a matter of prevailing trends rather than their own opinions they evade the important question of whether they believe that men and non-Jews ought to be allowed to edit these plays now. The fundamental moral principle at stake is whether a person should be prevented from doing something solely on the grounds of that person's race or sex. The ethical inadequacy of Maguire and Smith's non-commital answer is apparent if we imagine it being said by an apolitical German scholar trying to survive the vicissitudes of the Nazi encroachments on academic freedom in the 1930s. Can a Jew edit The Merchant of Venice? "Probably, but not at this juncture . . .".

    The final section of Maguire and Smith's essay compares Harold Jenkins's single-volume 1982 Arden2 Hamlet to Ann Thompson and Neil Taylor's two-volume 2006 Arden3 edition. Maguire and Smith notice that the play's concerns with legitimacy and authority are echoed in the scholarly debates about the three early editions of Q1 (1603), Q2 (1604-5), and Folio (1623). In modern editions these anxieties take the form of Oedipal impulses towards predecessor editions. The unacknowledged progenitor of this highly effective way of thinking about creativity is, of course, Harold Bloom, and although they use his expression "The anxiety of influence" (p. 303) Maguire and Smith never mention Bloom or his book of that title. They are able to diagnose that the Arden3 Hamlet "tries hard not to be haunted by its forbears" and sublimates its "patricidal impulses", and that the edition necessarily "betrays something more anxious about the legacy of the past" than it is prepared to admit (p. 303). But Maguire and Smith are apparently unable to see the same psychical processes at work in their own silence about Bloom.

    Maguire and Smith rightly point out that Taylor and Thompson presenting all three Hamlets in their Arden3 edition does not really solve the difficult editorial problems. Taylor and Thompson's Q2-based main volume is pitched by its publisher as good enough for most purposes but actually conceals most of the interesting variants from readers on the principle that they can get these from the second volume containing the Q1- and Folio-based editions. Maguire and Smith end this section of their essay with a disturbing story of British A-Level students being asked in an examination to comment on Hamlet's speech "How all occasions do inform against me", which they had never encountered before because they studied the play from the Royal Shakespeare Company's Folio-based Complete Works of 2007, edited by Jonathan Bate and Eric Rasmussen, which omits it. In fact Bate and Rasmussen do include the speech in an appendix at the end of the play, just as the 1986-87 Oxford Complete Works did. It was particularly remiss of the students' teachers to use this edition without familiarizing themselves with what it contains and where.

    The other article on editorial matter this year concerns itself with the possibilities arising from new technologies (Sarah Neville 'Rethinking Scholarly Commentary in the Age of Google', Textual Cultures 12.i[2019] 1-26). Neville argues that we have new opportunities to flatten the knowledge hierarchy that is expressed and substantiated by scholarly editions. The fact that digital editions now often include facsimiles gives the reader a chance to see the original context for an editorial emendation, so the mechanics of editing are more on show than ever before. But this display actually increases the authority of the editor: the "reader of a digital scholarly edition can see at a glance that an editor s job does not end with the reproduction of primary documents to make them available to readers" (p. 8). That is, readers can see how much more than the emendation of error has been done in the editing.

    Neville thinks that digital editions perpetuate "the very same forms of citation and scholarly credit that traditionally accompany printed scholarly editions" (p. 11) and that these are supporting the "old forms of scholarly authority . . . in new ways" (p. 11). She objects too to the "traditional list of works cited" (p. 12) as a "traditional top-down hierarchy" that needs to be turned into "a lateral and contingent arrangement that makes room for readers participation in the production of knowledge" (p. 12). Neville objects to the "ideal of an author s genius" that editing promotes when we should in fact be pointing out how much the other agents (scribes, printers) contributed to the work. She wants digital editions to "highlight the ways that scholars use others' research to construct the credibility necessary to speak authoritatively about a text and its authors" (p. 13).

    Neville argues that scholarly editions do not make it plain enough that the editor is indebted to Scholar X for this bit of knowledge and Scholar Y for that bit. I am not quite seeing where she thinks our current referencing systems fail in making these debts plain. She also thinks that readers' responses to what they read might usefully be incorporated into the edition, not merely as comments on a forum but by changing the ordering or ranking of parts in responses to reader feedback. For the most part, Neville's practical suggestions seem like commonsense calls for making everything one does as an editor a bit less opaque to the novice reader.

    Neville writes that "[Joost] Daalder's scholarly authority in his note on ballads [in his digital edition of Thomas Dekker and Thomas Middleton's 1 The Honest Whore] is . . . reinforced by the appearance of a hyperlink regardless of whether a reader chooses to investigate it further. The affordance explicitly radiates outward to imply that digital projects are part of a larger ecology of knowledge relative to their analog forebears" (p. 20). I am not seeing why Neville thinks that a digital hyperlink does this and that a printed reference to another work does not. Perhaps lazy readers who only bother with resources that are linked digitally might value a hyperlink over a printed reference, but surely we do not want to judge everything from the perspective of the laziest amongst us, the "users of digital media [who] are primed to value resources that exist online over those that do not" (p. 20).

    Neville claims that today's readers "expect immediate access to cited material as a signal of its legitimacy" (p. 21) and that because of "algorithmic searching, and omnipresent smart phones, we are now in a position where more of us--indeed, potentially all of us--are subjects presumed to know" (p. 24). I should have thought that the important distinction between scholarly editors and their readers (even their scholarly ones) is that the former are paid to know and the latter are paying to learn. This inevitable hierarchy ought not to embarrass us: it is the whole point of reading scholarly critical editions instead of cheaper alternatives.

    The present reviewer published one relevant article this year ('Scholarly Method, Truth, and Evidence in Shakespearian Textual Studies', ShSurv 72[2019] 150-9). It argues that the common mistrust of mathematical approaches to knowledge creation in Shakespeare studies arises largely from failures of investigators to attend carefully enough to the language they use, and it offers some tips for distinguishing reliable from unreliable studies in this area.

    In an article from 2017 that was previously overlooked by this review, Matt Vadnais somewhat repeats what he has published before ('Speeches, Speech Order, and Performance in Shakespeare's Printed Playtexts', ShakB 35[2017] 521-39). He explains Shakespeare's use of two-handed exchanges and speech stems in making his plays easier to perform. Vadnais agrees with Stern that plays were put on with little group rehearsal but he thinks that the playwright made this possible not only by choosing cues carefully but also by writing in a way that limited how many speeches a player had to remember at one time, and this made it easy to anticipate and pick out the cues. In particular, the plays were written to conform to conventions that indicated who was going to speak next.

    Although Iago has 300 more lines than Othello in their play, the two roles have an identical number of speeches. So do Proteus and Valentine in The Two Gentlemen of Verona. And for several more pairs the numbers of speeches are almost identical: King John and Philip in King John, Antony and Cleopatra in Antony and Cleopatra, Viola and Olivia in Twelfth Night, Antipholus of Syracus and Dromio of Syracus in The Comedy of Errors, and Richard 2 and Bolingbroke in Richard 2. Limiting the number of speeches an actor had to give was a way to limit the number of cues he had to listen for. An inexperienced player might be given a substantial part (in terms of the total number of lines) so long as it did not have too many speeches, since the only essential requirement was that he remembers the lines where he gave others their cues. It did not matter too much to the running of the play if he forgot some of the substance of the speeches.

    Vadnais reports from his counting that the typical character in an average scene has 8 speeches and this number was pretty constant. He calls this the "speech burden" (p. 524). Q2 Hamlet has a mean average of 10.6 speeches per character per scene and the much shorter The Comedy of Errors has an average of 10.3 speeches per character per scene. Mostly characters are not in successive scenes, so for Vadnais the limiting of the speech burden in a scene gave the actor not too much to remember at any one time (p. 526). I presume he means that between his scenes an actor could refresh his memory about the cues for the next scene by looking over his script.

    Another way of sharing out the burden of acting was making sure that each player was responsible for the most speeches in at least one scene. That is, the lead characters did not keep getting the majority of the speeches in scene after scene: the workload was spread across the company, from highest to lowest. Still, in many plays one character has more than 150 speeches to say, so there had to be other tricks to lighten the load. In Folio Hamlet 2.2, the actor of Hamlet has to listen for 23 cues ending in "Lord", and almost all are versions of "my Lord". He must have had more than these cues to tell him when to speak. Vadnais notices that the first 12 cues that the Hamlet actor must listen for in this scene come in a two-player exchange, so he did not need the cues: each time the other fellow stopped it was his turn to speak. Vadnais counts that of the 1185 scenes in the Shakespeare canon, 242 (about one-fifth) are two-person scenes. Longer scenes that have more than two characters often begin or end with an extensive two-character exchange. Here Vadnais essentially repeats what he wrote in his essay reviewed in YWES for 2013. 

    Next Vadnais turns to the notion of speech stems (also described in his 2013 essay), in which everyone is taking it in turns to speak to one character, so that the actor playing this character knows that whenever someone else stops speaking it is his turn to speak. Shakespeare typically had one speech stem per scene. If we consider the script as sequences of speeches falling between one entrance or exit and the next, speech stems become even more noticeable: quite often the whole of such a sequence is one speech stem, and what changes with an entrance or exit is simply who has the current speech stem: that is, who it is that everyone else is talking to.

    Unlike the two-person scene, speech stems became apparent only in rehearsal, and indeed identifying them may have been a large part of the point of having collective rehearsals. Vadnais tabulates which plays make the most use of these techniques (two-hander exchanges and speech stems) and it turns out to be the ones that are in various ways the most challenging for the performers: the ones with exceptionally big parts such as the roles of Hamlet and Othello and those with a lot of work for boy players. Again as in his 2013 essay, Vadnais next attends to scenes that really did need collective rehearsals (fights, robberies, and the like) and that do not take advantage of the two-hander and speech stem expedients.

    In another article previously overlooked by this review, Tom Clayton starts out considering some important editorial principles as they apply to the three substantive editions of Hamlet (Q1, Q2, and the Folio) and takes as granted certain recent arguments about them ('Whither Hamlet's 'Words, Words, Words'? Notes on Dialogue and Designs in Hamlet', BJJ 25[2018] 214-41). The first argument is that Q1 is based on a text made by stenographic recording of a performance, as presented by Stern in an article reviewed in YWES for 2013. The second argument (more of an assumption since Clayton does not offer substantial evidence) is that Q2 is "intrinsically the most authoritative single text" (p. 215).

    Clayton claims that what readers want is a "literary reading-text, as full and accurate as the available textual evidence permits" (p. 216) and that this entails conflation of the competing early editions of Hamlet. Of course, for his claim to make sense Clayton must silently ignore the evidence for revision of the play, since conflating the 'before' and 'after' textual states resulting from an act of revision generates an incoherent composite. For Clayton, Q1 is potentially deceptive in that good actors can conceal its demerits. Q1 can be "performed with a solemnity that makes it seem much better than it is . . ." (p. 217). I would suggest that if we find a dramatic script reading poorly but playing well, perhaps the fault is not in itself but in how we read it.

    Clayton quotes from the New Oxford Shakespeare Modern Critical Edition and Critical Reference Edition of Hamlet, remarking that both follow Q2 on a particular reading. This is not, as Clayton seems to believe, a matter of editorial choice but rather a necessity, since the former is merely a modernization of the latter. That is, the editor's task was to make an original-spelling edition, the Critical Reference Edition, and then, without altering the readings, to modernize it to produce the Modern Critical Edition. It turns out that Clayton's remarks on the early texts are only really a preamble to a literary-critical consideration of certain passages in the play, which is outside the scope of this review.

    And so to the many articles about computational methods of authorship attribution and their application to particular cases. The most controversial of these was part of a widely discussed claim that the computational literary studies in general is largely self-deluding and that we might as well just confine our activities to reading texts rather than counting things in them (Nan Z. Da 'The Computational Case against Computational Literary Studies', Critical Inquiry 45[2019] 601-39). Da's target is the studies that run computer programs over large textual corpora to count things and then use visualizations and statistics to "make arguments about literature or literary history or to devise new tools for studying form, style, content, and context" (p. 602).

    Da complains that some studies turn the finding of nothing into a discovery of its own that they treat as if it is meanful. I would say that finding nothing is indeed discovery so long as you really have found nothing; the key distinction is between finding nothing because there is nothing to be found (an important discovery) and finding nothing because you looked using methods incapable of finding what is there. Da objects to the fact that text-mining projects, and the laboratories they take place in, are able to attract large amounts of external funding. After all, the software being used is almost always free to the investigator and computing power has never been cheaper. The obvious retort to this is that the grant money largely goes on staff time, which is spent doing things that we have not yet managed to computerize, such as bringing XML transcriptions up to a suitable standard for a particular kind of experiment.

    When we get to the detail, it emerges that Da's premise is mistaken because she is substantially ignorant of the field. She claims that essays that perform what she calls Computational Literary Studies (CLS) "are more or less all organized the same way, detecting patterns based in word count (or 1-, 2-, n-grams, a 1-gram defined as anything separated by two spaces) to make one of six arguments: (1) the aboutness of something; (2) how much or how little of something there is in a larger body, (3) the amount of influence that something has on something else; (4) the ability of something to be classified; (5) if genre is consistent or mixed; and (6) how something has changed or stayed the same" (p. 605). Da simply overlooks the many scholarly essays that do not count words, including those that count character n-grams, the frequency and distribution of punctuation marks, the variations in lexical density, the lengths of units such as sentences and speeches, and other features. Ignorance of the field is also suggested by Da's  remark that a 1-gram "is defined as anything separated by two spaces".

    Da develops further mistaken generalizations, such as: ". . . CLS papers make arguments based on the number of times x word or gram appears" (p. 606). Some do; many do not. Worse still: "The highest number of consecutive words (1-grams) that CLS work has looked at is three (trigrams)" (p. 606). There are plenty of studies that have looked at longer strings than that. It is hard not to conclude that Da's ignorance of the field she is writing about disqualifies her from the critique she is attempting. Da introduces the notion of 'stop words' without glossing it, and asserts that whether an investigator retains or eliminates such words strongly conditions the outcomes of experiments.

    Da's remarks about 'stop words' might reasonably have been written in the 1970s when limited computing power and storage forced those doing certain kinds of corpus linguistics to eliminate the most frequent words from their experiments, hence the notion of 'stop words' that were not counted or included in a concordance. But freed from these technical limitations, today's careful studies categorize words by precise frequencies and avoid the imprecise term 'stop words'. Indeed, many studies today are devoted exclusively to the most-frequent words in a corpus, for which the notion of 'stop words' is meaningless. That Da thinks that the most-frequent words in texts "have to be removed for text mining" (p. 623) again betrays her ignorance of the many studies that focus exclusively on these words.

    Da's critique contains some remarkable howlers about methods that ought to embarrass the editors of the journal Critical Inquiry. She writes that "After all, statistics automatically assumes that 95 percent of the time there is no difference and that only 5 percent of the time there is a difference. That is what it means to look for p-value less than 0.05" (p. 608). In fact the p<0.05 threshold is not a feature of statistics itself -- just a convention used by some people -- and in any case finding that something would happen any way (by random variation) less often than one time in 20 (what we mean by  p<0.05) is not the same as assuming that there is "no difference" 95% of the time. The p threshold has nothing to say about how often anything special is actually happening. Rather, it is a comment on how often a particular observation will be made even if nothing special is happening. That is, p<0.05 is simply a threshold for discarding those experimental observations that we know would come up anyway by chance one time in 20, or more often. That Da does not understand common statistical measures vitiates several of the critiques in this paper.

    The most remarkable howler of all is Da's claim about Hugh Craig and Arthur F. Kinney's book Shakespeare, Computers, and the Mystery of Authorship (reviewed in YWES for 2009). Da writes that "The most famous example of CLS forensic stylometry--the use of statistical text mining to differentiate style, genre, and authorship--is the argument by Hugh Craig and Arthur F. Kinney that the late works by Shakespeare were written by Christopher Marlowe, even though Marlowe died more than a decade before Shakespeare s last known work (in accordance with the Marlovian Theory, which argues that Marlowe faked his death in 1593 and continued to write in Shakespeare s name)" (p. 622n39). It is hard to imagine where Da got this idea about Craig and Kinney's book, since even a skim-read would have dispelled it.

    Occasionally Da lands a valid critique. She finds that many visualization tools, especially those for representing networks, are unsuitable for the kinds of connectedness found in literary texts. Often the networks one is shown are essentially meaningless. I agree about that. But her gaffes predominate. Da glosses the word 'entropy' as 'redundancy' and then as the inverse of 'redundancy', and then tries to give a simple explanation of it: "Entropy measure sounds sophisticated (and seems analogous to literary complexity), but what it does, actually, is measure the number of distinct pair [sic] of words and their distribution in the total number of bigram pairs" (p. 634). This is, of course, not what entropy means. Da goes on to explain that the variety of words matters in such measurements ". . . two thousand, twenty thousand, or two million distinct words make a difference" (p. 634). Of course, no known human language has two million distinct words and the entire Oxford English Dictionary has fewer than 220,000 entries.

    Da complains that the field of Computational Literary Studies is just not very good at linguistics and that efforts to employ proper Natural Language Processing are "halfhearted" (p. 635), mainly because computerized tagging is not very good yet and because even low levels of error in the tagging matter for literary analyses. "Tagging errors and imprecision in NLP do not sufficiently degrade the extraction of information in many other contexts, but they do for literature" (p. 636). If this were really true, Da would seem to have discovered a possible marker for literary writing: it is that class of writing for which even low levels of tagging error matter when you do Computational Literary Studies methods. Da seems not to notice the paradox she has caught herself in. If she were right about Computational Literary Studies being inapplicable to literary analysis, then she would simultaneously be wrong, since she would have shown that this approach is capable of distinguishing literary writing from other forms.

    According to Da, the tagging problem is insuperable. Having to laboriously tag all the features in a large corpus of poems before computational analysis can find them makes the exercise pointless. "Couldn't someone trained in poetry just find, read, and classify them?" (p. 637). The claim that Computational Literary Studies is necessary because we can never read all the works out there is, Da argues, misguided, since there is also not enough time available to spend "Looking for, obtaining copyrights to, scraping, and drastically paring down 'the great unread' into statistically manageable bundles . . ." (p. 638). Da seems not to expect that as machine-automated tagging improves (in tandem with the general improvement in how clever computers are getting) this problem will diminish.

