SHAXICON '95

by Donald Foster, Vassar College

What SHAXICON demonstrates is that the rare-words in Shakespearean texts are not randomly distributed either diachronically or synchronically, but are "mnemonically structured." Shakespeare's active lexicon as a writer was systematically influenced by his reading, and by his apparent activities as a stage-player. When writing, Shakespeare was measurably influenced by plays then in production, and by particular stage-roles most of all. Most significant is that, while writing, he disproportionately "remembers" the rare-word lexicon of plays concurrently "in repertory"; and from these plays he always registers disproportionate lexical recall (as a writer) of just one role (or two or three smaller roles); and these remembered roles, it can now be shown, are most probably those that Shakespeare himself drilled in stage performance.

SHAXICON electronically maps Shakespeare's language so that we can now usually tell which texts influence which other texts, and when. Moreover, when collated with the OED or with early modern texts in a normalized machine-readable format, SHAXICON provides an incomplete record of Shakespeare's apparent reading. The main value of this resource has less to do with biographical novelties, however, than with problems of textual transmission, dating, probable authorship of revisions, early stage history, and the like. And because SHAXICON is a closed system, human bias in measuring lexical influence of this sort is effectively eliminated. The evidentiary value of supposed "verbal parallels" is no longer a matter of private intuition or subjective judgment, but quantifiable, using a stable lexical index (and measurable against a virtually limitless cross-sample of machine-readble texts).

In 1991, I published a 3-part report in SNL (see Summer, Fall, Winter 1991) about SHAXICON (the database was not then completed, and not yet dubbed), in which I made (in a few cases, mistaken) projections concerning Shakespeare's apparent stage roles (based on entries for about a third of the final lexical sample). The few botched projections derived in part from key-punching errors--e.g., "Pand" (Pandarus of TRO) was often being entered for "CPan" (Pandulph of JN), and "QnElz" (R3) for QnEliz (3H6); and in part from unavoidable limitations, explained in the SNL series, concerning the variable "richness" of character-specific lexicons, which could not be measured until the whole canon was indexed. These problems have been eliminated.

The following list represents a corrected catalogue of those roles that Shakespeare is most likely to have acted. These assignments vary somewhat in statistical significance, depending on sample size, etc. A fuller report (with instructions on how to run cross-checks and fully automated statistical analysis) will appear in my "SHAXICON Notebook" (a written commentary that has yet to be completed). In the meantime, here follows a list of Shakespeare's most likely stage-roles, as statistically derived. Keep in mind that this catalogue cannot be proven to represent historical actuality. SHAXICON handily selects Adam of AYL and the Ghost of Ham as probable Shakespeare roles, both of which are supported by hearsay evidence from the 17th century; the remaining roles find no external historical confirmation (although Davies mentions that Shakespeare played some kings, and SHAXICON indicates that Shakespeare played king-roles in AWW, 1H4, 2H4, HAM, LLL, PER, and probably MAC). Having studied the evidence from every conceivable angle, I'd say that the assignments below are good bets, even despite the lack of archival evidence to back them up, for the disproportion in Shakespeare's persistent recall of these roles is quite striking relative to other roles in the corresponding texts. There are a few texts (principally ADO, MV, and Jonson's EMI) in which Shakespeare may have played two different roles in two successive seasons of the same theatrical "run." But the statistical weight of Shakespeare's selective recall of particular roles is in most instances pretty clear; in fact, when multiple roles are identified by SHAXICON as probably Shakespearean, they are in most instances roles that are easily doubled (exceptions and problems are are noted below).

MOST PROBABLE SHAKESPEARE ROLES, BASED ON THE POET'S PERSISTENT AND MEASURABLE RECALL OF PARTICULAR CHARACTER-SPECIFIC LEXICONS:

• ADO: Leonato; later switching to Friar (Q version registers higher lexical recall for Leonato, F1 version higher for Friar). This could be viewed as a problem, since the same actor cannot have played both roles simultaneously, yet Shakespeare clearly "remembers" both roles (unlike all other principal parts in ADO, which he "forgets").

• ANT: Agrippa, Philo, Proculeius, Thidias, and Ventidius, probably simultaneously [!] (thus requiring some accommodation at 3.2.1 for Vntd/Agri), and probably with Proculeius taking Agrippa's lines in 5.1 (hence the textual crux recently discussed on SHAKSPER).

