THE ATTRIBUTION 0F AUTHORSHIP:
A COMPUTERIZED METHOD EVALUATED AND COMPARED
WITH OTHER METHODS PAST AND FUTURE

THESIS FOR THE DEGREE OF PH. D.

MICHIGAN STATE UNIVERSITY
GEORGE W. ZIMMER

1968

mass J ‘71,’ W; F‘WT ; I e L f! :11?” "

This is to certify that the

thesis entitled

THE ATTRIBUTION OF AUTHORSHIP:
A COI~IPUTERIZED METHOD EVALUATED AND COE-‘ZPARED
WITH OTHER METHODS PAST AND FUTURE

presented by

George W. Zimmer

has been accepted towards fulfillment
of the requirements for

Ph. D. degree“, English

:waJMM

Major professor
Date W /é/) / ffi

0-169

C,’
K 2730
3/0 fl”? -

ABSTRACT

THE ATTRIBUTION OF AUTHORSHIP:
A COHPUTERIZED METHOD EVALUATED AED COMPARE
WITH OTHER EETJODS PAST AfiD FUTURE

George W. Zimmer

The proving of authorship by statistical means has a long
but inglorious history in the field of English scholarship. What
usually has happened is that an undisinterested scholar, out to
"prove" that (for example) there was a "Pearl-poet" to whom can be
attributed three or four additional Middle English poems, lists
elements that the anonymous Pearl has in common with the other
poems, and concludes that on the basis of his "statistics" the
poems must have been by the same author.

In his 1941 dissertation (University of Hinncsota) John W.
Clark takes great pains to disprove the attractive "Pearl-poet"
theory by examining the large quantities of data used by scholars
from 1876 on, and finds their data invariably faulty or misapplied.
Even granting them sufficient accuracy, Clark maintains their data
could prove mere influence of one poet on another as well as it
proves con on authorship of the several poems.

But basically the fault in the early attempts at proving
authorship lay in the inaccuracy of the data, which was neces-

sarily gathered "by hand." "Precision without accuracy” is the

J

downfall of pre c01~utcr Statistical analyses OI literature. he
filould e:— wect a co13uter- aid ed 3,030 ct to avoid this pitfall. One

fairly typical study, 3:2.312 gagig§?, by Alvar Ellegard, does

in nvolve a connuter, but the commuter is put to work on deta de-

insceuratulv by ~I‘llegr—wrd's

rived laboriously, intuition lly, an;

own hand. Therefore, his nethod, which purports to prove that the

Letters f Jrnits were written by Sir Philip Francis, contains the

 

traditional flaw of preconfiuter stulies o; t1e sane kind.

V

m, fipl— r‘r 1‘ ’3 '.- ‘9‘ x v\ '3" ', --. . . L ~v (a ' pa 7.- ‘» w l ' r~ -'.
lee tasi set in the are e p.oject has to Joel ey ouyective

‘ a A ~' I o in n '1 - v.4 vr '3 w --~-. ~~ 4- ‘-~ .. - wire
of vocaeulary. inc zeta ye.e CeiirCn only iron the inetn noene o
_f v . . m- ‘l ' ~ - 1--..-: 4- -\ ..v' '. v rs "- r. V a ' . r‘
five nineteenth-century :iite s, Lien no Oel of attfiJhtln, an

.l.

anonymou: or do btful noon t any of the five. The purpose was Do
test only the teat itLJlf. ”Prvcision without accuracy" was
avoided by lavinr the connuter do the first selecting of vocabu-
lary items to be subjected to analysis.

J.

Fed into the ET 3 Control Date computer were more than five-

hundrnd pounds of Ill cords, each one bearing a sil: le line of
ve13e carefully verified, and Mltfl the snellin: of certain poten-
tially a biqueus words ceiventior alize in order to dispel the
ambiguity: KAthhe eLVilinry verb; “A-_—t1e month. A :lossary
prorram written by Jan-s D. Clark of Hichiran tate University
gave me alphabetized lists for every text fed into the conduter.
A.search of the largest texts having disclosed not one content:
word significantly present in one author and not as significantly

present in at least one other, I determined to use the connon

words in my testing.

George W. Zimmer

Accordingly, I took the forty-five most common words and
word-groups (AM+ARB+IS+WAS+WERE=one word; to which I apply the
term allomorphs in my thesis) with the exception of the personal
pronouns, and made of them a forty-five point profile for each of
my 230 texts. If a writer subconsciously chooses one function
word in preference to another, the one he chooses will form a peak
on the profile-chart of his texts, while the other, less-prefer-
red, word will show up as a valley. When charts of texts even of
different word-populations are compared, two by the same author
should have more points in common than two charts of texts by dif-
ferent authors.

The comparing was done by enlarging the points on the 230
profile-charts to quarter-inch holes, and then laying one chart on
top of another and seeing where the holes coincided. Thus every
text was compared with every other text, and five indices of cor-
respondence tabulated for each comparison. A chart that finally
analyzed the half-million bits of information thus tabulated
showed that the method has possibilities: the best of four cri-
teria into which I combined my five indices of correspondence was
able to call the correct author sixteen times more frequently than
would chance guessing.

The test might be highly reliable in establishing which
author of only two wrote a doubtful text. ts value, however, for
attributing a truly anonymous selection to one of a larger number
of possible authors I will insist upon denying. To do this, a
statistical test must have the same accuracy as a chemical test;

it must work every time under laboratory conditions if it is to be

George W. Zimmer
assumed workable when conditions are less controlled.
Although the project was a failure, and the test is untrust-
worthy as it now stands, there is some possibility that by shar-
pening my procedures (perhaps by using the comauter at all stages)

and by strengtheninf the word-list by dronpinr deed iteus and add-

P.

as some overlooked before, a reliable test for prov “? authorship

could yet be discovered.

THE ATTRIBUTION OF AUTHORSHIP:
A COMPUTERIZED METHOD EVALUATED AND COHPARED

WITH OTHER METHODS PAST AND FUTURE

BY

George W. Zimmer-

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of English

1968

(37-2 I v or I

:2- ./ ./ 7
V. ,

Copyright by
GEOL CE W. Zlg-HLER

196::

THE ATTRISUTEOI or AETIORSII‘:
A constrsalzzn hBTnGD EYALUATED ALB cenrnazb

WITH crass nrrlous PAST Ale FLTan

IETRCDUCTION
One of the host teapting fields of scholarship is that

ninvolving the determination of authorship. If the historical-
biogranhical approach to literature has ever had validity (and
it shall here be considered a self-evident truth that it did and
does have validity) it is necessary to know with certainty which
author wrote vhich works. Our very important stereotypes of the
various artists are largely dependent upon their canons of works:
subtract a Timon of Athens from his canon and the percept "Shake-
speare" becomes something other than it was with giggn in he
canon. The authorship problem does, then, have its willing sol-

vers, whose methods, however, vary widely.

The simplest method of settling an item within or outside a
canon is by edict. By declaration of a scholar or a school of
scholars an item is placed within or outside a canon. The logic
for such edicts is partially as follows: a work can be placed in
a canon only by the best authority; I am the best authority;
therefore this work is declared to he part of the canon. Prai-
matically, at least, the edict method is one of the best, espec-
ially when the placement of the work is afrecd uuon by those
holars most interested in the subject. But the edict method

gives rise to the tatihr of exception by rival scholars or riva

1

schools. lien tee rule states that the burden of eiseroof re ts
upon the distenter who cannot therefore use the edict nethod. It
° 'W' -A ' '_ v c _'_- 0,: ” ~, 1 o 9!

is at tile UOlnL teat statiStics are erouent to play on tee cues-

astiflhf

“J
O
I .1
"D
1.:
:5

"Xethods usiné statistics” is a caterory cl
several different ways: "successful" and hon-successful; simple
and intricate; pseudo anm true; early, more recent and meet re-
cent, or pre-comwuter, early computer, recent computer. Generally,
the early atteiuts USs simple processes involvinf pseudo-statis-
tics and declare themselves successful. Kore recent attempts are
less certain about success, although, paradoxicalli, they use
scientifically sounder procedure.

To be dealt with in this thesis are saeeies of early aid
recent, simple and intricate methods. Each species will be given
its just share of criticism including the most recent which,
accordin“ to my oun classification, is also the most intricate
and non-successful.

Sections of the thesis vill show the deficiencies of several
types of statistical methods for provire au horslip, tith parti-

cular attention to (A) the old, pre-eonuuter methods tzat relied

1

and simplistic couparisons and that were so

'3.)

on pseudo-statistics
very positive in their results, and to (?) the computer-based
method that involved he in my task. In between there will be
short looks at hybrid methods: computer-aided and intricately

statistical, but "positive” in their conclusions.

This element of "positivity" cannot be treated too fully.

3

ml 1" .L V " 1 fi fi ,2 ~ , .3 A _!_‘u _ ‘f ‘ n

lee efiorts oi the scaolsrs in the iele Oi attrior tion 01
g a - c ‘ , - - . K - ‘ - ‘ ‘

authorsnie in the ore—coneuter axe nave alreaev oeen charac—

terized as simp-istic and pseudo—statistical. In general the

earlier the study the less scientific it i s; but in about
Li rect ratio to the lack of scientific riser, result of such

{[1

studies tend to be declared "positive.” -here are very human

most inst,nces the "positive” results supoort a preconceived
notion. Later, as scientific rifor stiffened somewhat, an
occasional scLolar would cla.i ‘.n to have had his oninion changed
by his res arcL, yet results would be set forth as positive,
even thoush negative evidence was present in ne wly as rreat
proportion as positive. There are the furtller hulan inclina-
tions to put an end to a piece of research, and to fill an
unreasonable demand for a "happy ending" however false that
ending might be. Such endings, which demand "positive" re-
sults, are easy to sell to a readership unwilling to study
the data or the techniques upon which the results are based.
My dis se mt tion presents in three chapters three sta es
or episodes in the quest for certainty by scholars in the
field of English. The first chapter starts in 19u1 and re-
fers back to the beginninms of a particular authorship prob-
lem, that of the so-ce lled Pe«? rl-poet. It ca: be t Qien as
typical of authorship pr031e .s, in that almost any argument
ad: uced for the com only-held owinion has been respected,
viiile one most carefu.lly pr-jarce attack on the attractive

theory has been all but totally ignored. The first chapter

quotes heavily from that attack (John W. Clark's doctoral
dissertation, University of Minnesota, 1941) in order to re:
present its tone of futility and dogged determination.

Also, in presenting Clark's dissertation at such great
length, I am repaying a debt I had not been aware of owing
until I recently returned to its voluminous pages and found
there many of the ideas and attitudes I had thought I had
formed independently.

My second chapter advances to the 1960's, with an occa-
sional flashback to Clark and the history of the authorship
problem. In it I examine two types of computer-aided pro-
jects .aving as their ends the determining of authorship, and
find both of them lacking integrity. It is suggested in this
chapter that the computer should be allowed to do all of the
work; especially in the first stages of a project, human
error should not be permitted to intrude.

The third chapter, a delineation of my own project, which
is a study of vocabulary for proofs of authorship, should fol-
low the rule of maximum computer use. But it does not follow
my rule. The discrepancy is owing to the fact that this disser-
tation was written backwards. Hy decisions on how to proceed
had been influenced by Clark and many other books and articles

which I deliberately put aside while assembling my own project.

 

1 Eminent among the writings are "Eras in English Poetry,"
Josephine Hiles, PXLA LIX 553-75, which from examination of
syntax posits three eras for each century of verse; Statistical
Study 9; Literary Vocabulary. George Udny Yule (Macmillan,
19hh), where formulae are given for determining the authorship

The project was finished and the results were tabulated before
the relationships among the various methods fully occurred to
me. I had intended at the outset to demonstrate an impossible
thesis: namely, that it is not possible to prove authorship by
statistical examination of vocabulary. The project, however,
proved the only thing that it was capable of proving: that the
particular method used here is not capable of proving author-
ship. The positive knowledge of this negative fact coupled
with comparison of the method here oriwinated with previous
methods whose authors evince far less positive inoxl; :e of the
shortcominfs of their work has now led me to a new hope that
the authorship problem may yet after all eventually be solved
by statistical means.
The results of my research i- §e were negative and could

ave been pro-enter in a few pages. Althoufih I an

._J

properly I
as reluctant as anyone else to write at length about little,
I have a purpose for so aoins in the sections of this thesis
outside Chapter III. The positive result I aim at is putting
an end to misguiuea work in the field, for which there may
have seen an evcus e a few ghe wis ago, but not new. Chapter

III, however, is no longer than it has to be, in contrast

 

of passagjes of 10,000 words or lore; Tvneiggig; fatheggtiegi
A Tevtooo" of I 10*ctic l Lir'rdstigs Gus ta're .:-.erd.an

- u. r-_---. H..-— u.

( -Outon S'Traveiase ITCU), which conce11t‘ates on vocabulary

items (types) and their occvr‘nces (tokens) a the ;::ea:1s of

(

determini1¢ authorsnin; and “enta-nne—Shakesneare ani the
Deadly Pa ra.llel, " Geor“e C. WT/lor, hilola,ic l ngrteggz
XiII (1Tb;>) 330-3, 1ich lists pseudo-ser_ously seventy-five
kinds of evidence tiat canon and influence scholbrs use, f

“fliCh "”VTflbV-i f3 "vocabulary” is number 31.

to the toe-explicit works I criticize. In heroin? with this
conciseness is my deterrination in Chawter III not to repeat,
or to excerpt from, my cnar s and tables (thereby forcing the
reader to consult the full tables in their proper context
each time reference is made to one) unless to show how such
xcerpting can be made to give the appearance of validity to
otherwise inconclusive results.

The fourth chapter is a retrospect of the three stages
of work in the field, pre-computer, early computer and re-
cent, with a look at some trends that started after this
thesis was begun. It will there be shown how the problem of

attribution of authorship will probably be solved, when and

if it is ever solved.

