Computers and the Humanities. Vol. 37

480 Pages·2016·3.01 MB·English

ComputersandtheHumanities 37: 1–2,2003. 1
TheACHPage
TEI Consortium Members Meet in Chicago
Oneconsequence ofbeinganincorporated legalentity,whetherornotfor-profit,is
thatyouhavetoholdannualmembers’meetings.TheTEIConsortium–thataugust
inheritor of the pioneering work in standardizing and expressing a coherent view
ofwhattextreallyisandhowitshouldbemanagedandrepresentedindigitalform
carriedoutoverthelastdecadeofthe20thcentury(partlyunderthesponsorshipof
theACH)–isnoexception.Itssecondannualmembers’meetingdulytookplacein
thepleasantsurroundings oftheNewberryLibrary,Chicagooverasunnyweekend
thislastOctober.
This year’s programme had a strong digital library theme full of controversy
and debate, with keynotes from Susan Hockey (an elder statesperson of the TEI,
and now a professor of library science at University College London) and from
John Price Wilkin (doyen developer of the digital library at the University of
Michigan), nicely complemented by thought-provoking contributions from Mark
Olsen (University of Chicago), Bill Kretzschmar (University of Georgia) and
Wendell Piez (Mulberry Technologies). All the presentations given are avail-
able from the TEI Members’ website; see http://www.tei-c.org/Members/2002-
Chicago/.
Attheconclusion ofitssecondyearofoperation asamembershipConsortium,
theTEIalreadyhassomething onwhichtocongratulate itself:thecomplete trans-
lation into XML of its Guidelines which was published in June 2002. Perhaps of
more long-term significance however is the fact that the technical work needed to
bringthatstandardfullyuptodatehasalsobegun.NewTEIworkgroupsreporting
at the Meeting include one focussing on character encoding issues, chaired by
ChristianWitternfromKyotoUniversity,andoneonstand-offmarkuptechniques,
chaired by David Durand of Ingenta PLC. The TEI editors Syd Bauman and Lou
Burnard also presented plans for moving the production of the Guidelines into
a new XML schema based environment. A newly formed work group on issues
relating to SGML to XML migration also met; its recommendations, due early
nextyear,willbeofconsiderable impact.
As a community-driven initiative, the TEI must focus on the interests of its
members as well as undertaking the necessary technical work to maintain the
standard. One interesting indication of how this community-focus may develop
was the notion of forming particular Special Interest Groups which might act as
advocacy groups, identifying training and support opportunities within particular
2 THEACHPAGE
areas, notably perhaps the digital library community. Another major talking point
throughoutthemeetingwasthechallengeofhowtobuildupmembership,atatime
when academic budgets are under increasing pressure. Members were realistic,
but optimistic: the TEI scheme has become part of the intellectual landscape. No
serious alternative has yet emerged as a solution to the problems addressed by
the Poughkeepsie Conference of 1987 which set up the TEI, though those same
problems are still being re-discovered. A major task for the new TEI is therefore
toexpanditsoutreachactivities, toensurethatitstrainingandsupportservicesare
widelyavailable andofahighprofessional standard.
With that in view, the meeting endorsed proposals to organize amajor training
event in the summer of 2003, in addition to next year’s inevitable annual meeting
in November. The investment made in the TEI by hundreds of organizations and
individuals world-widesuggests thatthemomentumwhichledtotheformationof
theTEIConsortiumwillcontinue torollitforwardastheonlywayofmaintaining
theTEIGuidelines.For,asBasilBuntingsaidofPound’s‘Cantos’,“theyresemble
theHimalayas:youcanignorethemifyoulike–butyouwillhavetogoanawfully
longwayround.”
LouBurnard
EuropeanEditoroftheTEIGuidelines
ComputersandtheHumanities 37: 3,2003. 3
Introduction: ACH/ALLC 2001 Proceedings
LORNAHUGHES1andJOHNLAVAGNINO2
1HumanitiesComputingGroup,NewYorkUniversity,USA
2CentreforComputingintheHumanities,King’sCollegeLondon,UK
We are delighted to present a selection of work presented at the ACH/ALLC
2001 conference (New York University, June 2001). Thanks to the contributions
ofmanydedicatedpeoplewhotookontheworkofdoingresearch,writingpapers,
reviewing papers, assembling the program, and making the endless arrangements
toprovidehousing,food,water,andInternetaccessforthedelegates,wewereable
to preserve our reputations as the Rebecca of Sunnybrook Farm and the Sarastro
(respectively) ofthehumanities-computing world.
