A number of studies published over the past thirty years mark a shift in scholarship away from individual dramatists and plays towards the playing companies for whom they wrote and were written, considered alongside ‘other contributors to a company's dramatic output, such as actors, sharers, playhouse owners (and the buildings themselves), audiences, and patrons’.Footnote 1 These include monographs and collections of essays attending to particular companies, such as the Children of Paul's, the Chamberlain's/King's Men, Queen Elizabeth's Men, the Children of the King's Revels, the Children of the Queen's Revels, the Admiral's Men, Queen Anne's Men, and Strange's Men,Footnote 2 as well as general surveysFootnote 3 and a dedicated Oxford Handbook on the subject.Footnote 4 The ‘repertory approach’, as it has come to be known, has proved fertile ground, enriching studies of early modern English drama by provoking scholars to move beyond canon- and author-centric analyses to address anonymous and collaborative plays, to reconstruct company repertoires and historical trajectories, and to situate the drama within complex networks of professional competition and rivalry, patronage and politics, actors, audiences, and performance spaces.
If it is now a critical commonplace to recognise the production of early modern plays in both the theatre and the print-shop as collaborative enterprises, for some it has become evidence of the impossibility of ‘authorship’ as individual labour: as David Scott Kastan remarks, ‘authorial intentions are almost never solely determinative’ and ‘inevitably get transformed by the intentions of others in performance and in print’, because ‘the specific qualities of drama…inevitably dissolve authorial intentions into the collaborative demands of performance’.Footnote 5 Jeffrey Masten has gone so far as to insist on the futility of determining the discrete shares of collaborating playwrights, given that ‘the collaborative project in the theatre was predicated on erasing the perception of any differences that might have existed, for whatever reason, between collaborated parts’.Footnote 6
Recent theatre historiography and textual criticism has not only sought to disperse the authority of the ‘author’, but also – perhaps owing to a kind of horror vacui – to promote the playing companies to that privileged ‘authorial’ space newly vacated by the playwrights. ‘The author’, Lucy Munro reminds us, ‘is a useful organising principle, but it is not the only one available.’ Citing Foucault's claim that ‘since the eighteenth century, the author has played the role of the regulator of the fictive’, Munro argues that ‘to a large extent, the main “regulator of the fictive” in the early modern playhouse was the playing company, not the author’.Footnote 7
Those engaged in repertory study of early modern drama frequently assert that individual companies cultivated a ‘house style’, setting them apart from one another. In their influential study, Scott McMillin and Sally-Beth MacLean argue that ‘each company would have had its own style, its own textual procedures, its own sense of purpose, and its own impact on audiences and other acting companies’, before proceeding to identify the ‘special characteristics’ which gave Queen Elizabeth's Men ‘its identity – its acting style, its staging methods, its kinds of versification, its sense of what constituted a worthwhile repertory of plays’.Footnote 8 In the same vein, Mary Bly and Charles Cathcart note how ‘a company preference for plural authorship coexists with the collective adherence to a consistent repertory style’ in the case of the Children of the King's Revels, revealing ‘a distinctive company imprint’,Footnote 9 a ‘constraining authority governing the tenor of the plays’ and imposing ‘managerial control of a theatrical product written by more than one person’.Footnote 10 Munro similarly maintains that the repertory of the Children of the Queen's Revels ‘was predicated not only on authorial whim, but also on commercial exigency and on the relationship between individuals within companies’, whose plays ‘were created not only by the dramatists, but also through the ideas and desires of the company's shareholders, licenser, patrons, actors and audience’.Footnote 11
The assumption that playing companies developed distinctive and recognisable styles – relocating the Foucauldian ‘author function’ from the playwright to the company and its agents – is not limited to the specialised discourse of repertory studies, but has become axiomatic in criticism of early modern drama more broadly. For example, in the course of dispelling thirty ‘great myths’ about Shakespeare, Laurie Maguire and Emma Smith liken the ‘duopoly between the Admiral's Men and the Lord Chamberlain's Men during the 1590s’, both with a ‘contrasting personnel and house style’, to the modern phenomenon of ‘rival studios’ associated with ‘particular stars and a particular style of film’.Footnote 12 In his study of Shakespeare's late style, Gordon McMullan similarly equates the ‘formation of an acting company’ with ‘an institutionalisation of the collaborative process’, and argues that ‘the nature of the early modern company repertory militates in several ways against the idea of individual style’, giving rise instead to ‘a company style’.Footnote 13
As Gabriel Egan observes, set against this critical trend to ‘emphasize the collaborative, socialized labours of the players, the scribes and compositors’ to the extent that their ‘effects upon the surviving script are treated as though they are nearly as important as the author's labour’ are the ‘extraordinary successes’ in computational stylistics in ‘distinguish[ing] quantitatively between the stints of different writers in one script’, demonstrating the importance of authorship ‘in the teeth of postmodernism's denial of it’.Footnote 14 Such statistical studies ‘might have revealed – were free to reveal – that authorship is insignificant in comparison to other factors like genre or period’ and ‘secondary to other forces in textual patterning’; instead, quantitative studies have consistently established authorship ‘as a much stronger force in the affinities between texts’.Footnote 15
This chapter considers the extent to which early modern playing companies constrained the style of the playwrights engaged to provide scripts for them and promoted a distinct ‘house style’. Throughout our study we use ‘style’ to refer to the consistent patterns in word usage. On the one hand, this may seem a very limited concept of style. It is based entirely on quantitative measures, and inevitably relies only on a selection of the possible candidates for measurement. On the other hand, as numerous previous studies have shown, similarities and contrasts according to word frequencies do reflect the common distinctions we look for as readers in literary works – by author, genre, and period. Patterns in word frequencies are also richly revealing from a more interpretive point of view about local and more extended factors in expressive language. Stylistics by word frequencies, in other words, can be checked against formal categories to show that there is a genuine correspondence and can also, through its objectivity and capacity to work at scale, help in the usual business of literary analysis. If a given category does not emerge in this sort of study, we cannot conclude that it is not there in some shape or form, but we can reasonably conclude that it is not a powerful factor – certainly not in the way authors, genres, and eras are.