CHAPTER I

A CLASSIC CRITIQUE OF PRE-COMPUTER STATISTICS

In March of 1941 John Williams Clark submitted for his
doctoral degree from the University of Minnesota his disser-

tation, The Authorship 9; Sir Gawain and the Green Knight,

 

Pearl, Cleanness, Patience, and Ernenwald in the Light _f

 

the Vocabulary. This work attacks the labors of all the

 

scholars who up to that time had attributed the five poems
to fewer than five poets. The first scholars interested in
the problem had taken their cue from the manuscript of four
of the poems: since the first four have come down to us in
one document, the tendency would be to attribute them to a
single writer. This attribution had been disputed several
times before Clark, but the important editors of the texts
had persisted in the one-poet theory, in support of which
various kinds of evidence were presented. Professor Clark
went to great pains to destroy his predecessors' arguments,
sometimes by accepting their evidence but re-interpreting
it, and other times by adding further evidence to show that
they had used distorted data. His 531 pages of hard—hitting
argumentation are comprehensive and attach every important
argument adduced for the common authorship of any two of the

five poems. Clark is thorough, completing and correcting

0:)

other supposedly scholarly work. He gives the history of his

problem:

The earliest considerable attempt at deciding the ques-
tion of single or multiple authorship came in 1:76, in
Moritz Trautnann's doctoral dissertation, Ueber Ver-
fasser und Entstehungszeit Einiger Alliterierender
Gedichte des Altenglischen; and the opinion there ex-
pressed-~that the four poems were written by a single
poet--was given currency a few years later by Ten Brink,
in his Geschichte der Epglischen Literatur, the English
translation of which in the 'eighties extended Ten
Brink's reputation, and the respect in which his pro-
nouncements were held, beyond the then somewhat restric-
ted circle of English scholars who willingly read German.
Since that time, as Henner says (ed. Cl (Tor Cleannesé]
p. Xi, n.1), "practically all those who have made ‘—
special investigations of, or edited any of these poems
. . ., have accepted the Opinion" that a single author
wrote Gaw, Prl, Cl, and Pat. This opinion--perhaps we
may say this pious opinion--was erected almost into a
dogma by being spread by Professor (later Sir) Israel
Gollancz (who has done more than any other scholar to
create the reputation--if not, indeed, the identity--
of "the Pearl-poet") upon the sacred page of the
Cambridge History 9: English Literature, of which the
first volume, containing the account of the four poems,
appeared in 1907. Sir Israel had published his
adherence to the doctrine of single authorship as early
as 1391 (in his first edition of Prl), and to the end of
his life continued to proclaim it; his most absolute af-
firmation of faith being perhaps one that appears in the
preface of his edition of Pat (1913): "It is now gener-
ally accepted, in respect of the four poems, that all
the evidences of dialect, vocabulary, art, feeling, and
thought conclusively point to identity of authorship
..." Italics mine. [Clark's]

The opinion that Erh, also, was written by the
author of the four poems (or of one or more of them)
was first advanced som what later (by Carl Horstmann,
in the editio nrincens of Erk, in Altenglische legpnden
(Neue Folgei, 16b15, and has never been so widely adop-
ted or, in general, so confidently urged, as the opinion
that the four poems are from a single hand.
. . . In this attribution of the authorship of Erk to
"the Gawain-poet,” Horstmann (has) been followed by most
scholars, notably R. V. Chambers, in Essays and Studies
by Members 3: the English Association, 19. 126, n.2, but
even Chambers does not express himself with assurance,
and there is still a considerable amount of more or less

 

 

 

 

half-hearted dissent. (Pages 3-4)

Clark sees his task as giving heart to the dissent. He
disposes in turn of each of the p oponents of single author-
ship, and of each of the theories based on dialect (sixty:
three pages), proSody (twenty-one), interests, attitudes and
opinions (eight), syntax and style (thirty-two), and parallel
passages (forty-one pages). After 192 pages, he is ready for
Part II, the examination of the vocabulary. The chief con-
tention that Clark advances in Part I is that similarities or
even identity of dialect or the other criteria "prove" only
that the writer of one poem lived in the same area as the
writer of another, or was influenced by him. The evidence,
he claims, cannot conclusively show either separate or common
authorship, but, if the evidence is to be admitted, it has
more force in proving diverse authorship. Always, however,
there is the strong desire on the part of the earlier
scholars to demonstrate the more attractive theory: to set
up a "Pea rl-poet" whose canon of works, together with those
undoubtedly lost, would make him a worthy contender for
honors com: only reserved for Ch aucer.

Part II of Clark's dissertation is aimed at those
scholars who, following Ten Brink's Geschichte der Englischen

o 1 q o o ' , ‘ o .,
Literatur and Sir Israel Gollancz in the 1907 Cambriuse

 

History of English Literature, sought to give a statistical

 

1 Elistory of Pnhli
M O

h Literature: I, Eng. tr., H. a.
{ennedy (Iew Io;k, ).

S
-,
J

10

foundation to their preconceptions. His strictures are par-

ticularly directed toward J. P. Oakden's Alliterative Poetry

 

. V. . I . .2 i .
in middle anglisn and nenry L. Savage, editor of S . Erken-

U.“—

 

wald,3 although some lesser sinners attract a share of

attention.

The vocabulary of the five poems early engafed
the attention of scholars intent on discovering whether
-—or, as it almost seems as if we must sometimes say,
on proving that--the poems were by a single author.
Trautmann published on the subject three times. . . .
Trautmann's conclusion was, in brief, that the vocabu—
laries of the five poems are so much more like each
other than they are like the vocabularies of any other
ME alliterative poems, that we must suppose the five,
and only the five, to have been by a single author.
There is no doubt about the exceptional degree to which
the vocabularies of the poems resemble each other; but
subsequent investigations both more extensive and more
exhaustive than Trautmann pretended to have made, or,
indeed, could have made in the absence of editions with
more or less complete glossaries and of HE , have shown
that the vocabularies of the five poems are nowhere
nearly so similar or so peculiar as Trautmann thought.
‘his is well shown by Savage, in his edition of Erk,
pp. liv-lv: ". . . the value of Trautmann's findings
has been somewhat reduced by the appearance of the later
volumes of the HED and the progress of scholarship;
yet," Savage adds cheerfully, ”the test of vocabulary
indicates an unusually close connection between the five
poems, and has strong affirmative bearing on the possi-
bility of common authorship.”

In other words, we are right back where we started
--the vocabularies of the five poems are rather strik-
ingly similar, but not by any means so similar that
common authorship is the only possible (or even, I may
add, the most probable) explanation. That this fact is
perceived by the advocates of the theory of common
authorship needs no further proof than that they all pay
their respects to the vocabulary, and then look else-

 

2 University of Manchester Publications, CCV and CCXKKVI
(Manchester, 1950 and 1935).

3 New Haven, 1926.

11

where for cogent arguments in favor of their View.
(Pages 193-4)

And, as Clark demonstrated in Part I, the arguments from
elsewhere are not cogent, either. Why, then, does he devote
the bulk of his dissertation to an examination of the vocabu-
lary, especially when "attempts to prove the common author-
ship of the five poems, on the basis of their vocabularies

. . . have clearly failed"? (Page 195) At this point, Clark
could take credit for examining the vocabulary, the most ob-
jective of all the criteria and the one at the same time that
provides the most massive data; but instead he uses the device
as a sort of tail to pin on one of his less preferred prede-

C GSSOI‘S:

that failure has been proclaimed by no one more empha-
tically than by some of the principal advocates of the
theory of common authorship themselves. But an opening
has been left for studies of certain special aspects of
the vocabularies; and that opening has been seen by the
indefatigable Mr Oakden, who, in the second volume of
his Allit. Poetgy in Middle English, investigated three
of these special aspects: (17—“Chiefly alliterative"
words (by which Oahden means, not words that usually
alliterate, but words "found but rarely or even not at
all, outside the alliterative poems"); (2) "synonyms
for man, knight," and (3) "synonyms to express movement."
. . . consideration of his findings will repay our
efforts by the further suspicions it will arouse as to
the validity of the theory of the common authorship of
the five poems, and will serve as an appropriate intro-
duction to the main part of this dissertation. (Pages
195-6)

 

 

Clark soon shows that Oahden's fifty-three "chiefly allitera-
tive" words reduce to but twelve that are thinly distributed
among the five poems, only one appearing in all five poems;

that at least one synonym (douth) demonstrates "a fundamental

difference in Sprachgeffihl" (page 199) fron poem to poem;
and that Oakden's "definitely poetic" words for "so" have a
characteristically non-significant distribution. 0“ some
little significance for this present dissertation is Clark's
explanation following his table of "go"-words for the five

poems:

The 31 ssaries [which Clark had thought he could rely
on for his examination of Oahden's assertions are com-
plete except where "frequent" occurs under Gaw, and
possibly also under Pat generally--here, as with the
synonyms of gap, hnirht, I have pieced out Bateson with
Gollancz, and can hope for getting nothing more than an
approach to the truth. All the zeros under Pat, how-
ever, are probably right; neither Gollancz nor Bateson,
apparently, leaves words completely unnoticed in his
glossary except by accident. (Page 213)

In this instance, Clark does not permit himself the dudgeon
to which he ascends when he attacks Oahden directly; instead,
he merely accepts the incomplete research for what it is not
worth. Later, we see him re-doing far longer lists of French
and horse words, with the goal of achieving more nearly per-
fect accuracy. With an impossible foresight, he and all the
other glossary-makers would have waited to let computers do
the work with complete accuracy.

Yet, even accepting the 'mperfect lists for vhat they
are, there must also be an accurate, honest reading of them.
In one note, Clark accuses Oahden of carelessness (page 222),
and of misreading "the glossaries, or silently 'correcting‘
them." The same note (no. 17, page 223) tabulates the etymo-

logies given for four instances of the word note:

line Golla cz Savage Oakden
8 0E2 OE OF

101 CE OF OF
33 OF OF CE
152 CE OF OE

. . . all three scholars pave; airee. It makes no
immediate difference who is rifiht, or whether arybody
is rirht; the point is the at neither 09‘; en nor anyon
else really 11:-01.s (and. pro bably the author or authors
of the poe2 did not know) how often this allevedly
"chiefly al ite1.tive" word from CE pgtg 11iea1s in
the Five 0 18

WA
() PM

0 O O O

. . . As I have shown above, the elitors of the poems
sometimes disecree on the deriva ion of .o1ds . . .
and yet neither Gollancz nor Savage (nor Crhicn, for
that 1atter) expresses the slightest doubt that t1e
truth is attainable and the he has attained it. This
sort of t-1i113 s co:.1’.':on in t1e edit iro1s of the Five
Poe 3. (Pa es 22-3 ,b)

The lefan is "pr cisien \ithout accuracy,” and the further
point that Clark makes, about even the author not beilg a‘are
of etymologies, is one to store w,.ay for future reference.
However, he himself acts on the counter theory that the
aut‘10r(s )11ad an a.a:e n=es of words, since the bulk of
01a: ck' 8 research is precisely in his lists of rrencn and
Horse words. Furthermore, Clark makes no claims to perfect
accuracy in 2:_ lists. This disclainer is in accord with the
footnote above, but not with his abjuring of the "precision
without accuracy" 510 01.

It is ever thus. If the size of a writer's da‘a is in-

.. .-
L

13

'.J
O
U}
D
’ )
‘L
a
‘4
P]
(‘1'
)
(I
O
"5
L)
O
r x,
C)
O
C)
t..
to
(‘1‘
13‘
Q
{.5
O
J
y.
.4
'

|-lo

pressive, the stat
not fail to imaros s those wdo lave no inclination
them.

Clark had t1e inclination, and he did teat Cahien's

figures on tie incidence of Old Norse words in

11.}.

Let us consider Oakden's statement . . . that Gaw con-
tains 256 03 words. What is an ON word? For that
matter, what is "a" word? Are £31122: a., and a311,
adv., two words or one, for our pu2poses? Much might
be argued in favor of either answer; and so long as we
give the same answer consistently, it makes no differ-
ence which we choose. But which has Oahden chosen? He
doesn't say, and I don't know how to find out. Again,
whichever he has adopted, does he stick to it? Again,
I don't know. What I do know is that Oahden's habits
of work, so far as I have observed them--and I have ob-
served them pretty extensively--do not inspire me with
confidence that he has given very careful consideration
to the problem. (Page 553)

. . . I do not claim . . . that my judfiment has been
infallible, or even that it has been better than the
judgment of Kr Oakden and of the editors of the several
poems. But I believe that I have shown, beyond reason-
able doubt, that Oakden (like most of the editors) has
overestimated the number of words, in the five poems,
of which we can say with confidence that they are prob-
ably ON. (Page 555)

The aspersion cast (page 353), Kr. Clark concludes:

Incidentally, the close similarity of Oakden's figures
and mine for each of the poems severally (except Gan)

leads me to believe that Oakden must, after all, have

been nearly as cautious (always with the exception of

Gaw) as I about calling a word OR; (pate 357)

thus he puts himself in the same pocket as his chief target:

such statements as Oakden's that Gaw contains 238 ON
words-~or mine that it contains 202--are perfect exam-
ples of precision without accuracy, the fact being, of
course, that no one knovs how many "distinct words" the
language contains, or how many of them are OK. (Page

558)

A weird sense of futility pervades the dissertation. The
Horse words disposed of, in a mere 120 pages, Clark turns to
the French words. His predecessor, Hartley Bateson,# had

worked out the proportions of French words in the two poems

 

h Ed., Patience (2nd ed., London, 1918).

d
“ qr}

Eggglyani Patigggg as 32.57 to 1?.92.

I

v

Beteson, like Oe”l>.
without
of 1H3 rel; or
Drooasly xiscallee OF are
miscallrfi QT, very few (f
arwe SG\”CTOj (temobrflal;7(;0c>r

wor;%131 t'1e Ei‘Je Pocnus 9
OF 019 e1t in P11 to
as rie3o colv close

.. 7
as env tnet coull

either ;8 beinfi nrecise
" " -~ ‘L ‘. '-\ r'v #1" ~ ..'
nu'lfiCbln. to Jive notice
.1— . ~v u
;b Since t e Lords
-~\.— A ..‘I ~ -: J-'1‘ \a '\A ‘ E‘s ‘ v " '
o. v.~11;1 g_t_ LLO:&;;l;OQ&01f
r t1én 3Q), eni since there
., fifl ‘- r
k.