    One chapter in a collection of essays that was overlooked by this review last year will be considered here because in it Matthias Bauer and Angelika Zirker share Da's scepticism about recent computational analysis of language ('Shakespeare and Stylometrics: Character Style Paradox and Unique Parallels', in Zwierlein, Petzold, Boehm & Decker, eds. Anglistentag 2017 Regensburg: Proceedings 39, pp. 31-8). They believe that authorship attribution studies simply cannot work because dramatists devise distinct idiolects for their characters and these vary more by character and genre than writers vary from one another in style, and, making this problem worse, writers naturally imitate one another. Bauer and Zirker seem not to have noticed that in blind validation runs where the best authorship attribution methods are set to work on texts for which the authors are already known the methods routinely pick out the correct author in  80-95% of cases. This would be impossible if Bauer and Zirker's premises were correct. 

    Bauer and Zirker critique John Burrows and Hugh Craig's essay "Authors and Characters'" in English Studies 93 (2012), asking the reader to look at certain pictures in it. Burrows and Craig's Figure 7 shows a scatterplot of a Principal Component Analysis of the frequencies of 100 words in the speeches of five sets of characters from: i) Jonson's comedies, ii) Jonson's late comedies, iii) Jonson's tragedies, iv) Chapman's comedies, and iv) Chapman's tragedies. Bauer and Zirker report that ". . . the comedy characters are virtually indistinguishable by author" (p. 35). Looking at the same picture that is not what I see. Admittedly, there is some overlap in the zones occupied by Jonson's comedies and Chapman's comedies, but the zones are nonetheless distinct: Chapman's dots generally lie to the north-west of the scatterplot and Jonson's dots generally lie to the south-east. It is hard to know how to respond to a bald assertion that contradicts the evidence of one's eyes. I do not agree that what Bauer and Zirker have shown is that "character and genre combined may easily trump author" (p. 35). Quite the opposite.

    Next Bauer and Zirker turn to Gary Taylor's chapter "Did Shakespeare Write the Spanish Tragedy Additions?" in the Authorship Companion to the New Oxford Shakespeare edited by Taylor and the present reviewer. After a long preamble and the reproduction of an entire table from Taylor's essay they find an error: using searches in Literature Online Taylor reported "do ye hear me" as a unique parallel between Addition One of The Spanish Tragedy and Henry Porter's Two Angry Women of Abingdon. But, as Bauer and Zirker point out, there are also "do you hear me" in Much Ado About Nothing, "Do you not hear me" in The Tempest, and "Didst thou not hear me" in 3 Henry 6. As Bauer and Zirker observe, Taylor's rules on how much variation he will accept in counting a match -- including variant spellings and grammatical variation -- ought to have made these Shakespearian phrases count as matches.

    In fact, Taylor does not indicate whether he manually treated 'you' and 'ye' as the same word by running multiple manual searches or relied on Literature Online's built-in variant spelling finder. Likewise for the matches that Bauer and Zirker find that include 'not' within the phrase: we do not know just how, in technical terms, Taylor applied the latitude he gave himself. In all, the whole point of Bauer and Zirker's study of Taylor's essay is to identify this one error. Bauer and Zirker end with the objection that authorship attribution work seems ill-equipped to deal with the problem that ". . . an author may wittingly or unwittingly imitate and allude to another author . . ." (p. 37). They conclude that stylometrics "should leave behind its fixation on author identification" (p. 37).

    The journal Digital Scholarship in the Humanities this year published seven articles relevant to this review, two of them in a special issue guest-edited by this reviewer, which will be noticed first. In the first, Gary Taylor presents compelling evidence that the play Arden of Faversham, first published in 1592, was written at least in part (the start of Scene 10) by the relatively obscure author Thomas Watson ('Finding 'Anonymous' in the Digital Archives: The Problem of Arden of Faversham', DSH 34[2019] 855-73). At the opposite end of the spectrum of fame, Shakespeare has recently been convincingly claimed as the author of at least the central part of this play, its third act.

    Arden of Faversham is especially hard to attribute because it comes from the period before 1594, for which we have fewer plays than later on and fewer still that are single-authored and securely attributed. Adding to the problems, Arden of Faversham probably has at least two authors and if one of them was Shakespeare then it was not the Shakespeare we know from his later, more mature, works but Shakespeare near the beginning of his career. Moreover, some candidate authors for writing Arden of Faversham left us no single-authored securely attributed plays, such as Thomas Achelley, Michael Drayton, Richard Hathway, and Thomas Watson.

     Taylor constructed a corpus of fifteen dramatists' complete output (not just plays) from EEBO-TCP. These fifteen men are "known, or at least suspected, to have written for the commercial theaters in the period up to 1592, when the first edition of Arden was published" (p. 857), and they are (in diminishing order of canon size): Anthony Munday, Greene, Nashe, Thomas Lodge, Shakespeare, Marlowe, George Peele, John Lyly, Kyd, Drayton, Robert Wilson, Achelley, Henry Chettle, Watson, and Hathway.

    Taylor took as his sample from Arden of Faversham the run of 34 lines from 10.1 to 10.34, being 274 words. This was selected because "(1) it owes nothing to the narrative sources of the play, and (2) it begins a long stretch of text that recent investigators, on both sides of the divide, agree was not written by Shakespeare" (p. 859). From this 274-word sample Taylor extracted every word 2-gram, 3-gram, and 4-gram and searched for them in EEBO-TCP. He also searched for "every collocation of two or more semantic words (nouns, verbs, adjectives, and adverbs) ten words before or after each other" (p. 859).

    Taylor was looking for matches that appear only in Arden of Faversham and the canons of just these fifteen candidate authors. Taylor counted maximally, so each subgram within a 3-gram (of which there will be two 2-gram subgrams) or a 4-gram (of which there will be two 3-gram and three 2-gram subgrams) added one more to the count of matches, but he also tallied what would be the result if we counted only the largest n-gram as a single match and ignored all the subgrams within it. To adjust for the fact that some canons are much larger than others, Taylor scaled his hit-counts to the canon sizes, producing a "hits per 10,000 words" metric for each author.

    Looking in the dramatic canons, Watson came out as having the most matches with Arden of Faversham lines 10.1-34 and by a huge relative margin. Turning to the non-dramatic canons, the same thing happened: Watson had by far the most connections to Arden of Faversham lines 10.1-34. In both approaches, Kyd, who is Brian Vickers's and Darren Freebury-Jones's preferred candidate, came second to Watson. The same thing happened when Taylor counted only works from 1585 to 1594.

    Next Taylor tried weeding out the links that are unique to one of his fifteen candidates but no longer unique when we consider all of EEBO-TCP from 1585 to 1594. This did not add to his candidate list, but it did remove phrases that are not, in fact, peculiar to one writer, and the result was still the same: Watson came out on top, whether Taylor considered only dramatic canons or canons in all writing genres. Taylor reserves judgement on whether Watson might have written the rest of the play, and whether he collaborated with Kyd and/or Shakespeare.

    The second of the two articles from the present reviewer's guest-edited issue of Digital Scholarship in the Humanities was by the late David L. Hoover and it shows, among other things, that the authorship attribution methods used by Brian Vickers do not work ('Simulations and Difficult Problems', DSH 34[2019] 874-92). Here we will attend to only those parts of Hoover's essay relevant to early modern drama and the utility of counting n-gram matches to attribute authorship. Hoover points out that Vickers's insistence on considering only plays from around the same time as the one he wants to attribute risks overlooking valuable evidence from a later period if the true author was much younger than Vickers's preferred candidate. Moreover, Vickers never calculates just what results -- how many shared n-grams -- we should expect to find if his candidate is not the true author. Hoover here extends his previous work replicating Vickers's methods but using nineteenth-century drama instead of early modern drama. His conclusion is that Vickers's method used to attribute Arden of Faversham to Kyd would misattribute the plays Hoover is testing it with.

    The point of Hoover using the Victorian drama is that it provides enough plays to do the kind of validation runs -- what Hoover calls simulations -- that are not possible with the early modern drama because, for example, Kyd's securely attributed canon has at most only three plays: The Spanish Tragedy, Soliman and Perseda, and Cornelia (a translation of Robert Garnier's French play Corn lie). Taking the validation still further, Hoover invents pseudo-authors by splicing together parts of different writers plays, and these pseudo-texts also get seemingly conclusive attributions by Vickers's method, which of course they should not. Hoover concludes that shared rare n-grams alone do not provide adequate evidence of shared authorship.

    In the first of the other five relevant articles from Digital Scholarship in the Humanities this year, Pervez Rizvi attempts to show that Gary Taylor's so-called microattribution method of authorship attribution does not work and that the phrases that its proponents think are authorial markers are found in many plays that they simply overlook ('The Problem of Microattribution', DSH 34[2019] 606-15). Rizvi claims to be replicating the microattribution technique described in Taylor's essay "Empirical Middleton: Macbeth, Adaptation, and Microauthorship" (reviewed in YWES for 2014) and to have found much data that Taylor did not consider.

    The first sign that Rizvi is not really replicating Taylor's method is the difference in the amount of evidence they uncover for links between a 63-word passage from Macbeth (4.1.140-148) and the other plays of the period. Taylor found 18 matches but Rizvi finds 759. Either Taylor failed to notice 98% of the available evidence, or he and Rizvi are in fact looking for different things. Rizvi reports that "The stated aim of Taylor is 'to analyse every word and every possible combination of words' in the passage to be attributed" (p. 606). In fact, Taylor made clear that he was referring only to lexical words and by including the much more common function words Rizvi naturally finds more evidence. This difference in their methods accounts for almost all the differences in their findings. Another, but less important, cause of the differences is that Taylor used the LION dataset while Rizvi uses transcriptions of plays that he put together from disparate sources. When replicating searches for matches between texts, one must start from the same data and use the same procedures for ensuring that you count as matches words spelt differently in different plays. Rizvi did not do these things.

    Rizvi does a type-wise counting of his token-wise lists of matches and produces a table showing that whichever way you count it, the 63-word passage from Macbeth has more matches with Shakespeare than with any other author. The explanation, of course, is that Shakespeare's canon is much the largest and if we look only for the most common words forming collocations then his works will dominate the results. Also, in these counts it matters what we do with multi-authored plays. There is no sense in counting a match as being to Shakespeare if the bit of the play in question is, say, Middleton's part of Timon of Athens or John Fletcher's part of Henry 8. But in fact, Rizvi's list of who-wrote-what -- given in the files "Authors-List per Play.xlsx" and "Authors-List per Play-Author Combination.xlsx" in the 'data' subfolder, downloaded from Rizvi's website -- lists Shakespeare as the sole author of 1, 2, 3 Henry 6, Henry 8, Pericles, The Two Noble Kinsmen, Timon of Athens, and Titus Andronicus. This boosts the number of Shakespeare hits and decreases the number of hits that count for other authors.

    Rizvi repeats his procedure using the evidence adduced in Taylor and John V. Nance's "Imitation or Collaboration?" (reviewed in NYWES for 2015). As before, and for the same reasons, Rizvi gets many more matches than Taylor and Nance did. Next Rizvi does this all again with the essay "Shakespeare and Who?" by Taylor, Nance, and Keegan Cooper (reviewed in NYWES for 2017) and yet again with Nance's essay "Shakespeare and the Painter's Part" in the New Oxford Shakespeare Authorship Companion. In each case, Rizvi applies quite different methods and gets quite different results from the studies he is attempting to replicate. It is incumbent on those doing replication studies to use the same procedures as those whose work they purport to replicate.  

    Rizvi published two further articles in the same journal this year, both concerned with the Zeta method of authorship attribution invented by John Burrows. The first finds fault with the way that investigators connected with the New Oxford Shakespeare have presented the results of Zeta experiments ('The Interpretation of Zeta Test Results', DSH 34[2019] 401-18). Typically, investigators offer a scatterplot in which a datapoint's x-axis value represents the degree to which a text contains many of the words that are favoured in one candidate's author's canon and many of the words that are disfavoured in a second candidate author's canon, and the datapoint's y-axis value represents the degree to which a text contains many of the words that are favoured in the second candidate's author's canon and many of the words that are disfavoured in the first candiate author's canon.

    Naturally, when texts from these two candidate authors' canons are plotted, the first candidate's works tend to score highly on the x-axis because they contain many of the words the first candiate favours and many of the words the second candidate disfavours, and they tend to score lowly on the y-axis because they contain few of the words the second candidate favours and few of the words the first candidate disfavours. This puts the datapoints for the first candidate's works in the bottom-right (or south-east) quadrant of the scatterplot. The opposite is true of the datapoints for the second candidate's words, since these tend to score lowly on the x-axis because they contain few of the words the first candidate favours and few of the words the second candidate disfavours, and they tend to score highly on the y-axis because they contain many of the words the second candidate favours and many of the words the first candidate disfavours. This puts the datapoints for the second candidate's works in the top-right (or north-west) quadrant of the scatterplot.

    Thus if the Zeta method is correctly finding the words that the two candidate authors favour and disfavour, we should expect to see two distinct clusters of data points: one for the first candiate clustered in the south-east quadrant and one for the second candidate clustered in the north-west quadrant. When a datapoint for a text to be attributed is put into the scatterplot, its closeness to one or other cluster tells us how like one or other candidate's works it is in the words it favours and disfavours.

    Rizvi rightly objects to the phrasing of an essay by Jack Elliott and Brett Greatley-Hirsch in the New Oxford Shakespeare Authorship Companion that implied that Zeta method was correctly performing its classificatory role simply because the datapoints for the candidate authors' works formed the two expected clusters. As Rizvi correctly points out, if authors have any word preferences at all we will see this clustering and on its own it proves nothing. The present reviewer, as one of the editors of the New Oxford Shakespeare Authorship Companion, takes responsibility for not correcting this misleading phrasing in Elliott and Greatley-Hirsch's essay. The mistake, however, is one of preambulatory phrasing and does not invalidate the results that Elliott and Greatley-Hirsch found when they started putting datapoints into the scatterplots to represent extracts from the play Arden of Faversham.

    How close to one or other candidate's cluster must the datapoint from a text-to-be-attributed fall for the text to be classified as being by one or other of the candidates? Indeed, how do we even measure closeness in this kind of experiment? Hugh Craig and Arthur Kinney established the methodology of averaging the coordinates of the datapoints in each cluster to find the cluster's centre, drawing a straight line between the two centres (strictly, 'centroids'), finding the middle of this line, and drawing at that middle point an infinitely long second line (a bisector) perpendicular to the first. This bisector line divides the scatterplot into two zones, one associated with each candidate author. Which zone a new text's datapoint falls into determines which candidate author the text is attributed to.

    Rizvi points out that all this method really does when we ask 'which side of the bisector does this segment fall?' is tells us which of the two centroids the segment is nearer. That is, the bisector contains no more information than the two centroids it was made from: we have thrown away all the data about the training segments except their average x/y positions. We do not, for example, capture in the centroids any information about the compactness or spread-outness of the cluster of segments that produced the centroid. We do not get the standard deviation, for example. This is indeed a good point. But what can we do about it? It seems unreasonable to reject on these principles a method that, in blind validation runs, is demonstrably capable of accurate authorship attributions. 

    Craig's adaptation of Burrows's Zeta method does not experiment with counting tokens, only types. In particular, the Zeta method will treat differently two authors of whom it is true that the first uses word w at a steady rate (appearing across all his segments) and the second uses the same word the same number of times but in bursts (appearing in only some of his segments). Indeed, this will be true even if the second author uses the word more often, but in bursts, since each segment is recorded as either containing or not containing that word. Rizvi points out that plays might plausibly differ in this regard simply because of their subject matter: "[In] Richard II . . . there is one story line and, for that reason, perhaps a tolerably consistent rate at which words are used. . . . [as opposed to] 1 Henry IV or 2 Henry IV, in which the main story, wherein high-born characters speak in verse, is punctuated by a comic subplot wherein low-born characters speak in vulgar prose" (p. 410). Again, Rizvi's critique is correct and calls for an improvement to, not an outright rejection of, a method that demonstrably works.

    A second major thrust of Rizvi's article is that investigators have misused results from Zeta when applying it to authorship attribution problems. In an article called "Language Chunking, Data Sparseness, and the Value of a Long Marker List" (reviewed in YWES for 2014), Alexis Antonia, Hugh Craig, and Jack Elliott argued that short word n-grams are better discriminators of authorship than long ones, based on their experiments showing that short n-grams produce higher Zeta scores. Rizvi objects that Antonia et al. overlooked an important inherent fact of language that will give short n-grams higher Zeta values than long n-gram without any reference to their discriminatory value for authorship attribution.

    The linguistic fact Rizvi is referring to is that long n-grams that we find in common between two texts will necessarily contain a number of short n-gram matches tucked inside them. If we find that the 4-gram "a number of short" occurs once in text A and once in text B, that is one 4-gram match. But this 4-gram match contains two 3-gram matches between texts A and B: "a number of" and "number of short". And the same single 4-gram match contains three 2-gram matches between texts A and B: "a number", "number of", and "of short". For this reason alone, the count for 2-gram matches (here, a count of three) will be higher than the count for 3-gram matches (here, a count of two), and both of these will be higher than the count for 4-gram matches (here, a count of one).

    Thus Rizvi concludes that all Antonia et al. were able to show was that short n-grams get higher Zeta values (because of these higher match-counts), not that these higher Zeta values translate into better authorship attribution discriminations. But this conclusion is demonstrably incompatible with the content of Antonia et al.'s article, since they did not simply use highest Zeta scores as a proxy for shared authorship, and they considered several complicating factors.