To test how strong repertory company performs as such a factor, we employ the robust quantitative methods of computational stylistics described in Chapter 1 to search a corpus of plays (for which the auspices of first performance are known) for stylistic patterns with which to generate distinct profiles for each repertory company. Just as plays of uncertain authorship can be attributed to playwrights on the basis of their stylistic affinity, plays of uncertain auspices may then be compared with the stylistic profiles generated for each repertory to determine whether it is attributable to that playing company – a procedure that, until now, has relied upon a scholar's ‘ear’ and familiarity with repertory company practices, such as casting and dramaturgy. For example, ‘as a means of identifying potential repertory members’, Roslyn L. Knutson has proposed to ‘apply the dramaturgical house style’ of Queen Elizabeth's Men – as enunciated by McMillin and MacLean's examination of the nine plays forming the company's ‘core’ canonFootnote 16 – to ‘plays contemporary with their first decade of playing’.Footnote 17
We first construct a corpus containing only those plays with well-attributed first companies and first performed between 1581 and 1594 – a range that spans the formation and dissolution of Queen Elizabeth's Men. Table 6.1 lists the resulting corpus of thirty-nine plays, along with each associated first company, genre, and date of first performance; Appendix A provides fuller bibliographical details for each play.
Table 6.1 Plays with well-attributed first companies, c.1581–94
| First Company | Author | Play | Date | Genre |
|---|---|---|---|---|
| Admiral's (Nottingham's) Men | Marlowe, Christopher | 1 Tamburlaine the Great | 1587 | Heroical romance |
| Admiral's (Nottingham's) Men | Marlowe, Christopher | 2 Tamburlaine the Great | 1587 | Heroical romance |
| Admiral's (Nottingham's) Men | Uncertain | A Knack to Know an Honest Man | 1594 | Tragicomedy |
| Admiral's (Nottingham's) Men | Peele, George | The Battle of Alcazar | 1589 | History |
| Admiral's (Nottingham's) Men | Marlowe, Christopher; others | Doctor Faustus | 1592 | Tragedy |
| Admiral's (Nottingham's) Men | Uncertain | Edward the Third | 1590 | History |
| Admiral's (Nottingham's) Men | Lodge, Thomas | The Wounds of Civil War | 1588 | History |
| Chamberlain's (Hunsdon's) Men | Shakespeare, William | The Two Gentlemen of Verona | 1590 | Comedy |
| Children of Paul's | Lyly, John | Endymion | 1588 | Classical legend |
| Children of Paul's | Lyly, John | Galatea | 1585 | Classical legend |
| Children of Paul's | Lyly, John | Love's Metamorphosis | 1590 | Pastoral |
| Children of Paul's | Lyly, John | Midas | 1589 | Comedy |
| Children of Paul's | Lyly, John | Mother Bombie | 1591 | Comedy |
| Children of the Chapel Royal | Peele, George | The Arraignment of Paris | 1581 | Classical legend |
| Children of the Chapel Royal | Marlowe, Christopher; Nashe, Thomas (?) | Dido, Queen of Carthage | 1586 | Tragedy |
| Children of the Chapel Royal | Uncertain | The Wars of Cyrus | 1588 | History |
| Derby's (Strange's) Men | Shakespeare, William; others | 1 Henry the Sixth | 1592 | History |
| Derby's (Strange's) Men | Shakespeare, William | The Comedy of Errors | 1594 | Comedy |
| Derby's (Strange's) Men | Uncertain | Fair Em | 1590 | Comedy |
| Derby's (Strange's) Men | Marlowe, Christopher; others (?) | The Jew of Malta | 1589 | Tragedy |
| Derby's (Strange's) Men | Uncertain | A Knack to Know a Knave | 1592 | Comedy |
| Derby's (Strange's) Men | Greene, Robert; Lodge, Thomas | A Looking Glass for London and England | 1588 | Moral |
| Derby's (Strange's) Men | Marlowe, Christopher | The Massacre at Paris | 1593 | History |
| Derby's (Strange's) Men | Uncertain | (The Rare Triumphs of) Love and Fortune | 1582 | Moral |
| Derby's (Strange's) Men | Kyd, Thomas | The Spanish Tragedy | 1587 | Tragedy |
| Oxford's Boys | Lyly, John | Campaspe | 1583 | Classical legend |
| Oxford's Boys | Lyly, John | Sappho and Phao | 1583 | Classical legend |
| Pembroke's Men | Shakespeare, William; others | 2 Henry the Sixth | 1591 | History |
| Pembroke's Men | Shakespeare, William; others | 3 Henry the Sixth | 1591 | History |
| Pembroke's Men | Marlowe, Christopher | Edward the Second | 1592 | History |
| Queen Elizabeth's Men | Uncertain | The Troublesome Reign of King John | 1591 | History |
| Queen Elizabeth's Men | Uncertain | 1 Selimus | 1592 | Heroical romance |
| Queen Elizabeth's Men | Uncertain | The Famous Victories of Henry the Fifth | 1586 | History |
| Queen Elizabeth's Men | Greene, Robert | Friar Bacon and Friar Bungay | 1589 | Comedy |
| Queen Elizabeth's Men | Peele, George | The Old Wife's Tale | 1590 | Romance |
| Queen Elizabeth's Men | Wilson, Robert | The Three Lords and Three Ladies of London | 1588 | Moral |
| Queen Elizabeth's Men | Uncertain | King Leir | 1590 | History |
| Queen Elizabeth's Men | Uncertain | The True Tragedy of Richard the Third | 1591 | History |
| Sussex's Men | Uncertain | George-a-Greene | 1590 | Comedy |
We project the word-frequency counts for the 500 most frequent words across the corpus for each 2,000-word segment into a two-dimensional space using Principal Components Analysis (PCA), treating the scores on the first and second principal components as Cartesian coordinates defining each segment as a point in the scatterplot (Figure 6.1). The relative distances between points within this space represent degrees of affinity, such that segments of similar stylistic traits will cluster tightly together, whereas dissimilar segments will be plotted further apart. With three exceptions, there are no tight discrete groupings of points on the scatterplot in Figure 6.1 – rather, points belonging to almost every repertory company are interspersed with one another around the origin of the graph. This suggests that segments of plays belonging to different repertory companies share similar stylistic traits. The exceptions include the cluster of interwoven points belonging to the Children of Paul's and Oxford's Boys, plotted as black and grey circles respectively to the bottom right of the scatterplot, and the selection of points belonging to the Admiral's (Nottingham's) Men, plotted as ‘x’ symbols, forming a tight cluster to the left of the origin. These outliers warrant further investigation.