C) -

J a»
’3
c:-
(5 O H; 1-“ t:
"J
o
(.1-
J.
o
"1

. r ‘ c’v"
t ; J .30 't8_ulfl

-'-~ ‘A - ,., F ~. ‘ n'~ -
o JC‘Lufli” onv—-es cl 1x3,jproouulf,
a _

Then, since he is unsure of the meaning of Sateson's ratio
of 5h.b7 to 19.92, 019 r; does it over again; ”and for food
measure, I have extenfied the inves 1~m~t on to tne other

poems." (Pane b6?) 2e laments:

iJ.
(D
l I
('1-
C)
rt-
0
:5
h.
H
1
C3
0
C)
k
d
}-I
C)
:3
O
:J
1.;

513»

e -or lamentations
a 1
F

1.1 O (‘2,
5"
('1‘ D
F"
L)
O
9
h)!
L)

" 1:;
O
’J
C.)
(.1-
i.)
1.5
v 3
Cf-
m
QJ
'._.l
O
C)

,C
b ; ness f uh ‘ "as, after
al_, a.self-iiiosed task; IU' I cannot for3~er to say
that I wish (a) that editors of IE texts hel sereed on
such cuestioas as the nroner alohebetical nosition of
2 (in both values), X (vowel), and u an} 3, end (6)
that HE noets 31i scribes had, as rtenus Ward said,
:nov .n how to en ll. The mere mech n'ca complications

1
, nto the connilation
of such a list 3 that below, are yond the nowers of
anyone wh s not evneriencei the: to in: ine. But the
1'03 wes at len jitw. 1‘“ ”1.31101" and I have the 111°C) rt of
gnoainj that I have discovered some facts 3 it, and
subsequent students-~1M1o e “at1e11t1c11 stunies were

not cut short, like mine, with plane sconetry--nev dis-
cover more. (Page L69)

introduced cse txo f cto:cs

:(T)

.
.4

'1."

But what has he discoverefi that is of any lest_nj worth?
His "statistics" are of no more value than those of the

th e

f...’

post as

1w

l V ' 1, L” 1 . ,1 .-.
scaolnrs he sttec s: tge nets are r tier ed u

many onnortunitics for erro as were nresent when the earlier

scholars worked without the aid of the ng_rd _E) lish.2ig-

nest unsta ti al

w

tionarv, and the data are handled in a

manner. Clark \HO‘rs ut ratios of French to horse weris for

16

each of the five poems. Pearl has 4.29 OF words to 1 Oh

 

word; Cleanness 3.56; Erkenwald 3.60; Gawain 3.56; and Pati-

 

ence 2.67 OF words to 1 ON word.

It is my duty to remind the reader once more that these
figures, like all those in this chapter, are only ap-
proximate, so that such a difference as that between Erk
and Gaw, or even that between Erk and Cl, is probably
not very significant; but such differences as . . . that
between Prl and Pat can certainly not be attributed to
the roughness of the basic figures. (Pages 478-9)

Perhaps not, but reason for tabulating the ratios in the first
place can be questioned, and the differences can hardly be
said to indicate anything of importance. Clark sees the

difference between Patience and Cleann 33 as showing that the

 

former was written first, "before the author had become so
well acquainted with French literature," (page 479) although
an increased sophistication about French writings could con-
ceivably have caused an author to eschew French terms in
favor of native (including, all unaware, horse) words.5
Again, Clark is precise though inaccurate when determin-
ing munbers of French words, counting as a single form the
noun and verb taxen from the same source-word, but counting
them as two if they are derived independently (affray, n. +
affrave, v. = one word; but afvaunce, n. + affve, v. = two
words). (Page 431) Furthermore, he counts separately the

full and aphetic forms of the same word: ”This is perhaps not

entirely reasonable, but it is convenient, and can hardly make

 

5 Later (page 555), Clark admits that it is "pointless and
unrealistic for us to pretend to know when a HF poet was
thinking of the native word and when he was thinking of the
foreign one."

17
any serious difference in any conclusions to be drawn from
the list," (page 430) the reason being, that Clark himself
does not take his lists very seriously. When you are out to
disprove a theory, any kind of statistics will serve. Kever-
theless, am happy that he did distinguish the full and
aphetic forms, because thereby he "proved" the greater pro-
portionate incidence of the full form in rimed poetry. His
care with the two kinds of Riddle English poetry no doubt
influenced my decision to consider verse forms in deriving
glossaries. Just as Clark wasted effort in verifying Bate-
son's ratios, so also did my meticulousness go for almost
naught. (See Chapter III.)

I find, upon re-reading Clark's dissertation four years
after I first borrowed it (during which four years my own
project received its form), that I am more indebted to it
than I would ever have cared to admit. The impression I re-
tained was almost entirely of his contentious tone stemming
from exasperation and frustration. I see now that it was he
who must have put into my mind the notion to settle upon the
inconspicuous words as possible indicators of writing habits.
From him came my technique of so dividing poems that the
effect of rime could be either emphasized or nullified. And,

finally, Clark's dissertation had, as I hope mine will have,

the value of pointing the way to a method and a point
of view, in determining questions of authorship, that
may, when brought to perfection, lead to more positive
and dependable conclusions than I have been able to
make them yield. (Page 572)

His method, owing to his negative point of view, yielded
Clark further frustrations. Positive thinking has a far
better chance at publication than does mere objectivity. Or
perhaps Clark's tone insulted too many too important
scholars.6 The dissertation was never published in its en-
tirety. Portions of it appeared in a variety of journals;7
but his powerful argument was completely ignored fifteen

years later by a supposedly demolished target when Henry L.

 

6 His argets include Gollancz, Savaf 3e, Oal;den (of course),
Iiss Iiary Serjeantson ("The Dialects of the west Ifi dlands in
Iiddle English," in Review pf En“lish Studies 5. 54. 136, 519
(1927)), J. R. R. Tolkien and E. V. Gordon edd., Sir Gawain
and the Green Knight (London, 1925), and Lane Cooper and his
doctoral students. O akden and Serjeantson are accused ofa
suboressio veri (page 55), as are Tolkien and Gordon (pages
114-5). Oakden is treated quite harshly throughout, often
through the medium of jokes and catch-words, the points of
which are not always clear to me. Perhaps most typical of
Clark's fits of ill-humor is the footnote on page 7h: "This
is neither Hr Chapman's first, nor his most ambitious, nor
his most fruitless contribution to the study of the four
poems. The only true and lawful claimant to those titles is
Mr Chapman's doctoral dissertation, produced in 1927 at Cor-
nell University under the Concordantifex Haximus, Professor
Lane Cooper, and entitled 'A Lexical Concordance of the
Middle English Pearl, Cleanne ess, Patience, and Sir Cawayp and
the Grene Knight.‘ In a fiVe-page preface (the only thing
the dissertation contains besides the concordance prOper), Hr
Chapman writes: 'The work is recorded on about no, 000 slips
...; and in this shape the copy in due time will be sent to
the printer...'" Clark is piqued because this concordance,
which he intended to rely upon, covers only six letters of
the alphabet. Howadays they use the computer at Cornell for
the production of concordances. A quality product results,
but without that useful byproduct, the Ph.D.

1
\
1‘

 

 

 

 

 

7 "Observations on Certain Differences in Vocabulary
Between Cleanness and Sir Gawain and the Green Knight,
Philological Quarterly, "VVIII_(1949), 261-73; ”Pa aphrases
for ‘God' in the Poems Attributed to 'The Gawa in-Poet, '"
Modern Langua e Fotes,L LVV (1950), 232- -6; "'The Gawain poet'
and the Substantival Adjective,’ Journal of En lis h and Ger-
manic Ph ilology, XLIX (19 50), 60-6 ; "On Certain 'Allitera-
tive' and 'Poetic' 1"ords in the Poems Attributed to 'The

 

 

 

19

(‘3

Savage“ again advanced the theory of common authorship as

being all but universally accepted.

 

Gawain-Poet,'" Modern Language Quarterly, XII (1951), 587-98.

 

8 Henry Lyttleton Savage, The Gawain-Poet: Studies i3 his
Personalipy and Background (Chapel Hill, 1956). John Conley's
review points out that "John W. Clark's sobering studies of
the vocabulary of the five poems are not even listed. Yet
Professor Clark shows that J. P. Oakden's Alliterative Poetry
23 Middle English, to which Dr Savage appeals, is far from being
trustworthy." Speculum, KKXII (1957), 858-61.

 

-AUT-1 ORSLIP P403 ELS IE TAB “ARLY COfiPLTSR AGE

The main quarrel that a statistician would have with

ClarZ: or his predecessors would be over the tin; samples that
they sow; times Zeafil with. Secondarily, the statistician

.L

mifht marvel over tleir lacv
takes into consideration ponulations and devia tio1:s. Final-

ly, he could ask for a kind of acctracy that either Clark

1

Q ' 1"

q .\ vs ~, - r a H ‘- ~ ‘A --. .. I ‘1' ‘ ‘w ‘ ,- ‘ (Y ‘6 3 . ‘
n01 1_e predecessors mere arepalen to QiOVldG. Pre-to.la «Hf

II scholars could not have foreseen the corp tor, a1d put off

,-
_l
.J

(‘1‘
x)
f)
O
3.)
(D
H
"J
'4.

heir researcEies until i lable. But tie author-

ship question continued to interest scholars even into the

nintee en- fifties ans -sittiee, who? a nu DC“ of stueies were

V O __ 0 ‘ _‘ - . p: . T . '1‘ {'1“ r .a_ I 4“ _'. 0
W10 Has Jinius° aid its consaniOh wold e, A statistical

N p.—.—- a... M " .- m-—.-=-—--—— m.-

.. , . . .. 1
metiod for De*e“11i.* Authorscia, oy Alver Elle ,ard (sta-

v‘w“ —.——«---

 

tistics by Per Sigurd Agrell), both leke heavy use of tie
computer. The latter voluae is a]: no t w ‘1olly concerned with

1

w w :. L...’ :-.: n ,. i r“ . u ,' 1 n a.
tables 01 seaeis ics Lorie, out ey co; 3 ter, Liti accompany-

ing explanations. Ihe pair or bocks is a conVinCirr perfor-

21
mance. Ellegard's history of the problem shows that Philip
Francis (1740-1818) had always been the prime candidate for
the honor of having composed the Junius letters, althoueh
Ellegard claims that he began his study favoring another
possible writer. Ellegard‘s central point seems to be that
direct testimony by the contemporaries of Junius can, now
that those witnesses are dead, have no chance of pointing
out the writer to blame for these incendiary pamphlets, and
that therefore a more objectively deductive method must be
used to determine their authorship.

Sir Philip Francis had long been the favorite candidate
among scholars who conjectured an author for the Letters.
Ellegard was persuaded by the biographical and other evidence
that the chances were avainst his having been the author, and
he set out to prove his contention by statistical means. His
method has several faint echoes of techniques that Clark 3-
anined in his critique of the scholarship on the "Pearl:
poet." Clark's unpatentable concepts of the favorite word2
and of the unconsciously-chosen expression5 alluded to at the
end of the previous chapter form the base of Ellegard's
system.

Simply, his system would compare corpora of Junie
writings with non-Junian writings by the presence or absence
of certain words or expressions. These plus- and minus-words

-- originally #58 of them, later reduced to 272 -- were

 

2 Clark dissertation, page 302.

3 Pages 9-11.

22
culled rather subjectively” from all of the Leggggg, from all
of the known, identifiable writings of Sir Philip Francis,
and from a "million-word sample" of contemporary writings.

A Junian plus-word is one used with higher frequency in the
Letters than in the "million-word sample" and a minus-word is
one that occurs with less frequency in Junius than in the
sample. The 458 words are on a sliding scale of positiveness
or negativity, with words used not at all (or almost not at
all) in the Junian letters designated the most minus. Elle-
gard himself culled the tables of occurrences of these #58
words or erpressions from double readings of all of the texts
involved (Junius, Francis, the million-word sample). From
these tables came the raw data which were fed to the computer
for manipulation by formula and for multiple grouping.

To repeat, the books by Ellegard and his statistician
are a convincing performance. I wrote a too-favorable review
of them for the iguana; _: inalisn and germanic Philology,
asserting my belief that Ellegard had solved for all time the
problem of the authorship of the Letters 2; Junius. I damp-
ened my praise, it is true, by mentioning a factual error or
two, and by not accepting completely either Ellegard's object—
ivity or his ability to scan so many texts for sis 453 items
as accurately as he would have his readers believe he did.

My review was in the mail to the Journal 9; English and

 

h Despite ilegard's claim of objectivity; see below,
gpages 26ff.

5 June, 1963, pages 688-9.

23
Germanic Philolosy before I saw the handling of the books by

the reviewer for the Times Literary Supplement,6 who brought

 

a nicer skepticism to the task.

I was then working on a series of projects for Professor
Arthur Sherbo at hichigan State University, taking over in
the midst of one project that had already been started with
another assistant. The goal of these investigations, as I
understood it, was to determine which of several 1000-word
texts belonged to Samuel Johnson and which were spurious.‘7
Therefore my interest in Ellegard's apparent success was
colored by my hepe for a like success, and slanted by my
involvement in a Johnsonian project.

Looking at the Ellegard books now that I have become
more blasé'in the face of ponderous scholarship and somewhat
conversant with, if not statistics, statisticians, I can well
believe that Ellegard's statistician was sneering at him, (he
is called "subtle” by the Times Literary Supplement reviewer)
while working up his tables of results. For statisticians
seldom put any credence in the statistics of laymen, particu-
larly When those laymen are scholars of literature. One
noted statistician with whom I talked at Michigan State Uni-

versity even refused to accept the widely accepted results of

 

6 "The Statistics of Style," January 5, 1965, P389 1-

7 Though the sampling was wide enough for an effective
analysis, having twenty-two texts (mostly Johnson's, but with
examples from the known writings of other possible authors of
the disputed texts), it was not deep enough: a thousand words
could not provide sufficient data for the relatively crude
tests to which they were to be put. There will be more about

these projects later.

24
the work done on the Federalist papers by Frederick Mosteller
and David L. Wallace,8 where the problem involves only two
possible writers.

The professional statistician's skepticism stems from
the tendency of dealers in authorship problems to vant to
sell their cause regardless of negative evidence. Thus,
while Ellegard claims to have been converted from an anti:
Franciscan stance, nevertheless his case is made to seem in-
evitably to lead to Francis. And although he may have chosen
his plus and minus test words from the Letters before he be-
lieved them to be by Francis, he still know Francis as a
prime candidate, and he could have been subconsciously influ-
enced in his choice of the 458 test words by his familiarity
with Francis' writings.