    The added complications considered by Antonia et al. included the fact that for any set of, say, three words such as 'the', 'cat', and 'sat', there are only three 1-grams that can be counted but there are six possible 3-grams a writer might construct from them: 'the cat sat', 'the sat cat', 'cat the sat', 'cat sat the', 'sat the cat', and 'sat cat the'. When we consider a set of four words, as in 'the', 'cat', 'sat', and 'on', there are 24 ways to combine them as 4-grams but only four 1-grams. (The number of combinations -- strictly they are permutations -- is given by the function n! where n is the number of words in the set and the exclamation mark stands for the mathematical procedure called 'factorial'. With n = 3 the value of n! is 3 2 1, which is 6, and with n = 4 the value of n! is 4 3 2 1, which is 24.)

    Thus there are more ways for authors to construct phrases that are common to two texts when we are considering 4-grams than when considering 3-grams or 2-grams, or 1-grams, but the occurrences of each of these ways will be rarer. That is what Antonia et al. refer to as "sparseness" in their essay title, and they explicitly track the complex interplay of "the rival principles of marker abundance and marker density" (Antonia et al. p. 154). Thus, contrary to Rizvi's characterization, Antonia et al. do not simply conclude that 1-grams are always best for authorship attribution. Rather, they conclude that ". . . no one style of n-gram outshines the others in providing authorial markers" and that "attributionists would be wise to keep an open mind about the usefulness of each" (Antonia et al. p. 159). 

    Rizvi ends with an attempt to replicate the experiment by which Jack Elliott and Brett Greatley-Hirsch, in their essay in the New Oxford Shakespeare Authorship Companion, conclude that Arden of Faversham contains some writing by Shakespeare. Rizvi finds that he gets different results and that the attribution of Arden of Faversham is not demonstrably correct. However, the replication has marked differences from Elliott and Greatley-Hirsch's experiment, most importantly in that Rizvi uses modernized and lemmatized transcriptions of early modern plays where the experiment was on original-spelling unlemmatized transcriptions. It is unclear whether the original experiment and the replication were the same regarding the exclusion of speech prefixes and stage directions. Rizvi thinks that his texts "may differ in minor ways from the texts used in Elliot and Greatley-Hirsch" (p. 414) but I would consider this an understatement. I would conclude that this loose replication is unable to tell us anything about the validity of the original experiment, but for Rizvi the conclusion is stark: ". . . all the Zeta test results in Elliot and Greatley-Hirsch (2017) are unreliable" (p. 417).

    Rizvi's other article this year about the Zeta method shows that the calculation of its value could be simplified without the loss of functionality or discriminatory power ('An Improvement to Zeta', DSH 34[2019] 419-22). The processes of determining the sets of words that each of two authors favour and disfavour contains unnecessary repetition. A sorted list that has at its top the 50 words most favoured by the first candidate and disfavoured by the second candidate will of necessity have at its bottom the 50 words most favoured by the second candidate and disfavoured by the first. That is, the classic Zeta method creates two lists but the second one need not be calculated because it is just the same as the first list but in reverse order. A second improvement recommended by Rizvi is that instead of producing a Zeta value between 0 and 2 the calculation could be reframed to produce a value between -1 and 1, which is how some other mathematical metrics used in computational stylistics (for instance correlation) are expressed.

    Last year we reviewed Helmut Ilsemann's "Christopher Marlowe: Hype or Hoax", finding no reason to support its conclusion that Marlowe did not write most of the plays we attribute to him. Ros Barber has written a response that comes to the same conclusion ('Marlowe and Overreaching: A Misuse of Stylometry', DSH 34[2019] 1-12). Barber rightly objects that Ilsemann used far too little writing for his ground truths, his samples of known-Shakespeare and known-Marlowe writing. In particular, it is unreasonable, she objects, to take 1 Tamburlaine as decisively typical and measure all the other supposed Marlowe plays by their difference from it. After all, the publisher Richard Jones in his remarks "To the Gentleman Reader" at the front of the 1590 first edition reported that he had removed the comic scenes. (Rather than quoting the 1590 edition, Barber uses a second-hand transcription from a modern critical essay, and she misquotes it by turning Jones's "what times" into "what time". The source, the modern critical essay, gets the quotation right.)

    Barber contends that comedy is found throughout Marlowe's plays. This matters because Barber three times quotes Joseph Rudman claiming that "genre trumps authorship" (once on page 6 and twice on page 10), which according to Rudman "has been shown in many studies". The essay by Rudman that Barber quotes was reviewed in NYWES for 2016 and it does not identify any of the studies that are supposed to have found this, and nor does Barber. Indeed, Barber asserts that no study has shown that authorship trumps genre, but in fact Arefin et al. "An Information Theoretic Clustering Approach for Unveiling Authorship Affinities in Shakespearean Era Plays and Poems" (reviewed in YWES for 2014) did exactly that.

    Barber, again rightly, objects that using only 1 Tamburlaine as the sole reference point for Marlowe's style is unwise because it is an early play and Marlowe's style changed over time; also, collective fashions in language change over time. Barber also shows that Ilsemann was wildly inconsistent in his selection of known-Shakespeare plays, switching from play to play for different experiments. As Barber describes them from her personal correspondence with Ilsemann, his procedures make no sense. She speculates that it was probably because Marlowe's Edward 2 has lots of English noble family names and aristocratic titles -- as do Shakespeare's history plays used by Ilsemann but 1 Tamburlaine does not -- that in Ilsemann's experiments Edward 2 came out looking like Shakespeare's writing.

    Ilsemann used Solimon and Perseda as his known-Kyd play rather than The Spanish Tragedy because the latter is now thought to have some Shakespeare in it. As Barber rightly points out, to avoid this problem Ilsemann could simply have used one of the editions of The Spanish Tragedy prior to the 1602 edition in which the additions first appeared. Like the present reviewer in last year's NYWES, Barber complains that Ilsemann's attributions require us to ignore strong external evidence (such as title page ascriptions) about who wrote what.

    In his own article this year, Ilsemann argues that the English translation of Robert Garnier's play Corn lie as Cornelia that we have thought was by Kyd -- because the title page and dedication say so -- is in fact by Marlowe ('Forensic Stylometry', DSH 34[2019] 335-49). Ilsemann starts with the premise that Cornelia was "never meant to be performed" (p. 335), which in fact is more than we know, and observes that writing a closet drama would be an odd thing for Kyd to do after his release from imprisonment and torture. But publishing under his own name one of Marlowe's plays that, because of Marlowe's death, fell into his hands, would make sense since Kyd was destitute.

    Ilsemann sets out to investigate Cornelia's authorship. He remarks that he decided, as he prepared texts for comparison to Cornelia, that "plays by Chapman, Chettle, Greene, Munday, Lyly, Lodge, Peele, Rowley, and Shakespeare could be eliminated" (p. 336) but he does not indicate why and I cannot infer what he is getting at. Instead, he compared Cornelia to Solimon and Perseda and The Spanish Tragedy (now both widely believed to be Kyd's), the two parts of Tamburlaine, and Nashe's Summer's Last Will and Testament.

    Regarding methodology, Ilsemann remarks that Rolling Delta -- that is, John Burrows's Delta technique applied with a rolling window moving across the text -- is able to detect collaborative authorship, which is true but rather glosses over details of such things as how large the window is and how much it is rolled forward each time it is moved. This matters because Ilsemann immediately starts relating his experimental results and they do not make obvious sense.

    Ilsemann's Figure 1 is labelled "Cornelia -- delta attribution with MF1W", where MF1W means "most frequent single word". Above the figure is a second label that reads "Cornelia attributions with 10000-word windows". But 10,000 words is a huge window since, by Ilsemann's reckoning (given on the same page), the play contains only 15,622 words in all and the other plays he tested it against have at most 17,867 words. If Ilsemann's window really was 10,000 words then the crucial variable is how far he rolled it forward for each test. The x-axis of Figure One is labelled "words" and runs from 4000 to 11,000 and there appear to be 23 data points across this axis, starting at the x value of 4000 and ending at the x value of around 10,500. From this figure and Ilsemann's silence on what he did, I cannot be sure what his experiment actually consisted of.

    Perhaps the data point for x=5000 represents the result of the window resting over words 1 to 10,000 of each play (that is, about two-thirds of the play) and the result is then assigned to word 5000 because that is the word falling at the mid-point of the window. The data points seem to be spaced at 250-word intervals across the x axis so presumably 250 words is how far the window was rolled forward between each count. If my inferences are correct, this is an unusually large window-to-roll ratio of 40:1. More commonly such experiments use 1000- or 2000-word windows rolled forward perhaps 500 words each time (for a ratio of 2:1 or 4:1). Aside from a remark that it is recommended by the creator of the R Stylo software package not to use one smaller than 5,000 words, Ilsemann does not discuss the size of his window.

    For detecting co-authorship, small windows have the advantage of being able to pick out authorial stints that are as small as a few speeches or a single scene, but at the cost of producing large quantities of data to process, perhaps 30 to 40 windows per play. Measuring across two-thirds of a play at once (10,000 words) seems a peculiar experimental choice. Ilsemann's Figures Two and Three are the same as Figure One but with the counts being made not of single words but of character (not word) bigrams and trigrams. All three figures show that of the plays tested the one closest to Cornelia is Marlowe's 1 Tamburlaine with Locrine as a close second when counting character bigrams. In his essay called "Christopher Marlowe: Hype or Hoax" (reviewed in NYWES for 2018) Ilsemann concluded that Locrine is Marlowe's work, so for him this is corroboration.

    Thus Ilsemann finds that Cornelia is essentially Marlowe's work and speculates that Marlowe must have left it behind in his papers that Kyd had access to and Kyd took it over. But did Kyd modify Cornelia before he published it as his own? To answer this, Ilsemann gives a description of a method and its results that I can make no sense of:

The outcome can be seen in Table 1, a matrix that displays text segments of 250 words vertically (Column A), beginning at 500 words, where B5 is the first measuring point of the 1,000-word window. The first 1,000 words of the play were attributed to Marlowe. The next window covers 1,500 words and its first measuring point (C6) at 750 words returns Marlowe as well. Total 750 words are also the second measuring point of the 1,000-word window marking words 251 1,250 of the text. Accordingly, the first measuring point of the 5,000-word window at 2,500 words can be found at J13. (p. 337)

In this description Ilsemann uses the terms "segment" and "window" but they are apparently not the same thing. He describes how one "window covers 1,500 words" and another is a "1,000-word window" and another is a "5,000-word window". Looking at Table 1 while reading this explanation does not enable me to make sense of it and hence I must simply report that I have no idea what experiment Ilsemann carried out.

    Ilseman's conclusion is that there is some writing by Kyd and some by Nashe in Cornelia. Recall, though, that all the plays he tests are by Marlowe or Kyd, or Nashe, with no other candidates under consideration. Thus if the value of the feature that Ilsemann is counting in Cornelia were merely to randomly fluctuate then his experiment must, of necessity, deviate from pointing to Marlowe by pointing to either Kyd or Nashe. That the cells of his data table that point to Kyd and to Nashe fall into clumps -- a whole run of Kyd cells or a whole run of Nashe cells -- might be thought to point away from random variation in the feature that Ilsemann is counting. But in fact such clumping can arise from an author's random fluctuations of style so long as the unit of writing in each test is relatively small. This is because the count across one small stretch of text (corresponding to one cell in Ilsemann's table) need not be independent of the count for the previous small stretch of text (corresponding to the previous cell).

    For instance, if our units of text are, say, verse lines then counts of phenomena such as rhyme or stichomythia will appear in runs covering adjacent units, as when a character starts singing a song (generating a run of units showing rhyme) or two characters start to exchange verbal barbs (generating a run of stichomythia). This is not to say that the runs in Ilsemann's table are caused by merely random variation -- he may be onto something about the authorship of Cornelia -- only that it is possible to think of other explanations beside authorship that might produce his results.

    Halfway through his article, Ilsemann turns to the supervised machine learning methods called Support Vector Machines and Nearest Shrunken Centroid, of which he offers no explanation. It is not clear that Ilsemann understands the experiments he is conducting, since his account of what a 'rolling window' experiment involves is curiously back-to-front. He writes that ". . . segments of 5,000 words are consecutively attributed with an overlap of 4,750 words, so that every 250-word section is given an attribution" (p. 339). In this example the window is 5000 words wide and the amount that it rolls forward between each test is 250 words. If we denote each word in the play by a subscript that states its ordinal position in the text -- so that in Richard III's opening soliloquy "Now" is word1, "is" is word2, "the" is word3, and "winter" is word4 and so on -- then in Ilsemann's example the window first covers the play from word1 to word5000 and the attribution is made and then it rolls forward by 250 words to cover the play from word251 to word5250.

    Each time, it is the 5000 words of the window that are tested, not the 250 words by which the window rolled forward, so it is wrong to claim, as Ilsemann does, that "every 250-word section is given an attribution". Again Ilsemann presents his results without explanation as a series of lists and the upshot is that Cornelia is wholly or mainly Marlowe's except when Delta is classifying by most-frequent-word, where it comes out as nearly half (17 out of 41 segments) by Kyd, and when Nearest Shrunken Centroid is classifying by most-frequent-character-trigrams, where it comes out as nearly half (again 17 out of 41 segments) by Nashe.

    Ilsemann runs two more tests using the R Stylo software package to produce a "bootstrap consensus tree" and a "Cluster analysis" (p. 342), neither of which he explains. Ilsemann's results show that at this point he introduced three new texts into his analysis: Nashe's prose satire Pierce Penniless and Greene's plays James 4 and Friar Bacon and Friar Bungay. Ilsemann makes no mention of why they appear now in his analysis and not before. Ilsemann shows his bootstrap consensus tree as Figure Four but gives no indication of what its various connecting lines mean other than that the tree "places Cornelia together with other plays by Marlowe like the two Tamburlaines" (p. 342).

    Summarizing his work, Ilsemann writes that ". . . a vast number of reference texts have been consulted, and it was in this way that Kyd s Cornelia appeared, contrary to all expectations, to be of Marlovian provenance" (p. 343). I cannot discern where Ilsemann consulted a vast number of texts, since the core of the present study is a comparison of Cornelia with just two Marlowe plays, two Kyd plays, and Nashe's only play. Ilsemann appears to be aware of the problem of building new attributions using as one's ground truth texts that have themselves only just been attributed by new methods -- "the danger of circularity is always present" and "Once a text has been wrongly attributed, its samples cannot be used to identify other texts" (p. 343) -- but he appears nonetheless to fall into exactly this error in using here as part of his evidence for Cornelia being by Marlowe the fact that it tests like Locrine, which was freshly attributed to Marlowe by Ilsemann just the year before. But from Ilsemann's perspective the problem is rather that all our attributions based on the known Marlowe canon are unreliable because "most of his plays have been wrongly attributed" (p. 343). In an appendix Ilsemann offers more evidence for Marlowe's authorship of Locrine and it is of the same kind presented earlier in the article that I cannot understand.

    In the first of two substantial articles this year, Brian Vickers claims to find lots of unique matches between the Arden of Faversham Quarrel Scene (Scene 8) and the Kyd canon that show that Kyd authored it, and between Titus Andronicus 4.1 and the Peele canon that show that Peele authored it ('Is EEBO-TCP / LION Suitable for Attribution Studies?', EMLS 21.i[2019] n. pag.). In both cases his list of unique parallels contains many that are not unique, as this review will show.

    Vickers misunderstands what the Text Creation Partnership (TCP) dataset is, remarking that for books it has not yet processed "users will still find illegible words, replaced in some cases by question marks" (p. 2). In fact, if TCP has not yet processed a book there will be no searchable text at all, and not a poor quality one as Vickers believes. He also thinks that in its "previous existence as CD-ROM" (p. 2) the TCP dataset did not have a search engine. He appears to be confusing Literature Online (LION) with TCP since only the former ever existed in CD-ROM form and of course LION has always had a search engine. Something goes wrong with Vickers's phrasing when he writes that "the early version was involved several procedures" (p. 2).

    Vickers surveys MacDonald P. Jackson's work on the authorship of the Quarrel Scene in Arden of Faversham. He faults Jackson's assumption that the play was first performed in 1590 and his acceptance of the Oxford Complete Works's chronology that assigns dates for 2 Henry 6, 3 Henry 6, The Taming of the Shrew, and The Two Gentlemen of Verona that are earlier than those given in Wiggins's Catalogue of British Drama. According to Vickers this affected Jackson's chronological filter when searching LION and made him value links between Arden of Faversham and these four plays by Shakespeare that probably were written after Arden of Faversham and that hence have its phrases in them because of "recollection of a play [that is, Arden of Faversham] seen in performance" (p. 9).

    Jackson's later version of his own method began to discount qualitative matching of phrases and just used the numbers of matches, although he still discounted homonyms such as 'weeds' meaning clothes and 'weeds' meaning plants. Vickers claims to have caught Jackson violating his own rule about homonyms in treating "top bough" in Arden of Faversham as a match for "highest bough" in Dekker's work since in context in Dekker "'bough' has quite different connotations" (p. 9). (Perhaps so, but the issue is not connotation but denotation: in both cases 'bough' denotes part of a tree and that is quite a different case from the two meanings of the two distinct words spelt as "weeds".)

    Vickers has other trivial objections of this kind and in each case he treats connotations as denotations so that, for example, a collocation of 'gale' and 'shake' should not (according to Vickers) be treated as a match for the collocation of 'whirlwind' and 'shaken' (p. 10). He repeats his claim that other links can be explained by mere literary influence. Vickers also objects to Jackson's method of counting, so that one line in Arden of Faversham can provide the Arden of Faversham half of each of a set of links to multiple other plays.

    More substantively, Vickers reckons that Jackson's searches for phrases in the Quarrel Scene of Arden of Faversham simply overlooked "over 60 close matches with Kyd" (p. 10), listed in Vickers's appendix. Vickers does not explicitly claim that these are matches unique to just the Quarrel Scene and the Kyd canon, but this is implied by his use of a "newly available marked up corpus of 527 early modern plays prepared by Pervez Rizvi, which allows users to search for n-grams and collocations in all the texts" (p. 11). Vickers's list of phrases uniquely shared by the Arden of Faversham Quarrel Scene and Kyd's canon contains many that in fact are not unique to those corpora.