Figure 6.1 PCA scatterplot of 2,000-word non-overlapping segments of plays with well-attributed first companies, c.1581–94, using the 500 most frequent words, labelled by company.
When the points are re-labelled according to authorship rather than repertory company (Figure 6.2), with segments of uncertain attribution plotted as ‘.’ symbols, discrete clusters become more discernible. The two outlier groupings previously identified in Figure 6.1 as segments from the Children of Paul's and Oxford's Boys are now revealed to constitute a single Lyly cluster, plotted as black triangles. The third outlier grouping, constituting a selection of segments from the Admiral's (Nottingham's) Men, are shown to belong to Lodge, Marlowe, and Peele.
Figure 6.2 PCA scatterplot of 2,000-word non-overlapping segments of plays with well-attributed first companies, c.1581–94, using the 500 most frequent words, labelled by author.
This group is made up of Marlowe's Tamburlaine plays, Peele's Battle of Alcazar, and Lodge's Wounds of Civil War. Although classified in Annals as ‘history’ plays, both Alcazar and Wounds share generic and stylistic traits with the ‘heroical romance’ of Tamburlaine and belong to a family of plays aptly described by G. K. Hunter as the ‘sons of Tamburlaine’.Footnote 18 Critics have long recognised resemblances between Wounds and Tamburlaine, particularly in Lodge's portrayal of Sulla, although it is unclear which play came first.Footnote 19 Described as ‘out-Tamburlaining Tamburlaine’,Footnote 20Alcazar is much indebted to Marlowe,Footnote 21 and Peele elsewhere links ‘mighty Tamburlaine’ with ‘King Charlemagne’ and ‘Tom Stukeley’ in A Farewell, a poem celebrating the departure of ‘noble’ John Norris and ‘victorious’ Francis Drake on their counter-Armada to the Iberian coast in 1589.Footnote 22 The similarities between the plays are demonstrated by the greater relative frequency of words associated with battle and conquest (arms, bloody, fight, foes, soldiers, sword, war) and the display of power (honour, march, mighty, power, princely, proud, royal) that they share (Figure 6.3).
Figure 6.3 PCA biplot of 2,000-word non-overlapping segments of plays with well-attributed first companies, c.1581–94, using the 500 most frequent words, highlighting selected generic markers.
The PCA uncovers further patterns in the use of personal and possessive pronouns (Figure 6.4), with this outlier grouping characterised by a greater relative frequency of the first-person plural genitive our and the second-person singular informal genitives thy and thine, as well as a corresponding relative infrequency of first-person singular pronouns (I, me, my, and mine) and second-person plural or formal singular pronouns (ye, you, and your). Many of these favoured pronouns occur in short succession within these plays; for example: ‘That he hath given our foe into our hands’, and ‘Thy love, thy loyalty and forwardness, | Thy service’ in Alcazar;Footnote 23 ‘I hope our lady's treasure and our own | May serve for ransom to our liberties: | Return our mules’, and ‘Go, stout Theridamas, thy words are swords, | And with thy looks thou conquerest all thy foes’ in 1 Tamburlaine;Footnote 24 and ‘Forsake our friends, forestall our forward war | And leave our legions full of dalliance’, and ‘Draw forth thy legions and thy men at arms, | Rear up thy standard and thy steeled crest’ in Wounds.Footnote 25
Figure 6.4 PCA biplot of 2,000-word non-overlapping segments of plays with well-attributed first companies, c.1581–94, using the 500 most frequent words, highlighting personal and possessive pronouns.
In other words, the PCA reveals not only authorial patterns in the samples, but also patterns in genre, such as the clustering of segments with a higher relative proportion of words typical of heroical romances and their ‘vast territorial scope’, ‘wandering hero[es]’, ‘great exploits of love and war’, ‘generation of emotions of sublime awe, wonder, [and] horror’, ‘gratification of wish-fulfilling fantasy’, and ‘providential design (however qualified)…in tension with [their] chronicle-history basis’.Footnote 26 When the points are re-labelled according to genre (Figure 6.5), segments categorised as ‘heroical romance’ (plotted as black circles) all appear to the left of the graph (the bottom left in particular), whereas ‘classical legend’, ‘comedy’, ‘moral’, ‘pastoral’, and ‘romance’ segments are plotted to the right, with ‘classical legend’ occupying the bottom right almost exclusively.
Figure 6.5 PCA scatterplot of 2,000-word non-overlapping segments of plays with well-attributed first companies, c.1581–94, using the 500 most frequent words, labelled by genre.