I can cite no authority for the caveats in the preceding
paragraph. The notions expressed formed in my mind as my
belief about Ellegard and others changed under the influence
of one or both of the statisticians to whom I talked at Mich-
igan State University. Some statisticians --—anc again I ca -
not cite texts or give names -- would further object to the
way the analysis in Ellegard's books groups writers and texts.
The million-word sample contains texts which are then re:
grouped into the writings of individual authors, and compared
with the million-word sample. There is a bias factor in the

formula used to evaluate the deviations of individual authors'

 

8 Inference and Disputed Authorship: The Federalist,
(Addison-Wesley, Reading, Mass., 196h).

 

25
writings from the standard of the sample. Katurally, «rancis
is found to deviate farthest from the sample and to be the
closest to the Junian standard.9 To one of the statisti-
cians with whom I spoke, a text is a text; and it loses its
testability when divided or when combined with other texts.
Ellegard's system repeatedly combines texts.

The 272 test-words are themselves combined also. The
changes are rung on the combinations of groups of groups of
plus— and minus-words. It is not enough that there is a
descending scale of Junianity attaching to groups of plus:
words; in order to test many texts in a variety of ways -- he
fills forty-five pages with tables ---Ellegard combines his
groups into the very Junian, the somewhat Junian, and the
slightly Junian. Of course, the first super-group excludes
non-Junian texts better than the other two. That this is a
species of circular argument is not pertinent, since the
results, by such manipulation, are rendered so much more
positive.

Putting aside the subtleties read between the lines by
the Times literapy Supplement reviewer of the Appenlix
written by Ellegard's statistician, Per Sigurd Agrell, per-
haps the main mark against the theory, even assuming that it
proves Francis the most likely author among hose authors

considered, is the contention that only a few of the several

 

9 The possibility that Francis might have consciously
imitated Junius is not given weight.

26
hundred thousand Englishmen capable of having written the
Letters were entered into Ellegard's process. Ellegard here
rightly falls back on biographical details- Francis was
available and knowledgeable as the letters were being written,
and they stopped when he went to India. Yet the fact remains
that almost any number of secretaries or mistresses of Oppo-
sition members in Parliament could have had access to the
information displayed, and might have had the literary skills
associated with Junius.

But if the "'one new fact' demanded by Dilke"1O is pro-
vided by Ellegard's finding that Francis' style is the
closest to the Junian of all the feasible and known contend-
ers, perhaps the question is after all solved. TLe method,
however, is not convertible to other problems of contested
authorship where the biographical details are not so clear-
cut. It would be almost worthless where the data-providing
subject matter is l'mited, where texts contain fewer than
two thousand words, or where the writings are in verse.

4y principal criticism of Ellegard's procedure concerns

his method of gathering data.

My procedure was as follows. After a cursory
reading of the whole material --all the Junius Letters
--, in order to get a general impression of the language
of the time, I carefully combed the Junius material for
words and constructions which seemed to me to be used
with remarkable frequency in it. I then did the same
for the comparative sample of a million words, noting
not only the words which struck me as remarkably fre-
quent in the various texts, but also those words which,

 

10 Who Has Junius?, page 119. Charles Wentworth Dilke
wrote in the Athenaeum in the mid-1300's of his disbelief in
the theory of Franciscan authorship of the Letters.

 

 

27
though not particularly frequent, I did not remember
having seen in Junius.

In this way I obtained a preliminary list of Junian
plus words --from my reading of Junius -- and of Junian
minus words ---from my reading of the comparative mat-
erial. After I had got the whole list of plus and minus
words by heart, I read through the whole tex material
again, registering each occurrence of each word included
in the list. When this had been done, the total number
of occurrences was added up for each word, in Junius on
one hand, and in the comparative material on the other.
After this, it only remained to calculate the distinct-
iveness ratio, and the final testing list could be
drawn up.

In order to minimize as far as possible the number
of occurrences lost by inadvertence, each page of text
was read through twice ovpr. Even so, however, mistahes
have certainly been made.

At the very time I was reading Ellegard's boon for review, I
was having trouble maintaining accurate counts of my own.

The old HISTIC computer seemed not to be as truthful as we
thought it should be, in giving us counts of words in senten-
ces and of words of certain lengths. I we ndered if the pro-
gram for alphabetized lists of words was also playing us
false. The text I checked was supposed to have thirty-four
instances of the word "of" according to the count of the
MISTIC computer. I then discovered how difficult it is to
find as few as thirty—four "of's" in a text of a mere thou-
sand words; and I knew the number that I was trying to find,
and was circling each "of" as I found it. It took about
eight readings. Ellegard's assurance that two readings would

not result in mistakes that would have significant bearing

 

11 A Statistical Method . . ., paces 22-3.

. 12 .
on his results is one statement that I cannot accept.
Ellegard was aware of the means available for avoiding

all mistakes, in deriving an unbiassed testin list, as well

C)

as in counting the items from that list in the separate

texts.

There are two ways of guaranteeing an unbiassed testing
list. One is to examine completely the vocabulary of
all the texts investigated, and draw up the lists of
plus-expressions and minus-expressions wholly on the
basis of such complete investigation. The other way

is to select a sample of expressions according to a
well-defined objective criterion, which can be reason-
ably assumed not to favour or disfavour any particular
candidate. The selection may be either random, or
systematic.

I have followed neither course. To make a complete
investigation would have been a Herculean task: it will
have to wait until the whole prgcess can be carried out
by electronic computer. . . .

Ellecard's choice was wrong. Unlike John W. Clark in 1941,

he could have chosen the computer and he should have waited

for suitable programs for sifting and culling. Instead, he
repeats the error that Clark is so vociferous about of having
precision without accuracy. His indices to the fifth and sixth
decimal place, and his forty-five pages of perfectly worked-out
tables are all based on shifty data. The precision and
accuracy must begin at the beginning or there will obtain
inexorably the dictum of the computer operators: Garbage in,

garbage out.

Computer-use for authorship problems was very much in

 

12 g Statistical Method . . ., page 23.

13 Who Was Junius?, page 113.

29
the atmosphere in the early sixties. The first phase of the
Johnson problem initiated by Professor Sherbo ground to a
halt: the thousand-word texts we were using were just not
large enough to show distinctiveness in such aspects as sen-
tence—length, or numbers of x-lettered words. We were look-
ing for a determination of which of several disputed texts
were Johnson's by means of a statistical process suggested by
George Udny Yule.”L Even the excellent glossary program
which gave us the occurrences of every vocabulary item in
alphabetized lists failed to yield promising results, pre-
sumably for the same reason: the shortness of the texts.

We then struck out in a new direction. Retaining the twenty:
two thousand-word texts while finding some way to increase
the kinds of possibly distinctive data was the immediate task.
White Knight that I am, I invented my own system.

The system sought to multiply the data by counting, not
words, but groups of words. The repetition of patterns
might be the clue to an author's writing habits. Raw words
would, of course, not serve this new purpose, because there
is so much variety in the selection of content-words that
the longer patterns would almost never be repeated. The
raw words were given coded designations indicating their

"part of speech" and their "funstion in sentence.” Thus the

 

14 g Statistical Study 2; Literary Vocabulary (Cambridge,
Macmillan, 19AM), and "On Sentence-length as a Statistical
Characteristic of Style in Prose; with Application to Two
Cases of Disputed Authorship," Biometrika, XXX (1938), pages

563-90.

 

30
expressions "in the house," "on the town," and "over the rain-
bow" would all have the same coded appearance: PJ DE KP, for
"Breposition introducing adiective phrase," "the Qeterminer
'thE,'" and "goun object of Ereposition." All 22,000 words
had to be so coded (by hand!) before the computer program for
selecting patterns could begin to work. By this time the
casual reader will recognize the same old fallacy of pre-
cision without accuracy in the work. r‘he coding had to be
done by hand, and I made the usual claim of consistency to
refute in advance any argument that my data might be deficient.

The program ran, the like patterns were collected, and
the output was analyzed. The results were nothing if not
inconclusive.

Not so, however, with the results of a very similar in-
vestigation that was being carried on at the same tine at
Columbia University by Louis T. Kilic. Hr. Milic's disser-
tation also examined prose of the eighteenth century by tabu-

15

lating patterns of words; but his coding of words was based
on the structural grammar of C. C. Fries rather than on the
traditional parts of speech, and his sample texts ran to four:
thousand words instead of our one thousand. Perhaps, I

reasoned, his results would tend to be more positive than

ours o

 

15 A Quantitative Approach to the Style of Jonathan Swift
(The Hague, Houton, 1967). Er. Hilic and I exchanged several
letters and talked long distance once as he was putting the
finishing touches on his dissertation. I can at this moment
understand his state of mind the day my call went through to
him. By the time his work would have been available, I had

already abandoned the system.

 

 

31
Despite Milic's apparent success, (his "study claims to
have produced a method of identification by internal evidence,
free of the usual uncertainties, using statistical methods
and computer technology")16 I decided to reject the process
we had discovered independently when it came time to under-
take my own dissertation project. What course I took, how
it failed, and why I think yet that it is, in the main, the

right course form the substance of the remainder of this

thesis.

 

16 Dissertation Abstracts,XXIV (1964) 3730.

CHAPTER III

DISCOVERING A TEST FOR PROVIMG AUTHORSHIP:
A STATISTICAL TREATHEET OF MOST OF THE LONG POEMS OF LORDSWORTH,

KEATS, SHELLEY, BROWSING, AI-ID TJINYSOI‘E

The project I now undertook started out to be the impossible
one of proving that authorship cannot be proved by a statistical
ex mination of vocabulary. Translated into possibilities, it
meant that I would devise the best test I could and use it under
ideal conditions with the hope that it would work but with the
expectation that it would not.

Of all the segments of the initial project with Professor
Sherbo, I had confidence only in the "glossary" program. The
tests for sentence— and word-length were almost patently unwork-
able when used on our short, thousand-word texts. With alpha-
betized word lists, however, texts could be compared for every
item of vocabulary.

Such was my intention: to omit consideration of no word
out of fear of the charge of conscious or unconscious prejudice
that Ellegard was subject to when he drew up his list of #58
items. I also left the Johnsonian milieu with its doubtful
texts so that I could concentrate on the test itself and not
on any immediately practical application thereof.

To the best of my limited knowledge, there had been no
completely objective tests of authorship. Even if the researcher

32

KN

3
is without an axe to grind, so to speak; that is, even if he
does not hold a prior belief that a certain author is to be
credited with the disputed work, he nevertheless does begin
with a strictly l'mited set of candidates, and omits from
consideration the possible stray contributor, or the truly
anonymous writer who was not known in his own time to have
written anything. Always, in such research, there is the task
set of attributing something of doubtful authorship to a
:nown writer. And although controls are purportedly used, the
methods are never tested entirely apart from the problem for
which they were designed. Having a goal in mind can cause an
experimenter to color his data, even unconsciously, by select-
ing items for analysis any other way but at random.

Another abuse of scientific methodology occurs when com-
paratively scant data, never gathered with perfect accuracy,
are formulizated and magnified into imposing tables of figures
to the third and fourth decimal place. This could be called
the Gold Bug distortion, whereby a mistake of inches near the
trunk of the tree amounted to many feet when the final line
was projected. The deeper you dig in a wrong location, the
more foolish you appear in retrospect. Bllegard's two volumes
on the Junius problem are a good example of this sort of
abuse.

I take pains to avoid both of these pitfalls. Hy purpose
is to seek a method of proving authorship by examination of
vocabulary usage. I avoid the first trap by choosing to ex-

amine only known works by known poets. And, secondly, I allow

the perfectly accurate countind machine to anass xv initial
data, from which I subtract, by wholly objective means, the
usable oarts.

Ce n a poem by Keats be distinpuished from one by Shelley,
hordsv.'orth, Browning, or Tennyson throufih the use of a test
involvinf the poets' choice of certain voca *oule ry items?

With the aim of discovering such words, I set about to feed

.1

every noon of more than t7xo thousa ' words by these five
ninetee nth-century poets into the 3600 Control Data computer
then newly installed at nichigan State University. I did not
have enotéh time to submit every poem of 01 e t1an t o thou and
words by the five poets, but my ozuis sions were completely by
chance (see Appendix A for a list of the tejzts). If an author—
ship test by VVCGJUlc ry does exist, it probably will not work
on poems of much less than four the rs and words, but my inten-
tions included the deternininfi of how s all a text can be
tested successfu ully.

Each line of each poem vas punched on an l;fi card and
carefully ve7i1ied see Specimen 2 in Appendix F). Not
knowing which words would enter into my analysis, I soucht
to eliminate ambiguity 3y Iollo““fi certain conventions of
spelling. The aLIiliaries "may” and ”niqht" were to he sepa—
1e nouns of the saze s; e,lling oy appandint an
to the nouns. The Bri ti sh "round" was always spelled "around”
when it meant "a:?o nd. Contractions were spelled out so that
both parts of the word could be counted while the ”word”

0 v n I _‘ ._f f‘ - C“ . r. ' .A, J-
itse 11 would rerieter only once: "itis" 1or “it' ,” and ”cahht"

for "can't." fly failure to somehow differentiate ”to" tie
preposition and ”to" the infinitive sifnal is only slightly
mitigated by the contention that all ”to‘s" are equal inas-
much as the poet is choosing in either instance the same
two-letter word.

Mr. James D. Clark, of the Department of Psycholoqr at
an State Unive-sity, wrote the program for data re-
trieval. Mr. Clark's glossary prepran yielded me alphabet-
ized lists of words and their occurrences in more than two
hundred texts (see Specimen 3 in Appendix F). The possibil-
ities for expanding the number of "texts" are nearly
limitless, since halves, thirds, fourths, or sixths can be
combined in many ways. No text is made up of parts from
different poems, however.

With the computer output I was ready to follow Ellegard's
retrospective advice to select words entirely objectively
(see quote, pages 26-7). I conned the lists for words sig-
nificantly present in texts by one poet and not so iresent in
texts by the others. he such words seen to exist; that is
to say, if a word is used in several poems by one writer, it
will be used with about the sane frequency by at least one other
writer. It would be necessary, I decided, to work with those
words apnearing in practically every text. These are the words
automatically excluded from nest concordences: the non-content
or function words. And since they all agdear in aluost every

1 ‘-

tevt, my treat ent of then would have to consider their rela-

‘5

—: a. L ’H ‘ a 'P J' \' ", ~. A, '11 ‘3’ '. , '0 ' -\ 1.
tiers to one 3R0bu‘l: does a writer's re~cateu c oice Oi LL-

56
diminish his uses of "a” while at the same time, perhaps, his
"and's" are impinging upon his "or's"?