    His example #2 is 'my' near 'brain' and as well as the matches he finds in the Arden of Faversham Quarrel Scene and Soliman and Perseda there is "my head, my brain" in Marlowe's Massacre at Paris, "my busie braines" in Greene's Orlando Furioso, "my husband . . . his braines" in Marlowe's 1 Tamburlaine, "my braines" in Anonymous's The Wars of Cyrus, "my very braines" in Lyly's Galathea, "my giddie braine" in Marlowe's Edward 2, and literally dozens of other instances in literary works of the period.

    To take another example, Vickers's #11, the collocation of 'harvest' and 'corn' is indeed common to the Arden of Faversham Quarrel Scene and Kyd's plays, but it also appears as "corne in haruest" in Wilson's The Cobbler's Prophecy, as "sow his corne, and in haruest" in Richard Tarlton's News Out of Purgatory, as "a large haruest for a little corne" in Lodge's Rosalynde, and other less canonical literary works. Some of Vickers's hits are obviously commonplace phrases: his #17 is the phrase "for his pains" that also appears as "for his paines" (and "for your paines" and "for thy paine") in Anonymous's A Knack to Know a Knave, and as "for his paines" in Greene's Friar Bacon and Friar Bungay.

    Vickers's claimed parallel "of his end" (#21) also appears in Peele's The Battle of Alcazar. Vickers's claimed parallel "for my sake" (#25) also appears, as any student of the drama would expect, in dozens of works including Marlowe's Edward 2, Lodge's The Wounds of Civil War, Wilson's The Cobbler's Prophecy, Marlowe's 1 Tamburlaine and The Massacre at Paris, Marlowe and Nashe's Dido Queen of Carthage, Anonymous's Fair Em and Troublesome Reign of King John, Shakespeare and others' The Contention of York and Lancaster, Wilson's Three Lords and Three Ladies of London, and many other literary works.

    Vickers's example #60 is "an orator" which unsurprisingly also turns up in Marlowe's Dido Queen of Carthage, Shakespeare and others' The Contention of York and Lancaster, Lodge's The Wounds of Civil War, and other literary works. Vickers's long list of parallels between the Arden of Faversham Quarrel Scene and the Kyd canon is not wrong: these are parallels, but they are not unique parallels and the same phrases can be found in many plays from the same period (1585-95) and in many more non-dramatic literary works in LION.

    Actually, Vickers's list is strictly wrong in the sense that he overlooks at least one piece of evidence that supports his case: "I now I" appears not only in the Arden of Faversham Quarrel Scene and in Kyd's Soliman and Perseda but also in Kyd's The Spanish Tragedy as "I, now I know thee, now thou namest my Sonne". Whatever Vickers's precise search methodology -- he does not describe it in detail -- the fact that it misses at least one parallel with the Kyd canon indicates that it is unreliable. Vickers reports that he used the software called "WCopyfind 4.1.1, a free program developed at the University of Virginia by Dr Lou Bernard" (p. 11n27). This itself is an authorial misattribution, since WCopyfind was developed by Lou Bloomfield not Lou Burnard. (Burnard has confirmed this in an email to the present reviewer.)

    Vickers next turns to Titus Andronicus 4.1 and the recent work by William W. Weber and Anna Pruitt showing that it is in fact Shakespeare's not Peele's. Again Vickers offers his own list of matches, in this case between Titus Andronicus 4.1 and the Peele canon, in an appendix, and this time he is explicit that they are "unique" to those two corpora (p. 15). They are not.

    Vickers's #15 is "arms in", which also occurs in Lodge's The Wounds of Civil War, Kyd's Cornelia and Soliman and Perseda, Greene's Orlando Furioso, and Shakespeare and others' The Contention of York and Lancaster. Vickers's #16 is "more than one", which also occurs in Greene's Orlando Furioso. Vickers's #17 is "the fact", which also occurs twice in Anonymous's The True Tragedy of Richard 3, Robert Wilmot, Christopher Hatton, Henry Noel, G. Al., and Roderick Stafford's Tancred and Gismund, and Kyd's The Spanish Tragedy.

    Vickers's 24# is "the camp", which also appears in Marlowe's Edward 2 and 2 Tamburlaine, Marlowe and Nashe's Dido Queen of Carthage, Anonymous's The Wars of Cyrus, Jack Straw, and Fair Em, Greene's Orlando Furioso, Kyd's Cornelia, and Shakespeare and others' Richard Duke of York. Vickers's #26 is "sweet niece", which also appears in Lodge's The Wounds of Civil War. Vickers's #30 is "may this", which also appears in Anonymous's Fair Em and The True Tragedy of Richard 3, Marlowe's 1 Tamburlaine, and Kyd's The Spanish Tragedy.

    Vickers's #40 is "gentle lord", which also occurs in Marlowe's Edward 2 and Marlowe and Nashe's Dido Queen of Carthage. Vickers's #51 is "their blood", and it occurs also in Kyd's Cornelia and The Spanish Tragedy, Anonymous's The True Tragedy of Richard 3, Marlowe's 1 Tamburlaine and 2 Tamburlaine, Lodge's The Wounds of Civil War, Greene's Orlando Furioso, and Shakespeare and others' Richard Duke of York. Vickers's #52 is "they rise", and it occurs also in Lodge's The Wounds of Civil War. And so on.

    Vickers's central claim is that Jackson overlooked a lot of unique matches and that he, Vickers, is listing them to show Jackson's error. But Vickers's list, here regarding Titus Andronicus 4.1 as previously regarding the Arden of Faversham Quarrel Scene, simply is not a list of matches found only in the text under analysis and one author's canon. It is remarkable that Vickers believes that phrases such as "the fact" and "gentle lord" and "their blood" are unique to Titus Andronicus 4.1 and the Peele canon, and it is inexplicable that the experience of searching for them did not disabuse him of this. The only possible conclusion is that something is seriously awry with Vickers's means for searching texts and he is being misled by it. Because of his error, Vickers concludes that LION is unsuitable for authorship attribution and his method based on anti-plagiarism software and Rizvi's "Collocation and N-Gram" dataset is superior. In fact the reverse is true.

    Darren Freebury-Jones published what look like three articles this year, but as we shall see substantial repetition between them makes the total more like two-and-a-half. In the first, he argues that Kyd wrote Arden of Faversham and King Leir (Darren Freebury-Jones ''Fearful Dreams' in Thomas Kyd's Restored Canon', Digital Studies / Le champ numérique 9.i[2019] 1-25). His method is to show that these two plays have "parallelisms of thought and language, versification characteristics, and overall dramaturgy" (p. 2) with the accepted Kyd plays of The Spanish Tragedy, Solimon and Perseda, and Cornelia.

    Freebury-Jones selectively quotes the numbers from Philip W. Timberlake's tables of plays' feminine ending rates in The Feminine Ending in English Blank Verse (1931) that support his claim and ignores those that do not. Thus he writes that ". . . Timberlake recorded an average of 10.8% feminine endings in King Leir, which corresponds to the 10.2% for Solimon and Perseda, and 9.5% for Cornelia . . . [and] Arden of Faversham averages 6.2% feminine endings". True, but the same tables show that the average for Kyd's The Spanish Tragedy is just 1.2%, which is lower than the rates of the authors Greene, Peele, and Marlowe whose low rates, according to Freebury-Jones, rule them out as contenders for authorship of King Leir.

    Of course, Freebury-Jones could argue that chronology is a confounding variable here and that The Spanish Tragedy was decidedly old-fashioned. But he makes no attempt to address the matter, which gives the unfortunate impression that he simply does not expect the reader to go and look up Timberlake's tables for herself and see the problem. The wider problem, of course, is that these feminine-ending rates vary considerably from play to play and between scenes within a play.

    Freebury-Jones next turns to the evidence from phrasal parallels between Kyd plays and King Leir found by Vickers, which according to Freebury-Jones now amounts to an "impressive number of unique matches between King Leir and The Spanish Tragedy alone" (p. 6). Freebury-Jones cites and quotes at length (pp. 6-7) three publications by Martin Mueller that he claims corroborate Vickers's attribution, all online and given in his References list as "Mueller, Martin. 2009a. 'Vickers is Right About Kyd'", "Mueller, Martin. 2009b. 'N-grams and the Kyd Canon: A Crude Test'", and "Mueller, Martin. 2014. 'Repeated N-grams in Shakespeare His Contemporaries (SHC)'", followed by their respective URLs.

    The first two of these references point to pages on Vickers's personal website and the third to a website at Northwestern University and at the time of writing (August 2023) all three give "Not found" (Error 404) errors. Three further online sources created by Mueller are cited by Freebury-Jones as ""Mueller, Martin. 2017a. 'Phrasal Repetitions Between Arden of Faversham and Other Early Modern Plays'", "Mueller, Martin. 2017b. 'Phrasal Repetitions in Early Modern Drama'", and "Mueller, Martin. 2018. 'Shared Unique Tetragrams Between Early Modern Plays'". The URLs for these do resolve to working webpages, but all three are simply tables of data. With no description of the experiments that produced these data, the references are meaningless.

    Freebury-Jones sketches some of his own searching of texts from LION but his account is so cryptically expressed -- "I . . . adjusted my figures according to composite word counts" (p. 8) -- that it is impossible to know what he actually did. Freebury-Jones claims that he "double-checked" his results using the dataset at Pervez Rizvi's Collocations and N-grams (CAN) website where there is meant to be an essay called "Which N-grams are the Best?". But Freebury-Jones gives an URL pointing only to the entire CAN website, and at the time of reviewing (August 2023) there is no essay of that title to be found by manually scrolling through all the site's pages and following all its internal links. Web-searching for the phrases that Freebury-Jones quotes from Rizvi's essay returns only copies of the essay here being reviewed (because it contains these quotations), not Rizvi's essay that is the alleged source.

    The same problem besets Freebury-Jones's quotations from another essay by Rizvi that is meant to be on the CAN website called "Arden of Faversham and the Extended Kyd Canon" but is nowhere to be found. Freebury-Jones's documentation of his sources would be unacceptable in an undergraduate essay. Again Freebury-Jones reports his own searching for 4-word phrasal matches between the agreed Kyd plays and King Leir, but provides only the total count of matches and not the phrases themselves. Then he does the same for Arden of Faversham, identifying total counts but not mentioning the phrases.

    At the end of the summary of his findings, Freebury-Jones refers the reader to an online appendix that details the rare 4-word (and longer) phrases that he thinks link the known Kyd plays to Arden of Faversham and King Leir (p. 10). Freebury-Jones's definition of rareness is, he claims, the same as used by MacDonald P. Jackson in his book Determining the Shakespeare Canon (reviewed in YWES for 2014), namely that the phrases must occur "not more than five times in drama of the period 1580-1600" (p. 8).

    The first three sections of Freebury-Jones's appendix can be ignored because they list the phrases in common between the accepted three Kyd plays The Spanish Tragedy, Solimon and Perseda, and Cornelia. Next come the links between The Spanish Tragedy and King Leir, and Freebury-Jones cites the two occurrences in each play of the phrase "and yet me thinks". It is unclear why Freebury-Jones believes that this phrase meets his criteria, since it also occurs in Marlowe's Edward 2 (first performed 1592) as "and yet me thinks I should commaund", in Anonymous's The Taming of a Shrew (first performed 1594) as "And yet methinkes the foole doth looke asquint", in Jonson's Every Man Out of His Humour (first performed 1599) as "And yet mee thinkes you should take your leaue", in Jonson Cynthia's Revels (first performed 1600) as "and yet me thinks you are tedious", in Dekker, Chettle, and William Haughton's Patient Grissel (first performed 1600) as "And yet me thinkes I am not halfe", in Marlowe's Doctor Faustus (first performed 1592) as "And yet me thinkes if that death were neere", in Greene's Friar Bacon and Friar Bongay (first performed 1589) as "And yet me thinkes this Farmers ioylly sonne Passeth", and in John Marston's Histriomastix (first performed 1599) as "And yet me thinkes he should be by".

    That is 8 occurrences in plays from 1580-1600, plus the 4 that Freebury-Jones found in The Spanish Tragedy and King Leir for a total of 12. Freebury-Jones's cutoff is supposed to be no more than 5 occurrences. This problem is not an isolated slip affecting only the phrase "and yet me thinks". Freebury-Jones also thinks that the phrase "Me thinks I should" meets his criterion of occurring no more than 5 times in plays from 1580 to 1600, but in fact it appears in Munday's Fedele and Furtunio (first performed 1584) as "Me thinkes I should spowt Lattin", in Marlowe's Edward 2 (first performed 1592) as "Me thinkes I should reuenge me of the wronges" and "me thinkes I should commaund", in Lyly's Mother Bombie (first performed 1591) as "me thinkes I should not Mother Bombie tolde me", in Peele's The Old Wives Tale (first performed 1590) as "me thinkes I should haue seene you cast away", in Porter's The Two Angry Women of Abingdon (first performed 1598) as "me thinkes I should Wife it as fine as", and in Jonson's Cynthia's Revels (first performed 1600) as "mee thinkes I should wish my selfe all manner of".

    That is 7 occurrences in plays from 1580-1600, plus the 2 that Freebury-Jones found in Solimon and Perseda and King Leir for a total of 9.  Freebury-Jones's cutoff is supposed to be no more than 5 occurrences. (All my dates of first performance come from the online Database of Early English Plays hosted by the University of Pennsylvania.) Something is clearly awry with Freebury-Jones's method of searching texts as it is bringing in irrelevant phrases, but it is hard to say just what the problem is.

    Freebury-Jones turns from his data to his conclusions from it. He finds that Arden of Faversham "was written solely by Kyd, and that verbal links with Shakespeare's plays are indicative of Kyd's influence on Shakespeare, as opposed to Shakespeare's authorship" (p. 10). For this he offers the present work plus his past publications, all of which have been reviewed in past roundups of YWES and NYWES. The second half of Freebury-Jones's article turns first to pre-digital studies that ascribed Arden of Faversham to Kyd and his own fresh literary critical approach that considers Kyd's depiction of ominous dreams.

    Kyd's plays share with Seneca's the idea the dreams could be prophetic, while the dominant Elizabethan idea was that they could not. Freebury-Jones details this trope from Kyd's plays (and in particular the detail of someone telling the dreamer to pay no attention to the experience), and in particular the dramatization of a character waking up and thinking that the events of the dream are still happening. Of course, Caesar's wife Calpurnia has a prophetic dream in Julius Caesar, and he is told to ignore it, as does Clarence in Richard 3, and Hermia wakes up believing in her dream in A Midsummer Night's Dream, as does Richard in Richard 3. Freebury-Jones shows the same things happening in King Leir and Arden of Faversham and using some of the same words. Freebury-Jones addresses Clarence's dream in Richard 3 and Calpurnia's in Julius Caesar but argues that Shakespeare was inspired by Kyd's use of dreams in the first instance and that the second is too late ("around a decade after Kyd s accepted plays"") to count in this discussion. Freebury-Jones concludes by alerting the reader to the forthcoming complete works edition of Kyd that he and Vickers are general editing.

    Freebury-Jones's second article this year has substantial passages that duplicate word-for-word what Freebury-Jones wrote in the article just reviewed called "'Fearful Dreams'" ('The Diminution of Thomas Kyd', Journal of Early Modern Studies 8: Beyond Books and Plays Cultures and Practices of Writing in Early Modern Theatre[2019] 251-77). On page 253 of this second article, the 286 words from "Edmond Malone was . . ." down to ". . . feminine endings in King Leir" appeared in identical form on pages 4 to 5 of ""Fearful Dreams'". Likewise in this second article, on pages 253 to 254 the 337 words from "the case for Kyd's authorship . . ." down to ". . . Kyd's own pen' (48-49)" appear in identical form on pages 11 and 12 of "'Fearful Dreams'".  And again, this second article's pages 256 to 258 have 709 words from "Martin Mueller has created . . ." down to ". . . quite solid evidence" that repeat the material from pages 6 and 7 of "'Fearful Dreams'". If these two articles were reviewed for the United Kingdom's periodic research audit called the Research Excellence Framework, their repetition of material would make one or other of them ineligible.

    As he did in "'Fearful Dreams'", Freebury-Jones selectively cites Timberlake's tables of feminine ending rates as if these show that only Kyd could have written King Leir because 10.8% of its lines end this way. Freebury-Jones does not identify particular pages in Timberlake's book as his sources for various numbers, which makes it hard to check his logic since the book presents the same data in various different ways. Nor does Freebury-Jones mention that Timberlake himself could see problems in attributing plays by this metric, since the 10.2% rate for feminine endings in Solimon and Perseda "goes beyond anything else that can safely be ascribed to Kyd" and Timberlake wondered whether the play might have been "revised by another hand before it was finally published" (Timberlake p. 51).

    Freebury-Jones surveys some historical claims for Kyd's authorship of King Leir and Arden of Faversham and then what he sees as the unfairness of the New Oxford Shakespeare's attribution of the latter to Shakespeare, with a long exposition on Freebury-Jones's own recent publications that, he believes, showed where MacDonald P. Jackson went wrong in his authorship attribution work. As in his essay "'Fearful Dreams'", Freebury-Jones quotes two publications by Martin Mueller ("Vickers is right about Kyd" and " N-grams and the Kyd canon: a crude test"), referring to them as "Mueller 2009a" and "Mueller 2009b" respectively, but for both he gives the same URL of "http://www.brianvickers.uk/wp-content/uploads/2016/05/Martin-Mueller-on-Brian-Vickers-and-the-Kyd-canon.pdf" and at the time of writing (August 2023) this produces a type 404 "Page not found error".

    Next Freebury-Jones turns to Rizvi's Collocations and N-Grams dataset and repeats what he has argued elsewhere about it providing support for Kyd having written more than just The Spanish Tragedy, Solimon and Perseda, and Cornelia. At this point (page 259), the reader is more than a third of the way into Freebury-Jones's essay without anything having been claimed that the author has not claimed before using the same evidence and, in large part, the same words. The next two pages are devoted to Freebury-Jones's response to MacDonald P. Jackson's book Determining the Shakespeare Canon, which repeats the critique made in Freebury-Jones's essay "A Raven for a Dove" reviewed in NYWES for 2016 and which critique he recycled in his essay "In Defence of Kyd" reviewed in NYWES for 2018.