Given the remarkable similarity between segments belonging to different repertory companies revealed by PCA (Figure 6.1), failure to classify segments by repertory company using Random Forests, a machine-learning technique outlined in Chapter 1, is unsurprising. After a randomly selected one-third of the segments are withheld by the algorithm to be tested later, 500 decision trees are populated using the remaining two-thirds of the training dataset, trying twenty-two random word-variables at each split in the decision trees. When the randomly withheld segments are reintroduced and classified using the decision trees, the resulting classification error rate is 38 per cent. The threshold for what constitutes an ‘acceptable’ classification error rate will depend upon the particular conditions of an experiment. The higher the error rate, the weaker the relationship between the variables and the classes. In the present experiment, 38 per cent is an unacceptable error rate for differentiating between nine classes. Table 6.2 gives the confusion matrix of the Random Forests classifications for each 2,000-word segment using the 500 most frequent words across the corpus. The only repertory company to have all of its segments correctly classified is the Children of Paul's (0 per cent error rate); however, this error rate ignores the fact that all eleven Oxford's Boys segments are incorrectly classified as belonging to Paul's (100 per cent error rate). The Chamberlain's (Hunsdon's) Men and Sussex's Men repertories are similarly misclassified in their entirety (eight and four segments respectively), with segments belonging to the three largest companies represented in the corpus – the Admiral's (Nottingham's) Men, Derby's (Strange's) Men, and Queen Elizabeth's Men – producing misclassification rates of 26 per cent, 23 per cent, and 32 per cent respectively. In other words, when trained on two-thirds of segments belonging to every repertory company from the period, the algorithm misclassifies almost one-third of all segments belonging to Queen Elizabeth's Men.
Table 6.2 Confusion matrix for Random Forests classification of 2,000-word non-overlapping segments of plays with well-attributed first companies, c.1581–94, using the 500 most frequent words
| Admiral's (Nottingham's) Men | Chamberlain's (Hunsdon's) Men | Children of Paul's | Children of the Chapel Royal | Derby's (Strange's) Men | Oxford's Boys | Pembroke's Men | Queen Elizabeth's Men | Sussex's Men | Misclassification (%) | |
|---|---|---|---|---|---|---|---|---|---|---|
| Admiral's (Nottingham's) Men | 37 | 0 | 0 | 0 | 11 | 0 | 0 | 2 | 0 | 26 |
| Chamberlain's (Hunsdon's) Men | 0 | 0 | 0 | 0 | 7 | 0 | 0 | 1 | 0 | 100 |
| Children of Paul's | 0 | 0 | 35 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Children of the Chapel Royal | 7 | 0 | 0 | 1 | 6 | 0 | 0 | 3 | 0 | 94 |
| Derby's (Strange's) Men | 4 | 0 | 0 | 0 | 53 | 0 | 1 | 11 | 0 | 23 |
| Oxford's Boys | 0 | 0 | 11 | 0 | 1 | 0 | 0 | 0 | 0 | 100 |
| Pembroke's Men | 4 | 0 | 0 | 0 | 8 | 0 | 10 | 11 | 0 | 69 |
| Queen Elizabeth's Men | 4 | 0 | 0 | 0 | 17 | 0 | 0 | 44 | 0 | 32 |
| Sussex's Men | 0 | 0 | 0 | 0 | 2 | 0 | 0 | 2 | 0 | 100 |
Mindful that Random Forests and similar machine-learning techniques fare better with classification problems where there are fewer classes, we repeat the tests on a sub-set of the corpus comprising only the Admiral's (Nottingham's) Men, Derby's (Strange's) Men, and Queen Elizabeth's Men. These are the three largest repertory groups in our sample, with fifty, sixty-nine, and sixty-five segments respectively. We project the word-frequency counts for the 500 most frequent words across this corpus sub-set for each 2,000-word segment into a two-dimensional space using PCA, treating the scores on the first and second principal components as Cartesian coordinates defining each segment as a point in the scatterplot (Figure 6.6).
Figure 6.6 PCA scatterplot of 2,000-word non-overlapping segments of plays associated with the Admiral's (Nottingham's) Men, Derby's (Strange's) Men, and Queen Elizabeth's Men, c.1581–94, using the 500 most frequent words, labelled by company.
As before, points belonging to all three repertory companies are interspersed with one another around the origin and are plotted all over the graph, suggesting a high degree of shared stylistic traits. Two outliers emerge: a selection of Admiral's (Nottingham's) Men segments similarly cluster in and around the bottom left – the same ‘heroical romance’ segments belonging to Lodge's Wounds, Marlowe's Tamburlaine plays, and Peele's Alcazar – and a loose series of three smaller clusters of Queen Elizabeth's Men segments towards the bottom-right edges of the scatterplot. When the points are re-labelled according to known authorship rather than repertory company (Figure 6.7), with segments of plays of uncertain provenance plotted with ‘.’ symbols as before, the outlier clusters to the bottom right of graph are revealed to belong to Robert Wilson, plotted as unfilled grey circles.
Figure 6.7 PCA scatterplot of 2,000-word non-overlapping segments of plays associated with the Admiral's (Nottingham's) Men, Derby's (Strange's) Men, and Queen Elizabeth's Men, c.1581–94, using the 500 most frequent words, labelled by author.
Contrary to our initial intuition, PCA of the smaller sub-set of the corpus reveals a higher degree of overlap between segments belonging to different repertory companies. Perhaps this is due to the absence of John Lyly's segments, which constituted a significant stylistic outlier and point of difference. As such, we anticipate similar – if not higher – error rates when using Random Forests to classify segments in the corpus sub-set by repertory company. As before, after a randomly selected one-third of the segments are withheld by the algorithm, 500 decision trees are populated using the remaining two-thirds of the data, trying twenty-two random word-variables at each split in the decision trees. This time, when the randomly withheld segments are classified using the decision trees, the resulting classification error rate is 32 per cent, slightly lower than before. Table 6.3 gives the confusion matrix of the Random Forests classifications for each 2,000-word segment using the 500 most frequent words across the corpus sub-set. The algorithm misclassifies 32 per cent, 29 per cent, and 37 per cent of segments belonging to the Admiral's (Nottingham's) Men, Derby's (Strange's) Men, and Queen Elizabeth's Men respectively. This is an increase in error rate of 5–6 per cent when compared with the classifications made using the full corpus. In other words, when trained on two-thirds of segments belonging to all three companies with the largest surviving canon of plays from the period, the algorithm misclassifies approximately one-third of all segments.