Inoneoftheconference’shighlights, theRobertoBusaAwardwaspresentedto
ProfessorJohnF.BurrowsoftheUniversityofNewcastle,Australia.Thistriennial
award was instituted by the ACH and ALLC “to recognize outstanding achieve-
ment in the application of information technology in humanistic research”. The
citationforthepresentation oftheawardread:
The Association for Computers and the Humanities and the Association for
Literary and Linguistic Computing grant the Roberto Busa Award for 2001
to John F. Burrows, for exemplary contribution to scholarship in humanities
computing. Hisimaginativeapplicationofstatisticstotheliteratureofthe17th
tothe20thcenturieshasinspiredagenerationofcolleaguesandstudents.More
than anyone else he has bridged the gap between literary criticism and statis-
tics, enriching both areas and making the latter a part of mainstream literary
scholarship. In doing so he has helped to put humanities computing on solid
ground.
Professor Burrows’s Busa Award lecture is followed in this issue by a small
selection of other work presented at the conference. This year’s theme was
“Digital Media and Humanities Research”, and a comparison with past confer-
ence programs and proceedings shows how large a place questions of publication
have come to occupy in the field: where once the principal focus of research
was in scholarly analysis, today as much work goes into publication, and it is
almost routine for scholars to assume that they are addressing an unlimited audi-
ence around the world. No collection this small can represent the breadth even of
what went on at the conference, but we hope that this sample will serve as some
indication ofwherethefieldwasatthestartofthemillennium.

ComputersandtheHumanities 37: 5–32,2003.
5
©2003KluwerAcademicPublishers. PrintedintheNetherlands.
Questions of Authorship: Attribution and Beyond
A LectureDelivered on theOccasion oftheRobertoBusaAward
ACH-ALLC 2001,New York
JOHNBURROWS
CentreforLiteraryandLinguisticComputing,UniversityofNewcastle,Australia
Just here, in Washington Square, not long after the inception of this great univer-
sity, Henry James’s Catherine Sloper is last seen when, “picking up her morsel of
fancy work, she had seated herself with it again – for life, as it were”. By1901, a
coupleofgenerationslater,shemightwellhavefoundintellectual andimaginative
enrichment by enrolling as a woman student at NYU. By 2001, she might even
have joined ourranks. Hadshe taken up our sortof work(for life, asitwere), she
wouldhavefoundherself,aswedonow,amongfriendsfromallaroundtheworld.
In an occasional lecture of this kind, you are very much at risk. Given so free
a hand, I might move at a stately pace from acknowledgements to reminiscences,
and then onward to a wealth of geriatric platitudes. I shall try not to overdo the
reminiscences. Likeeverybodyelse,Ishallconfinemyplatitudes tothoseIoptim-
istically regard as the priceless wisdom of old age. I shall even try to show you
something new. But my debts, to begin with them, are many and profound. I am
most grateful to you all for this award. When Iconsider the wonderful things that
so many of you are doing, opening up far-reaching new avenues in humanities
research, I am astonished at being chosen as successor to Father Busa. I do feel
bound, of course, to accept your decision. After this news first reached us, one of
our daughters brought out an old cartoon from The New Yorker. It shows a man
reading hiswifealetterhehasjustreceived. Hesays, “ThisistheoneIhavebeen
waitingfor.ItisaletterofapprovalfromeveryoneIhaveeverknownforeverything
Ihaveeverdone”.IhaveyourletterandIthankyou.
In an era when we all rely increasingly on institutional support for our
researches, I have been generously treated by my university, by the Australian
ResearchCouncil, and(inthevitalearlystages) byStJohn’s College,Cambridge.
I have had excellent advice over the years from first-rate statisticians and
programmers.1 Myresearch-colleagues inNewcastleandelsewherehavegivenme
their support and encouragement for up to twenty years.2 Only my family have
surpassed theminpatience andenthusiasm.
When your choice falls upon an old man, you must expect some tales of those
far-off days when our public universities were still recognizable as universities;
whengender wasatermweusedingrammar;andwhenacomputerlesspowerful
6 JOHNBURROWS
than a modern laptop would fill a room. In those days – il y avait une fois – I
spent asabbatical leave inOxford working onJane Austen’s literary vocabulary. I
wascompiling alittleconcordance ofMansfield Park,focussing onwhatItookto
be a set of interesting words. My method was to read attentively, to underline my
chosenwords,andthentoentertheminanelaboratecard-index.Onmostmornings
oftheweek,wedroveouryoungestdaughtertoaday-nursery inBanburyRoad.A
little waydowntheroad, the OxfordUniversity Computing Centrewasalready in
business and Susan Hockey and Lou Burnard were soon to begin helping people
like meto use smart new mark-up systems like COCOA.But computers were not
yet on my horizon and I did not even know that the Centre existed. I was not to
benefitfromSusanandLou’shelpandgoodadviceforsomeyearsafterthat.