• AWW: King of France

• AYL: Adam; adding old Corin the Shepherd in two revivals of AYL.

• COR: Shakespeare role uncertain. Shakespeare's writing after Coriolanus registers disproportionate recall of the Sicinius role, but without the hugely lopsided excess in lexical recall that obtains for the designated Shakspeare roles in most other plays.

• CYM: 1.Gent (I.i), Philario (I.iv, II.iv), and Jupiter (V.iv)

• EMI-F (Jonson): SHAXICON indicates that F1 may represent a major Elizabethan revision of Q1, followed by a minor Jacobean revision (as per established textual scholarship on EMI). SHAXICON also confirms that Shakespeare probably knew the play in performance: in 1598, and again in 1604, words from EMI come pouring into Shakespeare's writing, forming very distinct peaks of lexical influence just when we know that EMI was, indeed, acted by the King's Men (and again in 1612-13). But lexical influence by character (entirely independent of general lexical overlap) gives mixed signals: after both Q1 and F1 EMI, Shakespeare's dramatic and non-dramatic writing registers disproportionate recall of the Lorenzo-Kno'well role (esp. the F1 Old Kno'well); but Judge Clement (esp. the Q1 Clement) seems to influence some of the post-EMI non-dramatic texts; and the Thorello-Kitely role has an inexplicably high correlation with Shakespearean texts both before and after the production of EMI. It is possible that Shakespeare alternated roles in at least one version of EMI, but the senior Lorenzo/Kno'well figure is the one showing the most pervasive and persistent lexical influence on his own writing. One suspects that Jonson's extensive and multiple revisions in EMI are responsible for the slightly fogged picture presented by the character-distributions for this play.

• ERR: Egeon (I.i, V.i) and Dr. Pinch (IV.iv).

• 1H4: King Henry.

• 2H4: King Henry (and perhaps Rumor, but only briefly).

• H5: SHAXICON indicates that Shakespeare probably played the French Constable and Exeter in the Quarto version (in 1599, while also playing Exeter in a revival of 1H6). In F1 Henry V, Shakespeare appears to have performed Bishop Ely and Montjoy, and probably, on some occasions, the Chorus. (The influence of the Chorus-role is less strongly marked than that of Ely and Mountjoy, but still pronounced than for other roles in the play). The Chorus-role is easily doubled with Montjoy--but tripling with Ely raises a problem at I.i.0, when the Chorus walks offstage and Ely walks on. Shakespeare may possibly have performed Ely and Mountjoy in some productions, the Chorus and Mountjoy in others.

• 1H6: Bit parts only, most probably including Vernon or Exeter, but possibly Bedford, Lucy, Mortimer, Suffolk, Warwick, or Winchester (insufficient pre-1H6 sample, and uneven figures thereafter).

• 2H6: Suffolk (also Suffolk in the "bad" 2H6-Q, which appears certainly to antedate the F1 version, as has been argued by Steve Urkowitz).

• 3H6 Warwick (Old Clifford in the "bad" 3H6-Q, which appears certainly to antedate F1 version, as has been argued by Steve Urkowitz).

• H8: Prologue and 1.Gentleman; or none (statistically uncertain, due to insufficient post-H8 lexical sample).

• HAM: Ghost, 1.Player, Mess-Gent. of 4.5 (and perhaps also role in the Mousetrap, most probably Lucianus). In F1 Hamlet, the Mess-Gent role is partly folded into Horatio; but given Shakespeare's persistent recall of the role even after 1601, it seems likely that the F1 variant in this instance represents a casting-change made later than the bulk of Shakespeare's F1 revisions.

• JC: Shakespeare role(s) uncertain, due to apparent revision and shortening. Most probably, Decius; and, somewhat less probably, Flavius.

  Note: Decius-Flavius doubling is not possible in the F1 version unless F1 has been shortened from an earlier version. In F1, at I.ii.0, Flavius and Decius enter as mutes; but the very text of JC I.ii offers some evidence that the text has, indeed, been shortened at this point (e.g., in the same scene, at I.ii.285, Casca reports that "Murellus and Flavius, for pulling scarfs off Caesar's images, are put to silence"; but, if we may believe the F1 stage direction at I.ii.0, Casca was on stage with Murellus and Flavius moments earlier--from I.ii.0 to at least I.ii.214--and Casca hasn't heard boo about Caesar's images in the interim). SHAXICON thus seems to confirm the view that JC-F1 is a shortened text (albeit with some added bits (e.g., the second account of Portia's death, which are indexed in SHAXICON under JC-b). I am inclined to accept the assignments of both Decius and Flavius to Shakespeare, but there is room for doubt.