Table 6.3 Confusion matrix for Random Forests classification of 2,000-word non-overlapping segments of plays associated with the Admiral's (Nottingham's) Men, Derby's (Strange's) Men, and Queen Elizabeth's Men, c.1581–94, using the 500 most frequent words
| Admiral's (Nottingham's) Men | Derby's (Strange's) Men | Queen Elizabeth's Men | Misclassification (%) | |
|---|---|---|---|---|
| Admiral's (Nottingham's) Men | 34 | 15 | 1 | 32 |
| Derby's (Strange's) Men | 6 | 49 | 14 | 29 |
| Queen Elizabeth's Men | 6 | 18 | 41 | 37 |
These preliminary results suggest that repertory company is not a useful principle for stylistic discrimination, and that authorship and genre are stronger signals in the stylistic affinities between plays belonging to the same period of composition.
Internal Stylistic Cohesion: Queen Elizabeth's Men and the Children of the King's Revels
If we cannot accurately discriminate between plays of the same period of composition on the basis of company, is it still possible to measure adherence to a ‘house style’ within individual repertories analysed in isolation? This much has been argued for the repertories of Queen Elizabeth's Men and the Children of the King's Revels. When the ‘plays are brought together as one textual group’, McMillin and MacLean assert that ‘sameness rather than variety is a leading characteristic’ of Queen Elizabeth's Men's repertory, ‘especially when it comes to such basic theatrical characteristics as casting, doubling, staging, and dramaturgy’.Footnote 27 In relation to the comedies associated with the Children of the King's Revels, Bly goes a step further to argue that ‘while the plays are readily distinguishable from most early modern romantic comedies, they are barely distinguishable from each other’. For Bly, it is this ‘homogeneity’ in the Children of the King's Revels repertory – not just within the comedies – that suggests the existence of ‘a cohesive decision-making body that directed the tenor of the plays’.Footnote 28
PCA is an ideal computational method with which to test the internal variance within these individual repertories. If a company's output is stylistically cohesive – as has been claimed for both Queen Elizabeth's Men and the Children of the King's Revels – the greatest variance should be within the plays themselves. That is, we would expect PCA to differentiate between different segments of the same plays and split them apart. However, if a company's output is not stylistically coherent, the greatest variance should be between the plays. That is, we would expect PCA to cluster segments of the same plays together, and to differentiate between these play-clusters.
We project the word-frequency counts for the 500 most frequent words across the 8 plays known to have been written for and first performed by Queen Elizabeth's Men (listed in Table 6.1) for each 2,000-word segment into a two-dimensional space using PCA,Footnote 29 treating the scores on the first and second principal components as Cartesian coordinates defining each segment as a point in the scatterplot (Figure 6.8). The greatest variance is indeed between, and not within, the plays: segments of the same play cluster together with varying degrees of density, from a very tight cluster of 1 Selimus segments, plotted as unfilled up-turned triangles, through to a sparser, but still clearly delineated, cluster of The Famous Victories of Henry the Fifth segments, plotted as unfilled black circles.
Figure 6.8 PCA scatterplot of 2,000-word non-overlapping segments of plays associated with Queen Elizabeth's Men, c.1581–94, using the 500 most frequent words, labelled by play.
When the scatterplot is relabelled according to author, with segments of unknown authorship plotted with ‘.’ symbols, it becomes clear that authorship is one of the factors explaining the variance between the plays, with well-defined clusters for each of the three known playwrights (Figure 6.9).
Figure 6.9 PCA scatterplot of 2,000-word non-overlapping segments of plays associated with Queen Elizabeth's Men, c.1581–94, using the 500 most frequent words, labelled by author.
Another stylistic factor in the groupings of the play segments using PCA is genre, demonstrated when the scatterplot is relabelled accordingly (Figure 6.10). Segments classed as ‘moral’ cluster together in the north of the graph; ‘comedy’, ‘heroical romance’, and ‘romance’ segments cluster to the west and east of the origin; and ‘history’ segments group together in the southern region of the graph with one exception – segments belonging to King Leir – which may be distinguished from the others as a ‘legendary history’.Footnote 30
Figure 6.10 PCA scatterplot of 2,000-word non-overlapping segments of plays associated with Queen Elizabeth's Men, c.1581–94, using the 500 most frequent words, labelled by genre.
In each of these graphs, an outlier segment of Robert Wilson's The Three Lords and Three Ladies of London is plotted away from the rest of the play and warrants further examination using a biplot (Figure 6.11). This segment contains the climactic Armada scene, in which the titular lords of London vanquish the Spanish enemy. At the ‘centre of the contest’ are shields on both sides, and ‘the battle is marked out by the advance and retreat of these blazons’. McMillin and MacLean describe this scene as ‘a ballet in which England defeats Spain – virtually a dance of herald against herald, page against page, lord against lord – which ends when the Spanish shields can be battered apart and the English held up in triumph’.Footnote 31 The verbal and physical sparring between these groups during this episode – particularly in reference to their shields – is reflected in the use of demonstrative (that, those, this, these) and relative (which, whom, whose) pronouns with greater relative frequency than in other segments of the play. For example: ‘Then know, Castilian cavalieros, this: | The owners of these emblems are three lords, | Those three that now are viewing of your shields.’Footnote 32
Figure 6.11 PCA biplot of 2,000-word non-overlapping segments of plays associated with Queen Elizabeth's Men, c. 1581–1594, using the 500 most frequent words, highlighting demonstrative and relative pronouns.
Perhaps it is unsurprising that the repertory of Queen Elizabeth's Men, with its range of authorial voices, genres, and dates of composition over a fourteen-year period, demonstrates a high degree of internal stylistic variance. Can the same be said for the four comedies written for and performed by the Children of the King's Revels, a company in existence for only a year (1607–8)? These include Edward Sharpham's Cupid's Whirligig (1607), Lewis Machin and Gervase Markham's The Dumb Knight (1608), John Day's Humour Out of Breath (1608), and Lording Barry's Ram Alley (1608). Bly's claim that these comedies are ‘barely distinguishable from each other’ incorporated a fifth play, John Day's Law Tricks, which we have excluded from our analysis because it properly belongs to another company and predates the formation of the Children of the King's Revels.Footnote 33
We project the word-frequency counts for the 500 most frequent words across the four plays written for and first performed by the Children of the King's Revels for each 2,000-word segment into a two-dimensional space using PCA (Figure 6.12). With so many shared attributes – all the segments come from comedies written and performed within the space of a year for the same company by novice playwrights – one might expect PCA to discern a high degree of stylistic cohesion. In fact, this is not the case. As with Queen Elizabeth's Men, the greatest variance is in fact between, and not within, the plays, with segments associated with each comedy forming distinct clusters: Ram Alley to the north, Humour Out of Breath around the origin, The Dumb Knight to the east, and Cupid's Whirligig to the south.