    There follows a return to the argument from feminine endings, for which Freebury-Jones again recycles and cites three of his previous publications. Then comes a page-and-a-half on Marina Tarlinskaya's work on early modern prosody and the inadequacy, as Freebury-Jones sees it, of Jackson's engagement with it, and then a page of potted summaries of conclusions reached by A. F. Kinney, Brett Greatley-Hirsch, Jack Elliott, Peter Kirwan, and Lene Buhl Petersen.

    Finally, two-thirds of the way into the essay, Freebury-Jones begins to offer his "findings for a single word: 'But'" (p. 264). Freebury-Jones counts how often this word appears in the first foot of a verse line, which he finds occurs once in "20 lines in The Spanish Tragedy, 23 lines in Soliman and Perseda and King Leir, and 19 lines in Arden of Faversham, which we might compare to Shakespeare's rate of once every 29 lines in Henry VI Part Two, and 46 lines in The Taming of the Shrew". Freebury-Jones goes on: "It also seems worth pointing out that of the 45 instances of 'But' in scenes Four to Nine of Arden of Faversham, which Jackson gives to Shakespeare, 25 occur at the start of verse lines, at a rate of one every 21 lines" (p. 265).

    This is a fascinating claim by Freebury-Jones. From my own counting prompted by his, I broadly agree with Freebury-Jones about Kyd's The Spanish Tragedy, finding that 1-in-20 of its verse has "But" in the first foot (I make it 131 instances in the play's 2745 verse lines). I also broadly agree about Arden of Faversham, finding that 1-in-19.5 of its verse has "But" in the first foot (I make it 110 instances in the play's 2145 verse lines). But breaking apart Arden of Faversham I find that Scenes 4-9 have this feature at the rate of 1-in-23 verse lines while the rest has it at the rate of 1-in-18 verse lines. And Freebury-Jones's count for The Taming of the Shrew (which I make 1-in-42 not 1-in-46) is not Shakespeare's typical rate but near the top of his range.

    My counts for this feature in Shakespeare's plays up to year 1600 are: The Two Gentlemen of Verona 1-in-22; Titus Andronicus (Shakespeare's parts) 1-in-32; The Taming of the Shrew 1-in-41; Richard 3 1-in-43; The Comedy of Errors 1-in-33; Love's Labour's Lost 1-in-45; Richard 2 1-in-41; Romeo and Juliet 1-in-33; A Midsummer Night's Dream 1-in-30; The Merchant of Venice 1-in-35; 1 Henry 4 1-in-37; 2 Henry 4 1-in-41; Much Ado About Nothing 1-in-26; Henry 5 1-in-26; Julius Caesar 1-in-42; and As You Like It 1-in-27. It does not seem worth testing The Merry Wives of Windsor since it has so little verse.

    The lowest Shakespeare score on this test is for The Two Gentlemen of Verona, which the New Oxford Shakespeare chronology assigns to the same year, 1588, as Shakespeare's contribution to Arden of Faversham. It would take a lot more work deriving counts for other dramatists for this metric to be demonstrably useful for authorship attribution. Unfortunately, Freebury-Jones devotes just 287 words of his essay to this topic, before returning to his major theme of the history of various attributions and samples of critical opinions about the various allusions that early modern dramatists made to one another, such as Nashe's "the Kidde in Aesop".

     Freebury-Jones reads through a series of such allusions, principally by Greene, finding them appropriate to Kyd writing the play Fair Em. Then Freebury-Jones spends four pages responding to Terri Bourus's book Young Shakespeare's Young Hamlet, which conveys her argument that Q1 Hamlet reflects Shakespeare's first attempt to write a play of this story in the late 1580s, and that the allusions in the 1590s to a play of this title are to Shakespeare's play and not to what used to be called the Ur-Hamlet, written by Kyd. Freebury-Jones defends the view that Q1 Hamlet is based on a memorial reconstruction of the script better represented by Q2 and that the version that is alluded to several times in 1589 and the 1590s was Kyd's lost earlier version. He concludes that it is time for "recounting Kyd's stock", bearing in mind what he considers to be the overwhelming evidence that this stock consists of more plays than the three -- The Spanish Tragedy, Solimon and Perseda, and Cornelia -- that most scholars attribute to him.

    The last of Freebury-Jones's three articles this year argues that Arden of Faversham and Kyd's Solimon and Perseda share an unusual phrasing in their stage directions that is rare anywhere else ('Corresponding Stage Directions in Plays Attributable to Kyd', ANQ: A Quarterly Journal Of Short Articles, Notes, And Reviews 32.i[2019] 16-7). The unusual expression "Then they" appears in five stage directions in Arden of Faversham and six in Kyd's Solimon and Perseda. Freebury-Jones claims that "No other publicly performed play of the Elizabethan period matches this count . . ." (p. 17). This is true. The phrase comes up in other plays' stage directions but not so frequently: once each in Anonymous's King Leir, Captain Thomas Stukely, A Warning for Fair Women, and Mucedorus, and once each in Peele's The Battle of Alcazar and Nicholas Udall's Ralph Roister Doister.

    Freebury-Jones claims that only in Arden of Faversham and Solimon and Perseda and King Leir does the phrase "Then they" appear at the beginning of a stage direction. I disagree: Ralph Roister Doister has a stage direction "Then they sing agayne" (STC 24508, sig. B3r). Moreover, points out Freebury-Jones, the more specific stage direction "Then they fight" occurs only in Arden of Faversham and Kyd's Solimon and Perseda. I agree.

    Freebury-Jones ends by noting that Arden of Faversham has 62 instances of a stage direction beginning "Here enters" and Fair Em has 4, suggesting to him that the latter has the same author as the former, namely Kyd. If anything, I would have thought that, although the phrase is somewhat rare, the large difference between 62 and 4 instances tells against common authorship. Forms of "Here enter" in stage directions are not particularly unusual. The anonymous play Common Conditions has 34 instances of "Here enter" or "Here entereth" stage directions, and Appius and Virginia, anonymously published in 1575 and perhaps by Richard Bower, has 16.

    An article by Brian Vickers from 2016 was overlooked in previous reviews and will be noticed now (''The Misuse of Function Words in Shakespeare Authorship Studies': Online Essay Hosted on the Website of the 'Electronic Text Reuse Acquisition Project (ETRAP') at Https//www.etrap.eu/wp-content/uploads/2016/12/2016-11-30-vickers-misuse.pdf:'[2016]). The article's main contention is that 2 Henry 6 is wholly by Shakespeare and studies pointing to its co-authorship are methodologically flawed. Vickers writes that "Despite the general approval within the scholarly community of using function words for authorship attribution . . . a central plank of its theoretical justification is mistaken, which makes it unsuitable for the study of Shakespeare and other early modern dramatists" (p. 3). Vickers cites a number of investigators who assume that rates of function-word usage are unconscious and hence immune to authorial control. He disagrees: ". . . the choice of function words is determined by a larger set of conscious choices, the language user's intended communication of meaning" (p. 6).

    This seems a debate hardly worth having since no matter what the cause, function-word frequencies have in many experiments across a range of literatures, including early modern plays, been shown to be distinctive of authorship and to yield reasonably reliable results in blind authorship-attribution validations. In what follows, Vickers schools the reader in the meaning of the grammatical category of prepositions.

    Oddly, Vickers writes of a "group of pronouns, including up, down, along, across . . .". The Oxford English Dictionary does not recognize these as pronouns. This does not seem to be a typo, since four times on page 7 Vickers refers to words of this type (including over, under, above, beyond, at, on, for, over, from, and to) as pronouns where all other sources I can find categorize them as prepositions, adverbs, and conjunctions. It is hard to account for the oddness of the faulty grammar lesson that Vickers gives his readers, nor indeed its purpose since the investigators he scolds for counting function word seem, from his quotations of them, at least as knowledgeable on the subject as he is.

    Next Vickers addresses genre as a potentially confounding variable in authorship attribution, and he thinks it particularly influential regarding plays because "a dramatist does not write in propria persona" (those last three words are Latin for "in his own person"). It is interesting to know that in certain genres it is harder to tell one writer from another, but again this is rather an irrelevancy. If in blind validation, based on texts of known authorship, a method is able correctly to attribute authorship to an acceptable degree of accuracy, the presence of confounding variables that might be limiting that accuracy does not undermine the reliability of the results obtained; it merely suggests ways that the method might be improved by controlling for those variables.

    By raising these objections to others' scholarship -- and John Burrows is his repeated target -- Vickers seems to want to give the impression that this whole subject is more complex than others have realized, but this is a rhetorical manoeuvre rather than an intellectual one. Burrows's  methods demonstrably work despite the complications. At great length, Vickers picks through the bits of Shakespeare that seemingly are in propria persona such as the dedications of Venus and Adonis and Lucrece to Henry Wriothesley, illustrating his (Vickers's) skill at literary criticism but adding nothing to his argument.

    Having named the various rhetorical devices Shakespeare uses in his dedications, Vickers asserts that when his dramatic characters use them they serve to individuate the characters, one from the next. But that is all Vickers can do: assert that regarding the things that stylometry measures the variation in character idiolects frustrates the attempt to distinguish Shakespeare's writing from others' writing. Vickers is unable to demonstrate that these character idiolects actually frustrate authorship attribution by function-word analysis since, in fact, there are plenty of studies that demonstrably achieve reliable attribution by counting this feature in his plays and those by other dramatists. Yet according to Vickers "This is self-evidently impossible" (p. 16). It is hard to know how to reply to an assertion that what has been achieved time after time in countless replicated studies is impossible.

    Vickers presents no evidence that the authorship attribution studies he dismisses are faulty, other than the fact that sometimes they produce results he does not favour, such as detecting "Marlowe as co-author of [Shakespeare's] 2 Henry VI" and confirming "the theory that the Quarto and Folio texts of King Lear are separate compositions" (p. 19). Vickers gives the impression that the studies that came to these conclusions simply took on trust the validity of their own methods, but they did not. Rather, they reached their conclusions about 2 Henry 6 only after the methods had been blind-tested in multiple validation exercises that demonstrated beyond doubt that when presented with texts for which we know the correct author the method also and independently pointed to that author. Regarding King Lear the dispute is about authorial revision not authorship attribution, so Vickers's mentioning of it here is merely gratuitous. Vickers is silent on the validation experiments of Burrows and Craig and Kinney and others, and he needs to be since these show that his claims of methodological error are groundless.

    At this point (p. 21) Vickers brings in a new confounding variable: chronology. It is no use, he argues, trying to judge the Henry 6 plays by using the accepted Shakespeare canon as one's reference point -- as Craig and Kinney do in their book Shakespeare, Computers, and the Mystery of Authorship (reviewed in YWES for 2009) -- since the Henry 6 plays were written significantly earlier than the rest of the canon. According to Vickers, Craig's mistaken identification of Marlowe's hand in 2 Henry 6 arises because it "was Shakespeare's first play" and fails to meet the norms of his later plays because "Shakespeare was as yet in the process of establishing norms of usage at every level of his art" (p. 21). But in fact the Henry 6 plays are not so much earlier than the other Shakespeare plays. They were likely preceded by the sole-authored The Two Gentlemen of Verona and by Titus Andronicus, of which almost everyone agrees that we can isolate the parts Shakespeare wrote from those written by George Peele.

    To support his claim that 2 Henry 6 was Shakespeare's first play, Vickers adopts Wiggins's chronology in Volume 3 of his Catalogue of British Drama in which 2 Henry 6 and 3 Henry 6 are dated to 1591, Titus Andronicus, The Taming of the Shrew, 1 Henry 6, and The Comedy of Errors to 1592, Richard 3 to 1593, and The Two Gentlemen of Verona to 1594. Vickers neglects to mention that Wiggins put large error bars on these "Best Guess" years, giving the limits as from 1587 to 1592 for 2 Henry 6 and 3 Henry 6, from 1584 to 1594 for Titus Andronicus, from 1589 to 1592 for The Taming of the Shrew, from 1589 to 1593 for The Comedy of Errors, from 1591 to 1597 for Richard 3, and from 1587 to 1598 for The Two Gentlemen of Verona. Given these overlaps in Wiggins's chronology, any of these plays might by his reckoning have been Shakespeare's first and 2 Henry 6 might have been preceded by any or all of Titus Andronicus, The Taming of the Shrew, The Comedy of Errors, Richard 3, and The Two Gentlemen of Verona. (In fact, Wiggins gives his "Best Guess" year for The Comedy of Errors as 1593 in his summary List of Entries (3:xi-xiii) but as 1592 in the actual entry for this play (3:206)).

    Vickers appears to believe that by this chronological argument he has disposed of Craig's evidence for the mixed authorship of 2 Henry 6, and with it "[John Dover] Wilson's unjustified" division of it, and "the doubtful arguments of Paul Vincent" pointing the same way (p. 22). What follows is Vickers picking over some details of Craig's analysis arising from the fact that his 2000-word segments do not align with scene or act intervals. Craig of course chose these fixed segments precisely because we do not know the units of co-authorship, which may have aligned with natural dramatic breaks in some cases and not in others. In making the claim that perhaps Marlowe provided the scenes of Jack Cade's rebellion, Craig was analyzing broad trends and did not mean to imply perfect alignment between his artificial units and the real stints of writing.

    Against the claim that the Cade episodes are extraneous to the wider plot,  Vickers offers literary criticism showing that "These scenes are not 'detachable' from the play, since they dramatize the accelerating decline of England into chaos" (p. 25). There is nothing wrong with literary criticism, of course, but Vickers is particular about who does it. When the data from his tests were puzzling, Vickers claims, ". . . Craig fell back on impressionistic literary criticism" (p. 26).

    Vickers next gives his reader a long lesson in linguistics (pp. 30-36), since "scholars should possess a theory of language" (p. 30) before attempting authorship attribution. Vickers starts with Plato and ends in the mid-twentieth century when the structuralist linguistics of Ferdinand de Saussure was in the ascendant. It is fatal to Vickers's point that he closes his account just when everything in linguistics was about to change, twice. The first revolution absent from Vickers's survey is the transformational generative grammar introduced by Noam Chomsky in his 1957 book Syntactic Structures.

    Chomsky always insisted (and still does) that natural-language generation cannot be accounted for by a finite-state Markov process. Yet everywhere -- and this is the second revolution -- we now see Markov-chain probabilistic approaches to language generation producing impressive results, particularly those using the newly developed machine-language transformers such as the Generative Pre-Trained Transformer (GPT) from the group OpenAI. Lacking any awareness of these developments, Vickers's essay reads more like an historical document from before the 1950s than a recent research paper introducing new developments.

    Having established that linguistics research supports his approach of searching for phrases rather than individual words in order to attribute authorship, Vickers describes his approach using the software WCopyfind and 'InfoRapid Search & Replace'. First he detected all the trigrams (and longer) shared between 2 Henry 6 and 3 Henry 6, 606 of them in all, and then he went searching for these in "a database containing all the plays performed in the public theatre between 1579 and 1596" and recording just those, amounting to 130, found once in the database (pp. 36-37). Of those 130, 22% were matches to the Shakespeare canon.

    The logical flaw that vitiates this approach has been repeatedly pointed out in responses to Vickers's work, most comprehensively in Jackson's essay "One-Horse Races" in the Authorship Companion to the New Oxford Shakespeare. It turns out that we can reproduce this procedure for almost any two substantial texts and find a surprising number of phrases that are common to these two and rare or absent everywhere else. This is simply a feature of large texts, irrespective of authorship. Only in Vickers's interpretation is this phenomenon treated as evidence of shared authorship.

    For Vickers, "The remarkable number of matches [between 2 Henry 6 and 3 Henry 6] establishes conclusively that Shakespeare wrote the whole play [of 2 Henry 6], with no other hand intervening" (p. 38). Here Vickers drops the claim about matches with the rest of Shakespeare's canon and lists at length the phrases common to 2 Henry 6 and 3 Henry 6 (pp. 40-54). Since no one disputes Shakespeare's hand in these two plays, that they share a lot of phrases is hardly surprising. That the two plays are part of the same story involving many of the same characters -- so for instance both make reference to Henry 6 being crowned at the age of nine months, and to news being sought from France, and to disagreements about who should wear "the English crown" -- makes extensive parallels of phrasing even less surprising than it would be for other plays.

    In the final section of his essay, Vickers turns to prosody. He reproduces Timberlake's tables of rates of feminine endings in 2 Henry 6 which show a range from 0 (in the shortest scene) to 25% of all lines in scene 4.10. Among them are scenes that Craig attributes to Marlowe: 1.1 (6.3%), 1.2 (8.7%), 2.3 (14.5%), 4.2 (2%), 4.4 (5.6%), 4.7 (21%), 4.8 (20.5%), 4.9 (21.7%), 4.10 (25%), and 5.2 (10.5%). Vickers remarks that ". . . since the mean total for Marlowe never exceeds 3.7 per cent, it is evident that he had no hand in this play" (p. 55). But a mean average is not the salient metric here, since the values for these scenes range over more than an entire order of magnitude (2% to 25%). The highest values found in this play, those over 20%, are as anomalous for Shakespeare as for Marlowe, according to Timberlake's counts.

    The mean average rate of feminine endings for Love's Labour's Lost given in Timberlake's book is 6% (p. 98), the mean average for Romeo and Juliet is 6.3% (p. 105), the mean average for A Midsummer Night's Dream is 5.6% (p. 106), and the mean value for 1 Henry 4 is 4.3% (p. 111). In no scene in these four plays does Shakespeare come near the feminine-ending rate of 20-25% found in 2 Henry 6 -- in fact he seldom exceeds 10% -- except in scene 3.4 of Romeo and Juliet in which Timberlake counts 8 of the 23 lines as having feminine endings (34.7%).

    Obviously, rates of feminine ending use in short scenes will be subject to larger random fluctuations than rates in long scenes. In the immediately preceding scene in Romeo and Juliet, 3.3, Timberlake counts just 5 feminine endings across its 128 lines (3.9%), but this difference between them -- 3.9% versus 34.7% --  should convince no one that scenes 3.3 and 3.4 of Romeo and Juliet have different authors.