• JN: Cardinal Pandulph.

• LLL: Ferdinand (possibly with one brief stint as Boyet).

• LR: Albany. (The Albany role reduced in the revised F1 version. This is one of several designated Shakespeare roles that appears to have been cut or reduced ca. 1612. It is doubtful that Albany was performed by Shakespeare subsequent to the 1612 revision.)

• MAC: Duncan (and perhaps one or two bit roles after Duncan's assassination, including the nameless Lord of III.vi) -- but perhaps Banquo. The evidence of Shakespeare's role in this equivocating play is itself equivocal, probably as a result of late revision, but possibly because Shakespeare alternated roles, playing Banquo in some productions, Duncan in others.

  That MAC was revised ca. 1612 seems altogether likely from the evidence of SHAXICON (principally in I.v.1-30,. I.v.71-3, IV.iii all, and V.ix.1-19, indexed in SHAXICON under MAC-b). Simon Forman's eye-witness account of MAC as acted in 1611 suggests that the ur-MAC had a larger Duncan-role than in the F1 version. It has recently been argued on SHAKSPER that there was an Elizabethan MAC (not extant) on which the 1606 version was based; if these theories of revision are correct, even in part, they would provide a satisfactory explanation for the irregularities in the SHAXICON data for MAC.

• MM: Escalus; possibly switching to Friar Peter in a late revival, later than 1610.

• MND: prob. Theseus, but with very irregular figures. SHAXICON traces enormously high Theseus-"influence" on the post-1594 poems, but relatively slight Thesus-"influence" on the post-1594 plays. Although Shakespeare's post-MND writing registers lexical recall of the Theseus-role that is significantly higher than for any other role in MND, there is inexplicably high recall of the Helena role as well, especially in texts written 1602-4.

• MV: Shakespeare seems to have played Antonio in all productions; but Morocco is a second "remembered" role, especially as manifest, albeit intermittently, in the lexicon of the post-1594 poems and in the 1595-6 plays. No other role in the play comes close to these two parts in lexical "influence" upon the poet's post-MV writing. Morocco is doubtful: Shakespeare's averaged recall of the Antonio role, after 1594, is a full 55% higher than would obtain in an exact distribution, while his averaged recall of the Morocco role is only 2% higher than expected. But Morocco tends to register its strongest influence on Shakespeare's writing when Antonio doesn't, and vice versa; perhaps Shakespeare alternated roles. He cannot easily have played both simultaneously, at least not in the Q1 or F1 text.

• OTH: Brabantio (and possibly a second small role after Brabantio's farewell in I.iii; but this possibility has not yet been tested). The Brabantio role is reduced in the Q1 version (as revised and cut from the antecedent F1 version ca. 1612); this is one of several designated Shakespeare roles that was cut, omitted, folded into another character ca. 1612, all of them evidently signalling Shakespeare's retirement from the stage. SHAXICON identifies a final "run" of OTH in 1611-13, first in the F1, then in the Q1 version; but it is doubtful that Brabantio was performed by Shakespeare later than 1612.

• PER: SHAXICON suggests that PER is a very early play (ur-PER), the palimpsest of which is imperfectly represented by acts I-II of PER-Q. The play was clearly revised by Shakespeare, with altogether new or greatly re-written acts III-V. This revision may have taken place as late as 1606/7 (as customarily dated), but a date of 1600 appears more likely. SHAXICON offers no support for the view of the Oxford editors that PER-Q represents a Wilkins-Shakespeare collaboration, yet it leaves open such a possibility insofar as Wilkins could be shown to have tinkered some with acts I-II. (This could be tested by indexing other texts by Wilkins, and running lexical cross-checks.) Shakespeare appears to have acted both Antiochus and (at least when doubling was needed) Simonides, and he may have performed or read Gower's part from time to time, most notably ca. 1608/9 (cf. notes on H5-F1, another script for which Shakespeare registers sporadically high recall of the chorus-role, especially ca. 1608/9--perhaps the company was short-handed in that year). With a reduced cast, he may have simultaneously performed Antiochus, Simonides, Cleon (without the dumb shows) and Gower, all of which register higher than expected lexical recall, the last two only intermittently. Because of its episodic structure, Pericles can be performed with a small cast, and the SHAXICON data suggest that Pericles was indeed performed, on occasion, with at least one player performing multiple roles.