Another memorable episode occurred when I gave my first paper on the
outcomeofthesehand-counts. ItwasataconferenceattheUniversityofAdelaide
and the Chair of my session was an elderly Professor of English. As I walked
towardsourroomforthesession, Icameupbehindhimandoneofhisoldfriends.
Theypaused brieflyatourdoor.Thefriendglanced intotheroomandsaid,“Well,
Iwishyoujoyofthatone”.Thenhewentoffhappilytohearadifferentpaper.My
Chairman, however, rose handsomely to the occasion. After a few introductory
remarks, he said, “And now Dr Burrows will take us back to the roots of our
subject”. He then sat down beside me and slept peacefully for half an hour. As
I finished my presentation, he stirred himself and said, “Dr Burrows has taken us
backtotherootsofoursubject.Arethereanyquestions,orshallwebreakfortea?”
Morethanaquarterofacenturylater,Iamstilldiggingbusilyaway.Idohopethat
youwillnotallgotosleep.
My hand-counting of words taught me three things. The many words I had
singledoutfellintosuchrevealingpatternsthatitbecameclearthattheschemewas
wellworthpursuing and ought tobeenlarged. Itwasclear thatJane Austen wrote
soexactlythatshewasaperfectsubjectforthisapproach. Andthetaskbecameso
onerousthathand-counting wouldnotdo.So,in1979,webeganmakingmachine-
readable versions of Jane Austen’s novels. From the first, we had the full support
ofJohnLambert, thenDirectorofComputing Servicesinouruniversity. (Afterhe
retired in 1991, John joined my research-centre as its programmer. He continued
enthusiastically and effectively in that role until his death last month.) But, back
therein1979,thepreparatoryworkmovedonmorerapidlyinCambridge,whereSt
John’sCollegemademewelcomeforamemorableyearandwhereJohnDawson’s
generosity andmanyskillscameintoplay.
For what seemed a long time, it was all input and I despaired of intelligible
output.Inthosedays,however,scholarswerestillfreetopursuetheirideaswithout
being forced into premature publication. It was also feasible, at that time, to seek
research-funding forprojectsbasedmoreonwhatmightbedonethanonwhatone
hadalreadydoneandeventosucceedagainsttheopposition ofsomeassessors. At
thetimeofmyfirstapplication,IhadyettopublishinthisfieldandIwasfacedwith
anassessorwhobluntlydeclaredthatitwouldbeagrossmisuseofpublicfundsto
QUESTIONSOFAUTHORSHIP:ATTRIBUTIONANDBEYOND 7
supportwhatIproposed.Fortunatelytherewereotherassessorsandthecommittee
didnotagreewithhim.TheprojectwasfundedandoursacrilegiousassaultonJane
Austen went ahead atabetter pace. Ourable and good-hearted research-assistant,
AlexisAntonia,joinedtheteamandIemployed myfirstprogrammer.
Enoughofthesefondmemories.Itistimeformetotakealeapoftwentyyears
and examine some of the longer-term results. Ihave chosen, as my title indicates,
toconsiderthematterofauthorship,mychiefpreoccupationinrecentyears.3 After
glancing atthe present state of theart in this area of computer-assisted research, I
wouldliketodescribeapromisingnewapproachforwhichyouarethefirstpublic
audience. I shall turn, finally, to contemplate some possible developments for a
futureinwhichmanyofyouwillplaymorepartthanI.
Theideathattextsofdoubtfuloriginmightbeattributedtotheirtrueauthorsby
counting up the occurrences of salient features originated long before the advent
of the computer. But the computer has enabled us to pick out less conspicuous
features and to count them far more swiftly and accurately than our predecessors.
If only in the initial gathering of data, the heroic labors of scholars like Thomas
Mendenhall or G. Udney Yule can soon be surpassed by anyone with the ability
to run a user-friendly program on texts downloaded from an online archive. That
is not to say, of course, that our analyses of the results are always as judicious as
thoseofthepioneers.
