1

Stylometric Analysis of Early Modern Period
English Plays

Santiago Segarra,1, Mark Eisen1, Gabriel Egan2, and Alejandro Ribeiro1

1Dept. of Electrical and Systems Engineering,, University of Pennsylvania, Philadelphia, USA
2School of Humanities,, De Montfort University, Leicester, UK

Abstract—Function word adjacency networks (WANs) are used
to study the authorship of plays from the Early Modern English
period. In these networks, nodes are function words and directed
edges between two nodes represent the likelihood of ordered
co-appearance of the two words. For every analyzed play a
WAN is constructed and these are aggregated to generate author
profile networks. We first study the similarity of writing styles
between Early English playwrights by comparing the profile
WANs. The accuracy of using WANs for authorship attribution
is then demonstrated by attributing known plays among six
popular playwrights. This high classification power is then used
to investigate the authorship of anonymous plays. Moreover,
WANs are shown to be reliable classifiers even when attributing
collaborative plays. For several plays of disputed co-authorship, a
deeper analysis is performed by attributing every act and scene
separately, in which we both corroborate existing breakdowns
and provide evidence of new assignments. Finally, the impact
of genre on attribution accuracy is examined revealing that the
genre of a play partially conditions the choice of the function
words used in it.

Index Terms—Authorship attribution, word adjacency net-
work, Markov chain, relative entropy.

I. INTRODUCTION
Stylometry involves the quantitative analysis of a text’s lin-

guistic features in order to gain further insight into its underly-
ing elements, such as authorship or genre. Along with common
uses in digital forensics [1], [2] and plagiarism detection [3],
stylometry has also become the primary method for evaluating
authorship disputes in historical texts, such as the Federalist
papers [4], [5] and Mormon scripture [6], in a field called
authorship attribution. One of the most notorious collection of
texts for which such disputes exist is the collection of dramatic
works produced in England during the Early Modern era,
covering the 16th through mid-17th century. Due to factors
such as inaccurate pressing information and undocumented
collaborations, the precise authorship of many of these plays–
including works by famous playwrights William Shakespeare
and John Fletcher–remains highly contested.

Stylometric analysis of the work from this time period dates
as far back as the nineteenth century in F. G. Fleay’s analysis
of verse features in Shakespeare’s plays [7]. Similar analyses
based on the manual counting of linguistic features continued
throughout the early twentieth century [8]–[10]. Computer-
based techniques for counting the frequency of various stylistic
features, such as rare words or phrases, have become very

Supported by NSF CAREER CCF-0952867 and NSF CCF-1217963.

common over the past few decades. The premier work done in
evaluating authorship in Early Modern era drama includes the
work of MacDonald P. Jackson [11], [12], Brian Vickers [13],
and Hugh Craig and Arthur Kinney [14], each of whom studied
the works of Shakespeare and his contemporaries extensively
using computational stylometry techniques.

The techniques used in authorship attribution began almost
a century ago by examining sentence lengths in texts to
determine authorship [15]. In [4], stylometric analysis first
began to consider function words as important stylistic mark-
ers, producing unprecedent results. As such, function words
have continued to be common in analysis techniques [16],
[17] due to their context independence and universal usage
in English language texts. These methods rely mainly on the
frequency of usage of function words. Numerous other stylistic
features have since been used in authorship attribution studies,
including the study of vocabulary richness [18], [19] and the
use of part of speech taggers [20].

Our method for attributing texts, developed in [21], also
measures function word usage to distinguish author styles.
Rather than only considering word frequencies, however, we
consider a more complex relational structure between an
author’s usage of function words. We construct word adjacency
networks (WANs) with function words as nodes, and edges
containing information regarding the use of two function
words within the same sentence or phrase. We interpret each
WAN as a Markov chain that assigns transition probabilities
between the appearance of two function words. We can then
measure similarity between WANs by using a measure of
relative entropy. Markov chains have previously been used
in [22] and [23] for the purposes of authorship attribution,
though neither consider the use of function words. Results
in [21] show an increase in attribution accuracy compared to
the most common frequency-based methods. We employ this
new technique then to further analyze and add insight into the
authorship disputes of Early Modern English dramatic works.

We first present an overview of the construction and com-
parison of WANs in Section II. We discuss in Section III the
main playwrights used in our analysis and the construction
of their profile networks as well as a measure of similarity
between profiles in Section IV. In Sections V-A through V-E
we perform a stylometric analysis of the complete canons of
our five primary playwrights, followed by a summary of results
in Section V-F. We are able to demonstrate high attribution
accuracy between six candidate authors. An analysis of a set



2

of plays published anonymously or without a clear author is
performed in Section VI. We then examine the use of WANs
in determining authorship of plays known to be written by
multiple authors in collaboration. This is first done by ana-
lyzing entire plays in Section VII and then through extensive
interplay analysis of a set of particularly controversial plays in
Section VIII. Our results largely corroborate existing theories
regarding these plays as well as, in some cases, propose
new authorship breakdowns. We conclude in Section IX by
providing a brief analysis of the use of WANs in distinguishing
between the three most common dramatic genres of the era:
comedy, tragedy, and history.

II. WORD ADJACENCY NETWORKS

When doing authorship attribution, we are given a set of
candidate authors A and a set of known texts written by
these authors T , and the objective is to correctly attribute
a collection of texts U of unknown authorship among the
authors in A. More precisely, the idea is to construct a function
r̂U : U → A relating every text in U with its rightful
author in A. In [21], [24], we propose an authorship attribution
method based on function word adjacency networks. For each
text, a word adjacency network (WAN) of function words,
i.e. words that convey only grammatical relationships, can be
constructed. Formally, from a given text t we construct the
network Wt = (F,Qt) where F = {f1,f2, ...,ff} is the
set of nodes composed by a collection of function words and
Qt : F × F → R+ is a similarity measure between ordered
pairs of function words.

The similarity function Qt measures the directed co-
appearance of two function words. I.e., once we encounter
a particular function word, Qt indicates the likelihood of
encountering another one in the few words following the first
one. More precisely, to compute Qt we first divide the text t
into sentences sht where h ranges from 1 to the total number
of sentences. We denote by sht (e) the word in the e-th position
within sentence h of text t. Moreover, we consider that two
words in the same sentence are related if they are at most
D ∈ N positions apart and the relation between words decays
with their position difference according to a discount factor
α ∈ (0,1). In this way, we define

Qt(fi,fj)=
∑
h,e

I
{
sht (e) = fi

} D∑
d=1

αd−1 I
{
sht (e + d) = fj

}
,

(1)
for all fi,fj ∈ F .

We then generate a profile network Wc = (F,Qc) for every
author ac ∈ A using the WANs from those texts in T known
to have been written by the corresponding author ac. Formally,
if we denote by T(c) the subset of T written by ac, then the
similarity function Qc of the profile is computed as

Qc =
∑

t∈T (c)
Qt. (2)

The similarity function Qc depends on the amount and
length of the texts written by author ac. This is a problem

since we aim to compare profiles of different authors. Thus, we
apply the following normalization to the similarity measures

Q̂c(fi,fj) =
Qc(fi,fj)∑
j Qc(fi,fj)

, (3)

for all fi,fj ∈ F . In (3) we assume that the total texts
written by author ac are long enough to guarantee a non
zero denominator for a given amount of function words |F |.
If this is not the case for some function word fi, we fix
Q̂c(fi,fj) = 1/|F| for all fj . In this way, we achieve
normalized networks P̂c = (F,Q̂c) for each author ac. The
network P̂c estimates an ideal network Pc which captures the
stylistic fingerprint of author ac. Since the similarities out of
every node sum up to 1 in the network P̂c, it can be interpreted
as a discrete time Markov chain (MC). Thus, the normalized
similarity Q̂c(fi,fj) between words fi and fj is a measure
of the probability of finding fj in the words following an
encounter of fi for texts written by author ac. Similarly, we
can use normalization (3) to build a MC Pu for each unknown
text u ∈ U.

In order to perform the attribution, we need a way of
comparing the generated MCs. By construction, every MC has
the same state space F , facilitating the comparison. Indeed, we
use the relative entropy H(P1,P2) as a dissimilarity measure
between any two chains P1 and P2. The relative entropy is
given by

H(P1,P2) =
∑
i,j

π(fi)P1(fi,fj) log
P1(fi,fj)

P2(fi,fj)
, (4)

where π is the limiting distribution of P1 and we consider
0 log 0 to be equal to 0. From [25], if we denote as w1 a
realization of the MC P1 then H(P1,P2) is proportional to
the logarithm of the ratio between the probability that w1 is
a realization of P1 and the probability that w1 is a realization
of P2. In particular, when H(P1,P2) is null, the ratio of
probabilities is 1 meaning that a given realization of P1 has
the same probability of being observed in both MCs. Thus, H
is a reasonable dissimilarity measure between MCs. Utilizing
(4) we construct the attribution function r̂U by assigning the
text u to the author with the MC most similar to Pu, i.e.

r̂U(u) = ap, where p = argmin
c

H(Pu, P̂c). (5)

Notice that the relative entropy in (5) takes an infinite value
when any word transition that appears in the unknown text
does not appear in the profile. In practice we compute the
relative entropy in (4) by summing only over the non-zero
transitions in the profiles,

H(P1,P2) =
∑

i,j|P2(fi,fj) 6=0

π(fi)P1(fi,fj) log
P1(fi,fj)

P2(fi,fj)
.

(6)
Because the calculation of relative entropy in (6) only adds rel-
ative entropy for nonzero transitions, profiles built from fewer
total words will on average contain less nonzero transitions and
will thus sum together fewer terms than larger profiles. When
attributing an unknown text among profiles of varying size, we
avoid this potential biasing for smaller profiles by summing



3

only over transitions that are nonzero in every profile being
considered,

H(P1,P2) =
∑

i,j|Pc(fi,fj) 6=0
for all ac∈A

π(fi)P1(fi,fj) log
P1(fi,fj)

P2(fi,fj)
.

(7)
In the following sections, expression (7) is used to compare
the Markov chain representations of WANs when performing
attributions following rule (5).