Figure 6.12 PCA scatterplot of 2,000-word non-overlapping segments of Children of the King's Revels comedies using the 500 most frequent words, labelled by play.
When the points are relabelled according to author (Figure 6.13), the scatterplots look identical, for the simple reason that each of these four comedies is known to have been written by different hands – with one possible exception. In his epistle, Day refers to Humour Out of Breath as ‘a poor friendless child…yet sufficiently featured too, had it been all of one man's getting (woe to the iniquity of Time the whilst)’.Footnote 34 Scholars have interpreted this remark as Day's acknowledgement of his unnamed collaborator, most probably Edward Sharpham, who, ‘woe to the iniquity of Time’, died in 1608 when the play was printed.Footnote 35 The plotting of the outlier first segment of Humour Out of Breath away from the rest of the play and closer to the Cupid's Whirligig cluster is perhaps suggestive of Sharpham's early involvement, but not conclusive.
Figure 6.13 PCA scatterplot of 2,000-word non-overlapping segments of Children of the King's Revels comedies using the 500 most frequent words, labelled by author.
Far from being ‘barely distinguishable’, this analysis suggests that authorship emerges as a stronger signal of the stylistic affinity even between segments of plays with a shared genre, repertory company, and date of composition.
Authorial Constraint: Richard Brome, 1629–40
In order to look beyond any potential author, genre, and period effects, we construct a corpus of nine well-attributed, sole-authored plays belonging to the same period of composition (1629–40), genre (comedy), and author (Richard Brome), but associated with three different repertory companies (Beeston's Boys, the King's Men, and King's Revels Company). With such a corpus, we should be able to determine whether Brome adapted his authorial habits to suit different ‘house styles’ – that is, whether we can systematically distinguish Brome's comedies written for Beeston's Boys from those written for the King's Men or King's Revels Company. The plays in this corpus are The Court Beggar (1640), The Damoiselle (1638), and A Mad Couple Well Matched (1639) for Beeston's Boys; Covent Garden Weeded (1632), The Northern Lass (1629), and The Novella (1632) for the King's Men; and The City Wit (1630), The New Academy (1635), and The Sparagus Garden (1635) for the King's Revels Company.Footnote 36
We project the word-frequency counts for the 500 most frequent words across the Brome corpus for each 2,000-word segment into a two-dimensional space using PCA (Figure 6.14). With three exceptions, there are no discrete groupings of points on the scatterplot – rather, points belonging to all three repertory companies are plotted around the origin of the graph and intermingle with one another. As before, this suggests that segments of plays belonging to different repertory companies have common stylistic attributes. The exceptions include outlier clusters of Beeston's Boys segments to the bottom-left of the scatterplot, King's Revels segments to the top-left, and King's Men segments to the far right.
Figure 6.14 PCA scatterplot of 2,000-word non-overlapping segments of 9 comedies by Richard Brome, 1629–40, using the 500 most frequent words.
When the points are re-labelled according to play-title rather than repertory company (Figure 6.15), segments belonging to the same play are shown to cluster together in varying degrees, with segments from The City Wit, Covent Garden Weeded, and A Mad Couple Well Matched constituting the outliers identified in Figure 6.14.
Figure 6.15 PCA scatterplot of 2,000-word non-overlapping segments of 9 comedies by Richard Brome, 1629–40, using the 500 most frequent words, labelled by play.
With potential author-, genre-, and date-effects minimised – if not zeroed out – by the composition of the corpus, we may have expected adherence to a particular ‘house style’ to emerge as the strongest signal discernible by PCA. Examination of a series of biplots suggests that the PCA instead reveals underlying thematic and grammatical patterns in the comedies as more important stylistic factors, trumping any affinity there may be between plays of the same repertory company. In thematic terms, The City Wit, Covent Garden Weeded, and A Mad Couple Well Matched are not so much outliers as exemplars of three distinct comedic subjects which, in combination, characterise the plays in this corpus: wealth, wine, and women (Figure 6.16).
Figure 6.16 PCA biplot of 2,000-word non-overlapping segments of 9 comedies by Richard Brome, 1629–40, using the 500 most frequent words, highlighting thematic markers.
The City Wit, a city comedy following the tricks and travails of a London merchant as he recovers his fortune, exemplifies plays concerned with wealth and is characterised by a high relative frequency of words associated with financial prosperity and monetary exchange (e.g. dear, fortune, given, poor, pound, purse, rich, and worth). Wine, or rather the tavern community that renders ‘conviviality and sociability as purchasable, at least in terms of the raw materials of food, drink, and entertainment’,Footnote 37 is another thematic topic in these comedies, exemplified by Covent Garden Weeded, which features no fewer than two taverns, one of them playing host to the Brotherhood of the Blade, ‘a fraternity of hooligans modelled on those known to have been active in Caroline London’.Footnote 38 Words associated with drinking and camaraderie (e.g. civil, company, drink, gentlemen, men, together, welcome, and wine) occur with relatively higher frequency in this play. As objects of sexual and financial desire, as well as significant characters within the plays themselves, women provide another thematic focus in this group of Brome's comedies. A Mad Couple Well Matched offers a ‘plentiful lady-feast’, to quote Alicia, a ‘light wife’ who joins a dramatis personae that also includes a pregnant ‘whore’, a ‘rich vintner's widow’, an old crone, a steward who spends most of the play disguised as a woman, and a ‘lady’ who employs a bed-trick to evade the unwanted advances of her husband's nephew.Footnote 39 As the exemplar of this theme, A Mad Couple Well Matched contains a relatively high frequency of words associated with women (e.g. ladies, lady, ladyship, madam, whore, wives, women, and widow).