Afteralivelyperiodofexperimentandcontroversyduringthe1970sand1980s,
whenM.W.A.Smithwasastalwart gatekeeper, wehaveentered aphaseofquiet
butworthwhileprogress intheareaofattribution. Wehavenotescaped thebattles
thatsoexcitethemediawheneveraputativespecimenofShakespearianaislocated.
Onmatterstobeconsideredlater,wehavenotyet,asJosephRudman(1998,2000,
p. 170) would wish, fixed upon a single method of analysis or identified a “veri-
fiably unique style”. Butour methods are increasingly reliable, our use of them is
evermorerigorous, andwehavevastnewcorpora tostrengthen ourcomparisons.
Most of the recent and current work focuses on phenomena that occur very
frequently rather than on those whose rarity is their hallmark. The most common
wordsofthelanguagehavebeengivenmoreattentionthaneverbefore,whetherin
studieswherethewordsareallowedtochoosethemselvesonnoothergroundthan
theirrelativefrequencyorelseinstudieswheredistinctions betweenlexicalwords
andfunction-words areattempted, theformerthen being discarded astoosubject-
specific. In a searching comparative study, Richard Forsyth and David Holmes
(1996)haveshownthatverycommonstringsofcharacters(oftenoverridingword-
boundaries) can yield somewhat more accurate results than the common words
themselves. The case for still maintaining an interest in the common words and
for allowing them to choose themselves rests upon two grounds. Such procedures
involve the least possible intrusion by the investigator and they offer the most
transparently intelligible results.
The precise form of a given authorial problem has a strong bearing on the
choiceofasuitablemethodofanalysis.Amongthemainanalytical toolscurrently
8 JOHNBURROWS
in use, artificial neural networks (Waugh, 2000) and discriminant analysis (Craig,
1999a)arebothattheirbestinclosedinquiries wheretheonlyquestioniswhether
Specimen X belongs to Set A or Set B. Although such problems do arise, this
degree of closure is usually confined to the last stages of a larger inquiry, when
most candidates have been eliminated. Neither of these approaches is as service-
able in more open games, especially because they lack the transparency so useful
for exploring the evidence. Cluster analysis (Craig, 2000) is preferable to either
of these methods for such exploratory purposes and can also yield corroborative
evidence.Itsdisadvantages arethatthedetailedevidenceremainsopaqueandthat,
forjustthisreason, tworathersimilarspecimenscanturnawayfromeachotherin
theearlyiterationsoftheprocessandendupmuchmorewidelyseparatedthanthey
should.Thelifeoftheliterarystatisticianisfullofsuchlittledisappointments. But
there ismuchtobesaidforusing methods that allowsuch outcomes tobestudied
andtheircauseseasilyunderstood.
It is partly for this reason that, as David Holmes (1998, p. 114) has said, prin-
cipalcomponentanalysis(pca)iscurrentlythefirstportofcallincomputer-assisted
studies of authorship. It has been put to good use, both in authorial and in quite
other studies, byagrowing number of scholars. See,for example, Baayen (1996),
Binongo (1995), Binongo and Smith (1999), Burrows and Craig (2001), Burrows
andLove(1999), Craig(1999b), Forsyth(1999), McKenna(1999), Tabata(1994),
Tweedie(1998).Someofthisworkdealssuccessfully withFrenchandLatintexts.
The fact that pca displays the phenomena most responsible for a given outcome
meansthattheevidenceismoretransparent thaninthemethodsmentioned above.
And, especially with the shrewd use of control-specimens evident in some recent
studies, pcacanyieldextremelyaccurate inferences.
In turning to its chief limitation and proposing a remedy, I do not wish to
displace pca but to complement it and so to consolidate it in the role for which
it is best fitted, in the middle stages of the game. The crucial point is that pca is
not intrinsically a test of authorship but only of comparative resemblance. This
offersusgreatversatility, butcancreateasubtletrap.Figure1,forexample,treats
of two sets of English Restoration verse, written by Shadwell and Tate respec-
tively. It also shows that entry X, representing a further specimen, lies closer to
Tate’s entries than to Shadwell’s. But entry X actually represents Absalom and
Achitophel, a poem that is undoubtedly the work of John Dryden. This does not
mean that the test has failed or even that the result is genuinely misleading. It
simplymeansthat,though XmaybemorelikeTatethanShadwell,theauthorship
of X is not adequately tested here. Properly stated, the original question here is
not “Who is the author of X?” but “Do the entries in this scatter-plot fall into any
intelligible pattern?” Since the two main sets of entries do fall into authorial sets,
furtherauthorialinferencesarenotinappropriate. Butmorestringenttestingwould
beneededbeforeanysuchinferencescouldbetakenseriously. Muchofourrecent
progress hasresteduponourincreasing reluctance toacceptfacileresults.