III. AUTHOR PROFILES

The stylometric analysis in this paper focuses on the attribu-
tion of plays written during the English Early Modern period
stretching from the late 16th century to the early 17th century.
William Shakespeare is the most prominent playwright active
in this period but there are several other authors that were also
active during this time. For most of the paper, we focus on the
authors listed below where we also detail the number of plays
that are currently attributed to each of them and the period
during which they are presumed to have written said plays1:

(1) George Chapman (1559-1634), active circa 1596-1620.
Considered sole author of a total of 13 plays plus 2
collaborations.

(2) John Fletcher (1579-1625), active circa 1605-1625. Sup-
posed to have written 47 plays, being sole author in 22
of them while Francis Beaumont and Phillip Massinger
were his main collaborators in the rest.

(3) Ben Jonson (1572-1637), active circa 1596-1637. Pre-
sumed sole author of 17 plays plus 1 collaboration.

(4) Christopher Marlowe (1564-1593), active circa 1586-
1593. Putative sole author of 6 plays and 1 collaboration.

(5) Thomas Middleton (1580-1627), active circa 1603-1625.
Believed to have written 26 plays, 14 of them as sole
author and 12 in collaboration.

(6) William Shakespeare (1564-1616), active circa 1589-
1614. A total of 38 plays are attributed to Shakespeare
and collaborators.

In the above list, we do not consider as plays minor dramatical
compositions such as masques, entertainments and pageants.
Chapman, Fletcher, Jonson, Marlowe, Middleton, and Shake-
speare are included in our analysis since they posses large and
well studied canons compared with their contemporaries.

The WAN attribution algorithm developed in [21] and
briefly reviewed in Section II uses known texts of a given
author to construct a profile against which unknown texts are
compared. Since profiling accuracy increases with the length
of the texts considered when building the profile, we build
profiles from all texts of sole authorship for a given author that
have little or no history of authorship dispute. The full list of
plays used to build the six profiles is reported in Table I. When
building profiles for a given author, we generally subscribe to
the information provided in [26] to determine texts of sole
authorship. An exception to this is Middleton, who’s profile

1Information compiled from the Database of Early English Playbooks
(DEEP) [26] and the database of catalogued plays in Chadwyck-Healey Liter-
ature Online (LION) [27]. Whenever inconsistencies in authorship information
arise, we consider [26] as accurate.

is built using the texts attributed to him in the 2007 Oxford
Collected Works of Middleton [28], which contains a more
recent and accepted study of his canon. Two plays included
in Middleton’s corpus in [28]–The Revenger’s Tragedy and
The Second Maiden’s Tragedy–were published anonymously
and have long history of disputed authorship [29]–[32]. To be
safe, we do not include these plays in Middleton’s profile but
provide an analysis of their authorship in Section VI.

Notice that each profile is built from a different number
of texts. Marlowe, the least prolific writer of the ones here
considered, is accepted as the sole author of 6 plays that
totalize 103,160 words. Shakespeare, the most prolific sole
author, is the undisputed sole author of 28 plays, totaling
679,256 words. Due to this difference, we compute the relative
entropy between the WAN of an unknown text u ∈ U and each
profile using (7) rather than (6).

In order to generate faithful representations of authors’
styles, we remove artifacts introduced by modern transcrip-
tions by using the earliest editions available of each text in
the LION database [27], with the exception of Shakespeare’s
First Quarto editions. Although Shakespeare’s canon was first
published in full in 1623, there exist earlier editions for a
number of his plays known as First Quartos. As there is
currently no scholarly consensus on which editions are more
authoritative, to be consistent we use 1623 editions for all
Shakespeare texts. When using original transcriptions we have
to account for the fact that many words had multiple accepted
spellings during the Early Modern era. E.g., the word ‘of’ is
also spelled as ‘off’, ‘offe’, or ‘o’ whereas the word ‘with’
may also appear as ‘wid’, ‘wyth’, ‘wytt’, ‘wi’, ‘wt’, and
‘wth’. Many of these alternate spellings are used infrequently
and thus do not contribute highly to the WAN of a text.
Nevertheless, we correct the WANs so that the occurrence of
any of the alternative spellings is treated as the same word.
We emphasize that spelling preferences carry little information
about the authorship of a play. Indeed, spellings in printed
editions were not necessarily those of authors as they were
often selected by printers to accommodate the fixed length of
lines in printing presses [33]. In addition, we remove speech
prefixes, or the character name preceding each speech, to avoid
cases in which character names are abbreviated to function
words (e.g. Anne abbreviated to ‘An’).

For the WANs in this work we use the optimal parame-
ters determined in [21], α = 0.75 and D = 10. Because
punctuation marks were often added by publishers rather than
the authors themselves [34], we instead delimitate sentences
at the end of character speeches. The WANs are built with
the 100 most common function words in the Early Modern
period, listed in Table II. This number is chosen based on a
training period to find the optimal number of function words
in which we attribute all texts with undisputed authorship, i.e.
those plays listed in Table I. A list of the most common Early
Modern period alternative spellings is given in Table III. For
the cases where one alternative spelling can be assigned to
multiple conventional spellings, e.g. ‘yt’ can be associated with
‘it’ and ‘that’, we assign every appearance of the alternative
spelling to the most common usage.



4

TABLE I: Plays used to build author profiles

William Shakespeare
Antony and Cleopatra (ANT) All’s Well that Ends Well (AWW)
As You Like It (AYL) The Comedy of Errors (ERR)
Coriolanus (COR) Cymbeline (CYM)
Hamlet (HAM) 1 Henry IV (1H4)
2 Henry IV (2H4) Henry V (H5)
Julius Caesar (JC) King John (JN)
King Lear (LR) Love Labour’s Lost (LLL)
The Merchant of Venice (MV) The Merry Wives of Winsdor (WIV)
A Midsummer Night’s Dream (MDB) Much Ado About Nothing (ADO)
Othello (OTH) Richard II (R2)
Richard III (R3) Romeo and Juliet (ROM)
The Taming of the Shrew (SHR) The Tempest (TMP)
Troilus and Cressida (TRO) Twelfth Night (TN)
The Two Gentlemen of Verona (TGV) The Winter’s Tale (WT)
Christopher Marlowe
Dr Faustus (DRF) Edward II (E2)
The Jew of Malta (JEW) The Massacre at Paris (MAS)
1 Tamburlaine (T1) 2 Tamburlaine (T2)
John Fletcher
Bonduca (BON) Chances (CHA)
The Faithful Shepherdess (TFS) The Humorous Lieutenant (HUM)
The Island Princess (ISL) The Loyal Subject (LOY)
The Mad Lover (TML) Monsieur Thomas (THO)
The Pilgrim (PIL) Rule a Wife and Have a Wife (RAW)
Valentinian (VAL) Wife for a Month (WFM)
The Wild Goose Chase (WGC) The Woman’s Prize (WPR)
Women Pleased (WPL)
Ben Jonson
Alchemist (ALC) Bartholomew Fair (BAR)
Catiline’s Conspiracy (CAT) Cynthia’s Revels (CYN)
The Devil is an Ass (DIA) Epicoene (EPI)
Every Man in His Humour (MIH) Every Man Out of His Humour

(MOH)
The Magnetic Lady (MAG) The New Inn (NEW)
Poetaster (POE) The Sad Shepherd (SAD)
Sejanus’s Fall (SEJ) The Staple of News (SON)
A Tale of a Tub (TUB) Volpone (VOL)
George Chapman
All Fools (ALL) Sir Giles Goosecap (SGG)
Bussy Dambois (BDA) Caesar and Pompey (CAP)
The Conspiracy of Charles Duke of
Byron (CDB)

The Tragedy of Charles Duke of By-
ron (TDB)

The Gentlemen Usher (GEN) A Humorous Day’s Mirth (HDM)
May Day (MAY) Monsieur D’Olive (MDO)
The Blind Beggar of Alexandria
(BBA)

The Revenge of Bussy Dambois
(RBD)

The Widow’s Tears (WID)
Thomas Middleton
Your Five Gallants (FIV) A Game at Chess (GAC)
A Mad World My Masters (MAD) A Chaste Maid in Cheapside (MAC)
Hengist King of Kent (HEN) Michaelmas Term (MIC)
More Dissemblers Besides Women
(DIS)

No Wit, No Help Like a Woman’s
(NOW)

The Phoenix (PHO) The Puritan Widow (PUR)
A Trick to Catch the Old One (TCO) The Widow (WID)
The Witch (WTH) Women Beware Women (BEW)

IV. SIMILARITY OF PROFILES

We compute the relative entropy between every pair of
author profiles for the six authors introduced in Section III
using expression (7); see Table IV. Every entry in the table
represents the relative entropy between the corresponding
authors in the rows and columns. In this table, as well as
in the remaining of the paper, relative entropies are multiplied
by 100 to facilitate their display. We use the term centinats,
or cn for short, to denote the resultant unit of measure of
relative entropy. The 4.7 in the Chapman row entry and
Shakespeare column entry indicates a relative entropy of 4.7cn
between Chapman’s and Shakespeare’s profiles. Note that, as
expression (7) is not symmetric, the values in the table are also
not symmetric, although they are similar in most cases. E.g.

TABLE II: List of function words used in WANs

a both in no past this while
about but into none shall those who
after by it nor should though whom
against can like nothing since through whose
all close little of so till will
an could many off some to with
and dare may on such until within
another down might once than unto without
any enough more one that up would
as every most or the upon yet
at for much other them us
away from must our then what
bar given need out therefore when
because hence neither over these where
before if next part they which

TABLE III: Common alternative spellings for function words

Conventional Alternative
it yt t
of off offe o
that thatt thate yat yt
with wid wyth wytt wi wt wth

TABLE IV: Relative entropy between profiles.

Shakespeare Fletcher Jonson Marlowe Middleton Chapman

Shakespeare 8.9 4.7 8.9 6.8 4.8

Fletcher 7.4 7.3 14.7 8.0 8.4

Jonson 4.1 7.9 11.1 6.7 5.4

Marlowe 10.1 17.4 13.0 16.5 12.9

Middleton 5.8 8.2 6.3 14.1 6.6

Chapman 4.7 9.6 5.8 11.4 7.3

4 6 8 10 12 14 16 18
4

6

8

10

12

14

16

18

Fig. 1: Asymmetry of dissimilarities in Table IV.