I had thirteen of my texts (see Appendices A and 3) of

1

about eight thousand words scanned for the words common to
all thirteen. After the disqualification of personal pronouns
as too dependent upon content,1 forty-five words remained.
Fourteen of the forty-five have variant forms, which were
carefully combined to make single items (see Appendix C for
the allomorphs of these fourteen items). It was not possible
at this stage to distinguish the usages of several ambiguous
forms, such as "as" and "like," but even deliberate refusal
to distinguish them could be justified by the argument that
the poet did after all choose the word in question, and pro-
bably unconsciously, since most of the forty-five items tend
to take lisht stress in their verses. This justification
could perhaps extend to the single item "to," which might
have been separated into its use as proposition and use as
infinitive signal.

The forty-five key words having been determined upon by

purely objective means, it remained to find a way of using

 

1 This decision is based on an experience that Professor
Sherbo shared with me. An xamination of three 12,000-word
texts from consecutive issues of The Gentleman's hagazine of
the 1740's for clues to identify the author of the doubtful
middle text showed that the single word to vary significantly
in usage from text to text was the feminine pronoun. One of
the articles was about the Queen of Spain and also used the
feminine pronoun for certain abstractions. Another article
also used the feminine pronoun for abstractions, but was not
concerned with the Queen. The third had no feminine pronouns,
although some of the same abstractions were referred to.

37

then for comparing texts. It will be remembered (Appendix A:
list of texts) that the texts vary in length from one thousand
words to eicht thousand. I imagined that there could con-
ceivably be distinct patterns of usage for these test or key
words. Such patterns would have to have been determined en-
tirely by unconscious selection by the poets. If, or since,
they were beyond the ability of the poet to control, these
patterns should be so much the more effective for use in es-
tablishing an authorship test that would distinguish a poet
from his imitators.2 And if such a test exis ed, it could
possibly be used on texts of vastly unequal length. So,
rather than concentrate on comparing texts of commensurate
size, I decided to compare each text with every other.

Accordingly, I made profiles of all the texts by graphing
the forty-five words. I gave the word most frequently used
in a text the value of IO %, plotted at the top of the graph.
Each of the other forty-four words were given proportionate

positions (see Specimen 6, Appendix F). In order to compare

 

2 My interest in the subject somewhat antedates my quick
Masters Thesis (University of Detroit) written in the summer
of 1959 in which I "traced" "evidence" of "influence" of
Shelley on four subsequent poets, mainly through their use of
common vocabulary items. The kind of item I then concentrated
on is typified by the word shiev used by Shelley, of course,
and by Francis Thompson in what must have been a conscious
attempt to resemble Shelley. A word would not have to be as
outrageously "poetic" as skiey for an admirer to borrow it;
the other conscious borrowings would, however, tend to be the
slightly out-of-the-ordinary. Apart from the function words
in "turns of phrases" so borrowed, practically all of the
borrowings would be content words. And no imitator would
think to conform his own usage of all function words to the

patterns of his master.

\N
C

the graphs visually, I enlarged the forty-five points to
quarter-inch holes (see Specimen 7, Appendix F). Preperly
positioned, one on top of the other, all the points of com-
parison between two graphs were immediately visible and
ready to be counted. A gross count of simple coincidences
would not have justified the use of graphs, since such data
could be compiled merely by having the computer examine the
charts of numbers that lay behind the graphs, and letting it
do the ratios at the same time. Perhaps, I reasoned, the
best profile similarities were skewed out of recognition by
the lack of coincidence between the two leading words, that
is, those given the value of 100%. Frequently it did happen
that the greatest numbers of closest correspondences were
found only after searching for them.

This searching added but little time to the comparison
of each pair of graphs. Each comparison of two graphs took
approximately one minute. After taping Graph #1 to a dark
board, I positioned Grap #2 atop it and counted (1) the
number of holes that corresponded at all, (2) the total that
corresponded closely (that is, that showed more than half a
hole-diameter), (3) the number of correspondences above an
arbitrary 15$ line. For the fourth and fifth indices of
correspondence, Graph #2 was slid up an inch toward the top
of the board while I looked for the greatest number of addi-
tional correspondences above the 15% line on Graph #1. Then
(Eraph #2 was slid down an inch (two inches, really) while

correspondences above the 15% line on Graph #2 were sought.

\JJ
Ki)

Sometimes more than a minute was consumed in finding the
scores for the fourth and fifth indices - approximate and
close correspondences above the 15% line. Time was also
consumed in taping the bottom graph to the board and re-
moving it, in taking out and putting away the sets of

graphs and the sheets onto which I was writing the five
indices. I wrote five index numbers (sometimes seven) for
each of 230 x 229 x % comparisons of graphs, and I counted
about fifty correspondences for each comparison of two
graphs. The 1,320,000 bits of information thus counted

were recorded on a triangular chart made up of three hundred
individual 8%" by 12" sheets, and measuring twenty by twelve
feet (see Specimen 8, Appendix F: a part of one of the 8% by
12 inch sheets).

A computer, which would not have had to tape graphs to a
board, or would not even have had to use graphs at all, could
have completed the counting in a matter of minutes once it
had been programed and the material had been prepared for it.
I justify my performing this long phase without the aid of a
computer by the fact that I was not.exactly certain what I
was looking for or how I would be able to use the data I was
compiling. At one stage of the comparison phase, when it was
about one-fourth finished, I took notes on how a certain text
(#56 by Shelley; see Appendix A) compared positively with
other texts: that is, with what texts did it yield high in-

dices of correspondence. I was most exhilarated when, seven

40
times out of seven, high indices actually did point to other
texts also by Shelley. But the eighth, tenth, and eleventh
comparisons of the twelve made resulted in false identifica-
tions. So promising did the system appear at this t'ne that
I attempted to present an explanation of it at the April 30,
1966 meeting of the Hidwest hodern Language Association in
Iowa, where I learned that it is nearly impossible to present
unconvincing facts convincingly. For by the time of the con-
ference, when my facts should have been more firmly positive,
because I was by then dealing with larger texts, they were
more inconclusive than ever before or since: I knew they were

not as positive as the Shelley #56 figures indicated, but

I—l

neither could I say that they would be negative until had
finished the three hundred pages of my 20' x 12' chart.

With all the indices tabulated, the final step was to
test the results. If there is a profile or a set of profiles
made up of points on a graph representing the proportional
occurrences of forty-five common words selected objectively,
and typical for each of my five poets, then surely the follow-
ing test will find it. I reduced the five indices for each
comparison down to four criteria, three positive and one
negative. The first positive criterion consists of the five
indices added together, the second is merely the third index
(correspondences above the 15% line), the third adds the last
two indices (the moveable ones), and he last is the same as

the first, except that the low total is the test. That is to

say, if two graphs have no points of correspondence then the

1.5.. 1
poems behind the graphs should not be by the same poet. Any
of the positive criteria should identify texts as being by
the same poet, since what I looked for in each case were the
extreme examples (combination of indices distorts the data in
my favor; it was done because there was no way of predicting
which criteria or indices would assay out). For each of the
230 texts, I rejected all but these texts of the remaining
229 that were most like it. By expectation, at least above
some limit of about 3,500 words, each text should have selec-
ted matching texts; each Keats text should have selected
{eats texts, each Shelley, other She_leys. Beyond this, the
negative criterion should never have selected texts by the
same poet.

The chart of the last analysis (see Appendix E for a
summation of this chart), showing the results of one of the
grossest possible tests of the validity of my method, must
measure 230 by 230 squares. host of these squares will not
contain an entry because only the extreme examples of merely
four criteria are tabulated for each row of 230 squares. Let

page) what might be a ran-

J.

me present here (in Figure 1, next
dom sample of the chart of the last analysis. Divided into a
hundred equal parts of twenty-three squares on a side, one such
part of the chart, the twenty-third (counting from top left),
contains twenty-one criteria of correspondence, every one of
which indicates texts that are indeed by the same poet, or, in
the case of the negative criterion, texts not by the same poet.

These are exactly the results I had hoped for. The next step

568 11

58W 311

61W KEY TO CRITERIA

- total of 5 indices . 0
3rd index: correspond-
55w ences above the 15% line

- 4th + 5th indices
low total of indices

o\o\
bu
£7:
03h:
u:

ONO
\lO“
2222
H
OH
Ill

FIGURE 1

Square 23 of the Chart of the Last Analysis

This is the only square of the hundred into which I divided the chart
of the last analysis where there are no misses. Thirty-one of the hundred

have more misses than hits.

#3

is to use a finer test, broadening the criteria until negative
results are reached.

Rather, that would have been the next step, had Square
#23 been typical. A look at the final tabulation (again, see
Appendix B) will show that Square #23 is the only one of the
hundred that is unanimous in supporting the original theory.
The remaining ninety-nine squares range down to a perfectly
negative correlation (Square #44, Appendix E), with all the
criteria miss'ng the mark. Hot even the negative criterion
had reliabili y. 3y rifhts, the negative criterion should
never have appeared when texts by the same poet were being
compared. Yet it did, forty-five out of 295 times, a ratio
only slightly better than pure chance. When a text selected
more than one other text for positive or negative correlation
all pairings were listed. For example text #54 (Browning)
correctly chose five Wordsworthian texts in the twenty-third
square alone for negative correlation. In each of these five
instances all of the indices added up to only ten correspond-
ences. Text #Sh selected non-Browning texts sixteen times in
229 trials, but selected another Browning text once. This is
only three or four times better than guessing. Likewise the
positive criteria, although far better than pure chance, which
would score a hit approximately every fourth or fifth time,
still did not have the consistency needed for an effective
test (see Appendix G).

The present experiment, therefore, is a failure. It did

not discover a way of proving authorship by means of words

44
selected objectively and graphed without regard to text length.
Improvements in the method are beyond the scope of the present
study, which has now lost some of its innocent objectivity.

I have some faith that by working over the data that went into
the positive results I can cull out certain of the dead vocabu-
lary items and build a stronger test. It is encouraging that
two of the positive criteria, the second and the third, were
right roughly four times as often as they were wrong, and that
the first criterion was almost sixteen times as effective as
guessing; that is, it was right three or four times as often as
it was wrong. Especially by concentrating on the relatively
few data of Magic Square #23, something viable might be con-
ceived (some of the magic of this square perhaps derives from
the fact that seventeen of its twenty-three texts are from

Wordsworth, thirteen from The Excursion alone). Future mining

 

of these statistical lodes, however, will have to avoid the
pitfall of circular reasoning, and any testing device so de-
rived will itself have to be tested against poems not on my
list. Until a test is devised that will work on any hundred
out of a hundred texts of known authorship, I, for one, will
refuse to accept its results when it is used with texts of

doubtful autho*ship.

CHAPTER IV

COLCLUSICI: OUTLOOK

The time has come for me to fulfill the promise made in
the Introduction to point out the way that the attribution of
authorship problem may be solved. To recapitulate: John W.
Clark concluded that the problem probably will never be
solved, since it is impossible to tell whether close similar-
ities between texts indicates common authorship or mere in~
fluence of one author on another. This almost certainly will
remain true of disputed texts such as his four or five poems,
whose author(s) left no definitely attributable corpora to be
used for wcomparison. Circular arguments, however much pleasure
they give the disputant, cannot be said to solve anything; and
when one compares data from eawain with data from Pearl, and

data from Patience and Cleanness with data from the other

 

two, all without knowing how many authors are involved or what
their known works will yield for data, it seems more than a
little vain. If the names of the poems were applied to cigar-
ettes respectively noted for "manly flavour," "taste beyond
price," "slower burning," and "less smoke per puff," it still
could be that all the brands are rolled in the same shed in
Lexington, although not necessarily by the same machine.
Almost as a postscript to the problem taken up in Chapter

One comes in 1966 A Concordance to Five Kiddle English Poems,

 

a computer—derived work by Barnet Kottler and Alan M. Markman

#5

46
(University of Pittsburgh Press). The concordance relies
heavily on the work of Sir Israel Gollancz, whose editions of
four of the poems are the basic texts, that have henner's,
Savage's, and Tolkien and Gordon's for variants, and whose
edition of the Pgagl is one of three variant texts backing up
the 1953 Gordon edition. The concordance is a volume that
would have saved John W. Clark much of his labor, thereby
perhaps depriving us of one of the most truculent pieces of
scholarship existing.

In the Kottler and Karkman Concordance all the words of

 

the five poems are listed somehow: those in Appendix I are
frankly "Partly Concorded," which means that some of the line
numbers are given for their occurrences, although why all are
not given is not explained. Of special interest here is the
decision not to concord a list of 152 words -- a list

strange in that it includes words used thousands of times
("the," "and") and words used only once ("foul") -- including
all but fifteen of my forty-five test words. TLe list of 152
merely gives the numbers of occurrences of each item in the

five poems as a group. Four more pages added to the book's

 

total of 794 could have presented the numbers of occurrences
of these words in each of the five poems. But again the pre-
sumption on the part of the compilers is that there will be
more interest by authorship scholars in the less common words.
It is nevertheless an improvement over the Hatthew Arnold
concordance and older concordances generally in which the more

common words are merely listed without any tally whatever.

#7

1

Moreover, at the University of Pittsburgh, accordin' to

C

Kottler and Harkman, retrieval of further information is
possible, since their input data is kept stored on majnetic
tape. But before that tape denagnitizes, it might be best
to get everything of possible use in print.

But to return to the recapitulation. Ellegard's con-

0.‘

clusion was that the Junius problem lad been solved. Against
him it can be maintained that circular reasoning was used to
show that Francis' style is, not identical with, but closest
to that of Junius. Horeover, the data going into his calcula-
tions is suspect because it was not objectively or accurately
derived. This last objection also is sufficient to vitiate
the ole ms that a project like Milic's might have to validity.
Yet, if there is still anyone who wants to know who the
real Junius was, or whether the supposed author of Shakespeare's
other works was the same one who wrote The Two Noble Kinsmen,
there may be hope. I believe the hope to lie in the method of
profiles. If I had to try the project again, I would choose
five other writers, and work with only half of their works. I
would have a computer count and combine allomorphs for the
forty-five items plus a few more: an item for all personal
pronouns, and an extra form for "to" would find places in the
new list. I would then have all of the operations for deriv-
ing my first three indices done by computer. .
I would consciously use the circular device to find out

which items of vocabulary are most usable in obtaining positive

results: wherever "hits" were registered, the profiles would

be scrutinized for points of correspondence. In this way
maybe half of the items could be eliminated as having no
bearing on the distinctiveness patterns of writers.