    To be a useful metric for authorship attribution, something we count must occur at a broadly consistent rate across the works in a candidate's canon and occur at markedly different but also broadly consistent rates in other candidates' canons. The rates of feminine-ending use in early modern drama meet neither of these criteria. Only by selectively picking from Timberlake's data is Vickers able to give the impression that Shakespeare must have written all of 2 Henry 6.

    Vickers devotes the final page and a half of his essay to a sketch of Ants Oras's work on where pauses fall within verse lines, mainly by direct quotation of 448 of Oras's words. Vickers claims that "The graphs that follow" (p. 58) support his claim about the authorship of 2 Henry 6, but in the PDF version of the essay downloaded by this reviewer no graphs follow. Instead a note reads "[For Oras's graphs see the attached Appendix (pdf)]"; of course a PDF cannot have an attachment. The essay being online, the reviewer took a look around the folder where the essay itself resides, <https://www.etrap.eu/wp-content/uploads/2016/12/>, but found nothing that looked like Oras's graphs. Instead the folders appears to contain photographs of Vickers delivering his essay as a talk in Germany in 2016.

    Another previously overlooked article must be noticed now (R. Aljumily 'Hierarchical and Non-hierarchical Linear and Non-linear Clustering Methods to 'Shakespeare Authorship Question'', Social Sciences 4.iii[2015] 758-99). Aljumily starts with an overview of the history of stylometry for authorship attribution that is accurate about what it covers but omits mention of the key contributions by Frederick G. Fleay, Timberlake, Oras, Lake, Jackson, Roger V. Holdsworth, and Ward E. Y. Elliott and Robert J. Valenza.

    Then comes the description of the method to be used here. Aljumily's corpus is 9 works by Francis Bacon, 6 plays by Shakespeare (5 histories and a tragedy), 7 plays by Fletcher, 7 plays by Marlowe, 4 plays by Kyd, and "nine disputed works" (p. 766). The 6 plays by Shakespeare are Cymbeline, 1 Henry 4, 2 Henry 4, Henry 5, King John, and Richard 2 (apparently Aljumily considers Cymbeline a history play), and the 9 disputed plays are Edward 3, Hamlet, 1 Henry 6, 2 Henry 6, 3 Henry 6, Henry 8, King Lear, Richard 3, and Titus Andronicus.

    It is not clear why Aljumily thinks it possible to comment on the theory that someone other than the man from Stratford-upon-Avon wrote the Shakespeare plays using experiments that treat Hamlet, Richard 3, and Titus Andronicus as less securely known to be by Shakespeare than the 6 plays he chooses for his Shakespeare corpus. That is, the premises of his experimentation are unclear. As well as the uncontroversial attributions of The Spanish Tragedy and Solimon and Perseda to Kyd, Aljumily gives him Arden of Faversham and the 1605 anonymous Part One of Jeronimo, with the Wars of Portugal, which purports to be a prequel to The Spanish Tragedy but which almost no one believes is by Kyd.

    Aljumily seems not to know that Shakespeare's history plays are his least distinctive works and hence ought not to dominate a 'known Shakespeare' set used to provide a ground truth. Aljumily also seems unaware of the standard reference works for dating plays, relying on "University of Virginia Library and The Project Gutenberg E-Book library" (p. 768) rather than the most up-to-date sources, which are Alan B. Farmer and Zachary Lesser's online Database of Early English Plays hosted by the University of Pennsylvania and Wiggins's Catalogue of British Drama from Oxford University Press.

    The method employed in this study was to search the corpus to count the frequencies of "135 function words, 100 word bi-grams, and 24930 letter tri-grams" (p. 768). The counts from the 42 plays produced three matrices: one containing 42 rows 135 columns for the function words, one containing 42 rows 100 columns for the word bigrams, and one containing 42 rows 24,930 columns for the letter trigrams. Aljumily claims that this process resulted in "42 vectors in a high dimensional space" and that the frequency counts for the three kinds of feature "were stored in these vectors" (p. 768). Aljumily means that the first matrix was treated as representing coordinates in 135-dimensional space so that each of the matrix's rows (one per play) contains the set of 135 numbers needed to define a single point in that 135-dimensional space.

    So where do "vectors" come into this? A vector is a line in space that has a direction (that is, it points somewhere) and a magnitude (its length). It takes two coordinates, the start and the finish point, to specify a vector. If we choose as the start point the origin (x = 0, y = 0, z = 0 in three-dimensional space) and a given set of coordinates as the end point, then the coordinates can be said to specify the vector that runs from the origin to the point that the coordinates define. The same applies for the 135 coordinates that define a point in 135-dimensional space, although our minds cannot easily visualize such a space.

    Aljumily reports that the value in each cell in each matrix is "the number of times" (p. 768-9) that a particular feature appears in a given text, and later describes how he normalized these absolute counts to compensate for the texts being of different lengths. First, however, he discarded the columns that have the lowest values so that only the most frequent features remain, leaving 60 function word, 30 word bigrams, and 40 character trigrams, which he helpfully lists. To normalize for varying document lengths, Aljumily multiplied each cell's value by the mean-average of all the texts in the corpus and then divided the result by the total of the feature counts for the row (representing one text) that the cell is in.

    Once the numbers for each text are normalized, it is possible to treat them as the coordinates of points in multi-dimensional space and see which texts' points are close in space to which other texts' points. Texts whose points cluster with other texts' points are, by definition, alike in their scores for the features being tested. Curiously, and for reasons that Aljumily does not make clear, there appears to be an inscrutable aspect to the software that he uses to seek 'clustering' of the texts' points: "In the current application, the function that generated DFW, Dbigram, and Dtrigram may not be known, and the strong suspicion must be that the generating function was nonlinear, but this is not certain" (p. 772).

    Previously Aljumily used DFW, Dbigram, and Dtrigram as labels for the three matrices he created -- one for function words, one for word bigrams, and one for letter trigrams -- so it is unclear what he means by "the function that generated" each of them: he generated them himself by the processes he describes. Possibly he means that some kinds of authorial functions, that is some habits of writers' style, shaped the numbers in these matrices and that these functions are uncertain. The phrase "In the current application" is also mysterious: he presumably means 'in the uses to which I am here putting these methods'. As is often the case in such studies, obscurity in the explanation of the investigator's method casts doubt over the entire investigation.

     Aljumily describes the various mathematical processes by which the 'clustering' of the 42 points representing the 42 texts was measured and visualized, including Hierarchical Cluster Analysis, which produces dendrograms that look like family trees of relatedness, Principal Component Analysis, which reduces points in high-dimensional space to points in two-dimensional space for plotting on an x/y scatterplot, the technique called Self-Organizing Map Unified Distance Matrix (SOM U-Matrix), which visualizes in a kind of 'heatmap' the distances between neurons in a neural network 'trained' on the input data, and the technique called the Voronoi Map which shows a tessellation (a map of irregular tiles) that for a set of points on a two-dimensional plane depicts the regions on the plane that are nearer to each point than they are to any of the other points.

    It is not clear why Aljumily chose these methods of visualization -- he offers no rationale -- and the impression given is of an investigator simply trying whatever visualizations are offered by the software he is using. Aljumily does not mention that the SOM U-Matrix relies on training a neural network on the experimental data and seems to use the word "lattice" (p. 774) to refer to the neural network when describing how the method works.

    Aljumily presents the visualizations he created. The captions for the various figures, and the absence of instructions on how to 'read' them, suggest that Aljumily does not understand what they show. For instance, Figure 4 has a list of 15 plays along its y axis but no label on its x axis and is simply captioned "An assessment of clustering tendency test for Shakespeare DFW", with no mention of which method it derives from. In the section of the article in which he describes the results of the analysis -- which at first seems to show that Hamlet and King Lear are not Shakespearian in style (p. 776) -- Aljumily introduces the idea that maybe the variable genre confounds his results and he applies a solution: ". . . we added two plays of a similar genre and the same time period into the corpus". These two plays are Fletcher and Philip Massinger's The False One and John Webster's The White Devil.

    It should be obvious that it is illegitimate to modify the corpus being tested while the experiments are being run, and that doing so is not a reasonable response to the initial results seeming wrong (Hamlet and King Lear not looking like Shakespeare). Aljumily presents a further 26 visualizations arising from his experiments, none of them adequately described. Typical conclusions include "This experimental result is suggestively significant, but not enough to draw firm conclusions for the problem in question" (p. 776) and "Given the lack of congruence, the only possible result to suggest is, therefore, that DisLear [King Lear] is not excluded from the possibility of having another author's or a collaborator's style" (p. 786).

    Aljumily's set of supposedly disputed texts includes the sole-authored Shakespeare plays Hamlet, King Lear, and Richard 3, and the co-authored Shakespeare plays 1 Henry 6, 2 Henry 6, 3 Henry 6, Henry 8, and Titus Andronicus, while his supposedly known-to-be-by-Shakespeare set is Cymbeline, 1 Henry 4, 2 Henry 4, Henry 5, King John, and Richard 2. By what he calls his "validated and fully objective and replicable mathematically-based methodology" he finds that ". . . the hypothesis that Shakespeare is the author of the disputed plays traditionally attributed to him is falsified in favor of alternative author(s)" (p. 794). Aljumily claims that these alternative authors cannot be sought by his method, and seems not to notice that his method was itself entirely concerned with testing whether the disputed plays are more like the known work of Shakespeare than the known works of Bacon, Fletcher, Marlowe, or Kyd, the other authors represented in his corpus.

    Aside from the many methodological problems described above, the key problem with this analysis is that Aljumily devised his set of "disputed" and "known Shakespeare" with no explicit rationale of selection. If the man from Stratford-upon-Avon was the fraud that the anti-Stratfordians believe, then the plays in Aljumily's "known Shakespeare" set cannot be trusted any more than the "disputed" plays can, and the approach he takes would on principle be unable to prove or disprove the claim he sets out to test.

    The last of this year's essays on authorship attribution shows that Henry 8 was written by Shakespeare and Fletcher (not Massinger) and that Thomas Merriam's tweaks to the accepted boundaries between the co-authors' stints are right (Petr Plecháč 'Relative Contributions of Shakespeare and Fletcher in Henry VIII: An Analysis Based on Most Frequent Words and Most Frequent Rhythmic Patterns', arXiv 1911.05652[2019] n. pag.). Plecháč starts with a survey of the claims and evidence that Henry 8 was co-authored by Shakespeare and Fletcher. Because some previous studies have also pointed to Massinger, Plecháč considers his possible contribution too. The method was to count the frequencies of the 500 most frequent words (which is a standard approach) and also to count "the frequencies of 500 most frequent rhythmic types" (p. 3), which is unusual.

    By rhythmic types Plecháč means patterns of stress in a line, which can be represented by a string of binary digits so that the regular iambic pentameter "the VIEW of EARTHly GLORy MEN might SAY" is represented as 0101010101 (where "0" means an unstressed syllable and "1" means a stressed one), while "till this TIME POMP was SINgle, but NOW MARRied" is represented as 00110100110. Plecháč's source texts were the XML encodings from the Early Print project at Washington University of St Louis and to derive the rhythmic patterns he used the open-source software called Prosodic, which has a really quite remarkably high success rate at coming to the same decisions as human experts about where the stresses fall in a line. (Not that humans necessarily agree, of course: the second example above could alternatively be stressed as "till THIS time POMP was SINgle BUT now MARRried".)

    To take the numbers produced by counting these phenomena and derive classifications (such as 'by Shakespeare' and 'by Fletcher') from them, Plecháč uses the machine-learning method of Support Vector Machines (SVM). On a two-dimensional scatterplot of x/y values we might find that all the dots representing data about one author's works fall into a cluster in one corner of the picture and all the dots representing data about another author's works fall into a cluster in another corner of the picture. In order to decide which of these two authors we should attribute a work of unknown authorship to we need to define a boundary between the clusters.

    We might calculate a line representing the best boundary between these two clusters by drawing a line that joins the two centres of the two clusters and then bisecting this line and projecting a new line, our boundary, at 90 degrees from this middle. (This is the method used in connection with the Zeta classifier, as critiqued above by Rizvi.) There are other ways to define the best boundary between clusters and in two-dimensional SVM we find the widest 'channel' (defined by two parallel lines) that can be run between the clusters so that no point in a cluster falls into the channel; the boundary is then the line running down the centre of that channel. In other words, we maximize the gap between the boundary and the nearest data points on either side of it. The same process can be applied in multi-dimensional space, which arises from counting multiple features of each text, where instead of lines we have parallel planes forming the channel and the boundary in its middle.

    As his sample Shakespeare plays Plecháč uses 53 scenes from the late plays Coriolanus, Cymbeline, The Winter's Tale, and The Tempest. For Fletcher he uses 90 scenes from Valentinian, Monsieur Thomas, The Woman's Prize, and Bonduca. For Massinger he uses 46 scenes from The Duke of Milan, The Unnatural Combat, and The Renegado. Plecháč ran cross-validation in which the classifier was trained on scenes from all the plays except one and then used the classifier to attribute the authorship of scenes from that one held-out play. (So, never did the validation of a scene's classification use in its training some other scenes from the same play.)

    The validation was run 30 times with random selections of scenes for training and classification and with the two tests (frequencies of function words and frequencies of rhythmic patterns) used as separate classifiers and also used together as one composite classifier. The results are remarkably good: used in combination the tests got the authorship attribution right between 96% (worse case) and 100% of the time. The word frequencies turned out to be more reliable than the rhythmic frequencies and the worst results were obtained using Massinger's rhythmic frequencies.

    Because Plecháč has more training scenes than he needs to build a classifier he built 30 classifiers by drawing scenes randomly from the Shakespeare, Fletcher, and Massinger sets. This allowed him to run his tests 30 times -- once per classifier and each time based on different combinations of training scenes -- and treat each result as a 'vote' he could count. The conclusion is that Massinger had nothing to do with writing Henry 8. James Spedding's division of the play between Shakespeare and Fletcher was right except that Plech c goes against Spedding in finding 3.2 as Shakespeare not Shakespeare+Fletcher, and 4.1 as Shakespeare not Fletcher. (But these results are refined in a second experiment.)

    Since we know that authorial stints do not necessarily coincide with scene breaks, Plecháč next tried a rolling-window approach. He used 100-line windows advancing 5 lines at a time and rather than give a simple verdict for each section tested he used his classifier to generate a probability distribution over all the candidates. In validation using 8 plays, 4 by Shakespeare and 4 by Fletcher, this method gave strong distinctions between the authors. Applied to Henry 8, the method showed clearly where the authorship flips from Shakespeare to Fletcher and from Fletcher to Shakespeare. The split was this:

Shakespeare 1.1, 1.2
Fletcher 1.3, 1.4, 2.1, 2.1, 2.2
Shakespeare 2.3, 2.4
Fletcher 3.1
Shakespeare then Fletcher 3.2. (The crossover is around TLN 2200 as Merriam claimed.)
Shakespeare then Fletcher 4.1, 4.2. (The crossover is not at the scene boundary.)
Shakespeare 5.1
Fletcher 5.2, 5.3, 5.4, 5.5. (There is some slight evidence of Shakespeare in 5.4.)

There are three articles by Merriam that Plecháč's conclusion somewhat corroborates: "Taylor's Method Applied to Shakespeare and Fletcher" and "Though This Be Supplementarity, Yet There is Method In't", both reviewed in YWES for 2003, and "Henry VIII, All Is True?" reviewed in NYWES for 2018.

    Five articles this year are on matters textual but not primarily about authorship. MacDonald P. Jackson shows that he was right in his claim of nearly 50 years ago that Sonnets (1609) was set by two compositors working in the shop of its printer George Eld, to judge from its two distinctively different patterns of punctuation ('Bibliographical Principles and George Eld's Quarto of Shakespeare's Sonnets', The Library 20[2019] 216-23). An edition of Sonnets by Carl D. Atkins (overlooked by YWES for 2007) claimed that Jackson is wrong, referring readers to Atkins's essay "The Application of Bibliographical Principles to the Editing of Punctuation in Shakespeare's Sonnets" (reviewed in YWES for 2003).

    Jackson's evidence for two compositors in Sonnets was the presence of two incompatible habits of punctuation: one man (called Eld Compositor A by Jackson) preferred to end a quatrain with a colon rather than a comma, and another (called Eld Compositor B by Jackson) preferred, although not so strongly, to end a quatrain with a comma rather than a colon. Jackson found that the spelling preferences that led Alice Walker to hypothesize an Eld Compositor A and Eld Compositor B working on Eld's 1609 edition of Troilus and Cressida also aligned with alternations between contrasting punctuation habits in Sonnets: when the punctuation he was counting changed, the spellings changed too.

    Other typographical habits in Sonnets -- the locations of the sonnet-numbering headings and the signatures -- also match the punctuation and spelling transitions. This is most unlikely to be coincidence. Jackson also applied the so-called psychomechanical test of looking for spaces set or omitted after commas in short lines, and again found a correlation with his Compositor A/B distinction. Jackson acknowledges that other scholars have shown the limitations of deciding compositorial stints by spacing after punctuation (particularly D. F. McKenzie's celebrated essay "Printers of the Mind") and by spellings (particularly Pervez Rizvi's essay "The Use of Spellings for Compositor Attribution in the First Folio" reviewed in NYWES for 2016), but insists that where multiple independent features all point the same way the results are reliable. Moreover, subsequent studies have found fresh evidence for Jackson's attribution of compositor stints.

    Atkins thought that Jackson's compositorial stint assignments were incompatible with the setting of Sonnets by formes since it would require a change of compositor within a forme, but Jackson points out that such a change was entirely normal. Atkins thought it fatal to Jackson's attribution ot stints that quite a few of the sonnets start on one page (supposedly set by one compositor) and end on the next (supposedly set by the other compositor). Surely, Atkins reasoned, this would leave each compositor with no idea how to punctuate the poem since he had not seen and made sense of the whole of it.