5

the relative entropy between Shakespeare’s and Chapman’s
profiles is 4.8cn rather than 4.7cn in the opposite direction. In
general, dissimilarities between profiles in both directions are
highly correlated as can be observed in Fig. 1. In this figure,
the coordinates of every point correspond to the dissimilarities
in both directions for every pair of profiles. The arrangement of
the points along the diagonal implies that a high dissimilarity
in one direction is associated with a high dissimilarity in
the opposite one. Hence, this correlation allows us to speak
about the similarity between two authors without specifying a
direction.

The entropy-based dissimilarities in Table IV dispel the
Marlovian theory of Shakespeare authorship [35]. If Marlowe
and Shakespeare were the same person, we should observe the
dissimilarities between Marlowe’s and Shakespeare’s profile
to be smaller than the distances between each of the other
profiles. However, the relative entropies between Marlowe’s
and Shakespeare’s profiles average 9.5cn in both directions
which is larger than the dissimilarity between Shakespeare and
all of the other authors. Shakespeare’s profile is, on average,
closest to Jonson profile – average relative entropy of 4.4cn
– although still sufficiently different so as to assert that they
belong to different authors, as verified by the attribution of
plays in Section V. The highest dissimilarity among any pair
of profiles occurs between Marlowe and Fletcher with a mean
of 16.1cn. As will be seen in Section V, the relative similarity
between two profiles affects our ability to distinguish between
them when attributing a text.

V. ATTRIBUTION OF PLAYS

We attribute the plays written by Jonson, Middleton, Chap-
man, Marlowe, and Shakespeare among the 6 author profiles
introduced in Section III. The attribution of Fletcher’s plays is
performed in the discussion of collaborations in Section VII.
When attributing any given play, profiles are built using the
plays listed in Table I excluding the play being attributed. We
do not report raw relative entropy values between the play
being attributed and the author profiles, but instead subtract
from these values the relative entropy between the play and
a profile containing all available texts. Intuitively, the profile
containing all of the texts represents the writing style of an
average playwright from this period. This is done to make
the figures easier to view but does not change the results in
any way. Each raw relative entropy value is discounted by
the same constant value, thus preserving relative distances.
As a result, both negative and positive relative entropy values
are possible. A negative relative entropy value indicates that
the play’s WAN is more similar to the author profile than to
the profile of the average playwright while a positive relative
entropy indicates the opposite.

A. Ben Jonson

In Fig. 2 we present the attribution of the 17 plays believed
to have been written solely by Jonson, plus one collabora-
tion. In the horizontal axis we present the plays to attribute
and the vertical axis represents the entropy distance (7) in
cn from these plays to the different profiles identified with

TABLE V: Thomas Middleton plays to be attributed in addi-
tion to those listed in Table I.

The Bloody Banquet (BAN) The Changeling (CHG)
A Fair Quarrel (AFQ) The Family of Love (FAM)
The Patient Man and The Honest
Whore (THW)

Match at Midnight (MAM)

The Old Law (TOL) The Roaring Girl (TRG)
Anything for a Quiet Life (AGL) The Spanish Gypsy (TSG)
Wit at Several Weapons (WEA)

distinct markers and discounted by the distance to the average
playwright.

Among the mentioned 18 total plays, including the collab-
oration, an accuracy of 94% is achieved, correctly attributing
17 of these plays to Jonson, i.e., the entropy distance of every
play to the profiles achieves its minimum for Jonson’s profile.
The play, Eastward Ho, is accepted as a collaboration between
Jonson and Chapman plus a third author, John Marston, whom
we have not profiled. Here Chapman is not well ranked,
suggesting that his contributions were minor compared with
Jonson’s. The relative contributions of both Jonson and Chap-
man in Eastward Ho are analyzed further in Section VIII-A.

The misattribution in Fig. 2 for plays solely written by
Jonson occurs for The Case is Altered, which is misattributed
by a small margin. Mixed authorship has been suggested due
to the irregularity in the structure of the last two acts [36].
The play’s content has also been compared to A Comedy of
Errors, written by Shakespeare, who is also here the closest
author [37]. Another play, Sejanus His Fall, is attributed to
Jonson but only by a small margin. It has been pointed out
that this play might contain elements of a second author, with
both Shakespeare and Chapman a possible candidates [38],
[39]. Our analysis indicates that the play is closer to the style
of Shakespeare than to the style of Chapman. Sejanus His Fall
is also one of only two tragedies Jonson published–the other
being Catiline His Conspiracy–possibly biasing results against
a profile built almost entirely with comedies. The relationship
between genre and attribution is explored further in Section
IX.

B. Thomas Middleton

In Fig. 3 we present the attribution of 28 plays, 14 of
which are generally believed to have been written only by
Middleton and 12 in collaboration. We also include in our
set two plays originally assigned to Middleton–The Family of
Love and Match at Midnight–but not included in his corpus
in [28].

Among the 14 plays believed to be solely written by
Middleton, we attribute 12 to Middleton obtaining an accuracy
of 85.7%. The first misattributed play, A Game at Chess,
is attributed to Shakespeare by a very small margin, likely
due to random error. This is also true in the case of Hengist
King of Kent, noted for being the sole history play Middleton
produced. Additionally, although [28] does not find evidence
of Middleton in The Family of Love or Match at Midnight, our
results show that he is at least a stronger candidate in these
plays than the other five authors. The low relative entropy



6

−2

0

2

4

6

8

10

R
e

la
tiv

e
 e

n
tr

o
p

y 
to

 p
ro

fil
e

(c
e

n
tin

a
ts

)

 

 

AL
C

BA
R

CI
A

CA
T

CY
N

DI
A

HO EP
I

M
IH

M
O
H

M
AG

NE
W

PO
E

SA
D

SE
J

SO
N

TU
B

VO
L

Shakespeare
Fletcher
Jonson
Marlowe
Middleton
Chapman

Fig. 2: Attribution of Jonson plays. We attribute the 16 plays in Table I plus The Case is Altered (CIA) and Eastward Ho
(HO). A single misattribution occurs for The Case is Altered.

−4

−2

0

2

4

6

8

10

12

R
e

la
tiv

e
 e

n
tr

o
p

y 
to

 p
ro

fil
e

(c
e

n
tin

a
ts

)

BA
N

CH
G

DI
S

AF
Q

FA
M

FI
V

G
AC

TH
W

M
AD

M
AC

M
AM HE

N
M

IC
NO

W
TO

L
PH

O
PU

R
AQ

L
RO

A
TS

G
TC

O
W

ID
W

EA
W

TH
BE

W
M

AC
M

EA TI
M

Fig. 3: Attribution of Middleton plays. We attribute the 14 plays in Table I, the additional 11 plays in Table V and 3 collaborations
with Shakespeare: Macbeth (MAC), Measure for Measure (MEA) and Timon of Athens (TIM). Only 2 sole authored plays are
misattributed. Also, the attribution of Shakespeare’s plays reveal that Middleton’s contribution was minor.



7

−1

0

1

2

3

4

5

6

7

8

9

R
e

la
tiv

e
 e

n
tr

o
p

y 
to

 p
ro

fil
e

(c
e

n
tin

a
ts

)

AD
M

AL
L

BB
A

BD
A

CD
B

CA
P

TD
B

HO G
EN

HD
M

M
AY

M
DO RB

D
SG

G
W

ID

Fig. 4: Attribution of Chapman plays. We attribute the 13
plays in Table I plus The Tragedy of Chabot Admiral of
France (ADM) and Eastward Ho (HO). Out of 15 plays, 10 are
attributed to Chapman. Collaboration with Jonson in Eastward
Ho can also be observed.

value of −3cn between Middleton’s profile and the WAN of
The Family of Love adds evidence to the claim that Middleton
contributed to this play [40].

Among the 12 collaborative plays, 7 are attributed to Mid-
dleton. Thomas Dekker and William Rowley were Middleton’s
usual collaborators. As neither of these authors are profiled,
each of the plays written with these authors is attributed here to
Middleton with the exception of The Bloody Banquet, which is
marginally attributed to Shakespeare over Middleton. Another
misattributed play, The Spanish Gypsy is usually considered
to be a collaboration between Middleton, Dekker, Rowley,
and John Ford [40], [41] which may explain why Middleton
is ranked second behind another author. We also attribute
Middleton’s three collaborations with Shakespeare. Measure
for Measure, Timon of Athens, and Macbeth are correctly
attributed to Shakespeare. Moreover, for these three plays,
Middleton is ranked very poorly being the fourth preferred
candidate in all of them. This supports the accepted idea that
Middleton’s contribution in these three plays is minimal [42],
[43]. We examine these plays in closer detail in Section VIII-C.

C. George Chapman

Chapman is considered to be the author of 15 plays, 13 as
a sole author and 2 in collaboration. In Fig. 4, we attribute
these plays among the 6 profiles. In total, 10 of the 15 plays
are attributed to Chapman. In the cases of plays written in
collaboration, The Tragedy of Chabot, Admiral of France is
attributed to Chapman while Eastward Ho is attributed to
Jonson, as discussed in Section V-A. Notice that of the four
remaining misattributions, three are assigned to Shakespeare
with Chapman as the second preferred candidate. This is
consistent with the fact that in Table IV, Chapman’s profile is
most similar to Shakespeare. Thus, cases of random error will
therefore most likely attribute to Shakespeare.

−12

−10

−8

−6

−4

−2

0

2

4

6

8

R
e

la
tiv

e
 e

n
tr

o
p

y 
to

 p
ro

fil
e

(c
e

n
tin

a
ts

)

 

 

DI
D

DR
F E2

JE
W

M
AS T

1 T2

Shakespeare
Fletcher
Jonson
Marlowe
Middleton
Chapman

Fig. 5: Attribution of Marlowe plays. We attribute the 6 plays
in Table I plus Dido Queen of Carthage (DID). A single
misattribution occurs for the collaborative play Dido Queen
of Carthage.