Armed with this stripped-down list, I would turn the
computer loose on the other half of my writers' works. I
would fully expect the assaying power to be great --much
higher than 50%; but I would demand that it test out every
time before I would claim for it any efficacy in proving the
authorship of doubtful writings.

For if we insist on having recourse to the techniques
of science, we must abide by the strict laws of scientific
proof. No chemical test is valid unless it always work
under controllably identical conditions. If there be an
"essence of Keats" underlying subject matter, theme, word
order, thought, emotion, and style, that essence ought to
be detectible. And if it is truly his essence, then it will
be found in every one of his works of appreciable length.

Given the existence of this "essence," ”profile," or
"handprint," will it ever be possible to feed into a computer
a newly-discovered work and get a positive identification or
non—identification within seconds of time? I can answer that

easy question, "Absolutely yes.”

Words
in
Code Poet Text

APP E1 IDI ii A

The Texts Used

Poem Title

Division

 

1 T 976
2 if 985
5 T 936
h T 100h
5 T 1007
6 T 1013
'7 K 1051
8 T 1128
9 T 1142
10 T 1151
11 W 1175
12 W 1269
13 W' 1377
14 T 1395
15 S 1415
16 K 1455
17 K 1478
H3 T 1h79
19 S 1536
20 T 1625
21 W 1623
22 T 1650
5 T 1650
21+ T 165:?
25 T 1653
26 T 16MB
27 5 161,6
8 s 1652
29 S 1632
50 S 1752
*51 I( 1747
32 S 1761
33 S 1831}
34 w 1896
55 S 1922
56 S 1968
37 T 1939
38 T 1991
59 W 1992
LLO K 2055
1:.1 K 2057
1+2 W 2055
15 S 2062
hlv, B 2062
[3-5 B 2075

Two Voi
I!
H
n
H
H

C88

Hyperion
Locksley Hall

H
H

White Doe of Rylstone

In Kenoriai

Queen
The Eve
11
Two Voi
Queen H

Locksley Hall+60 Years

Rab

of St. Agnes

ces
ab

White Doe
Locksley t 60

fl
1!
II
N

5rd Third
2nd Lines
5rd Lines
2nd Third
1st Jhird
1st Lines
Book III

Odd Lines
Odd Couplets
1stHflf
Canto VI
CmfloV
Canto IV

7th 206 Lines
Part II

2nd Half
1st1kdf

2nd Half
Part I

Even Couplets
Canto I
1stfiflf
Odd Lines
Even Lines
2nd Half
Odd Couplets

Queen Kab Part VI
" Part III
" Part VIII
" Part IX
Isabella
Daemon of the World Part I
Queen Nab Part V
Michael 1st Half
Queen Hab Part VII
" Part IV
The Princess Part I
Oenone
Michael 2nd Half
The Fall of Hyperion 2nd Half
" 1stHflf
The Borderers Act IV
Lines Written Among the Euganean Hills
Pippa Passes Hisht
H Iloor1

#9

 

Words
in

Code Poet Text Poem Title DlViSlon
#6 W 2096 White Doe Canto I
1+7 W 2111'r The Excursion Ist ‘alf, Boo: VIII
Lu?» W 2160 " 1st Third, 300-; II
(,9 W 217’} ‘.'-."hite Doe Canto VTI
50 W 2193 " Canto III
51 S 2199 Mask of Anarchy
52 W 2219 The Excursion 2nd Half, Book VIII
55 T 2251; Locksley Hall
5b, B 226:9; Pippa Passes I-Iorning
55 1'.’ 2270 The Ez'tcursion 5rd Third, Book II
56 S 2278 liellas Rimed Lines
57 W 2279 The Prelude 2nd Half, Book III
53 W 2295 The Excursion 2nd Third, 3001; II
59 W 2516 " 5rd Fourth, Book IV
60 W 2552 The Prelude 1st Half, ‘dool: III
61 W 255L:. The Excursion 5rd Third, Book III
62 VI 2559 " 1st Fourth, Book IV
65 K 2357 0tho the Great Act II
61;. W 2560 The Excursion 2nd Third, 3001-: III
65 W 2565 " 2nd Fourth, Book IV
66 w 236.6 " 1st Third, [3001: I
67 1'! 256.9» " 4th Fourth, Book IV
68 W 2575 " 181: Third, Book V
69 K 2575 Lamia Part II
70 S 2536 The Cenci 1st Half, Act V
71 S 2537 Daemon of the World Part II
72 S 2597 The Sensitive Plant
75 W 2&05 The Excursion 1st Third, Book III
713, S 21+L;.9 The Cenci 2nd Half, Act V
75 W 21.52 The Excursion 2nd Third, Book I
76 W 21:56 The Prelude Book XII
77 W 21:65 The Excursion 2nd Third, Book V
78 T 21+71 Pelleas and Ettarre 1st 'Ialf
79 T 21477 Balin and Balan 1st Half
80 W 2485 The Excursion 5rd Third, Book V
81 S 21491 Letter to I-Iaria Gisborne
82 S 2506 Hellas 1813 Third
85 K 2550 0tho the Great. Act IV
8L1. T 2555 Pelleas and Ettarre 2nd Half
85 T 2555 Balin and Balan 2nd Half
86 W 2575 The Excursion 1st Third, Book VII
87 w 2584 " 2nd Third, Book VII
88 W 2585 " 5rd Third, 3001; VII
89 W 2585 " 5rd Third, Book I
90 T 2615 Merlin and Vivian 31; Third
91 T 2627 " 5rd Third
92 S 265i; Hellas 5rd Third
95 K 2655 0tho the Great Act III

Words

in

Code Poet Tex

 

94

95
‘96

97

9’7)

99
100
101
102
105
104
105
106
107
108
109
110
111
112
115
114
115
116
117
118
119
120
121
122

3
124
125
126
127
129
129
150
151
152
133
154
135
156
137
158
159
110
1191
142
1 1+3

BLUUJHCUEEEIDU)VH8tDCJB=SHBB+3 m

A
=:-
V

WCUUJwtafijfi

NLJJHFEll—BUJUJ“

H4
V

raqutntaw

2653

2674
2695
2636
2693
2702
2704
2707
2709
2724
2725
2725
2729
2751
2754
2736
2 53
2749
2752
2754-
2760
2767
2775
2778
2905
2310
2712
2817

_ 3)
()3 C)\'] \1 (D (Db-J [‘0

J

C) .

3’)

J—*O<3\30)

,_ ') (4'3 1

,—
\—

JQ -'

OG\G\

I 7:) ..
‘
J

\O\’)\:) ..
...J
O

mruh)RHvr%DJh)merun)
O\O
\n

2922

Poem Title

51

Division

 

Hellas

In Hemoriam

herlin and

In Kenoriar

Guilt and Sorrow

In Memoriam

The Pope (Ring and Book)
11

Vivian

In Memoriam
Hyperion
Charles the First
11
Husings hear Aquapendente
Guilt and Sorrow
The Pope
In Memoriam
The Pope
11
In Kenorian
Guilt and Sorrow
In'fienoriam
The Prelude
Guisseppi Capnsacchi (R 6 E)
The Princess
The Pope

Guissppi Caponsacchi
H
11

The Cenci
The Princess
The Excursion

0tho the Great
The Excursion
ll
11

Gu1iss ejapi Caponsacchi
11

The Lover's Tale

1!

H
Half Tone (Bin? and Book)
Hyperion
IIal f Roz": e
Tw Voices
half Rome
The Idiot
Half Rome
The Lover's Tale

”3
— '7
Joy

2nd

6th
2nd
2nd
2nd

th
LL 1: I1
2nd
4 VI

.0 ‘r
#004;

1st
2nd

Third

412 Lines
Third
412 Lines
Half
412 Lines
Sijzth
Sizflni
412 Lines
I
half
Half

3 Rimes

st
5rd
5th
6th
1st
1st
7th

Sixth
412 Lines
Sijith
Sixth
412 Lines

Half
412 Lines

Book XIII

5th

Sixth

art III

5rd
6th
2nd
1st

Act

Sixth
Sixth
Sixth
Si .:4.

II

Part VII
Third, Book VI

1st
Act
2nd

V

Third, 300

k VI

1st Half, Book IX

2nd
4th
5rd
5rd
st
2nd

2321(I

2001:

1st
5rd

4.1; h
L1. th

EIal f, Book
Si::t}1
Siszth
Fourth
Fourth
Fourth
Fourth
II
Fourth

Fourth

1
Fourth
7“ ‘V 4"
£Oarbfl

‘7
1‘.

 

Words
in

Code Poet Text Poem Title Division
144 K 5056 Lamia Part I
145 T 5060 The Princess Part VI
146 S 5061 Prometheus Unbound 4th 400 Blanks
147 T 5067 The Last Tournament 1st Half
143 S 5071 Prometheus Unbound 5rd 400 Blanks
149 K 5077 Sleep and Poetry
150 T 5114 The Last Tourn'ment 2nd Half
151 S 5151 Prometheus Unbound 1st 400 Blazfizs
152 S 5150 " 2nd 400 Blanks
155 T 5268 Loclzsley Hall Sixty Years After
154 1'] 5521 The Prelude 3001; XIV
155 S 5525 The Cenci Act I
156 S 5554 Queen liab Parts III 8.: VIII
157 S 5555 " Parts II 8: VII
158 T 5415 The E-iarriaf‘e of Gersint 1st Half
159 T 31mg N 2nd Half
60 ‘3.’ 5444 An Evening? Wall:
161 W 5464 The Prelude Book III
162 W 5470 " Book II
163 T 3501 The Ping:
164 ‘52" 55273 The Prelude 33001: IV
165 W 5553 The Brothers
166 T 5567 The Princess Part II
167 K 5605 Endymion 1st Half, Book I
161°; II 5622 " 1st Half, 13001: I".
169 B 5637 Pompilia (Ring and Book) 2nd Fourth
170 B 5759 " 5rd Fourth
171 K 5751 Endymion 2nd Half, Book I
172 T 5755 Gareth and Lynette 2nd Third
175 5 576-1 Prometheus Unbound Act III
E
3%
I77 K 5777 0tho the Great Act I
173 T 5779 Gareth and Lynette 5rd Third
179 K 3798 Endymion 1st Half, Book II
180 B 5801 Pompilia hth Fourth
181 S 5811 The Cenci Act III
182 T 5812 Gareth and Lynette 1st Third
185 B 5820 Pompilia 1st Fourth
184 K 5826 Endymion 2nd. Half, Book III
185 S 5857 Prometheus Unbound Act I
186 T 5868 The Passing of Arthur
18 K 5872 Endymion 2nd Half, Book II
188 S 538 The Cenci Act IV
189 K 573.92 Endymion 2nd Half, 3001*; IV
190 T 35925 Geraint and Enid 1st Half
191 K 5961 Endymion 1st Half, Book III
192 S 5‘6") Adonnis
195 T 4025 Geraint and Enid 2nd Half

Words

 

in
Code Poet Tex Poem Title Division
194 VI 4245 The Borderers Act III
195 V! 4266 The Prelude Boo: IX
196 'r 4275 The Coming of Arthur
197 s 1.3-5.011. The Triumnh of Life
‘I’
I99 I? 4407 The Princess Part V
200 “I 4455 The Prelude Book X
201 S 4492 Rosalind and Helen 1st Half
202 5 454-7 " 2nd Half
205 ID 4562 The Princess Part IV
*
’2‘05 1': 13-5711. The Prelude 3001; v
206 VI 4575 The Borderers Act I
207 ’T 4672 In Hemoriau 2nd Half, Even Lines
203 ID 4690 " 2nd Half, 3 Rimes
209 IT 4744 ” 1st Half, Even Lines
210 'T 4760 " 2nd Half, A Times
211 T ‘4761 " 1st Half, 3 Times
212 V1 4770 The Prelude Book I
215 T? 4773 In Hemoriam 2nd Hilf, Odd Lines
*
215 T 4323 In Hemoriam 1st Half, A Rimes
216 T L545 " 1st Half, Odd Lines
217 VI 4375 The Borderers Act II
2V3 S 4922 Julian and fiaddalo
219 V1 493’ The Prelude Book VIII
220 S 5023 Peter Bell the Third
221 VI 5107 Descriptive Sketches
222 S 5559 Swellfoot the Tyrant
225 S 5592 Alastor
224 {P 5411 In Hemoriam 2nd 824 Lines
225 T 5512 " 1st 8: 7th 412 Lines
226 W 5525 The Prelude 3001: VII
227 VI 5659 " Book VI
228 (P 5727 Guinevere
229 {P 589 Lancelot and Elaine 1st Half
250 T 5919 " 2nd Half
251 I( 6153 The Cap and Bells
252 h! 615 Tour of the Alps
255 IT 6615 Aylmer's Field
234 '1 7175 Peter Bell
255 1? 7229 Enoch Arden
256 CD 7536 The Holy Grail
257 I3 7951 Bishop Blougram's Apology

 

* Text #51, Isabella, suffers from defective computer print:
out. It was retained anyway, as a control.
* Eight texts were temporarily misplaced when the charts were
being made.

54

Q? In the listing on the preceding pages, the column
headed Division has arabic numbers when the division is
mine; roman numerals indicate the divisions I found in
the poems. The original project planned was to make much
of comparing within poems to see whether, for example,
similarities were greater between the first and second
halves of a poem than between its odd and even lines.

Such divisions are magnificently easy to accomplish with
IBM cards, as are the subsequent combinations of glossaries.
The computer, just like anybody else, is able to alphabet-
ize fifty vords much more than twice as fast as one hundred.
Dividing the poems, therefore, not only gave greater flexi-
bility to the project, but it also saved computer time.

The poems on this list were, for the most part, key-
punched from the following editions:

Poems of Robert Browning, ed. Donald Shallcy, Houshton
Hifflin, 1956.