    Jackson responds that with sonnets the compositors could easily just follow the rule that lines 4, 8, and 12 (the ends of quatrains) are likely to need punctuation heavier than a comma. After all, the same problem occurs in the breaking of speeches and even single sentences across pages and compositor stints in the printing of plays. And, this problem of sonnets running across page breaks applies even if only one man does the type setting. With setting by formes, he might well have to set the last few lines of one sonnet on one page long before he sees and sets the first lines of the same sonnet on another page.

     Jackson is also able to point to a new piece of evidence adduced by Atkins that actually supports Jackson's case. The penultimate line of Sonnet 7 on B2v ends with a colon, and Atkins points to lots of other pages containing sonnets that have a colon there. Jackson notes that this habit is strongly correlated with Eld Compositor A: he set half as many pages as Eld Compositor B but set 15 of the 17 penultimate lines that have this feature.

    Misha Teramura argues that William Warburton's reordering of the lines of the Fool's prophecy in King Lear is correct ('Prophecy and Emendation: Merlin, Chaucer, Lear's Fool', Postmedieval: A Journal of Medieval Cultural Studies 10[2019] 50-67). The prophecy appears only in the Folio version of the play and goes like this:

When Priests are more in word, then matter;
When Brewers marre their Malt with water;
When Nobles are their Taylors Tutors,
No Heretiques burn d, but wenches Sutors;

When euery Case in Law, is right;
No Squire in debt, nor no poore Knight;
When Slanders do not liue in Tongues;
Nor Cut-purses come not to throngs;
When Vsurers tell their Gold i'th'Field,
And Baudes, and whores, do Churches build,

Then shal the Realme of Albion, come to great confusion:

Then comes the time, who liues to see t,
That going shalbe vs'd with feet.

The blank spaces between lines shown above are not present in the early edition: they are added here to help illustrate the structure that Teramura discusses. The prophecy comprises 4 lines describing morally bad behaviour, 6 lines describing morally good behaviour, and then two contrasting conclusions arising from that behaviour, one catastrophic and one mundane.

    Teramura claims that the prophecy is a response to 'Geoffrey Chaucer's Prophecy', "widely credited as the medieval poet's dire warning for England" (p. 51). Warburton moved the mundane conclusion to the end of the first description, that of morally bad behaviour, and left the catastrophic conclusion at the end of the second description, that of morally good behaviour. Teramura surveys other responses that did the reverse, putting the catastrophic conclusion at the end of the morally bad behaviour and the mundane conclusion at the end of the morally good behaviour.

    What matters is whether one thinks this really is a prophecy, a warning of what will happen if society turns bad, or else is an ironic statement of what follows because society already has turned bad. And, adding a further complication, the play's theme of topsy-turvydom gives reason to think that the Fool might in any case say the opposite of what we expect. Some editions move no lines around, taking the view that the speech's disorganized structure is part of its point.

    Teramura quotes the prophecy printed by William Caxton at the end of Chaucer's Anelida and Archite, which got taken by subsequent editions as part of the Chaucer canon. This prophecy is clearly some kind of a source for the one in King Lear. That the prophecy really was Chaucer's writing was, on the evidence presented here, widely believed at the time. Teramura surveys responses to the prophecy in King Lear and gives her reasons for thinking Warburton's emendation correct, since it provides "arguably the most compelling intertextual engagement with its source and the darkest political message" (p. 60).

    Putting the mundane conclusion after the morally bad behaviour creates an intentional anti-climax. This makes the point that things already are morally bad. Putting the catastrophic conclusion after the list of morally good behaviour indicates that "so thoroughly are these vices woven into the fiber of Albion's identity that to imagine the state without them is necessarily to imagine Albion undone" (p. 63).

    Heidi Craig traces the major trends in the publication of Shakespeare's plays in the middle of the seventeenth century ('Missing Shakespeare, 1642-1661', ELR 49[2019] 116-44). There were only three editions of Shakespeare during the Interregnum: the quartos of The Merchant of Venice (Q3 1652), Othello (Q3, 1655), and King Lear (Q3 1655). Craig's theory is that with no chance to see new plays in the theatre, publishers reckoned that readers would want to read previously unpublished plays to satisfy their thirst for new works, so instead of reprinting Shakespeare they focussed on new first editions.

    The long gap between the Second Folio of 1632 and the Third Folio of 1663 was due in part to a rights dispute between, on one hand, Mary Allot, widow of Second Folio publisher Robert Allot, and Philip Chetwind, Mary Allot's second husband and Third Folio publisher, and on the other hand the stationers Andrew Crooke (Robert Allot's former apprentice) and John Legate. Rights in the 16 plays first published in 1623 First Folio were co-owned by Isaac Jaggard and Edward Blount, and Craig details how the inheritors of these men came to dispute the right to reprint the Folio.

    Craig traces the various Shakespeare rights holders in the 1620s-1660s. Lots of playscripts got sold to publishers with the Interregnum closure of the theatres. Publishers needed playscripts only for previously unpublished plays, since for previously published plays they could use a previous edition, but had to negotiate the rights with its holder. The market had been flooded with Shakespeare in the previous decades and there was little appetite to reprint his plays, and apparently quite a lot of unsold stock left over from earlier editions.

    Humphrey Moseley was the Interregnum's most active publisher and "single-handedly shaped the era's field of printed drama" (p. 129). Moseley's 1647 Beaumont and Fletcher Folio made much of the fact that none of its contents had been printed before. Moseley printed a series of New Plays collections by various dramatists -- Richard Brome, James Shirley, Massinger, Middleton, and Lodowick Carlell -- because novelty was his speciality. We have this drive for novelty to thank for much of the drama surviving at all.

    William Leake's Q3 The Merchant of Venice of 1652 appears to attempt to capitalize on parliamentary debate about the readmission of Jews to England. And the timing of Othello Q3 of 1655 may have been judged to coincide with a major English victory over Barbary corsairs. Craig thinks that Iago's joke to Brabantio that "you ll have coursers for cousins" in the first scene is a pun on 'corsairs'. Leake's title page provocatively displayed his printer's mark, a crown of the old open style. When Jane Bell printed King Lear in 1655 she accidentally infringed the rights of Miles Flesher: she had the rights to King Leir not Shakespeare's play. Craig traces the complexities of the rights to King Leir, and other cases of stationers getting the two plays mixed up.

    Craig considers some misattributions in two lists of plays from 1656. Shakespeare is the most misattributed author by far. This cannot be because Shakespeare's name was thought to sell books: no title pages in the period 1642 to 1660 misattribute plays to Shakespeare, whereas a dozen misattribute plays to other dramatists. Some of the lists' misattributions to Shakespeare simply echo earlier title-page misattributions to him. Others seem to be misattributed just because Shakespeare was the archetypal origin for certain kinds of old plays.

    In the course of explaining why the Latin scene, 4.1,  in The Merry Wives of Windsor is so funny, B. J. Sokol adds to the body of evidence that the play was likely composed in the year 1600 ('The 'Rule of Three' and the 'Callback': How Comic Form in The Merry Wives of Windsor 4.1 May Help to Date its Folio Text', BJJ 26[2019] 97-112). The humour comes from the comedic rule-of-three and the use of a 'callback' -- returning in one joke to an element of an earlier joke, also known as 'using a hook' -- which together make this scene a masterclass in comic writing. Sokol turns to the French lesson in Henry 5 3.4, which also depends on "mistaken perceptions of only apparent bilingual cognates" (p. 104).

    Sokol finds The Merry Wives of Windsor version of essentially the same interaction to be the more comically sophisticated, which puts the date of The Merry Wives of Windsor after that of Henry 5. This is because ". . . it was characteristic of Shakespeare's career not only that he often later revisited themes or motifs that he'd used before, but also that his later versions typically excelled his earlier ones in terms of structural complexity and characterological depth" (p. 105). Thus if Henry 5 is dated after March 1599 by its Chorus's reference to Essex's expedition, The Merry Wives of Windsor is later still.

    Two relevant chapters in collections of essays were published this year, and both came from the same book. In the first, Patrick Spottiswoode offers three notes to the text of Othello that arise from his editorial work in producing an edition for schools ('Othello: Three Notes for Dr Ralph to Query', in Menzer & Cohen, eds. Shakespeare in the Light: Essays in Honor of Ralph Alan Cohen, pp. 63-70). In the first, Spottiswoode observes that Shakespeare makes much more of the Venetians' arrival in Cyprus than does his source Cinthio (Giovanni Battista Giraldi). Cyprus was the home of Venus and Spottiswoode sees Desdemona as a kind of Venus conquering her Mars in that upon arrival Othello ignores the Governor of Cyprus, Montano, to speak at length to Desdemona.

    Iago and Cassio independently notice that Desdemona has overmastered Othello: "Our general's wife is now the general" and she is "our great captain's captain". Spottiswoode suggests that Cassio's speech about how the wild seas, for love of her beauty, abstained from wrecking the boat that brought Desdemona to Cyprus ("Tempests themselves . . . letting go safely by | The divine Desdemona") is indebted to Edmund Spenser's description of Venus: ". . . thy smyling looke doest pacifie | The raging seas, and makst the stormes to flie; | Thee goddesse, thee the winds, the clouds doe feare, | And when thou spredst thy mantle forth on hie, | The waters play and pleasant lands appeare" (The Faerie Queene

    In the opening scene, Iago tells Roderigo to wake Brabantio with news of his daughter's marriage to Othello and "Though that his joy be joy, | Yet throw such chances of vexation on 't | As it may lose some colour". That is, Brabantio should be happy to have such a husband for his daughter, but Roderigo should make the event less attractive to him; thus "lose some colour" is roughly equivalent to the modern phrase 'take the shine off'. The quarto and Folio agree on the reading "loose some colour" but Spottiswoode wonders if "colour" is a misprint for 'choler', since 'loose some choler' (in the sense of 'loose' as 'release') "is much more in keeping with what Iago seeks to achieve throughout the play" (p. 68). Spottiswoode's last suggestion is that the handkerchief is so important to Desdemona and to the action of the play that Desdemona should be visibly carrying it for her entrance in the Senate Scene, 1.3.

    The second relevant chapter from this book concerns the moment in Macbeth when Banquo's Ghost turns up to the feast and Macbeth raves about "such sights", causing Ross to ask him what he means (George Walton Williams 'Reading Lady Macbeth's Line, 3.4.117', in Menzer & Cohen, eds. Shakespeare in the Light: Essays in Honor of Ralph Alan Cohen, pp. 29-32). The Folio prints this moment as:

    Rosse.  What sights, my Lord?
    La. I pray you speake not: he growes worse & worse
Question enrages him: at once, goodnight.

It is usual to suppose that Lady Macbeth's "I pray you speake not" is directed at Ross, but Williams reports a production at the Blackfriars replica theatre in Staunton Virginia in which Macbeth tried to answer Ross's question and Lady Macbeth put her hand over her husband's mouth and said to him "I pray you speake not". Williams thinks that this moment parallels that of Lady Macbeth's earlier feigned swoon by which she cut off Macbeth's account of his killing of Duncan's servants. In both moments, Lady Macbeth intervenes to save her husband from saying too much.

    And so to relevant pieces in the journal Notes and Queries. Three are by Thomas Merriam. In the first, Merriam adds to the body of evidence he has been collating to show that Shakespeare contributed not only the Additions to the play Sir Thomas More but also the original version that survives in Anthony Munday's handwriting (''Six-word Collocations in Shakespeare and Sir Thomas More' -- Revisited', N&Q 264[2019] 415-6). Merriam builds on his essay 'Six Word Collocations in Shakespeare and Sir Thomas More' (reviewed in YWES for 2009). In his Arden3 edition of the play, John Jowett rejected this evidence because, variously, the matches were not unique to Sir Thomas More and Shakespeare or the matches were inexact or because the dating allowed the author of Sir Thomas More to have simply read the matching phrases in published Shakespeare works.

    Now, using Pervez Rizvi's database on collocations across all the drama, Merriam has found a further 8 five-word matches, 1 six-word match, and 1 eight-word match unique to Sir Thomas More and Shakespeare that are not possibly the result of someone simply borrowing from Shakespeare's published works. Merriam claims that the matches are word-for-word and here they are with punctuation removed and with, after each one, my verdict on it:

"that did I my lord" in Sir Thomas More and "that I did my lord" in Hamlet. This is not a word-for-word match.

"with him for he will" in Sir Thomas More and Julius Caesar  I agree that this is unique.

"to the law but God" in Sir Thomas More and 2 Henry 6. I agree that this is unique but the passage in 2 Henry 6 appears in Scene 3, a likely Marlowe section.

"God save the king God save the king" in Sir Thomas More and 2 Henry 6. I agree that this is unique and this section of 2 Henry 6 (Scene 20) is probably Shakespeare's.

"in the behalf of a" in Sir Thomas More and As You Like It. I agree tha this is unique in the drama, but Munday himself uses this phrase in The First Book of Primaleon of Greece in 1595 (STC 20366, sig. Aa1r).

"the mayor of London and" in Sir Thomas More and 1 Henry 6. I agree that this is unique but this section of 1 Henry 6 (Scene 1.4) is not by Shakespeare and the phrase occurs in a stage direction whereas in Sir Thomas More it is dialogue.

"come hither come hither come hither" in Sir Thomas More and As You Like It. I agree that this is unique, but in As You Like It it is part of a song, "Under the Greenwood Tree", that may have been widely known before Shakespeare used it.

 "good morrow to the sun" in Sir Thomas More and Cymbeline. I agree that this is unique.

"speak my mind I think" in Sir Thomas More and 2 Henry 6. I agree that this is unique and the part of 2 Henry 6 (Scene 9) is probably Shakespeare's.

"good night good night God" in Sir Thomas More and Othello. I agree that this is unique.

Merriam does not offer a comparison for how often we would find such unique collocations-in-common between works that we are sure are not by the same writer, which would give us a baseline for judging how significant it is that we find these ones. It is known to be the case that any two long works will have some collocations-in-common that no other writing has, simply because that is an inherent quality of long texts.

    Merriam does, however, point out that no such collocations-in-common are found between the Munday-handwriting original text of Sir Thomas More and the known works of Munday, nor between Sir Thomas More and Chettle's The Tragedy of Hoffman, which is relevant since Chettle is sometimes proposed as Munday's co-author on Sir Thomas More, as he is in the New Oxford Shakespeare. Merriam offers some further unique collocations-in-common where the matches are not quite word-for-word or else the Shakespeare play was in print so someone could have got the phrase from there. (Or, I would add, from hearing it in performance.)

    In his second note, Merriam observes that several distinctive phrases in the preface entitled "To the Great Variety of Readers" in the 1623 Shakespeare Folio that is signed by John Heminges and Henry Condell are found also in Jonson's works, as scholars have previously noticed ('Doubtful Authorship of the Preface of the 1623 Shakespeare First Folio', N&Q 264[2019] 464-6). Merriam asserts that some of them are unique to those preliminaries and the Jonson works, but he does not say what searches he did to reach this conclusion. Merriam traces the remarks of the various critics who noticed this before him.

    In his last note this year, Merriam remarks that in Othello Cassio tells Montano that Othello is married to a woman who "in th'essentiall Vesture of Creation, | Do's tyre the Ingeniuer", and that this last word is a problem ('A Crux in Othello', N&Q 264[2019] 447-50). The sense of modern 'engineer' is not quite right for one who creates descriptions praising a woman, which the preceding context demands. Also, 'tyre' seems to mean not only 'to exhaust' but also links back as 'attire' to 'Vesture' meaning clothing. Merriam thinks the line alludes to the Incarnation by means of the Virgin Mary, who clothes in flesh, as it were, God the creator. The idea is like Pericles's description of Marina in his remark that she "begett'st him that did thee beget".

    Florence Hazrat proposes that an idea in The Taming of the Shrew comes directly from the source play and that some surprising speech assignments that editors routinely emend may in fact be correct ('The Taming of the Shrew and Editorial Engagement: New Evidence from English Sources and Continental Adaptations', N&Q 264[2019] 410-5). The wooing-of-Bianca plot in The Taming of the Shrew comes from George Gascoigne's play Supposes (first performed 1566), which is a translation of Ariosto's Italian play I Suppositi (first performed 1509). It has not been noticed before that Shakespeare directly lifts Gascoigne's association of the musical scale with beating someone: "I will teach you to sing sol fa" is the threat in Supposes and in The Taming of the Shrew Petruccio threatens Grumio with "I'll try how you can sol-fa and sing it".

    A Dutch-language translation of The Taming of the Shrew was published in the Netherlands in 1654 and a German-language adaptation of the play appeared in 1672. In Folio The Taming of the Shrew the following lines are routinely treated as if they are assigned to the wrong speakers:

How fiery and forward our Pedant is,
Now for my life the knaue doth court my loue,
Pedascule, Ile watch you better yet:
In time I may beleeue, yet I mistrust.

Mistrust it not, for sure Aeacides
Was Aiax cald so from his grandfather.

I must beleeue my master, else I promise you,
I should be arguing still vpon that doubt,
But let it rest, now Litio to you:
Good master take it not vnkindly pray
That I haue beene thus pleasant with you both.

Modern editions usually give the first three lines, "How fiery . . . better yet", to Hortensio (who is pretending to be Licio the music teacher) on the grounds that Lucentio is the one pretending to be a Latin teacher (a Pedant or Pedascule). Then they give the line "In time I may beleeue, yet I mistrust" to Bianca and reassign "Mistrust it not . . . from his grandfather" to Lucentio, since as a Latin teacher he would best know the classical myth alluded to. Then they give the last speech to Bianca, since Hortensio-as-Licio can hardly say "now Litio to you".

    Hazrat disputes the reattribution of these lines as diminishing the wit of Bianca, who is capable of Latin allusions, and she maintains that Lucentio should say "I mistrust" about Bianca rather than her saying it about him. Hazrat offers in support of retaining the original Folio distribution of speeches the fact that the Dutch version has it. (I would have thought that it might simply have inherited this error from the Folio.)

    Hazrat makes the same argument about another group of lines wrongly assigned in the Folio, the first 10 of 4.2. She acknowledges that the reassignment of these lines that appears in modern editions first appeared in the 1632 Second Folio (which she misnames the "second quarto", p. 414) and also appears in the Dutch and German versions, from which she deduces that these are based "on a text later than 1631" (p. 414).