TABLE VI: William Shakespeare plays to be attributed in
addition to those listed in Table I.

1 Henry VI (1H6) 2 Henry VI (2H6)
3 Henry VI (3H6) Henry VIII (H8)
Macbeth (MAC) Measure for Measure (MEA)
Pericles (PER) Timon of Athens (TIM)
Titus Andronicus (TIT) Two Noble Kinsmen (TNK)

D. Christopher Marlowe

In Fig. 5, we present the attribution of 7 plays believed
to have been written by Marlowe, where Dido, Queen of
Carthage is the only collaborative work, with Thomas Nashe
as coauthor. We achieve an accuracy of 100% in attributing
Marlowe’s sole works. Dido Queen of Carthage is attributed
to Shakespeare by a small margin, with Marlowe as the second
best candidate.

In the case of sole authorship plays, each is attributed to
Marlowe by a substantial margin and with relative entropies
between −6cn and −13cn. These large negative values sug-
gest that the plays are much more similar to Marlowe’s profile
than they are to the profile of an average playwright. This
difference may be a result of the fact that Marlowe’s plays
were written at least a decade before most of the other authors
considered, thus indicating a shift in writing style during the
one or two decades that separate Marlowe from the rest.

E. William Shakespeare

In Fig. 6 we present the attribution of 38 plays believed to
have been written by Shakespeare, 30 of which are attributed
solely to Shakespeare in [26]. Note that 2 of the 30 sole
authored plays, 2 Henry VI and 3 Henry VI are not included
in Shakespeare’s profile in Table I because they have a strong
history of disputed authorship [14].

All of the 30 plays usually considered to have been written
only by Shakespeare are correctly attributed. However, excep-



8

−2

0

2

4

6

8

10

R
e

la
tiv

e
 e

n
tr

o
p

y 
to

 p
ro

fil
e

(c
e

n
tin

a
ts

)

 

 

AW
W
AN

T
AY

L
ER

R
CO

R
CY

M
HA

M
1H

4
2H

4 H51H
6
2H

6
3H

6 H8 JC JN LRLL
L
M

ACM
M M

V
W

IV
M

NDAD
O
O
TH

PE
R R2 R3

RO
M
SH

R
TM

P
TI

M TI
T
TR

O TNTG
V
TN

K
W

T

Shakespeare
Fletcher
Jonson
Marlowe
Middleton
Chapman

Fig. 6: Attribution of Shakespeare plays. We attribute the 28 plays in Table I and the additional 10 plays in Table VI. All plays
are attributed to Shakespeare. Marlowe’s distance to a play is highly dependent on whether the analyzed play is a history play
or not, emphasizing the impact of genre in attribution.

tional situations arise for the plays 1 Henry VI, 2 Henry VI, and
3 Henry VI, in which Marlowe is ranked uncharacteristically
high. The fact that Marlowe is ranked second for these plays
is noteworthy, since Marlowe’s profile is very dissimilar from
Shakespeare’s in Table IV. Consequently, he ranks poorly in
the attribution of most plays. In addition, the relative entropy
between Marlowe’s profile and the WANs of most plays is
between +6cn and +10cn, while the relative entropy between
Marlowe’s profile and these plays’ WANs is around +2cn.
Similarly, the relative entropy between Marlowe’s profile and
the WANs of Henry V, King John, Richard II, and Richard III
is around +4cn. These seven plays have in common that they
are history plays, a genre in which Marlowe wrote Edward
II and Massacre at Paris, comprising a third of his profile.
Thus, there is a genre bias of history plays towards Marlowe.
Focusing on the Henry VI saga, where the first part is a
known collaboration of Shakespeare with Nashe, we see a
particularly strong signature of Marlowe in the three plays
compared to Shakespeare’s other history plays. Moreover,
these plays were written during Marlowe’s most fertile years
and Marlowe had collaborated with Nashe in 1589 – two years
before the Henry VI saga – when writing his play Dido Queen
of Carthage. This supports the hypothesis that there was an
unknown collaborator in these plays [44], [45] and points at
Marlowe as a probable candidate. These collaborations are
covered in greater detail in Section VIII-D.

Among the 8 plays of accepted collaboration with others,
besides the mentioned collaboration in 1 Henry VI, we can
find the three collaborations with Middleton already analyzed

in Section V-B. From the poor ranking of Middleton in the
attribution pattern, we can conclude that Middleton’s revisions
and contributions were minor. There are also two collabora-
tions with Fletcher, namely Henry VIII and The Two Noble
Kinsmen. We attribute both to Shakespeare, with Fletcher
the second preferred author in the latter. In the case of the
former, on the other hand, Fletcher is not well ranked and
his contribution is not evident from the attribution of the
entire play. Shakespeare’s collaborations with both Fletcher
and Middleton are analyzed further in Sections VIII-B and
VIII-C, respectively.

F. Summary of Results

In total, we attribute correctly 71 out of the 77 plays we
consider that are traditionally attributed to single author and
listed in Table I, yielding an accuracy of 92.2%. Furthermore,
if we only consider attributions between authors that are more
than 5cn apart, then we fail only in 3, yielding an accuracy
of 96.1%. We utilize the high classification power for plays
of sole authorship to shed light on attribution problems of
anonymous plays written during the Early Modern period in
Section VI.

Of the 20 plays we consider that are generally accepted to
be collaborations, we attribute 17 to one of the contributing
authors, yielding an accuracy of 85%. Collaborative plays are
analyzed further in Sections VII and VIII.



9

TABLE VII: List of texts of unknown authorship.

Arden of Faversham (ARD) Edward III (E3)
Fair Em (FEM) Mucedorus (MUC)
The Nice Valor (TNV) The Revenger’s Tragedy (REV)
The Second Maiden’s Tragedy (SMT) Taming of a Shrew (TAS)

VI. ANONYMOUS PLAYS

In Fig. 7 we present the attribution of 8 anonymous plays
written during the English Renaissance. Authorship of some
of these plays have been more discussed and studied by
scholars than others. E.g., Edward III is commonly attributed
in part to Shakespeare [13] and our method supports this
theory. Indeed, this play was written during the early stages of
Shakespeare’s career and the Shakespeare profile is the closest.
Another play sometimes attributed in part to Shakespeare is
Arden of Faversham [13]. Again, our method supports this
theory. These plays are analyzed further in Section VIII-D.
In addition, the plays The Revenger’s Tragedy, The Second
Maiden’s Tragedy, and The Nice Valor are usually attributed
to Middleton [28], with the former two included in the 2007
Oxford Collected Works. Our method indeed attributes all
three works to Middleton. Furthermore, Fletcher is the second
attributed author of The Nice Valor, a play originally included
in the Beaumont and Fletcher folios of 1647 and 1679 [46],
leading some to believe that the play is a collaboration between
Fletcher and Middleton.

For the remaining plays, definite statements cannot be made,
but we can support or undermine existing hypothesis. For
example, Mucedorus may have been written by Shakespeare
as proposed by a number of scholars [47] since he is the first
ranked author among the six authors we profile. Fair Em has
also been assigned to Shakespeare [48] though there is no
scholarly consensus, with Robert Wilson, whom we do not
profile, often cited as a likely candidate [47]. The Taming of
a Shrew, the play generating controversies about the better
known Shakespeare play with similar title, is here attributed
to the Shakespeare profile. Note, also, that Marlowe is ranked
atypically high for this play–second behind Shakespeare. Both
Shakespeare and Marlowe have been proposed as candidates
for Taming of a Shrew [49], in the former case as a possibly
early draft of Taming of the Shrew. While our analysis points
to Shakespeare as a more likely candidate, observe that the
attribution of Taming of the Shrew in Fig. 6 ranks Marlowe as
the worst candidate, indicating that much more of his style is
evident in the early draft.

VII. COLLABORATIONS

In cases of multiple authors contributing to a single play,
we show how our method is still able to detect one or more of
the authors present in a full text by identifying the top ranked
authors in its attribution.

A. John Fletcher and collaborators

John Fletcher wrote numerous plays both by himself and
with collaborators. Consequently, his canon is an appropriate
text corpus to analyze the attribution of collaborative plays. In

0

2

4

6

8

10

12

R
e
la

tiv
e
 e

n
tr

o
p
y 

to
 p

ro
fil

e
(c

e
n
tin

a
ts

)

 

 

AR
D E3

FE
M

M
UC TN

V
RE

V
SM

T
TA

S

Shakespeare
Fletcher
Jonson
Marlowe
Middleton
Chapman

Fig. 7: Attribution of anonymous plays listed in Table VII. Our
method supports the usual theories of Shakespeare’s hand in
Edward III and Arden of Faversham. Also, Middleton’s style
in The Revenger’s Tragedy, The Second Maiden’s Tragedy, and
The Nice Valor can be observed, in accordance with current
authorship consensus.

TABLE VIII: Plays used to build profiles for Fletcher &
Beaumont and Fletcher & Massinger.

Fletcher & Beaumont
The Coxcomb (COX) Philaster (PHI)
The Woman Hater (TWH) Cupid’s Revenge (CUP)
A King and No King (KNK) Love’s Pilgrimage (PIL)
The Maid’s Tragedy (TMT) The Scornful Lady (TSL)
Fletcher & Massinger
The Custom of the Country (COC) The Double Marriage (TDM)
The Elder Brother (TEB) The False One (TFO)
John Van Olden Barnavelt (JVO) The Little French Lawyer (LFL)
The Lover’s Progress (LP) The Prophetess (PRO)
The Sea Voyage (SEA) Spanish Curate (TSC)
A Very Woman (TVW)

addition to the six profiles in the previous section, we include
two profiles built from plays written with Fletcher’s two most
frequent coauthors–Francis Beaumont and Phillip Massinger;
see Table VIII.

The attribution of Fletcher’s works are divided into two
plots. Fig. 8 shows the attribution of plays believed to have
been written solely by Fletcher and Fig. 9 shows the attribution
of plays believed to have been written in collaboration with
other authors. The set of plays presented before the first

TABLE IX: John Fletcher plays to be attributed in addition to
those listed in Table I.