The PCA aligns segments of other plays in the Brome corpus in relation to these themes, and not by repertory company. The Demoiselle, for example, combines the wealth and wine themes in its portrayal of Dryground, a gentleman who mortgages his estate to Vermin, an old usurer, in order to establish a ‘public ordinary, | For fashionable guests and curious stomachs, | The daintiest palates, with rich wine and cheer’ with the funds.Footnote 40 The investment proves sound, and the food and fare on offer quickly become popular among the gallants of London. Dryground also raffles off the virginity of his ‘daughter’ – actually the cross-dressed son of Brookall, a gentleman brought to ruin by Vermin – recovering their collective fortunes in the process. In his study of Brome, R. J. Kaufmann concludes a chapter on ‘Usury and Brotherhood’ with a discussion of The Demoiselle, focusing on the play's treatment of usury as a practice that ‘literally subverts the family and the hierarchy of loyalties which makes for the good community’.Footnote 41 The results of the PCA reflect this combination of the wealth and wine themes, such that The Demoiselle segments are plotted as ‘bridging the gap’ between The City Wit and Covent Garden Weeded.
The PCA of Brome's nine comedies also reveals grammatical patterns, such as in the relative frequency of personal pronouns (Figure 6.17). Second-person informal singular pronouns (thee, thine, thou, thy) form a group of vectors pushing the segments out to the north and north-east of the origin, diametrically opposed to the second-person plural or formal singular pronouns (you, your, yours) pushing south and south-west. The Demoiselle, plotted to the north and north-east of the origin, employs a relatively higher frequency of second-person informal singular pronouns in its exchanges between parents and children and between characters of differing social rank. As both a father and a ruined gentleman clinging to his former elevated status, Brookall's dialogue stands out for its informal register. Indeed, at one point he claims to ‘speak more like a father than a beggar, | Although no beggar poorer’.Footnote 42 Plotted to the south-west of the origin in diametrical opposition to The Demoiselle is A Mad Couple Well Matched, which features a relatively higher proportion of first-person singular pronouns (I, me, mine, my), second-person plural or formal singular pronouns (you, your, yours), and third-person feminine pronouns (her, she), all registers appropriate to the elevated language of love and courtship associated with the theme of women. By contrast, in the opposite hemisphere first- and third-person plural pronouns (our, their, them, they, us, we), as well as masculine singular pronouns (he, him, himself, his), push out from the origin to the east and south-east respectively, in line with the language of tavern culture and the familiar register of camaraderie that populate the wine theme in that region of the plot.
Figure 6.17 PCA biplot of 2,000-word non-overlapping segments of 9 comedies by Richard Brome, 1629–40, using the 500 most frequent words, highlighting personal pronouns.
In keeping with earlier experiments in this chapter, we also attempt to classify Brome's nine comedies into their respective repertory companies – thirty-three Beeston's Boys, thirty-three King's Men, and thirty-five King's Revels Company segments in total – using Random Forests. After a randomly selected one-third of all the segments are withheld by the algorithm, 500 decision trees are populated using the remaining two-thirds of the data, trying forty-four random word-variables at each split in the decision trees. When the decision trees are then used to classify the randomly withheld segments, the resulting classification error rate is 24 per cent. In earlier experiments, 38 per cent was considered an unacceptable error rate when differentiating between nine classes. Likewise, when differentiating between only three classes (as in the present experiment), an error rate above 12 per cent – or one-third of 38 per cent – is unacceptable. Table 6.4 gives the confusion matrix of the Random Forests classifications for each 2,000-word segment using the 500 most frequent words across the corpus sub-set. The algorithm misclassifies 36 per cent, 21 per cent, and 14 per cent of segments belonging to Beeston's Boys, the King's Men, and King's Revels Company respectively. In other words, when trained on two-thirds of segments belonging to all three companies – segments sharing the same author, genre, and period of composition – the algorithm misclassifies over one-third of all Beeston's Boys segments, over one-fifth of King's Men segments, and one-seventh of King's Revels Company segments.
Table 6.4 Confusion matrix for Random Forests classification of 2,000-word non-overlapping segments of 9 comedies by Richard Brome associated with Beeston's Boys, the King's Men, and King's Revels Company, 1629–40, using the 500 most frequent words
| Beeston's Boys | King's Men | King's Revels Company | Misclassification (%) | |
|---|---|---|---|---|
| Beeston's Boys | 21 | 2 | 10 | 36 |
| King's Men | 4 | 26 | 3 | 21 |
| King's Revels Company | 4 | 1 | 30 | 14 |
These results suggest that, even in the absence of competing authorial, genre, and period factors, repertory company is a weak – perhaps even insignificant – principle for stylistic discrimination and classification. In the case of Brome, unique archival materials provide additional external evidence and support for this conclusion. Brome signed a three-year contract with the King's Revels Company in 1635, committing to produce three plays annually for the exclusive enjoyment of the company at their Salisbury Court theatre. While the contract no longer exists, theatre historians have pieced its contents together through meticulous examination of the requests proceedings bill of complaint filed in 1640 against Brome by Queen Henrietta Maria's Men, the occupants of Salisbury Court at the time, and Brome's subsequent answer to the complaint.Footnote 43 Eleanor Collins, the most recent commentator on this convoluted and confusing legal narrative, summarises the proceedings as follows:
[Richard] Heton and the [Salisbury Court] company claimed that Brome had broken the terms of his contract and was, by 1638, in arrears of the agreement by four plays. Furthermore, Brome had not observed the article of the contract that forbade him to ‘write any playe or p[ar]t of a playe [for] anye other players or playe howse’. While Brome had been required to ‘applie all his studdye and Endevours theerein for the Benefitte of’ the Salisbury Court company, he stood accused of delivering one of the company's promised plays to Christopher and William Beeston at the Cockpit theatre, where it would be performed by Beeston's Boys…Despite this breach, Heton and Brome both renewed their contracts in 1638…[O]nce more, Brome agreed to supply the company with three plays each year, but this time for an extended term of seven years. This raised the total number of plays that Brome was expected to contribute to the Salisbury Court repertory from nine to twenty-one. In addition, the new contract required Brome to produce the plays that he owed from the prior agreement. Once more, the court case suggests that he failed to do so. By the end of the year he was in arrears by one new play, and the Heton complaint records that, by this time, Brome had ‘wholly applie[d] himself unto the said Beeston and the Companie of players Acting at the playhouse of the Phoenix [or Cockpit] in Drury Lane’.Footnote 44