 

"u

 

ete Poems pf heats and Shelley, Kodern Library,
86, no do

 

Poems f Tennyson, ed. Jerome 3. Buckley, Xoujhton

The Poetical Works 9: Alfred, Lord Tennyspn, Kiss and
hnijht, Troy, I. Y., 1a37.

 

Poetical Works of Wordsworth, ed. Thomas Hutchinson,

Oxford University Press, 1904 ( 960).

 

 

APPEIDIK B

The Thirteen Texts

From Which were Derived the Forty-five Test Words

 

 

 

Eumber
Text Division of Words
Endymion 2nd Half of Part II, 1st Half of Part III 7855
Geraint and Enid 7950
Bishop Bloughram's Apology 7951
The Cenci Acts I and V 3158
The Prelude Books X and XI 7897
Hellas 7778
Merlin and Vivian 7927
The Excursion 2nd Half of Book VIII, Book X 7995
0tho the Great Acts III, IV, V 3025
The Pope (The Ring and the Book) 2nd Half 8191
The Princess Parts I, II, III 8554
In Memoriam 1st 1256 Lines 3174
In Hemoria“ 2nd 1256 Lines 3085

Note the discrepancy in word population between the last
two items. It often happens that the same number of verses
will contain widely different numbers of words. A statistician
would consider the word to be the unit, but the poet would more
likely think of the single line of verse as the unit. Lines of
regular verse, moreover, are more ”like" one another than are
individual words, which range from the zero-like content of an
initial "it" or "there" to the most connotative abstract or
concrete terms.

The project I finally determined upon ignores both measures
of size by touching only non-content words and discarding all
others. Herein is some of my justification for not distinguish-

ing text size during the main portion of the project.

55

for
from
have
he w
if
in
is

it
let
like
may
more
no
not
now
of
on
one
or

shall

so
some
that
the
then
ther
thes
to

u
upon
what
when

.n ' 1
..hi co.

‘0..-

1
O-v-J

V. 1 O
Ui'h

‘ e

H .d

c!" (i

g,

e

{.3
v

n‘jhjfi"~\ xv

, f"
£Lt & QAIJJ as U

Allonorphs

C1D.

 

cannot, cannt (can't), could

2 A 4— 1
co, done, dost, dotn

o
_-

had, hadst, hath, nas, hast, hating

into

an, are, art, be, been, being, isit, 'tis,

'twere, was, wast, were, wert
iSit ’ 'tis , ' t‘I‘i'aS , 't‘i‘u’ere

likes

.. , - ',.3 .2- -_',.., fl:—
nayst, mloflb, nibhtst

cannot, cannt

shalt, should, shouldst

this

whom, whose
’ ° 4L.
J11“: would, wouldst

56

words with the Allomorphs

of Fourteen

'twas,

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 1 087654321099 WWWWGMHW 987654.321

sasuumhhwww,wmpmaa.gaaaa.a11 . o m
A]. r)
2 .u h/
new 1| m .20 I o 0%“ o W m, .m 6.“. .5 ram
00, 0 ® 0 1 ”a RVQXMVANV CV
My“ .1 r «a, o C ¢ W "and
.r. @ s o m Sm a so. 3%»
m. h o 0 5 «o nomfldmaoo
1.1 o b. o W on» no» hoods”:
mm W 25 23 35.
a a has
., 1 0.3
W. o h a o 9% 83m. .2: .54
1.7 m1, amp .5. 23%
we. EM .fi 8253393.»
% o ”A Inoo Huaowvdvuu
mum w 9 oi» can chum»
o._, N m: 650. I."
m. 1 s and sum as
a Qow as. .5 8m
ow ¢ 3 *9 on» 32.5
T . :71 announcing
mm .01; no»: can bung
om, 3 Amy .3009? «Hon
Ml mm 13.360 “35636
% M .5 Bean. .55
m 5 3050» Film
MW 0 L211 7; voHomdo. 3.3 .I
8 > e tt 5 nf e m r d 3.... a .

tm1 1.1.mmsmnwamm unwumm:me111mvmm:m1Wam

n 1mcmmm e.m m. o m 11 ms:

nu mwwu mm s e

A3 Ru '8 3 a: .35. ..8 29:6 :3 no 83.33

A Summation of the Chart of the Last Analysis

 

APP EIJDIX E

1 17/5 3/2 12/3 10/7 9/7 12/7 15/7 8/4 10/2 5/1
2 19/8 16/4 14/5 12/5 10/4 8/2 11/1 12/2 11/4 13/2
3 16/7"11/6 [EZZQZ 16/2 8/3 26/7 10/5 7/3 9/2 8/2
4 14/6 17/4 16/4 14/6 16/3 13/5 14/3 11/2 21/3 22/5
3 13/7 8/5 10/3 3/3 17/2 11/7 4/3 7/4 4/2 9/3
6 13/8 13/9 18/3 13/8 15/7 17/3 7/5 9/3 3/1 6/5
7 12/8 21/4 11/2 14/4 14/8 10/3 11/3 17/8 6/2 12/6
8 18/12 12/4 11/3 12/5 14/7 16/5 20/7 19/9 7/3 9/5
9 11/5 14/5 17/9 20/6 15/4 9/4 18/6 14/5 16/4 26/8
10 11/9 13/9 10/6 20/6 17/6 14/10 19/8 13/4 20/6 14/4

Each of the "improper fractions" above represents the score

'0

for a square consisting of 25 x 25 smaller squares. hithin most
squares there is a possibility of as many as 1,597 indices of
correSpondence or of 529 indices of non—correspondence. The
full chart is 250 by 250 squares or twelve by twelve feet.

The top fifure in each square above is the number of cri-
teria pertaining to the more than five hundred pairings of
graphs represented by each square, and the lower number indi-
cates the times the criteria were mistaken.

See Table 1, page 42, for a close View of one portion of
this chart.

is anything but typical of my results.

Table 1 represents only 1% of the total chart; it

APPENDIX F

Specimens from Different Stages of the Project

(1) Program cards, the first seven of 115 in the deck of
the word count program built by Ja m1es D. C11? r1: of the Psycho-
logy Department, Xichigan State University.

(2) Input cards for this project. Each card contains
one line of verse, every word verified and, in some instances,
res pelled according to a convention (see pa-ges 54-5). The
punching of the cards was done almost wholly in my heme, on a
machine graciously rented to me by IBM. Blank cards were pur-
chased from the Computer Center at Iichigan State University.

(5) Output sh.eet, to show how the computer completes its
part of the prOJ ect.

(4) Listin eet with th as 45 test words in various
possible forms.3

(5) Gathering sheet, with the 45 iteins totalled.

(6) Broken-line graph, in which the broken line does not
really mean anfirtiing, ut merely takes the eye from point to
point (however, the broken line is the profile of a text).

(7) Same graph with the points expanded to %" holes.
This graph, and 229 others, were the core of my project; com-
paring each one with all the others consumed the better part
of a full year. The same job done by computer (that is, what
turned out to be the effective part of it) would have taken
less than two weeks, including all preparation.

(8) One leaf from the 12' x 20' chart containing the in-
dices derived from comparing graphs (see pages 58-9).

59

‘ognn 3,Mo

.c-

' '

Kazsooo
1.
Iooououjao;j{ooloooouno

. 9 u s I nunuuuuuunuuazmn

f'

I'I

-Spec1nen°1

‘_‘__._—._...7-

anunanannnunxnaaoucuuagncnau

'.p‘

coN<<Tzaoooonoaddsoooooe3

QflflSSflflilflflflflfllnflfllfl2222305317!

-~c...“0.‘..........‘.““

. ”L.

! .000...°°"1.“hiihthitfi33§12120DIIIIIOIllllllnlnnlaaa-naaa-ao-~-—ww~u ‘~ ‘

_ '1".".Illtlil‘u.nn uuuuu - f
‘ L cgmvhv NA“

’ VVVV? { 5 canon
. L‘-..

- ‘ ounuuuouonaooooooou

I OIOOOIOIOOOGEOIOIIIIIIO ‘.n..nnun”"n'u.

' onennoououoncocoouueuo ..9..".9.._“..u

f 12:32:22232uuuuununanunnalaaaananuasnanua?::::::‘1“""'"“““““I“,"‘"‘,

H1IHIIH1‘H1IH

 

222222222222222222222222

g...“ ....

2222222222

- -.---.A..‘:

. -.. .M...~

'1~

i J a
01 .NUT'M(4)\, qu(|

 

”gtnjlnilH 1H1|H
-4....... .--...._....__._...... n---_.---o.A..“.’,”,,22222.22222.2.?.Z1“3
lg‘. 1 {fl} ’ i ‘ ‘ 1
h idpliifiiq” LFoj(aoL:FVIIei.wAw(lzoo
- t I I
“WPBTCQAV W2
. 1H 1).

:2 1 NEWER

  
 

IIOIOII000280000000600.000000000000.000000BOIOIOOIIIIIIII

so u ‘“ ca- ““
........ " f‘.”.“.‘"f‘:‘.

F94),

&__.

 

[32"firfimTCpD2933"

ngonqqpoouolonoo

 

r...,
t 4
H

r-

.H
1 ?

ICC 685669

 

9' 0":

LA 8.“

.--. n— «u..- ...u-—.-.. “rod... ‘--*.-.~-— 4—

mg COMPUTER LABORATDPV 'IDENTIFICATION CARD

. . l
22 2 O 3 O 7 I 2'0"12!!““I"IflflflflnflaflflaiflflI823}!!!II.”Q0“..0..I§I”QQSQIISQCClflflO‘IIUIIDNnnflfllnlfll

.-

 

7'5- ."specinen.2m ‘ . '

 

 

 

       
     

 

.
o ‘ i
.. - W" ”‘2': .1 "F'-
,: T I -.. f. I‘- L“- ' "" ’ - 3'
1 ~ ; . x ’4
' H ' '5 1L I I s-
:4? J,:r ,:--n 'f:T 1 ”4:: IT IDLJE
.‘. l‘.
{E 2 2 2 2 ‘ 2 L w 296;
I f f I If (I. '1
:2 I- ,;;- ,3r::~,;, ,, ,,,~--- -i---........aaaIInIIInIIIIIIIOIIIUIIIIII
, . 1': 1': I 'I T! 3 ' '
Va- # 4:2}? - £¥;J}~ .3._ TA I 7%
I ‘ V f I 2' J J ' 1" " -’ ‘ fii'f'i T; :~.-j :7 L ,_;‘ I O -..--.“ - -_ ~--__- 2.. ..
I I l I !' rj ," f 4 ‘h‘ _- “ .“~
L L‘ 2. LL 7. ...--.— __-,._,__...-- ---
r g. '1 .. --‘~.-~‘.- ‘ ‘-__V _'-_~_ up,” __.in ..-- -r--""""“’"'-'--_-VU '
3 I, ‘Q. _ ...... I I t. 3; -~ .; 5? - ‘“' “"' 5
_. -,,. ... -.. ; '4. 1: T f; C ,J': ’ ,1” "“I 2 I1 I
. .; . i i.-' —' '- i. '2 ' . ~ . . ...-,v !
'.. g , i. {.7 z.' ! J '1" p 4) / 0 _,_,,-__.,-----v ...4 .. - ,. -—»~ I
1 ' , r r ._L_ 4.1:.-- - 3,; — ~ 9... “ ‘ Mom. 5
" J. 1 , — ';'.:‘ “T; l; I 1 __ "I! i -‘J'- L " I JV , .. 2""\
1"- " -.. H " ‘ ' k -{ o'r’~£’1 “'5’,’
I r P :: u 2 * H ’1 J;[ dfltrv D. . 07 f '.r
. , _ i ._.'".i.“ - -_. ,- __, . *.' ". "---. _- ...i.-.‘ -r - -.1.......--_.._...- _ -12“--. -_._---..--.... _ ' .-. -._.-.._..---... .i ..-- -- .- ---
1 '4 ' 151%- VUILL; 23d Inlmu ' ' J
; ;- r
2 (AIL; L 4i L
. 7 '. ' '
I 3 .3 u [1me STATEMENT J Whom

.‘II IIII.IIIIII IIIIIIiIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII IIIIIII
I.."flDI'IflI“l3!”fl§.fl..'."3339fll.flCa“.'0'....9‘.'u-..fl..~.uu..inn Nuiflfl'fl I
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII,
fizzzz22I22221222;222iIIIIIIIIIIIIII22212121222III:IIIIIIIIIIIIIIIIIIIIIIII212222
:3, 3,333333 338331 IIII. III.IIIIIIIIIIIIIII3IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII
dIII IIIIIIIIIIIIIIII;IIIIaiIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII
Ibis sssssss; IIIL IssssssIIIIIsssIIIssssssssssssssssssIIIIIIIIIIIIIIIIssssssssssss I

| IBIII
IIIIIIIIIIIIIIIIIfiIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII I
39IIIIIi'IIIIII IIII"jIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIs

1:11”

I IpIIII
IInununsunInaalnnzsrlnaulnualuaaouunuddaccaunuv.Iusll-ucnunuau-nn. unnunnl
yucca mil.” - - --H , ,-... -- .. . .--... ,- .- .._.__.. !

 

SSI LSILIIOIIOIIIllI.IIIIIIItI3HH"II'lttlilSiltlililt3llllllitiIl53lti‘6 l
222IIHHHIHHHI‘HI'UH.711”HHHHHHHHHHUHHTHHHHH”HM

 

 

“w. a— _- -..-..... .--...-._,, i. _,J

611'

nnonrm; a
4:3 <o_omm axe axfin:

H. p .m. >To<m «. >mor~>z A. >o>~z
mo . 1 , , p H a
m. popgzmq o. >.x V. pawm a. prr
» I. . p .;a 1. . . H u
o. prozm . I, pa. »rozm -.. I p». brnHym . Hm. >rmo
» ‘.. 4 I,; p .:;.; H p
pa. brqxoco: p». »rz><m pm. >2 Ho. >znzcm
w Lr. , k . , . N.
.m ham. azacm Imo. icxum any. 13x4: has. zocrc
m p m ., p n
5 Age. zznax keg. frazzrmw. bop. smoz: hon. xmccnxq
P . ,;, p. y »
gag. <m>x be». «nbxzfizo gem. «mbxm goo. «m4
9 i; ‘ ..H. : . g 9
bag. <3: . .
p .
qaq>r zczzmm an zcxam u cu; I.
H.