Solo
Beggars’ Bush (BB) The Captain (CAP)
The Fair Maid of the Inn (FAI) The Noble Gentlemen (TNG)
The Queen of Corinth (QOC) Wit Without Money (WIT)
Collaborations
Henry VIII (H8) The Knight of Malta (KOM)
The Maid in the Mill (MIL) The Night Walker (NW)
Four Plays in One (FP) Two Noble Kinsmen (TNK)
Wit at Several Weapons (WEA) Love’s Cure (CUR)
The Bloody Brother (BRO) Thierry and Theodoret (THI)
Wandering Lovers (WAN)



10

−3

−2

−1

0

1

2

3

4

5

6

7

R
e

la
tiv

e
 e

n
tr

o
p

y 
to

 p
ro

fil
e

(c
e

n
tin

a
ts

)

 

 

BB
BO

N
CA

P
CH

A
FA

I
TF

S
HU

M IS
L

LO
Y

TM
L

TH
O

TN
G PI

L
Q
O
C

RA
W

VA
L

W
FM

W
G
C

W
IT

W
PR

W
PL

Shakespeare
Fletcher
F & Beaumont
F & Massinger
Jonson
Middleton
Chapman

Fig. 8: Attribution of solo Fletcher plays. We attribute the 15 plays in Table I and the additional 6 plays in Table IX. Six
plays are not attributed to the sole Fletcher profile and, among these, two plays are attributed to collaborative profiles including
Fletcher.

red line include attributions of plays written with Francis
Beaumont. The second division shows the attribution of plays
written with Phillip Massinger and the third division shows
the attribution of plays written with a mix of other authors.
In both figures we omit the marker corresponding to Marlowe
since he is poorly ranked for every play. This is consistent
with Fletcher and Marlowe having the most dissimilar writing
styles; see Table IV.

In Fig. 8, 15 out of 21 plays are attributed to the solo
Fletcher profile. Of the six plays attributed to other profiles,
two of them, The Captain and Queen of Corinth are attributed
to one of the profiles for Fletcher and a collaborator. Beggar’s
Bush is marginally assigned to Shakespeare and Jonson. The
Faithful Shepherdess, The Noble Gentleman, and The Fair
Maid of the Inn are mistakenly assigned to Shakespeare as
well, with Fletcher and Massinger ranked second. For the
latter, existing theories attribute the play to a collaboration of
four authors, two of which are Fletcher and Massinger, with
Fletcher’s contribution being minor [50]. This would explain
the fact that the Fletcher and Massinger profile is ranked
second but the sole Fletcher profile is poorly ranked.

In Fig. 9, 7 of the 9 Fletcher and Beaumont plays are
attributed to the Fletcher and Beaumont profile, while Phi-
laster is assigned to the sole Fletcher profile. A single mistake
occurs for Love’s Cure, a play historically attributed to many
different authors [51]. Additionally, all of the 14 Fletcher and
Massinger plays are assigned to the Fletcher and Massinger
profile. One of the three Fletcher profiles are also listed as
the top candidate in 3 out of the 7 plays written by Fletcher

TABLE X: Plays used to build profiles for Robert Greene and
George Peele.

Robert Greene
Friar Bacon and Friar Bungay Orlando Furioso
James IV Alphonsus, King of Aragon
George Peele
The Arraignment of Paris Edward I
The Battle of Alcazar The Love of King David and Fair Bethsheba
Old Wive’s Tale

with other collaborators. Of the four mistakes, two are the
plays coauthored with Shakespeare and discussed previously
in Section V-E and further in Section VIII-B. These examples
demonstrate that our tool remains effective even in cases of
mixed authorship and, in many cases, favors profiles built from
multiple contributing authors over profiles built from a single
contributing author.

VIII. COLLABORATIONS – INTRAPLAY ANALYSIS

We examine the authorship of collaborative plays through
the attribution of its individual acts and scenes. In Section
VII we analyzed examples of detecting collaboration in full
plays by looking at the top candidate authors. This does not,
however, suggest any particular breakdown of which sections
of the text were contributed by which author. Instead, we
may attribute pieces of the play separate from one another
to gain deeper insight as to how the play was written. We also
see cases where we can detect collaboration through intraplay
analysis where we could not when attributing the full text.



11

−4

−2

0

2

4

6

R
e

la
tiv

e
 e

n
tr

o
p

y 
to

 p
ro

fil
e

(c
e

n
tin

a
ts

)

 

 

CO
X
CU

P
KN

K
CU

R
TM

T
PH

I
TS

L
TW

H
PI

L
BR

O
CO

C
TD

M
TF

O
JV

O
LF

L
PR

O
SE

A
TS

C
TH

I
W

AN TE
B
AV

W LP H8
KO

M
M

IL
NW TN

K FP
W

EA

Shakespeare
Fletcher
F & Beaumont
F & Massinger
Jonson
Middleton
Chapman

Fig. 9: Attribution of Fletcher plays written with collaborators. We attribute the plays listed in Tables VIII and IX. The first
division includes plays written with Beaumont and 7 out of 9 are correctly assigned to the Fletcher & Beaumont profile. The
second division includes plays written with Massinger and all 14 plays are assigned to the Fletcher & Massinger profile. The
third division includes plays written with other collaborators and 3 out of 7 are assigned to a Fletcher profile. Out of 30 plays,
a total of 25 are assigned to a Fletcher profile.

TABLE XI: Function words used in the attribution of indi-
vidual acts, determined in the training process. A total of 76
words are used.

a both like nor shall they when
about but little nothing should this where
against by many of since those which
all can may off so though who
an could might on some till whose
and for more once such to will
any from most one that unto with
as if much or the up without
at in must other them upon would
away into no our then us yet
before it none out these what

TABLE XII: Function words used in the attribution of indi-
vidual scenes, determined in the training process. A total of
55 words are used.

a for no some upon
all from nor such us
an if of that what
and in on the when
any it one them where
as like or then which
at may our these who
away more out they will
but most shall this with
by much should to would
can must so up yet

In the following sections we attribute plays of known or
suggested collaboration between the six original candidate
authors as well as two new authors: Robert Greene and George
Peele. The plays used to construct Greene’s and Peele’s profile
are listed in Table X. Additionally, we re-train the WAN
networks due to the fact that smaller WANS increase the
attribution accuracy of shorter texts. This is because shorter
texts are less likely to contain less common function words.
As a result, larger networks that contain these less common
function words are more prone to over-fit to features of specific
texts rather than author style. From the training period, we
achieve accuracies of 93.4% and 91.5% for acts and scenes,
respectively. Note that in the case of scene attribution, this is
the accuracy of binary attribution, whereas the act attribution
is performed between eight candidate authors. The words used
in the resulting networks are listed in Tables XI and XII.

The figures display for each act or scene the difference
in relative entropy when comparing the two top candidate
authors, reflected by both the color of the bars and the titles
above and below the plot. A longer bar in a particular direction
indicates a larger difference between the entropies of the two
candidate authors. For example, in Fig. 11, red bars extending
upwards indicate an attribution to Shakespeare while blue bars
extending downwards indicate an attribution to Fletcher. In the
attribution of acts, we identify the two top authors as the two
highest ranked, whereas the attribution of scenes we consider
the two authors most often cited as candidates. In many cases,



12

1 2 3 4 5
0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0

0.5

1

1.5

2

2.5

1.
1

1.
2

2.
1

2.
2

3.
1

3.
2

3.
3

4.
1

4.
2

5.
1

5.
2

5.
3

5.
4

5.
5

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Chapman

Jonson

Fig. 10: Attribution of acts and scenes of Eastward Ho. Note
that Act 3 is assigned to Shakespeare over both Jonson and
Chapman.

the acts and scenes will be attributed between the same pair
of authors. Cases in which an act is attributed to a third author
are marked in the figure captions.

A. Jonson and Chapman

We attribute both the individual acts and scenes of the
single known collaboration between Jonson and Chapman,
Eastward Ho, which also includes contributions from a third
author, John Marston. Fig. 10 displays the results of the act
and scene attribution. Every act is assigned to Jonson, with
the exception of Act 3 assigned to Shakespeare. Chapman is
ranked either third or forth in all acts except Act 3 in which
he is ranked second. These results are similar to the full play
attribution from Figs. 2 and 4, in which Jonson was the top
ranked author and Chapman was not well ranked. While these
results on their own do not support Chapman’s contribution, a
look at the scene attribution does reveal some of Chapman’s
possible contributions. Most of the play is still assigned to
Jonson, however Chapman is seen as a more likely candidate

1 2 3 4 5
0

2

4

6

8

10

0

2

4

6

8

10

1.
1

1.
2

1.
3

1.
4

1.
5

2.
1

2.
2

2.
3

2.
4

2.
5

2.
6

3.
1

3.
2

3.
3

3.
4

3.
5

3.
6

4.
1

4.
2

4.
3

5.
1

5.
2

5.
3

5.
4

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Fletcher

Shakespeare

Fig. 11: Attribution of acts and scenes of Two Noble Kinsmen.

1 2 3 4 5
0

2

4

6

8

10

0

2

4

6

8

10

1.
1

1.
2

1.
3

1.
4

2.
1

2.
2

2.
3

2.
4

3.
1

3.
2

4.
1

4.
2

5.
1

5.
2

5.
3

5.
4

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Fletcher

Shakespeare

Fig. 12: Attribution of acts and scenes of Henry VIII.

in scene 3.3 and 5.2 whereas the attribution of scenes 3.1-
2 is too close to make any conclusion. While there is not a
scholarly consensus on the scene breakdown, many attribute
Marston to Act 1, Chapman to Act 2 and 3, and Jonson to Act
5 [36]. Most scholars agree in particular about scene 3.3 being
written by Chapman [44]. Our results support the notion that
Chapman did not write Act 1 and Jonson wrote Act 5. We also
provide further evidence that Chapman wrote 3.3, as it is, in
our analysis, the single scene that is assigned to Chapman by a
margin larger than 2cn. We also, however, find more evidence
of Jonson contributing Acts 2 and 4 than Chapman.