The primary objective of the contract was to secure Brome's services as playwright ‘to stock [the Salisbury Court] playhouse with current drama’, thus enabling ‘the consistent accumulation of a repertory’ to supplement the company's existing corpus of old plays available for revival.Footnote 45 But this is not to suggest that Salisbury Court approached Brome because he was particularly industrious as a playwright. According to the contract, Brome's ‘best art and industry’ were valued equally.Footnote 46 Richard Gunnell, Heton's predecessor at Salisbury Court, ‘saw how well Brome's plays were being received at the Red Bull’, a rival theatre, ‘and asked Brome to compose plays for their company’ instead. Prompted by the success of Brome's plays for them, the Salisbury Court company ‘sought to enhance its reputation and employed Brome exclusively as their poet’.Footnote 47
Brome's ‘art’ – his craft, his style – was what attracted paying customers to the theatres; his ‘industry’ simply governed the rate at which his scripts became available. The contract's exclusivity clause attempted to exploit and maximise Brome's art and industry respectively by ensuring that his energies were focused on meeting the quota of scripts, and by associating his creative output with the theatre at Salisbury Court alone. If it were possible – or even desirable – for a playwright to adopt a company's distinctive ‘house style’ wholesale, and to radically constrain, subsume, or alter their authorial voice in the process, Salisbury Court might have sought a more industrious candidate than Brome, given his repeated failure to meet the stipulated quota. By the same token, if Brome had no distinctive ‘art’ of his own and was able to simply adopt and switch between recognisable company styles, why should Salisbury Court deny him the opportunity to ‘write, invent, or compose any play, tragedy or comedy, or any part thereof, for any other playhouse’,Footnote 48 so long as he met their needs and made his quota?
Conclusion
What conclusions might be drawn from these experiments? If previous studies have demonstrated that authorship ‘emerges as a much stronger force in the affinities between texts than genre or period’,Footnote 49 our results suggest that ‘repertory company’ may be added to this list of secondary forces. As such, our results provide further support for Egan's argument that ‘plays were relatively stable works’: our continued ability to cluster and ‘discriminate between writers’, even when analysing plays written for the same repertory company in isolation, militates against the notion of the early modern theatre as ‘a melting pot that blurred all boundaries’.Footnote 50
Our results also provide much-needed quantitative support for the scepticism expressed by critics of repertory studies about the existence of, and pursuit to identify, distinct company styles. ‘Just as author-centred criticism tends to assume a consistency of political or intellectual allegiance from one work to the next’, Tom Rutter notes the danger of ‘look[ing] for points of similarity between plays written by different dramatists for the same company’, since in ‘seeking to identify distinctive characteristics for the various playing companies repertory studies can end up hypostatizing them’.Footnote 51 Roslyn Knutson similarly inveighs against assuming that companies cultivated a ‘house style’, given the difficulties of distinguishing between evidence of ‘company ownership and company influence’ and gauging ‘how much of a house style is the result of the dramatists’ sense of identity rather than that of the company’.Footnote 52 Rutter has since published a study of the Admiral's Men repertory, in which he concludes that:
while a repertory-based approach may encourage the identification of a company style, perhaps the best response to the varied, innovative and ideologically unfixed drama of the Admiral's Men between 1594 and 1600, open to the influence of Shakespeare while shaping his own dramatic development, is to refrain from doing so.Footnote 53
Even if repertory company performs feebly as a principle for stylistic discrimination between plays, this does not in and of itself discount the possibility that individual companies cultivated a recognisable ‘house style’. It does, however, suggest that recent theatre historiography and textual criticism has exaggerated the extent to which repertory companies constrained the authorial habits of the playwrights they employed. It also raises important questions about the nature of such ‘house styles’, since they are not evident in the places where we find authorial, genre, and period styles – that is, in the language of the plays themselves. If not in the language of the plays, evidence for company styles may well be found in performance – in the vocal, physical, and expressive qualities of different actors and types of acting, in the incorporation of dance, music, song, tumbling, ‘wit’, and other feats, and so on. However, performative elements such as these are ephemeral and, as G. K. Hunter observes, ‘the evidence left in texts is much too sporadic for the point to be developed’.Footnote 54 ‘The case for the development of consistent or long-lived “house” styles’ is similarly undermined, as Siobhan Keenan remarks, by ‘the fact that playwrights often wrote for several companies without necessarily changing their writing style and that plays (and players) moved between companies’. ‘It might be more appropriate’, Keenan adds, ‘to speak of acting companies fostering occasional dramatic and staging specialities, rather than developing wholly distinctive company repertories and performance practices’.Footnote 55 For example, Derby's (Strange's) Men may have cultivated expertise in the use of pyrotechnics,Footnote 56 but this feature is not present in every play of their repertory,Footnote 57 nor is it absent from the repertories of rival companies.Footnote 58
As Munro notes, to recognise that ‘companies purchased plays from dramatists – sometimes buying a complete script, more often paying in instalments after discussions over an idea, a plot or a completed act’, with ‘near-complete control over those plays’ progress to the stage’, is not necessarily a question ‘of denying the playwright's agency’ but of ‘acknowledging the compromises which writers make when they engage with institutions such as the early modern theatre industry’.Footnote 59 Whatever these compromises were, however, our results suggest they were not enough for companies to constrain an author's habits or fashion a repertory with a style that was distinctive, internally cohesive, or even statistically detectable in the absence of other competing factors such as genre and period.
