T.Hassner,M.Rehbein,P.A.Stokes,andL.Wolf15 - PDF document

T.Hassner,M.Rehbein,P.A.Stokes,andL.Wolf15 Dicultiesincommunicationbetweenpalaeographersandcomputerscientistsisapre-vailingproblem.Thisisoftenreectednotonlyincomputerizedtoolsfailingtomeettherequirementsofpalaeographypractitionersbutalsointheterminologyusedbythetwodisciplines.Bettercommunicationshouldbefosteredbyjointeventsandlong-termcollaborations. Computerizedpalaeographictoolsareoftenblackboxeswhichputthepalaeographerononeendofthesystem,onlyreceivingasystemsoutput,withlittleopportunitytodirectlyinuencehowthesystemperformsortocommunicatewithitusingnaturalpalaeographicterminology.Thelong-termdesireistohavethescholaratthecenterofthecomputerizedsystem,allowinginteractionandfeedbackinordertobothne-tuneperformanceandbetterinterpretandcommunicateresults.Thisiscrucialifpalaeographyistobecomeatrulyevidencebaseddiscipline.Tothisendtheuseofhigh-levelterminology,naturaltopalaeography,shouldbeintegratedintocomputerizedpalaeographicsystems. Palaeographicdata,scarcetobeginwith,isevenmorerestrictedbyaccessibilityandindexingproblems,non-standardbenchmarkingtechniquesandthelackofaccuratemeta-dataandgroundtruthinformation.Multipleopportunitieswereidentiedforacquiringdataanddisseminatingitbothinthepalaeographicresearchcommunityandoutsidetothegeneralpublic. Palaeographicresearchislargelyrestrictedtothedomainofexperts.Makingpalaeographyaccessibletonon-expertsbyusingcomputerizedtoolshasbeenidentiedasaneectivemeansofdisseminatingvaluableculturalheritageinformationwhileatthesametimepotentiallygivingrisetocrowdsourcingopportunities,suchasthoseprovedsuccessfulinotherdomains.Inadditiontothesethemes,severalspecicrecommendationsregardingresearchinfra-structureandsupportweremade.Theseinclude:1.Acleararticulationofstandardsfordigitalimageacquisitionfollowedbyalldigitalimagingprojectswhenpossible.2.EU-wideharmonisationofcopyrightandlicensingpractices.Copyrightorcontractualuserestrictionsonphotographsofculturalheritageitemscreatemanybarriersforresearchers.Inmanycases,tax-fundedorstate-supportedresearchprojectsmustexpendsignicantnancialandhumanresourcesonnegotiatingandpayingforreproductionrights,evenifthoserightsarebeingobtainedfromstaterepositories.3.Ideally,setcopyrightappropriatelytoallowforlarge-scalestudiesofcollectionsofmanuscriptimages.Makinglargesetsofimagesmoreeasilyavailableataninternationalscalewouldgreatlyfacilitatethepursuitofsignicantnewresearchquestions.4.Encouraginganinterdisciplinaryresearchagendaincludingdisciplinesdealingwithcomputableimagesfromvariousperspectivessuchasmedicalimaging,cognitivesciences,CulturalHeritageImaging(CHI),orNaturalLanguageProcessing(NLP).Thismanifestoelaboratesontheexistingchallengesandlimitationsoftheeldanddetailsthelong-termrecommendationsthathaveemergedintheworkshop. 12382 T.Hassner,M.Rehbein,P.A.Stokes,andL.Wolf17 1IntroductionManuscriptsarethemostimportantwitnessestoandartefactsfromoursharedculturalheritageoftheEuropeanMiddleAges.Currentestimatesarethatclosetoonemillionmanuscriptbookssurvivealongwithcountlessarchivaldocumentsfromaperiodstretchingacrossmorethanamillennium.Cumulatively,thesedocumentsarethechiefsourcesofhistory,historyofscience,literature,andarthistory(duetothepresenceofmanuscriptdecoration)fromthatperiod.Moreover,thesemanuscriptsareimportantsubjectsofscienticenquiryintheirownright,astheybearwitnesstothehistoryofthebook,toscribalandmonasticculture,thehistoryofthedevelopmentofhandwritingsystems,languagesanddialects,thehistoryandgenealogyoftextsovertime,andtheevolutionofstrategiesfororganizingtextsandknowledge.Althoughoftentakenmorebroadly,palaeographyisinessencethestudyofoldhandwritingfrommanuscripts.Assuch,palaeographersareoftenaskedoneoffourquestionsregardingmanuscriptdocumentsfromthepast:whatwaswritten,whenwasthiswritten,wherewasitwrittenandbywhom