B. Shakespeare and Fletcher

In Fig. 11 we show the attribution of individual acts and
scenes of Two Noble Kinsmen, a known collaboration between
Shakespeare and Fletcher. Whereas in Fig. 9 the play is
assigned to Shakespeare with Fletcher as the second best
candidate, here Acts 1 and 5 are assigned to Shakespeare
while Acts 2 and 3 are assigned to Fletcher. Act 4 is assigned
to Fletcher with Shakespeare and Jonson close behind. A
closer look into the scene breakdown reveals more specific

1 2 3 4 5
0

2

4

6

8

10

12

14

16

0

2

4

6

8

1.
1

1.
2

1.
3

1.
4

1.
5

1.
6

1.
7

2.
1

2.
2

2.
3

2.
4

3.
1

3.
2

3.
3

3.
4

3.
5

4.
1

4.
2

5.
1

5.
2

5.
3

5.
4

5.
5

5.
6

5.
7

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Middleton

Shakespeare

Fig. 13: Attribution of acts and scenes of Macbeth.



13

1 2 3 4 5
0

5

10

15

0

2

4

6

1.
1

1.
2

1.
3

1.
4

1.
5

2.
1

2.
2

2.
3

2.
4

3.
1

3.
2

4.
1

4.
2

4.
3

4.
4

4.
5

4.
6

5.
1

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Middleton

Shakespeare

Fig. 14: Attribution of acts and scenes of Measure for Mea-
sure.

assignments. Shakespeare is assigned to scenes 1.1-4, 2.1, 3.1-
2, 4.3, 5.1, and 5.3-4, Fletcher is assigned to scenes 1.5, 2.2,
2.4-6, 3.3-6, and 4.1-2, and close ties in scenes 2.3 and 5.2.
The scene breakdown we propose largely supports the one
given by Hallet Smith in The Riverside Shakespeare [52].

The act and scene analysis of Shakespeare and Fletcher’s
other collaboration–Henry VIII–is displayed in Fig. 12. Recall
that, when attributing the full play, Shakespeare was the
top candidate while Fletcher was in fact ranked fourth, thus
revealing no evidence of collaboration; see Fig. 6 or Fig.
9. We see similar results in Fig. 12, in which Shakespeare
is assigned every act. Fletcher, again, is ranked poorly in
every act. A scene-by-scene analysis between Shakespeare
and Fletcher however, does reveal Fletcher to be a stronger
candidate than Shakespeare in several individual scenes. In
fact, the scene breakdown we observe–in which Shakespeare
is assigned scenes 1.1-2, 2.1-2, 2.4, 3.2, 4.1-2, and 5.1-2 and
Fletcher is assigned scenes 1.3-4, 3.1, and 5.4, and 2.3 and
5.3 ties between both authors–is aligned to that proposed by
Cyrus Hoy [46] and currently accepted by many scholars.
The primary area of disparity between the breakdown we
propose and the one given by Hoy is the authorship of Act
4. While Hoy assigns Act 4 to Fletcher, we find that there
is greater evidence that Shakespeare contributed this section.
Both scenes are attributed to Shakespeare by a significant
margin of at least 5cn. Another point of contention is the
assignment of 2.3–given to Shakespeare by Hoy–to Fletcher
by a small margin.

The attribution of Henry VIII shows a clear example of
using intraplay analysis to detect collaboration at the level of
scenes that may be undetectable when looking at entire plays
or acts. In this play, there are several individual scenes that
attribute to Shakespeare by a margin as wide as 7cn, such
as scenes 1.2, 2.4, 4.1, and 5.1, that bias the attribution of
complete acts in favor of Shakespeare, while the scene to scene
analysis provides a clearer perspective.

1 2 3 4 5
0

2

4

6

8

10

0

1

2

3

4

5

1.
1

1.
2

2.
1

2.
2

3.
1

3.
2

3.
3

3.
4

3.
5

3.
6

4.
1

4.
2

4.
3

5.
1

5.
2

5.
3

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Middleton

Shakespeare

Fig. 15: Attribution of acts and scenes of Timon of Athens.

C. Shakespeare and Middleton

We analyze in Figs. 13-15 Middleton’s contributions to
Shakespeare’s plays, Macbeth, Measure for Measure, and
Timon of Athens. The attribution of the full plays in Fig. 3 did
not suggest that Middleton made any significant contribution
to any of these plays. The intraplay analysis of Macbeth at
the level of acts and scenes, shown in Figure 13, supports this
conclusion. A total of two scenes are assigned to Middleton
over Shakespeare, namely scenes 1.1 and 5.1. Scene 5.1 is
attributed to Shakespeare by only a small margin of 1cn while
scene 1.1 is assigned by a more substantial margin of 3cn.
Scholars have often flagged scenes 1.2, 3.5, and 4.1 as scenes
revised or contributed by Middleton [43], although we do not
find evidence of this in our analysis.

The case of Measure for Measure favors Shakespeare’s
sole authorship even more; both the act and scene analysis
displayed in Fig. 14 find Shakespeare to be the sole author of
the play. If Middleton had indeed revised the original play as
proposed by scholars [43], [53], we do not find evidence it
was substantial.

Of the three plays, we find that Middleton’s contribution
was likely largest in Timon of Athens. While all five acts
attribute to Shakespeare, in Act 3 it is by a margin less than
1cn from Middleton; see Fig. 15. This is even more evident
in the scene analysis. Middleton is a stronger candidate in
scenes 1.2, 3.2, and 3.4, with close ties in scenes 3.1, 3.3, and
4.2. This assignment supports much of the claim of authorship
provided in [13], [43].

D. Shakespeare and Marlowe

Although there are no unanimously agreed upon collabora-
tions between Shakespeare and Marlowe, there exist a number
of plays with controversial authorship that have been the sub-
ject of scholarly treatment regarding Marlowe’s contributions.
Of these, we examine the three parts of Henry VI as well as
the anonymous plays Arden of Faversham and Edward III.

As suggested by the results in Fig. 6, the three parts of
Henry VI have been considered as possible collaborations
between Shakespeare and Marlowe [14], though others such as



14

1 2 3 4 5
0

1

2

3

4

5

6

7

0

1

2

3

4

5

6

7

1.
1

1.
2

1.
3

1.
4

1.
5

1.
6

2.
1

2.
2

2.
3

2.
4

2.
5

3.
1

3.
2

3.
3

3.
4

4.
1

4.
2

4.
3

4.
4

4.
5

4.
6

4.
7

5.
1

5.
2

5.
3

5.
4

5.
5

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Marlowe

Shakespeare

Fig. 16: Attribution of acts and scenes of 1 Henry VI.

Greene and Peele have also been suggested. The attribution of
the acts of 1 Henry VI, displayed in Fig. 16, suggests that Act
1 could have been written by someone other than Shakespeare.
It is here attributed evenly between Shakespeare and Jonson
with Marlowe the next preferred candidate. Although Jonson
is generally not considered a candidate for this play, it may
suggest a similar author we do not profile. The rest of the
play is assigned to Shakespeare and, in the case of Acts 3 and
4, by a wide margin from second candidate Marlowe. The
scenes are attributed between Shakespeare and Marlowe. In
line with the act attribution, three scenes in Act 1 (1.1, 1.5-6)
attribute to Marlowe rather than Shakespeare. Other scenes that
attribute to Marlowe include 3.2, 3.4, 4.2, 5.1-2. Scene 4.2 in
particular is attributed to Marlowe by a large margin of almost
6cn. These results support parts of the breakdown suggested
by Hugh Craig [14], namely the attribution of someone other
than Shakespeare in Act 1 as well as Shakespeare in scenes
4.3-7. Although Craig contends that Marlowe likely wrote the
scenes involving Joan of Arc, we find only half of the Joan
of Arc scenes (1.5-6, 3.2, 5.2) to be more like Marlowe than
Shakespeare.

The act and scene attribution of 2 Henry VI is shown in

1 2 3 4 5
0

2

4

6

8

10

0

1

2

3

4

5

1.
1

1.
2

1.
3

1.
4

2.
1

2.
2

2.
3

2.
4

3.
1

3.
2

3.
3

4.
1

4.
2

4.
3

4.
4

4.
5

4.
6

4.
7

4.
8

4.
9

4.
10 5.

1
5.

2
5.

3

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Marlowe

Shakespeare

Fig. 17: Attribution of acts and scenes of 2 Henry VI.

1 2 3 4 5
0

2

4

6

8

10

0

2

4

6

8

10

1.
1

1.
2

1.
3

1.
4

2.
1

2.
2

2.
3

2.
4

2.
5

2.
6

3.
1

3.
2

3.
3

4.
1

4.
2

4.
3

4.
4

4.
5

4.
6

4.
7

4.
8

5.
1

5.
2

5.
3

5.
4

5.
5

5.
6

5.
7

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Marlowe

Shakespeare

Fig. 18: Attribution of acts and scenes of 3 Henry VI. Note
that the relative entropy for scene 4.2 extends out of the view
of the figure to +30cn.

Fig. 17. Act 1 is assigned to Marlowe and the rest is assigned
to Shakespeare, with Act 4 being a close tie between them.
In the former case, Shakespeare is the third candidate author
behind Peele. The scene analysis assigns to Marlowe scenes
1.1, 1.4, 2.2, 2.4, 4.1, and close ties in scenes 1.3, 2.3, 4.3-
5, and 4.7-8. Scenes 1.1 and 1.4, in particular, attribute to
Marlowe by a wider margin of 4cn, increasing the likelihood
of his contribution, while the other two scenes in Act 1
show less clear indication of authorship. In comparison to the
breakdown offered by Craig, our results support the claims that
Shakespeare wrote all of Act 3 and Marlowe possibly wrote
scenes involving Jack Cade’s rebellion (4.3-9). Act 2, on the
other hand, is attributed to Shakespeare in the act analysis but
most of the individual scenes are attributed to Marlowe. The
WAN of scene 2.1, in particular, has a large relative entropy
to Marlowe’s profile and indicates a strong likelihood it was
written by Shakespeare.