QUESTIONSOFAUTHORSHIP:ATTRIBUTIONANDBEYOND 9
Figure1. Shadwell, Tate,andatestpiece. Text-plotfor the99most common words ofthe
corpus.
A test like this would serve better if we knew in advance that X was the work
of either of a given pair of candidates and not of anybody else. But, as I have
said, such cases are so unusual that even a stringent test in the closed form “A
vs B” is seldom of much use except in the end-game. The authorship of a group
of three Restoration poems, variously assigned to Aphra Behn and Rochester, is
of this form (Burrows, 1995). The extent of Tate’s additions to Dryden’s text in
TheSecond Partof“Absalom and Achitophel” isasubtler example(Burrowsand
Love,1999,pp.169–174).Wherethereareseveralcandidates,successiveiterations
in the closed form can offer a ponderous way forward. The selection of particular
“markerwords”onthegroundthatagivenauthormakesmuchmore(ormuchless)
use of them than other authors of the time is helpful in the later stages of these
more complex analyses, allowing us to take full advantage of the latent power of
principal component analysis.
10 JOHNBURROWS
And yet there is a conflict between what we wish to do and what our tests
permit. Whereas our tests are best fitted for cases in the closed form, we would
wish totackle more open cases –including those cases where Ais aprincipal but
uncertain candidate withno single obvious rival and even those cases where there
is no recognized candidate at all. The clandestine political and erotic verse of the
Restoration and(asHaroldLovepointsoutinaforthcoming book)theintellectual
journalism of the nineteenth and early twentieth centuries abound in open cases
of both kinds. Until wecan make progress with problems like these, our role will
remain strictly ancillary to the traditional work of scholarship, corroborating or
casting doubt on the product of other sorts of evidence but rarely opening fresh
ground. We are still bound, it appears, by Richard Bailey’s dictum (1979, p. 7),
proposed over twenty years ago and lately put even more strictly by Binongo and
Smith (1999, p. 464). We should confine ourselves, that is to say, to cases where
thechoice lieswithinanarrow rangeofwell-matched setsandweshould proceed
withonlytwoauthors’textsatatime.
Againstthisbackground,Iwishtosketchanewpathforward.AtthetimeIwas
told of this award and invited to deliver this lecture, I was just beginning to write
anarticleabouttheworkIhavebeendoinginthelasttwoyearsorso.Itwastobe
entitled “Delta:ameasureofstylistic difference and aguide tolikely authorship”.
(The term “Delta” was chosen to represent D for difference and also as a gesture
ofrespect forUdneyYuleandthose otherpioneers inourfieldwhotriedtoderive
simple expressions of stylistic difference. Udney Yule’s Characteristic K remains
oneofthemostremarkable oftheseattempts.)Aversionofmyarticle willstillbe
necessarybecauseapubliclectureisnoplaceforthethoroughexpositionofanew
technique.Butitseemedappropriatetoacknowledgetheimmensecomplimentyou
havepaidmebypresenting youwithafreshcontribution toourprocedures.
Thefirststep istoestablish afrequency-hierarchy forthemostcommonwords
in a large group of suitable texts. The texts are grouped in subsets representing
the work of numerous authors of an appropriate era. With texts of a bygone era,
itis usual and desirable to standardize spelling and toexpand contracted forms of
expression inorder toreduce the influence oftrivial oraccidental variations. (Just
suchvariations werestudied bysomeofthepioneers ofstylometry. Butwhenone
works with common word-counts, they are merely a distortion.) It has also been
our practice, in Newcastle, to tag some of the more common homographic forms
inorder to distinguish the different uses ofwords like soand that.(The effects of
taggingarebeneficialbutthecostishigh,theintrusionuponthedataisregrettable,
and the interchange ofinformation withcolleagues ismade moredifficult.) When
the word-counts have been made, the frequencies are standardized as proportions
of each authorial subset so that the larger subsets do not exert an undue influence
onthecomposition orranking ofthehierarchy.
Working on these lines, we have formed a database of verse by twenty-five
poets of the English Restoration period.4 These yielded the frequency-hierarchies
wehaveusedforseveralrecentstudiesofauthorshipbasedonprincipalcomponent