 

 

 

 

 

 

4}“ l

3.... 2?

y/ 9’

ix,“ 4 --- M

5.95;...“ o ‘ ‘_'__

r9432" O ._
Wu ..-- m

 

h-"Au

 

 

... -nm‘m°-n

 

 

:
‘
1r. .- ’L1\au 14 _E;

u, n'
-C‘. -n‘ 1*,
.‘w “a

. ' 5:22... I wag-mi

{max ‘7’
...-ova“.th Cum-s “an...

 

6.13;..{J o ....
“342-44: 0 -.. ‘
I -

A M

Lflwflg.

‘/-7‘.. 97¢.

_ ...-

44;... 5’

AMmemu m mans—an I W“m‘
I
.4v¢5 gf’

—‘l-‘"..

 

 

'46-

,. ~— _:.'.I7.. 288'},

1:2,}? 2 __ ‘ {Er-3’. [’4' ' I
U w— - ‘-

ff’;‘€?~dh.~ '1‘, /
' .1“
MM V‘."

 

..— --~ -. u Jo—wmwusmwavnpmam
7 , {j
I
V . ' 5
. i f- : 4". ..D
"Bu-fl?‘°- i ' -'- A‘NI-‘J- W‘bv—afl"81~’-\}"‘ :«VJ—U—Jm :‘I-fi‘fi'ilrwlt' -5 i

ova-J” O

r45-m .w ‘rM‘W-trx' Amati 3* 3"!“

 

Maj/4”” 4311......“

Wfluwrv JWMbMNNMuM-pd “9 aI-rvw . w .— w “—_ r. ‘u‘n

 

a” 9

u: w' —‘v"‘_~ N h- h.

@{LECJ

‘J on WI. “WI.
0

.~.’".- 2 ~24

4. ...............,...-...-
j

[

 

. ,", ... fl
“m’v‘”. O (a: 71:33.:

...;-go‘ma'numa ~ ... IO? M I'M

m-.. 3

4.1.4.44 0 Aw- -- .2533 {x o
aézflf’ Q_

444 '2’": {’6 -:~‘-/?-_.°3.~.:... ”...”...
.dlqfi" é

 

 

(Maui ““.na

 

”Mm- ow W“

 

 

_——‘O‘

firm—Fw- 6'- 0.r-*& "4&4 .1594? 0

£344.20
figs}: 40 _. - 0
W 5’ ..

' ....

I .'

- rug.‘ «.9 m." s smgs‘wuu

65

 

 

Specimen 4

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

56““ nm- .__ ‘- - _ ._ -...§-P eA-Cimeg—é a-” I —- #—
a .11., .zo-ra... a 10+} ...-.....L
-. 5.1 .1: ‘_ __ _ ..-...me ,. _ ”8135i _ 7,33: n...
8f t G: «g
a l). .{ ,_.‘:7;_u.fiv_= mail-rm. svca‘mrJE‘11-Mtwuva- - “"zg-IK“ -‘““ " " '1 ‘ ——— 731—1. 7' '-
m3 ’ '7“ 33:1"; ‘1'
lei)? .1, v .....- ‘ .. - . ‘ - .. -- bi? . V L-
..‘333 ‘5: 43: l: ’~ _ A __ _- - 98“ ......“ , 2H1 ¢ 3% s. . . .,
Q ..3. -’
gi J, l +/ A _. ‘_‘ ‘A .. cut—:5. .- A-A-mvr‘-+.’._ fl—wm.d-umwu.l.;n
for 9‘ 5581‘ 8
£22533 2. £32m 3
have .2. w v v ~ haw: 1' w
1309 .234 ‘ hm? L
53 6 if #v (o W
- #111 - ’27 "f" ____. -,.__ - ...??rmm - 1‘: ti _ -.
if .24." arl+‘l+?-J/+.5'+-Lrov :73 2,“ +~f¢g +tl+5‘+?—;‘-i
, -3 7.. __~>::I+/ u _ fifi _ m“; "‘5’ _- A A A w M :51} +1
19% O 3. :3 0
like: 3 ' like 8'
223‘ I 0 3 v :7 may ' 3 *7
m: 4. £939 9
m ~ 7 v . ... ‘ 2:13 ‘ ‘— ‘ “ A— L, ”Mm-aw“
. \ \
39% IO 1+7 5‘5" I 4")
ma j 510‘: I
ma ...“: V - ._ (31’ i..." W {7'
__ {an :L _ -.....5223- .. -3. .. m... -.....
“Tone ‘— 3116 5'
c3: 3 ’7’? 3 _
811871 ,7, ‘- 84472‘11 -..... - 2—
39 . 6 «‘59 L
M__ 63129 5" - A éfi'fi? -.. ‘ M5: _ -_ “...;-..”
mag; so that; 3-0
$133 ‘lo ., .-___,. - W--- 3319. - __ __ -fi _. __- g" - - .4...
.3583 9, than ‘1
......311:=:?.9-_~.5: --- -- ”if” - ... . é . _ -- ......
‘5}3838 3- .HJ Whose 1+“;
—- :9 J 7 , - --——- g a ‘ fl.” ~ “W J~2mnwnuw “gm—”3:3!
m I 119 V
. ape-n / “P93 I m ,__-
what .2; - w ‘31!“ 7-
ch63} J t - v - M3, - 1" ..
“£1: GEE—"r WETLQh 5’
v;- ‘m g as .3 - 5:125; --..--... - .3 +3
viii _3 14-9., mu. 1+ 2.
1.7133111 1‘ git-h - A I L__ ..-“-.- _..
yet, I 397: I M ._ “-
\ 40 I , \ ..." \
' ( s \ 'r " >~ A
'1 \ . -. ‘ , .
v. ’1' \ ; -- 3“»: J k“ a“. x. ‘ “
, g. = .2 ; I; - a) J

 

o
, 2X

    
       
     
      

    

..ul

2".
..L-.2.o

I}

 

   

 

 

 

- - H ;‘ w ..l
r
’_ )
._._. ‘-.._..- .
‘ fl
.- s v I ‘ { \
. . . ~ . n... 1" . . . --R\«'.r-.4-v~'-Y.'|n-~'Qu1'-"\ m- n;r —>~ .4. (M,
> ' ' '. I. .
a ‘ _ I I ‘
‘_ . ‘ - ,. . s . . .. . i
- ...... .. . -...- ' .. , ._ -.-.--
, , u ': ' u
2 I ‘ Q ‘
I '<:a-~- ~.-.—.~v '. ' -'
s

 

           
          

\ l '
-uw--._+..__....-‘}
‘ 3

-.-..--~ .3."

o~--.-4--.n, --. .nr->t~-\:r..\-. ...
:

  

_:,.,

   

“...,- -: ~111v”..-...‘.'é.‘

    
  

 
     
  

we huh-«....» ,b' 1 . ;

      

  
  
 
   
         
      
       

         

I I , ’

.. . -u-mt-Mp-n'u-‘uamu‘a- ..."

.‘ ' . - ' . .

g . f

,r.-~. ~s.‘ :--a- ..y ~~mg~p~um VM>'~¥_~-.I..¢.,‘-A-‘vv. . n . ...
. . z .

 

 

 

. s, ..
..-... +1.... a~.—........-W
, .

O
... --.._;....... .-r... -.. .. .-..

   

...

 

.
.... .. -.-;.._--. . ...-.- 5.....- 9 I -.....w
. , '5

 
 

s
"“'r ‘ Q ~..v4--a.-M- -- .J- .-
.

  
 
 

, 1
‘,
v'l! Fri-,vnu1v L no MA

 

7
~.. -..,
‘ w- ..-;-r ....um.

   

-
.

I . .- ‘x v

.' "h”“vsnv ~-..-._..;--,..

‘1 [ab-1Q .

 

SpeCimen_7

66

I|||

urn"
é-xc

Specimen 8

I
0 J,

a.
I

12, 9,2, 5

0-0

. ”4
3/;52’ ('15. -?_.’/0; I},

‘7)“2; 5, f

f2

67

 

APPEIDIX G

The Effectiveness of the Four Criteria

Criterion* Hiéges Ratio ‘xpectation Effectiveness
2 222/67** 3.3 .2 16.5
3 238/233 1.02 .2 5.1
II 184/168 1.1 .2 5.5
0 2&7/47 5.25 5.0 1.05

*For an explanation of the symbols 2, 3, II, and 0, see p. 40.

**The figures in this column are reasonably accurate. Whai
I made the count last year the fractions were: 222/6?, 236/233,
189/162, and 250/45.

Since the texts behi.d the graphs used in this project
came from five different poets, every fifth time two graphs
were compared the texts behind them were by the same poet. A
person guessing blindly the identity of the second poet in
each case could erpect to be correct one out of every five
times (more often with Wordsworth, less often with Keats).
Thus the expected ratio is one out of five (1/5), or .2.

Yet the positive criteria (2, 3, and II) guessed right
five to sixteen times as often as a blind guesser would. There-
fore the positive criteria can be said to be five to sixteen
times as effective as sheer guessing.

The negative criterion (0) is obviously worthless as an
indicator of identity since its effectiveness quotient is prac-
tically the same as the quotient for chance guessing.

The positive criterion (2) slows some promise. With
certain testing items removed --—I have no notion which ones --
and others added, and with a confining of the testing to texts
of more nearly equal size, the effectiveness quotient could no
doubt be raised considerably. Sixteen may be high enough for
some purnoses, as when only two possibilities are present in an
authorship disbute. One hundred would be better when more po-
tential authors are in the competition. But when almost anyone
could have written a doubtful text even a quotient of infinity
would not be a positive identification.

68

SOURCES CITED

Clark, John Williams. The Authorshin of Sir Gawain and the

 

Green Knight, Pearl, Cleanness, Patience, and Erkenwald in
the Light 2; the Vocabulary. University of Minnesota
doctoral dissertation, unpublished, 1941.

 

 

Cited from Clark:

 

Bateson, Hartley, ed. Patience. Second Edition. London,
1913.

Cha meers, R. W. "Long Will, Dante, and th e Pi_hteous
Heathen, Essays and Studies by Members 0 tile English

 

ASSOCiatiO_ I: 'C .(19 .24W “.....- "-"'

Chapman, Coolidge Otis. ”The Authorship of the Pearl, "

Publications of the Hoiern Lan1uage ssociation, KAIVII

W932), 346. "'"

------------, A le xical Concorde irce of the Iiddle Enflish
Pea rl, Cleanne ss, Patience, en:1 Sir "Gamayne and. the Grene
Knifht. A--, fi--, h0-- — Jy--, and Z-- only. Cornell
bniversity doctoral thesis. 1927.

 

 

 

 

 

Gollancz, Sir Is~ael, ed. Cl eanzess. Volume I, Introduction,
Text, and hotes. London, 1' 2 .

r

--------~---, and habel Day, edd. Cleanness. Volume II,
Glossary. London, 1935.

 

 

 

 

 

 

 

------------, ed. atijggg. London, 1913.

------------, ed. Pearl. Revised edition. London, 1921.

------------, ed. t Erkenwald. London, 1922.

norst1a“- Carl, e1. Altenrlische Legenden ( one Folgr).
i‘.CilDI‘U.IlL, 1“] o

Oakden, J. P. Alliterat ive Poetr3r in Ij_dile En"lish. Volume
I, The Dialect e ail Letric l Surv;y. Publications of the
University 01 huncheste1,3/ (19 3Q).

------------. ------ ----- - Voluue II, A Survs of the

 

.
{Ural-7i +1011. Publications of t‘he L 1i V01 Si CJ’ Oi “ENCiieS—‘L—ICT,
CU 5-LJ;‘JTI (1935) o

 

Savafie, Henry L., ed. t Erggnuald. Jew Haven, 1326.
Serjeantson, hary S. "The Dialects of the 1est midlands in

fiiddle Enflish,” gevieg of Englis11 Stufies, Ill (1927), 5;,
1fi6, 319.

 

70

Ten in., ,ernhard. istorv of EV: li: h literature

...-

 

... -..—-- ....- --.. ....--——

1.rolu eI. E33. tr., I. L. Lenne J. new York, 1;“?

)0

 

 

Tol“ien, J. R. R. and E. V. Gordon, edd. Sir Gawain and
the Egee_n grimgt. London, $25.

TrautIann, Korftz. Ueher Verteeser uni ITits teeuuCeneit
einiggr Alliterierender \O’ICJLC (1_es Alter 1129593.
Ialle, 1776

Conley, John. Review of I. I. Sa_vaCe's fie Cawain-Poet
Studies in his Person ali:z §_ud Rae'Croune. Chapel” hill,
1

1956. In—Sneculum, ”"(II ( 57), 853-61

 

 

Ellegard, Alvar. Who Was Junius° Stockholm, 1062.

' /

------------. A Statistical Method for Determining Author shin.
Stockholm, 1962.

 

Reviews of fillerard:

 

"The Statistics of Style,” Tigg§ Literary Sunnlement. Jan.
5. 1963. paée 1.

 

Zimmer, GeorCe W. Journal of ngH
1111 (1963). 6.8-0.

 

Herder}, Gustave. Tyne—‘1 Q‘rnn

-....-

metice il Linfuistics. 's-GreveheCe, 1060.

\J

\\\\\

1!

Kottler, Earnet, and Alan I Ta“3‘
‘W

Niles, Joseihine. "Eras in EnClishL Poetry, PrZJi
'tIe -Offiflfll Ian31u1Ce film Clcxriation1L .L{ (195 5),

 

Iilic, Louis I. A Cnentitative A“:POCCH to

 

 

 

 

 

 

 

 

 

Jonathan Swift. The La ue, 957.
------------. Dissertation Abstracts, III IV (1,61), 3730
Hosteller, Frederick, and David L Wallace. Infer e‘.1ce aid
Disnuted Authorshig; The Fe‘erelist. Readinf, Iass., 1’36
Taylor, Georfe C. ”IonteiCne-Shahes‘etre and the Deadl‘
Parallel," P‘ilolo; MCII aparterly, IIII (1943), 330-3.
Yule, GeorCe Ulry. The StTtIGELC“7 Str"r or IltCT““V fece‘u-
1613‘". Ca‘lbrififn, $131.1,

------------. ”On Sentence-lenCth as a ta tistical Chi racter-
stic of Style," xionetriha, III (193~), 563-90.

 

MICHIGAN STATE UNIV. LIBRnRIEs
‘ll||llllll“HIWIWWI”Nl1”IIWIHIHNNHI
31293010631491