The intraplay analysis of 3 Henry VI in Fig. 18 attributes
Act 1 to Marlowe and the rest to Shakespeare. Although Craig
has suggested that the part of the text most likely written by
other authors is Act 4, the act analysis alone here suggests
otherwise. However, the attribution of individual scenes shows
a different pattern. Here, Marlowe is assigned four of the eight
scenes in Act 4, while Shakespeare is attributed scene 4.3 by
a very wide margin of 30cn–caused by the presence of a rare
transition–which likely skewed the entire act in Shakespeare’s
favor. In addition to scenes 2, 5, 7, and 8 in Act 4, Marlowe
is selected as the more likely candidate in scenes 1.1, 2.3-4,
3.3, 5.2, and 5.7. Shakespeare, meanwhile, is assigned scenes
1.2-4, 2.1-2, ,2.5-6, 3.1-2, 4.3-4, 4.6, 5.1, and 5.3-6. Scene 4.1
is a close tie between authors.

We also perform in Fig. 19 the intraplay analysis on the play
Arden of Faversham, attributed to Shakespeare in Fig. 7. Every
act is attributed here to Shakespeare. Although not shown in
the figure, the second preferred candidate in all acts except
Act 5 is Jonson, who is not typically considered a potential
author due to the year it was written. The other commonly
considered candidates for authorship are Thomas Kyd and



15

Marlowe [14], [54]. The former is not profiled due to a lack of
a sufficient number of texts to build a profile and the latter is
not well ranked in Acts 1-4 but is close to the second preferred
candidate in Act 5. For this reason, we attribute the scenes
between Shakespeare and Marlowe rather than Shakespeare
and Jonson. The scene-by-scene analysis shows Shakespeare
as the most likely candidate for almost the entire play, with
many scenes attributed to Shakespeare by a margin of at least
4cn. The exception to this is scene 5.5, which is assigned to
Marlowe, and scene 5.2, a tie between candidates. Our results
support existing claims by MacDonald P. Jackson [12] that
Shakespeare at the very least wrote the middle of the play
(Act 3), however we also find him to be a likely candidate in
at least Acts 1, 2, and 4 as well.

An analysis is performed for Edward III, attributed to
Shakespeare in Fig. 7. As before, the two most commonly
cited candidates for co-authorship with Shakespeare are Kyd
and Marlowe [14], [55]. The act attribution of Edward III in
Fig. 20 shows Act 1 assigned to Marlowe. Acts 2, 4, and 5
are attributed to Shakespeare, as well as Act 3 by a small
margin of less than 0.5cn. A look into the scene by scene
attribution, however, shows that in addition to 1.1, Marlowe
is also assigned scene 3.1 by a clear margin of 2cn. Marlowe
is also assigned scenes 4.1 and 4.7-8, while the attribution of
scene 1.2 does not provide a clear candidate. While not shown
in Fig. 20, the relative entropy values in attribution of scene
4.3 is large between both profiles (+2cn and +6cn between
Shakespeare’s and Marlowe’s profile, respectively), suggesting
neither Shakespeare nor Marlowe, but possibly a third author
contributed the scene.

Timothy Irish Watt has suggested that Shakespeare wrote
scenes 1.2 and 2.1 while someone other than Shakespeare,
Marlowe, or Peele wrote scenes 3.1-4.3 [14]. Our results point
to Shakespeare as a likely candidate for scene 2.1, with his
profile being almost 4cn closer to the WAN of Edward III
than Marlowe’s profile. Additionally, along with scene 4.3,
we find scenes 3.2-3 and 4.1-2, 4.5 and 4.9 to be possibly
written by a third author due to comparatively large distance
between the scenes’ WANs and both profiles. Not displayed in
Fig. 20, the closest profile between Shakespeare and Peele for

1 2 3 4 5
0

1

2

3

4

5

6

0

1

2

3

1.
1

2.
1

2.
2

3.
1

3.
2

3.
3

3.
4

3.
5

3.
6

4.
1

4.
2

4.
3

4.
4

5.
1

5.
2

5.
3

5.
4

5.
5

5.
6

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Marlowe

Shakespeare

Fig. 19: Attribution of acts and scenes of Arden of Faversham.

1 2 3 4 5
0

1

2

3

4

5

6

7

0

1

2

3

1.
1

1.
2

2.
1

2.
2

3.
1

3.
2

3.
3

3.
4

3.
5

4.
1

4.
2

4.
3

4.
4

4.
5

4.
6

4.
7

4.
8

4.
9

5.
1

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Marlowe

Shakespeare

Fig. 20: Attribution of acts and scenes of Edward III.

1 2 3 4 5
0

1

2

3

4

5

6

7

8

1

2

3

4

1.
1

2.
1

2.
2

2.
3

2.
4

3.
1

3.
2

4.
1

4.
2

4.
3

4.
4

5.
1

5.
2

5.
3

R
e
la

tiv
e
 s

im
ila

ri
ty

 t
o
 p

ro
fil

e
 (

ce
n
tin

a
ts

)

Peele

Shakespeare

Fig. 21: Attribution of acts and scenes of Titus Andronicus.
Note that here the comparative relative entropies for Act 1
and its sole scene, 1.1, differ. The plot of scene 1.1 reports the
difference in relative entropy between Peele and Shakespeare
while the plot of Act 1 reports the difference in relative entropy
between Peele and the second ranked author, Marlowe.

each of these scenes has a relative entropy between +0.1cn
and +1.7cn, whereas all other scenes range from −0.3cn and
−3.5cn from the closest profile.

E. Shakespeare and Peele

Shakespeare’s play, Titus Andronicus, is commonly cited
to include additions by Peele [13], and is attributed act by
act and scene by scene in Figure 21. Act 1 is assigned to
Peele while the rest of the play is attributed to Shakespeare.
In the scene attributions scenes 2.1 and 4.4 are attributed to

TABLE XIII: Relative entropies between scene 3.2 of Titus
Andronicus and author profiles.

Shakespeare Fletcher Jonson Marlowe
0.47 5.69 2.76 0.27
Middleton Chapman Peele Greene
3.72 2.73 4.80 1.12



16

0

1

2

3

4

5

6

7

8

9

10

R
e
la

tiv
e
 e

n
tr

o
p
y 

to
 p

ro
fil

e
(c

e
n
tin

a
ts

)

 

 

AW
W

AY
L

BB CA
P

M
IH

PO
E

M
IC

PH
O

M
DO SG

G
AN

T
HA

M
TM

T
BR

O
CA

T
SE

J
BE

W
RE

V
TD

B
RB

D
1H

4
2H

4 H5 1H
6 R2 R3 H8 E2

M
AS E

3

Comedy
Tragedy
History

Fig. 22: Attribution of plays between genre profiles. The plays to left of the first red line include comedy plays. The plays to
right of the first red line include tragedy plays. The plays to right of the second red line include history plays.

TABLE XIV: Plays used to build profiles for genre profiles.

Comedy
A Shoemaker a Gentleman (William
Rowley)

Fair Maid of the West (Thomas Hey-
wood)

City Madam (Phillip Massinger) Humor Out of Breath (John Day)
Heir (Thomas May) Orlando Furioso (Robert Green)
Tragedy
Atheist’s Tragedy (Cyril Tourneur) Rape of Lucrece (Thomas Heywood)
Cleopatra (Samuel Daniel) Fleire (Edward Sharpham)
Broken Heart (John Ford) Spanish Tragedy (Thomas Kyd)
History
Duchess of Suffolk (Thomas Drue) Edward IV (Thomas Heywood)
Sir John Oldcastle (Robert Wilson) Thomas Lord Cromwell (S.W.)
Perkin Warbeck (John Ford) Fuimos Troes (Jasper Fisher)

Shakespeare by a small margin of less than 1cn, evidencing
possible contributions of Peele. Typical attributions of this
play, such as the one performed by Brian Vickers [13], assign
to Peele Act 1 as well as scenes 2.1 and 4.1.

Another scene of interest in Titus Andronicus in the context
of attribution studies is scene 3.2, also known as the “Fly”
scene. This particular scene is present in the 1623 Folio but not
earlier additions, suggesting it was a later addition to the play
and possibly added by another author. The relative entropies
for this scene are compared in Table XIII. The two top
candidates here are Shakespeare and Marlowe. However, the
scene only appeared in editions published long after Marlowe’s
death so our top candidate for this scene remains Shakespeare.

IX. GENRE ANALYSIS

In addition to using WANs to distinguish author styles, we
may also use them to distinguish between plays more generally
at the level of genre. There has been debate among literary

scholars as to whether the classification of fiction into genres is
something determined solely by the plot or whether it can also
be determined by the writing style itself. We demonstrate that,
to some extent, it is possible to sort a play into its appropriate
genre by considering only its writing style as encoded by
function WANs.

We build three profiles for each of the three primary genres–
comedy, tragedy, and history–using plays that were not written
by the six main playwrights studied in this paper. The complete
list of texts used in the genre profiles is displayed in Table
XIV. The profiles use at most one play from any particular
author to avoid biasing the results based on author similarity
rather than genre similarity.

In Fig. 22, the results are shown from the attribution of ten
comedy, tragedy, and history plays between the genre profiles.
A total of seven of the ten comedy plays–displayed to the left
of the first red line–correctly attribute to the comedy profile.
Note also that all three misattributions are attributed to the
tragedy profile. The attribution of ten tragedy plays, displayed
to the right of the first red line, results in only three plays
being assigned to the tragedy profile, with Hamlet a close
three way tie between all profiles. From the remaining six
plays, five are assigned to the comedy profile. However, the
attribution of history plays results in 90% accuracy; shown
to the right of the second red line. In our results, we find
that distinguishing between history and the other genres is
easier than distinguishing between comedy and tragedy. This
is interesting because it is common to consider history and
tragedy more thematically similar than comedy and tragedy.
Our results, by contrast, suggest that the writing styles of



17

comedy and tragedy are more closely linked than the writing
styles of either comedy and history or tragedy and history.