• R2: Gaunt (in I.i - I.iii, II.i), the Gardener (III.iv), the Lord (IV.i), and probably also the Groom (V.v). Troublesome dating: SHAXICON seems to indicate that R2 derives from an earlier play, and that R2 was revised immediately after 1H4 (but prior to publication of R2-Q1). This finding is at odds with all past textual scholarship on the play, which has been nearly unanimous in viewing R2 as a text begun and completed ca. 1595.

• R3: Clarence (in I.i, I.iv, and V.iii) and Scrivener (III.vi). Possibly also Third Citizen (II.iii) in a late revival.

• ROM: Friar Lawrence in ROM-Q1,-Q2, and -F1 versions, plus Chorus in ROM-Q2 version (The Chorus-role was evidently omitted altogether in a late revival of ROM, as is intimated by its lexical distribution in the SHAXICON files, and confirmed by its omission from ROM-F1).

• SEJ (Jonson): Macro (I.i, II.iii, III.i, IV.ii); probably also (but less well-marked) Sabinius (I.i, II.iii, III.i, IV.iii), with some accomodation for a costume change after IV.ii (but Jonson reports in F1 that he has revised Sejanus, which means that this problem at IV.iii.0 may not actually have come up in the performed text).

• SHR: Lord, and perhaps also Pedant.

• TGV: Antonio and the Duke.

• TIM: Poet in TIM-a (representing ur-F1 version, the parts of TIM-F1 customarily ascribed to Shakespeare); no role apparent in TIM-b (widely supposed to represent Middleton or late-Shakespearean revision. SHAXICON suggests that TIM-F1 is a late, unfinished revision (ca. 1613) of a play first acted in 1601. TIM-F1 appears not to be a collaborative text per se.

• TIT: probably Aaron or old Lucius, or possibly alternating between these roles (neither is as strongly marked statistically as most other roles identified in this catalogue).

• TMP: no Shakespeare role apparent; possibly Shipmaster or Antonio (insufficient post-TMP sample).

• TNK: no Shakespeare role apparent; possibly Prologue and Doctor (insufficent post-TNK sample).

• TNT: Antonio (later adding Valentine [I.i]).

• TRO: Uncertain. Possibly, in a brief run of TRO-F1 ca. 1602/3, the Prologue (perhaps also Agamemnon or Ulysses, or a second bit part; not yet fully tested). The Q1 version may not have been acted until about 1612.

• WIV: In WIV-F1, Shakespeare appears to have played Ford, but only in two evidently brief runs. In WIV-Q Shakespeare appears almost certainly to have played the Host. Though one of the "bad" quartos, Q1 Wives appears certainly to antedate the F1 version; but SHAXICON casts doubt on Shakespeare's authorship of Q1 Wives.

• WT: Shakespeare roles uncertain; probably Archidamus (I.i), Antigonus (II.i, II.iii, III.iii), and 3rd Gentleman (V.i) in 1610/11, switching to Polixenes in 1612; but the post-WT lexical sample is too small to speak with confidence.

WHAT YOU NEED TO USE SHAXICON:

1. Patience.

2. Disk space. In its present form, SHAXICON sucks up 40+ megs just for the raw data, plus another 20 megs or so for the commentary, help files, and graphics; plus another 20 megs or so for the software. But don't start erasing those electronic games just yet in order to make room for it. The main database for SHAXICON is now complete, purged of errors, and generally usable; but it's not yet ready for prime time: SHAXICON now runs on ETC Word-Cruncher, which is limited in its capabilities and requires way-too-much manual labor (keying in lexical searches, etc.). We're now using Excel for the summary figures and graphics, which is a big time-saver--but we're likely to change over, prior to publication, to a slicker and more fully automated database-management system so that SHAXICON is more user-friendly in ALL respects. I'm inquiring after Oracle, 4D, and Fox.

Address queries to Professor Donald Foster, foster@vaxsar.vassar.edu.