    In fact, they could be based on an independent good text of the play which lacked the First Folio's error, since the sharing of correct readings does not indicate a common source; only shared incorrect readings can do that. Hazrat sounds less convinced about retaining the Folio speech assignments in the second case than she does in the first, referring twice to the productive "ambiguity" (p. 415) that arises from leaving them unemended rather than defending them as actually correct.

    John V. Nance argues that links between Hamlet and plays from the 1590s do not disprove the claim, made in the New Oxford Shakespeare, that Shakespeare first wrote the play in the 1580s ('Kyd, Johnson, and the Problem of an Early Date for Q1 Hamlet', N&Q 264[2019] 435-40). Wiggins in his Catalogue entry for Hamlet identifies three verbal links between Q1 Hamlet and materials not available in print in 1588-89 when the New Oxford Shakespeare claims that the play underlying Q1 Hamlet was written. These are phrases in Kyd's The Spanish Tragedy (printed in 1592), Marlowe and Nashe's Dido Queen of Carthage (printed in 1594), and Jonson's Every Man in His Humour (printed in 1601). Nance deals with the Dido Queen of Carthage link elsewhere -- in a contribution to a special issue of the journal Critical Survey reviewed above -- so this essay is about the Kyd and Jonson links.

    The lines in Q1 Hamlet are Gertred's "I will conceal, consent, and do my best | What stratagem soe er thou shalt devise" and the corresponding lines in The Spanish Tragery are "And heere I vow, so you but give consent, | And will conceale my resolution" and "I will consent, conceale| And ought that may affect for thine availe, | Joyne with thee to revenge Horatioes death". But since The Spanish Tragedy was first performed in 1587 or earlier, Shakespeare perhaps heard the 'consent'/'conceal' collocation in performance and got it from there.

    The Jonson link consists of Every Man in His Humour having "out of the ayre . . . How simple, and how subtill are her answeres?" which sounds like Q1 Hamlet's "How pregnant his replies are . . . Will you walk out of the air, my lord? . . . shrewd answers". Every Man in His Humour was first performed  in 1598 so Shakespeare writing in 1588-89 cannot have heard it. But Nance argues that in context the "composition of thought" is not the same in the two plays, and there is a "difference in the nature of expression" (p. 438).

    Nance points out that the verbal link regarding these phrases is closer between Every Man in His Humour and Q2 Hamlet than between Every Man in His Humour and Q1 Hamlet. He suggests that in revising Hamlet to make the Q2 version, Shakespeare remembered the phrasing from Every Man in His Humour, especially since he may well have performed in Jonson's play, and he made Hamlet more like a humours play by strengthening its link to it.

    John Peachman suggests historical identities for the Young Man and the Dark Lady of Sonnets (''Surprised Into Sonneteering': Shakespeare's Sonnets and the Gray's Inn Revels of 1594-5', N&Q 264[2019] 418-28). The New Oxford Shakespeare chronology puts early the Young Man sonnets in 1594-5 when Shakespeare also wrote The Comedy of Errors, Love's Labour's Lost, Richard 2, and Romeo and Juliet. But Eliot Slater's work, corroborated by MacDonald P. Jackson, showed that while Love's Labour's Lost, Richard 2, and Romeo and Juliet have strong rare-word links with those sonnets, The Comedy of Errors does not.

    The earliest of Shakespeare's sonnets were the ones to the Dark Lady written in 1590-5, so what made him switch to writing about the Young Man in 1594? Peachman suggests that it was the Gray's Inn Revels of 1594-5 during which The Comedy of Errors was performed and that this occasion "surprised Shakespeare into sonneteering to a young man" (p. 420), specifically the Prince of Purpoole Henry Helmes.

    Peachman explains that the sonnets to the Young Man treat him with reverence (as if a monarch) but also in places with criticism (as if an ordinary man), and while Shakespeare could never insult a real aristocrat in that way the Prince of Purpoole was only a temporary monarch and could be treated thus. The core theme of the Gray's Inn revels was that of male friendship, and particularly this Inn's friendship with the Inner Temple. Peachman gives what is known of Helmes's life, including that around September 1598 he had a poem published in a book alongside one by Edmund Spenser; this is just when the Rival Poet sonnets were written.

    Helmes was also mentioned in the dedication of a book in 1599, and Peachman thinks this might be alluded to in "And therefore mayst without attaint o'erlook | The dedicated words which writers use | Of their fair subject, blessing every book" (Sonnets 82.2-4). The Gray's Inn revels on 20 December 1594 included "Lucy Negro", a notorious Clerkenwell prostitute, and Peachman favours Duncan Salkeld's suggestion that maybe she was the Dark Lady of the sonnets. Peachman admits, though, that the stylometric evidence puts composition of most of the Dark Lady sonnets before December 1594

    Also regarding biographical links in the works, Brett Jones believes that Innogen's "Richard du Champ" allusion in Cymbeline, the name she gives for her master whose dead body she lies beside, is not to Shakespeare's fellow Stratfordian Richard Field but to Richard Lichfield ('Richard Du Champ in Cymbeline: A Contemporary Reference?', N&Q 264[2019] 451-4). Field did use the Spanish name "Ricardo del Campo" in his later 1590s books, but why would Shakespeare use French for him over a decade later? And would not likening Richard Field to Cloten be insulting to Field? Jones suggests that in fact "Richard du Champ" refers to "the Cambridge Barber-Surgeon" (p. 451) Richard Lichfield who wrote a book called The Trimmings of Thomas Nashe in 1597 under the name "Don Ricardo De Medico Campo".

    Trimmings seems to have been read by Shakespeare, as he uses a few unusual words and phrases from it in his plays. But this is still 10 years before Shakespeare wrote Cymbeline. To form a closer link, Jones makes a tenuous set of connexions. Trimmings was a response to Nashe's Have With You to Saffron Walden. In 1609 a lost ballad called Have With You To Pimlico was entered in the Stationers' Register alongside a pamphlet that survives called Pimlico or Run Red Cap. Jones suggests that in Jonson's Epicoene the character Cutbeard -- a barber who (like Lichfield) uses Latin -- is a dig at Lichfield. Thus at the time of Cymbeline another playwright, Jonson, was also making a dig at Lichfield.

    In Epicoene, Morose hopes that Cutbeard will not have the "trimming" of any customers except chimney-sweeps, which links to Lichfield's book Trimmings and to the song in Cymbeline in which "chimney-sweepers, come to dust". For some reason, Jones misquotes the line from Cymbeline as "chimney-sweepers, turn to dust". Jones perhaps overstates his case that ". . . it seems more likely that Richard du Champ [in Cymbeline] would be a reference to Richard Lichfield than Richard Field" (p. 454)

    MacDonald P. Jackson offers evidence that the play Alphonsus, Emperor of Germany was probably written in the 1580s or 1590s ('The Date of Alphonsus, Emperor of Germany: The Evidence of Unique N-gram Matches', N&Q 264[2019] 512-4). The first edition of Alphonsus, Emperor of Germany is the 1654 quarto that attributes it to Chapman. The Stationers' Register entry of 1653 gives it to John (perhaps an error for George) Peele, and Francis Kirkman in his catalogues of 1661 and 1671 gives it to George Peele. The date of composition is uncertain; Peele died in 1596.

    Jackson uses Pervez Rizvi's Collocations and N-Grams dataset and details some of the terminology that arises from it, including the notion of 'formal' n-gram counting which includes in its counts the smaller n-grams found within larger ones. A key element is how we adjust for authors having different canon sizes. All else being equal, an author with a large canon is going to give us more words and phrases common across his canon and a text we are trying to attribute than will an author with only a small canon.

    Jackson notes Rizvi's attempt at a compensating formula to remove this bias: "Matches between plays are 'weighted' when their number is divided by the combined total number of word-tokens in the two plays being compared" (p. 513). In fact, there is no sound mathematical basis to Rizvi's weighting formula, which he has changed since first publishing his dataset. No one else has a better formula, though: the problem of how we adjust for varying canon sizes to get the most reliable attributions has not been addressed by any investigator.

    Jackson found that when using uniquely shared 3-grams in single plays by Chapman, Dekker, Field, Fletcher, John Ford, Greene, Heywood, Jonson, Kyd, Lyly, Marlowe, Marston, Massinger, Middleton, Peele, Shakespeare, Edward Sharpham, Shirley, and Webster each play shared most links with the other plays by the same author. The next strongest association was with other plays of the same period, and this is how he dates Alphonsus, Emperor of Germany.

    Using the plays with the top 12 shared 3-grams and those with the top 12 shared 4-grams links, Jackson finds that "Only three of the twenty-four dates are later than 1595" (p. 513), suggesting an early date for composition of Alphonsus, Emperor of Germany. No one author leaps out from these data, but an early date does. Using a series of other plays from 1629-1631, Jackson shows that they rarely (just 1 play in 72) share a lot of links with a play from before 1596. Thus if Alphonsus, Emperor of Germany were a Caroline play, we would expect it to share links with other Caroline plays and it does not. Therefore, no matter who wrote it, Alphonsus, Emperor of Germany is likely a play from the 1580-1590s.

    Perry McPartland thinks that in the first performances of The Winter's Tale the part of Autolycus was doubled with that of the bear impersonator ('References to the Doubling of Autolycus and the Bear in Shakespeare's The Winter's Tale', N&Q 264[2019] 454-7). McPartland refers to the opening scene of The Winter's Tale "in which a pair of sophisticated courtiers perform an ironic impersonation of stock-characters" and that it is "an impersonation which initially takes the audience in" (p. 454). I am not clear what impersonation he is referring to. I can imagine how such an interpretation of the opening scene could be argued for, but McPartland seems to think that this is a critical commonplace that he may merely allude to.

    Richard Proudfoot argued that Autolycus's line "my shoulder-blade is out" hints at his doubling with Antigonus, of whom the Clown reports that "the bear tore out his shoulder-bone". But since Autolycus goes on to say that he has been mugged by a notorious rogue -- "Some call him Autolycus" -- it would be fitting for the play's self-reflexivity if Autolycus's line about his shoulder-blade being out were being delivered by the same actor (that of Autolycus) who played the bear that tore out Antigonus's shoulder.

    Pretending to be someone else, Autolycus describes Autolycus as an "ape-bearer" which literally means a showman of apes but suggests that he himself does impersonation (aping). The Clown says that he has heard of Autolycus and knows him to be one who "haunts . . . bear-baitings", giving another link between Autolycus and bears. McPartland thinks that the Clown's line about maidens' immodesty, "Will they wear their plackets where they should bear their faces?", is spoken to Autolycus and is another link of Autolycus to bears. Likewise when Mopsa offers to sing a ballad "If thou'lt bear a part" and Autolycus replies "I can bear my part", McPartland sees this as a further reference to Autolycus playing the bear.

    When changing her clothes under Camillo's supervision, Perdita says "I must bear a part" and McPartland has to explain this away as merely a joke since the boy playing Perdita could not have taken the role of the bear. His last example is that the Clown says to the Old Shepherd (in the exchange with Autolycus pretending to be a courtier who can help them get forgiveness) that "He [Autolycus] seems to be of great authority . . . and though authority be a stubborn bear". Thus McPartland's argument relies on reading all these occurrences of the word 'bear' as referring to the bear that ate Antigonus.

    Since she has published several critiques of the authorship attribution methods used in the New Oxford Shakespeare to attribute parts of the Henry 6 plays to Marlowe, it might seem surprising that Rosalind Barber would publish a note adducing evidence that "adds to the argument that Marlowe had a hand" in 4.2 of 2 Henry 6 ('2 Henry VI and the Ashford Cage', N&Q 264[2019] 409-10). Dick the Butcher says of Jack Cade that "his father had never a house but the cage". In the sources Cade is Irish, but the play has Cade come from Ashford in Kent and Barber points out that there actually was a place called the Cage in fifteenth-century Ashford. It was under the town's market hall and was a cell for locking up troublemakers such as drunks.

    The Cage was not famous enough for Shakespeare to have heard of it, but Marlowe being from Canterbury, 12 miles away, would likely have known of it. Barber's support for Marlowe's authorship is best understood in the context of her support for the anti-Stratfordian view that Shakespeare's authorship of the works attributed to him has been a giant literary hoax.

    We know that Shakespeare used language manuals such as those by John Florio, John Eliot, and Claudius Hollybrand. Andrew S. Keener reckons that an overlooked one is Colloquia et Dictionariolum originally authored by Noel de Berlaimont, which went through almost 150 editions across Europe ('A New Multilingual Source for Shakespeare's Henry IV Part 1', N&Q 264[2019] 417-8).

    Colloquia et Dictionariolum contains a scene in which an overworked young servingman named Francis is kept frantically busy by being given multiple simultaneous tasks. In 1 Henry 4, Prince Hal says "I can drink with any tinker in his own language" and likewise keeps a young tapster called Francis busy. Keener sees Berlaimont's book as a previously unnoticed source for the scene.

     In 1953 Solomon Pottesman found a bookseller's list including "loves labor lost" and "loves labor won", suggesting that Francis Meres in 1598 had been right about Shakespeare having written a play with the latter title. The other books named in the list are mostly known, but scholars have found one line indecipherable. Misha Teramura offers her new transcription of it, based on figuring out what the other books around it have in common: they are almost all concerned with James I's accession on 1603 ('The 'Indecipherable' Line of the Love's Labour's Won Bookseller's List', N&Q 264[2019] 441-4). From this, Teramura interprets the fragments of letters in the indecipherable line to be the tops of the letters in the title Pancharis: The First Book by Hugh Holland, published in 1603. This does not help us with Love's Labour's Won, though, as Teramura acknowledges.

    Douglas Arrell reckons that the "ij stepells" (that is, two steeples) mentioned in Philip Henslowe's inventory of Admiral's Men's properties in 1598 are the pillars of Hercules needed for the play 2 Hercules ('Hercules' Pillars: A New Identification of Properties in Henslowe's Inventory', N&Q 264[2019] 519-21). Heywood's play The Brazen Age, which may be a reworking of the 2 Hercules that played at the Rose in 1595 according to Henslowe's Diary, requires Hercules to enter carrying two pillars.

    Heywood in An Apology for Actors gives what seems to be an account of 1 and 2 Hercules in performance and in it the pillars that Hercules carries are called "Pyramides". Arrell suggests that they were tall flat-sided pyramids (more like obelisks) because flat-sided properties were easier to make than rounded ones. Henslowe's properties and costume list of 1598 (now lost but transcribed by Edmond Malone) contains "ij stepells" and Arrell suggests that these are the pillars of Hercules. He gives examples of 'pyramid' and 'steeple' being synonyms in the period.

    Why does Shakespeare make Michael Cassio in Othello a Florentine and why does Iago, in his opening monologue, say that Cassio has all the theory of war but no practical experience? Alessandra Petrina notes that Niccol Machiavelli in The Art of War established Florence as a centre of the science of war ('An Allusion to Machiavelli in Othello', N&Q 264[2019] 444-7). Matteo Bandello, a known source for Shakespeare, told an anecdote about Machiavelli being challenged to drill soldiers in the way he had described in The Art of War and failing miserably. Petrina suggests that ". . . Michael Cassio, a Florentine, is Machiavelli, the Florentine . . ." (p. 446).

    Sonnet 77 refers to the "waste blanks" that the addressee might write his thoughts in. These are usually taken to mean the blank pages of a notebook, but Adam Barker points out that almanacs had blank reverse pages and an almanac, with days marked out, is also more suited to this sonnet's reflection on passing time ('The Book with 'Waste Blanks' Referred to in Shakespeare's Sonnet 77 is an Almanac Rather Than an Empty Notebook as Has Previously Been Assumed', N&Q 264[2019] 429-30).

    In 1.3 of 1 Henry 6, Joan likens glory to an ever expanding but weakening circular wave in water. Percy Simpson thought he had found a source for this in Silius Italicus's first-century AD History of the Punic Wars. Brian Vickers notes that we know now that Nashe not Shakespeare wrote Act 1 of 1 Henry 6, yet Nashe is not known to have ever used a Latin source directly so presumably we have yet to find his intermediary source ('Nashe, 1 Henry VI, and Silius Italicus', N&Q 264[2019] 408-9).

    Siobhan Keenan has discovered the date and place of Richard Burbage's marriage to Winifred Turner ('New Evidence About Shakespearean 'Star' Actor, Richard Burbage', N&Q 264[2019] 460-4). The record is the parish register of St Mary's Church Rotherhithe, where the marriage took place, and the date is 2 October 1600. Burbage is described in the record as a gentleman. The marriage was by licence rather than the slower, cheaper route of having banns called for three weeks; possibly that suited Burbage's sense of his own newly achieved importance.

    Finally, B. J. Sokol observes that of all the teachers depicted in Shakespeare, only Hugh Evans in The Merry Wives of Windsor is also a parson ('Why Does Shakespeare Give His Windsor Schoolmaster a Double Occupation as an Educator and as a Parson?', N&Q 264[2019] 430-5). Combining these occupations was rare in Shakespeare's time, although it was officially encouraged. Sokol thinks that Shakespeare made Evans deliberately anomalous in order to show that a teacher must, to be effective, have also "a fair degree of compassion such as would befit a pastor" (p. 435).

Books reviewed

Menzer, Paul and Amy R. Cohen, eds. Shakespeare in the Light: Essays in Honor of Ralph Alan Cohen, Shakespeare and the Stage. Fairleigh Dickinson University Press [2019]. 174 pages. 85. ISBN 978-1683931645

Shakespeare, William All's Well That Ends Well, ed. Suzanne Gossett and Helen Wilcox, The Arden Shakespeare. Bloomsbury [2019]. 440 pages. 10.99. 978-1904271208

Zwierlein, Anne-Julia, Jochen Petzold, Katharina Boehm and Martin Decker, eds. Anglistentag 2017 Regensburg: Proceedings, Proceedings of the Conference of the German Association for the Study of English. Volume 39. Wissenschaftlicher Verlag Trier [2018] 372 pages. 63.50. 978-386821767