X. CONCLUSION

Function word adjacency networks (WANs) were used to
analyze the authorship of texts written by popular playwrights
during the Early Modern English period. Word adjacency
networks were built for a large set of texts in the corpus
of the analyzed authors and were compared via the relative
entropy measure. The networks of every text known to be
written by a particular author were aggregated to form a
profile network. The profile networks were then compared
to one another to determine the general similarity between
author styles. Each text in an author’s corpus was compared
to every profile and attributed to the author whose profile
network produced the smallest relative entropy. An attribution
accuracy of 92.2% was achieved when attributing amongst
all authors and an accuracy of 96.1% was achieved when
attributing amongst authors that are more dissimilar than 5cn.
With this classification power, a selection of anonymous plays
were attributed amongst the author profiles. The classification
power was then further evaluated with respect to plays written
by multiple authors, both through the attribution of an entire
play as well as its individual act and scene components. The
act and scene components were individually analyzed in a set
of plays with highly disputed co-authorship, in which we both
corroborate existing breakdowns and provide evidence of new
assignments. The impact of genre on attribution accuracy was
also briefly examined to gain insight into the similarity of
writing styles with respect to a play’s genre. We overall find
function word adjacency networks to be simple yet effective
tools in distinguishing between playwrights from the Early
Modern era by considering relational structures between func-
tion words not previously considered in authorship attribution
studies from this time period.

REFERENCES

[1] Olivier De Vel, Alison Anderson, Malcolm Corney, and George Mohay,
“Mining e-mail content for author identification forensics,” ACM Sigmod
Record, vol. 30, no. 4, pp. 55–64, 2001.

[2] Efstathios Stamatatos, “A survey of modern authorship attribution
methods,” Journal of the American Society for information Science
and Technology, vol. 60, no. 3, pp. 538–556, 2009.

[3] Norman Meuschke and Bela Gipp, “State-of-the-art in detecting aca-
demic plagiarism,” International Journal for Educational Integrity, vol.
9, no. 1, 2013.

[4] Frederick Mosteller and David Wallace, “Inference and disputed
authorship: The federalist,” 1964.

[5] David I Holmes and Richard S Forsyth, “The federalist revisited: New
directions in authorship attribution,” Literary and Linguistic Computing,
vol. 10, no. 2, pp. 111–127, 1995.

[6] David I Holmes, “A stylometric analysis of mormon scripture and related
texts,” Journal of the Royal Statistical Society. Series A (Statistics in
Society), pp. 91–120, 1992.

[7] Frederick Gard Fleay, Shakespeare Manual, Ams PressInc, 1878.
[8] Philip Wolcott Timberlake, The Feminine Ending in English Blank Verse,

George Banta Publishing Company, 1931.
[9] Ants Oras, Pause patterns in Elizabethan and Jacobean drama: an

experiment in prosody, University of Florida Press, 1960.
[10] Marina Tarlinskaja, James Bailey, and George T Wright, Shakespeare’s

verse: Iambic pentameter and the poet’s idiosyncrasies, Lang, 1987.
[11] MacDonald Pairman Jackson, Defining Shakespeare: Pericles as Test

Case, Oxford University Press, 2003.

[12] MacDonald P Jackson, “Shakespeare and the quarrel scene in arden of
faversham,” Shakespeare Quarterly, vol. 57, no. 3, pp. 249–293, 2006.

[13] B. Vickers, Shakespeare, Co-Author: A Historical Study of the Five
Collaborative Plays, Oxford University Press, 2002.

[14] D Hugh Craig and Arthur F Kinney, Shakespeare, computers, and the
mystery of authorship, Cambridge University Press, 2009.

[15] G Udny Yule, “On sentence-length as a statistical characteristic of
style in prose: With application to two cases of disputed authorship,”
Biometrika, pp. 363–390, 1939.

[16] Shlomo Argamon and Shlomo Levitan, “Measuring the usefulness of
function words for authorship attribution,” in ACH/ALLC, 2005.

[17] Patrick Juola, “Authorship attribution,” Foundations and Trends in
information Retrieval, vol. 1, no. 3, pp. 233–334, 2006.

[18] David I Holmes, “Vocabulary richness and the prophetic voice,” Literary
and Linguistic Computing, vol. 6, no. 4, pp. 259–268, 1991.

[19] David L Hoover, “Another perspective on vocabulary richness,” Com-
puters and the Humanities, vol. 37, no. 2, pp. 151–178, 2003.

[20] Doug Cutting, Julian Kupiec, Jan Pedersen, and Penelope Sibun, “A
practical part-of-speech tagger,” in Proceedings of the third conference
on Applied natural language processing. Association for Computational
Linguistics, 1992, pp. 133–140.

[21] Santiago Segarra, Mark Eisen, and Alejandro Ribeiro, “Authorship
attribution through function word adjacency networks,” CoRR, vol.
abs/1406.4469, 2014.

[22] Dmitri V Khmelev and Fiona J Tweedie, “Using markov chains for
identification of writer,” Literary and linguistic computing, vol. 16, no.
3, pp. 299–307, 2001.

[23] Conrad Sanderson and Simon Guenter, “Short text authorship attribution
via sequence kernels, markov chains and author unmasking: An investi-
gation,” in Proceedings of the 2006 Conference on Empirical Methods
in Natural Language Processing. Association for Computational Lin-
guistics, 2006, pp. 482–491.

[24] Santiago Segarra, Mark Eisen, and Alejandro Ribeiro, “Authorship
attribution using function words adjacency networks,” in Acoustics,
Speech and Signal Processing (ICASSP), 2013 IEEE International
Conference on. IEEE, 2013, pp. 5563–5567.

[25] George Kesidis and J Walrand, “Relative entropy between markov
transition rate matrices,” IEEE transactions on information theory, vol.
39, no. 3, pp. 1056–1057, 1993.

[26] Ed. A. B. Farmer and Z. Lesser, “Deep: Database of Early English
Playbooks,” http:// deep.sas.upenn.edu/ , 2007.

[27] Chadwyck-Healey. ProQuest Information and Learning, “Literature
Online,” http:// lion.chadwyck.com.

[28] Gary Taylor and John Lavagnino, Thomas Middleton: The Collected
Works, Oxford University Press, 2007.

[29] Richard H Barker, “The authorship of the second maiden’s tragedy and
the revenger’s tragedy,” The Shakespeare Association Bulletin, vol. 20,
pp. 52–62, 1945.

[30] Cyril Tourneur and Lawrence J Ross, The revenger’s tragedy, ed, 1967.
[31] Anne Begor Lancashire and Thomas Middleton, The second maiden’s

tragedy, Manchester University Press, 1978.
[32] MWA Smith, “The authorship of the’revengers tragedy’+ tourneur, cyril

or middleton, thomas,” 1991.
[33] Philip Gaskell, A new introduction to bibliography, Clarendon Press

Oxford, 1972.
[34] Edwin J Howard, “The printer and elizabethan punctuation,” Studies in

Philology, pp. 220–229, 1930.
[35] Archie Webster, “Was marlowe the man?,” National Review, pp. 81–6,

1923.
[36] T. P. Logan and D. S. Smith, The New Intellectuals, University of

Nebraska Press, 1977.
[37] James Loxley, The complete critical guide to Ben Jonson, Psychology

Press, 2002.
[38] A. Gurr, The Shakespeare Company, 1594-1642, Cambridge University

Press, 2004.
[39] A. Barton, Ben Jonson: Dramatist, Cambridge University Press, 1984.
[40] David J Lake, The canon of Thomas Middleton’s plays: internal evidence

for the major problems of authorship, Cambridge University Press, 1975.
[41] H Dugdale Sykes, “John ford, the author of” the spanish gipsy”,” The

Modern Language Review, vol. 19, no. 1, pp. 11–24, 1924.
[42] MacDonald Pairman Jackson, Studies in Attribution: Middleton and

Shakespeare, Institut für Anglistik und Amerikanistik, Universität
Salzburg Salzburg, 1979.

[43] Stanley Wells, Shakespeare and Co.: Christopher Marlowe, Thomas
Dekker, Ben Jonson, Thomas Middleton, John Fletcher and the Other
Players in His Story, Random House LLC, 2009.

http://deep.sas.upenn.edu/
http://lion.chadwyck.com


18

[44] Edmund Kerchever Chambers, The Elizabethan Stage, vol. 3, Clarendon
Press Oxford, 1923.

[45] Patrick Cheney, The Cambridge Companion to Christopher Marlowe,
Cambridge University Press, 2004.

[46] Cyrus Hoy, “The shares of fletcher and his collaborators in the beaumont
and fletcher canon (v),” Studies in Bibliography, pp. 77–108, 1960.

[47] Terence P Logan and Denzell S Smith, The predecessors of Shakespeare,
vol. 1, University of Nebraska Press, 1973.

[48] CF Brooke, “Tucker, the shakespeare apocrypha: Being a collection of
fourteen plays which have been ascribed to shakespeare,” 1908.

[49] Stephen Roy Miller, The Taming of a Shrew: the 1594 quarto,
Cambridge University Press, 1998.

[50] Terence P Logan and Denzell Stewart Smith, The Later Jacobean and

Caroline Dramatists, vol. 4, University of Nebraska Press, 1978.
[51] Ernest Henry Clark Oliphant, Plays of Beaumont and Fletcher, Yale

University Press, 1927.
[52] William Shakespeare, Gwynne Blakemore Evans, and John

Joseph Michael Tobin, The Riverside Shakespeare, vol. 1, Houghton
Mifflin Boston, 1974.

[53] Gary Taylor and John Jowett, Shakespeare Reshaped, 1606-1623,
Cambridge Univ Press, 1993.

[54] W. W. Greg, “Shakespeare and arden of feversham,” The Review of
English Studies, vol. 21, no. 82, pp. 134–136, 1945.

[55] T. Merriam, “Marlowe’s hand in edward iii,” Literary and linguistic
computing, vol. 8, no. 2, pp. 59–72, 1993.


	Introduction
	Word Adjacency Networks
	Author Profiles
	Similarity of profiles
	Attribution of Plays
	Ben Jonson
	Thomas Middleton
	George Chapman
	Christopher Marlowe
	William Shakespeare
	Summary of Results

	Anonymous plays
	Collaborations
	John Fletcher and collaborators

	Collaborations – Intraplay analysis
	Jonson and Chapman
	Shakespeare and Fletcher
	Shakespeare and Middleton
	Shakespeare and Marlowe
	Shakespeare and Peele

	Genre analysis
	Conclusion
	References