.Answeringthesequestions,andindeedreadingthetextitself,arebasicprerequisitesforanykindofworkwithprimarysources,andthestudyofalmostalleldsrelevanttotheancientandmedievalpastthereforedependsonthem.Inthisrespectpalaeographyissometimesregardedasamereauxiliarydiscipline.However,palaeographyalsoextendsbeyondthis:itencompassesthehistoryofoneofhumanity'smostpervasivetechnologieswritingandthereforeraisesquestionsofculturalhistory,thedevelopmentandspreadofideas,andsoon,alongwiththedeepunderstandingofthetransmissionanduseoftextswhichitbrings.Misunderstandingsherecanleadtosignicanterrorsinscholarship,suchasbasinghistoricalargumentsoncharterswhichprovetobelateforgeries[47],orconductingstudiesofspellingandautomaticauthorshipattributionwithoutconsideringtheeectsoftextualtransmission,bothscribalandeditorial,andthechangesthatthisbrings[49].Palaeographyasadisciplinetypicallyinvolvesdicult,complex,andtime-consumingtasks,ofteninvolvingreferencetoavarietyoflinguisticandarchaeologicaldatasets,andtheinvocationofpreviousknowledgeofsimilardocumentarymaterial.Duetotheinvolvedreadingprocess,itisdiculttorecordhowthenalinterpretationofthedocumentwasreached,andwhichcompetinghypotheseswerepresented,adopted,ordiscardedintheprocess.Itisalsodiculttoacknowledgeandpresenttheprobabilitiesanduncertaintieswhichwerecalledontoresolveanalreadingofatext.Asaresult,palaeographicaldiscussiontendstowardsassertionsbasedonexperiencewithlittlesupportingevidencesometimesnoneatallandthishasledtoanallegedlyauthoritariandisciplinewhichdependsonfaith[10]ordogma[16]andisbasedoninformedguesswork[16].Itisperhapsnosurprisethatthedisciplineitselfsuersasaresult[10,3].Palaeographyasadisciplineis,however,ofhighrelevanceforsocietyandeconomy.Alloftheworld'swrittenheritagewaswrittenbyhanduntiltheinventionofprinting,andtextswrittenbyhandhaveremainedimportanteversince.Manuscriptsarehenceoneofthemajorsourcesofknowledgeofhumancultureandsociety,crossingthebordersofmodernnations,formostofwhatwecallhistory.However,unlikeprintedtextswhicharedistributedthroughlibraries,handwrittensourcesareoftenaccessibleonlytoaverysmallandhighlytrainedgroupofexperts,andhundredsofthousandsifnotmillionsofmanuscriptsarescatteredaroundtheworld.Theycanbediculttondanddiculttoread,areoftenwritteninanoldlanguage,andfrequentlydealwithasubjectmatterthatcanbeunderstoodonlybyexperts.Ontheotherhand,however,theycanbeavaluableresourcealsofor 12382 T.Hassner,M.Rehbein,P.A.Stokes,andL.Wolf19rightlyreluctanttoacceptitsresults[49,58,9,43].Themajorchallengeforcomputationalapproachesistoprovideasystemwhichpresentspalaeographicaldataquicklyandeasilyinawaywhichscholarscanunderstand,evaluate,andtrust.Thesuccessandimpactofresearchandinitiativesincomputationalmethodssofar([54]withpublications[40,15,13,30,5])hasshownthestrongneedtocombinescienticcomputingandpalaeographyinordertofurtherinvestigatetheinterdisciplinarymethodsandscienticelds.Itisalsoapparentthatnoinstitutionletaloneasinglescholariscapableofundertakingcomprehensiveresearchthatencompassesallthosemethods(andpotentiallymore).Thus,ajointeortisrequired,preferablyonaninternationallevel.Asbecameveryevidentduringscholarlymeetingsonthistopic[54,62,5],palaeographersandcomputerscientiststendtothinkindierenttermsandtendnottoagreeevenonverybasicnotionssuchasevidenceormeaning.Successfulcollaborationbetweenresearchersinhumanitiesandincomputerscienceisnotnearlyassimpleasdeneacomputationalproblemandndanalgorithmtosolveit.Theinputisoftenlooselydened,andtheoutputneedstobemorethanjustascoreonsomeabstractscale.Itisthereforecrucialtoidentifyacommonlevelatwhicheectivecommunicationcanbeestablished.2.2ChallengesDuringtheDagstuhlPerspectivesWorkshop,theunmediatedinteractionbetweenpalaeo-graphersandcomputerscientistsyieldedseveralveryclearquestionsandthemesforthefutureofresearchinDigitalPalaeography.Theseincludethefollowingfourchallenges:1.HowtooptimizecollaborationbetweenallthedierentdomainexpertsinvolvedinDigitalPalaeography.Barriersincommunicationbetweenpalaeographersandcomputerscientistsareaprevailingproblem.Thisisoftenreectednotonlyincomputerizedtoolsfailingtomeettherequirementsofpalaeographersbutalsointhedierentterminologiesusedbythetwodisciplines.Itwasrecommendedthatbettercommunicationshouldbefosteredbyjointeventsandlong-termcollaborations.2.Howtoensurethatpalaeographersremainincontroloftheirresearch,whilsttakingadvantageofthepossibilitiesofcomputerizedapproaches.Computerizedpalaeographictoolsareoftenblackboxesputtingpalaeographersononeendofthesystem,onlyreceivingasystem'soutput,withlittleopportunitytodirectlyinuencehowthesystemperforms,ortocommunicatewithitbyusingnaturalpalaeographicterminology.Thelong-termdesireistohavethescholaratthecentreofthecomputerizedsystem,allowinginteractionandfeedbackinorderbothtone-tuneperformanceandtointerpretandcommunicateresultsmoreeectively.Thisiscrucialifpalaeographyistobecomeatrulyevidence-baseddiscipline.Tothisendtheuseofhigh-levelterminology,naturaltopalaeography,shouldbeintegratedintocomputerizedpalaeographicsystems.3.Howtofacilitatesharing,notonlyofpalaeographicaldataandresults,butalsoofthemethodologiesinvolvedinpalaeographygenerally.Palaeographicdataisscarceandaccesstoitisrestrictedbycopyrightandindexingproblems,non-standardbenchmarkingtechniques,andthelackofaccuratemeta-dataandground-truthinformation.Duringtheworkshop,multipleopportunitieswereidentiedfortheacquisitionofdataandforitsdisseminationinthepalaeographicresearchcommunityandtothewiderpublic. 12382 20ComputationandPalaeography:PotentialsandLimits4.Howtousetheoutreachpotentialoeredbycomputerizedtechnologiestoenrichpalaeo-graphicalknowledge.Palaeographicresearchisanexpertdomain.Makingpalaeographyaccessibletonon-expertsbyusingcomputerizedtoolshasbeenidentiedasaneectivemeansofdissemin-atingvaluableculturalheritageinformationwhileatthesametimepotentiallygivingrisetootheropportunities,suchascrowdsourcingandotherswhichhaveprovedsuccessfulinotherdomains.2.3NeedsInthismanifestoweaddressboththetechnicalaspectsofthecollaborationbetweencom-puterscientistsandhumanistsaswellasconceptualtoolssuchasmidlevelfeaturesandontologies(discussedbelow)thatcanserveasmeansforeectivecommunicationamongpractitioners.Theemphasisofthisdiscussionisnotonthemostecientalgorithm,produ-cingthemostaccurateresults.Itisalsonotontheleastambiguousandmostmeaningfuldenitions.Instead,theemphasisisonthemosteectiveandfruitfulcommunication.2.3.1DataAcquisitionRepositoriesacrosstheEuropeanUnionhavebeenengagedinlarge-scaledigitizationeortsinrecentyears,resultingincollectionsofhundredsofthousandsorevenmillionsofdigitalimagesofmanuscriptbooksandmaterials.DigitalPalaeographyreliesontheexistenceofthesedigitalsurrogatesofmanuscripts.Moreover,someofthemostexcitingprospectsofthiseldcanonlybedemonstratedonsucientlylargecollections.However,enablingthisrstrequiresmodicationofbothpoliciesandacquisitionpractices.Specically,fromthecomputeruser'sperspective,obtainingdigitalcopiescallsforsuitableproceduresandforstandardization.Recently,Shwekaetal.havesuggestedspecicpracticesdrawingontheirexperienceinlarge-scaledigitization[45].Thesesuggestionsrangefromminimumresolution,totheusageofparticularrulersandbackground,andalsoincludesuggestedpoliciesregardingavailabilityandmanipulabilityduringviewing.Itisemphasizedthattakingintoaccountthepotentialusageofacomputersystemtoanalyzetheimagedoesnotdegradetheexperienceofthehumanviewer.Forexample,whileimageanalysisismucheasieronablue/greenbackground,forhumanviewingpurposes,suchabackgroundcanbeeasilyreplaced.Weproposethefollowing:1.Acleararticulationofstandardsfordigitalimageacquisitionfollowedbyalldigitalimagingprojectswhenpossible.Wheresuchstandardsalreadyexists(e.g.DFG-Praxisregeln`Digitalisierung'oftheGermanResearchCouncilDFG[17]orJISCGuidelinesintheUK[23]),theyshouldbecheckedagainsttherequirementsofpalaeographers(seealso[56])and,ifnecessary,extendedtoencompassandmeettheminfull.Theseincludepracticessuchas: Properuseofcolourbarsandgreycards. Appropriateuseanddocumentationofilluminationandequipment(e.g.lightingparametersincludingpositioning,hardware). Referencestosizeoforiginalobjectsusingsharedstandards. MetadatadescriptionsofdigitizedobjectsfollowinginternationallyacceptedstandardssuchasMIX/METS;ifonetakesseveralimagesofthesameobject(e.g.dierentlighting,multiplesizes,multispectral),itisimportantthatthecorrespondingmetadata 22ComputationandPalaeography:PotentialsandLimitstools,wherepalaeographerscancontinuallyintervene,inform,correct,understand,use,andreuseresultsproducedbyandprocessesimplementedbythesetools.Onlyinthismannerwillpalaeographybenetoptimallyfromtherespectivestrengthsofthehumanandcomputationalagents.Ideally,developingsuchsemi-automaticandinteractivetoolswillstimulatetheestablishmentofamutuallybenecialcontinuousfeedbackloopbetweenhumanandmachine,wherebyhumanswillbeinvolvedatalllevelsofreasoning,machineswillbeabletolearnfromhumaninput,andpalaeographersandotherswilllearnandcreatenewknowledgemoreeectivelythroughtheuseofmachines[22,25].Werecognisethatacriticalmassofdataisrequiredforperformingresearch,andthepre-attentiveperceptionofthedatabyresearchersisamajorfactorinbuildingnewhypotheses.Thiscriticalmassofdatacan,ontheotherhand,onlybeobtainedthroughusableandergonomictools.Hence,intooldevelopmentforDigitalPalaeography,focusgroups,usertestingandproperuserinterfacedesignisneededinconsultationwithhumanitiesscholarsasend-users(forwhichseealso[24]and[26]).Afurtherrequirementthatemergesfromhereistherecognitionoftool-developmentasacademicresearchtoencourageDigitalHumanitiesscholarstopublishtheirworkandmakeitusablebyabroaderaudience.Inthefollowing,weoutlinethespeciclevelsatwhichhelpfulcomputationaltoolscanbedevelopedaswellaspossiblewaysofkeepingthehumansintheloop.Alltoolsdevelopedshouldbecompatiblewithoneanotherandcombinableatwill(or,moreprecisely,aslongasthenotionsinvolvedarecompatible,thetoolsshouldbe).Theymightbeusedsequentially,orcontributetooneanother.Wehaveidentiedthefollowingcategoriesofmulti-levelcomputationaltoolsforDigitalPalaeography:1.Low-leveltools: Binarization Segmentation Alignment,matchingandregistrationoffeatures(forsimilaritymeasures)includingexpertfeaturesofhandwritingextraction(e.g.angles,curvatures,strokes) Physicalfeatureextraction Similaritymeasures(forcomparisonbetweencharacters,words,texts,fragments,documents,corpora)2.Mid-leveltools: Clustering Classication Characterrecognition Wordspotting Cross-modalitysearchengines,wheretheinputforthesearchesmightnotbeinthesameformasthedatasetthatissearched,e.g.: Searchforastringinatext/corpus Searchforanimageinatext/corpus Searchforastringinanimage/asetofimages Searchforanimageinanimage/asetofimages Searchforashape(shapewouldherebeahand-drawninpute.g.SVG,asopposedtoanimagethatwouldbeinarasterisedformat) Image-text(shape-meaning)correlation3.Databases,wherethedataisorganisedinawaythatallowsfastqueriesof(forexample): Metadata Transcripts Images 24ComputationandPalaeography:PotentialsandLimits3.1.2AccesstoDataThesecondobservationisthat,beyondcontextandmeaning,currentcomputationalcon-straintsareusuallyrelatedeithertoaccesstodataortodataretrieval.Theproblemofaccesstodataislargelypoliticalinnature,thereforedieringbetweencountriesandbylawsinvolved(e.g.intheUK,accessanduseofimagescanbedrasticallyrestrictedduetocopyrightandlicensingissues:seeDataAcquisitionabove).Incaseofdataretrieval,itislargelythedegreeofexibilityofthesearchtoolsthatlimitstheirusability,theirusefulnessandtherebytheiruse.Searchtoolsoftenpresentdicultiesofprecisionandrecall,andthisisusuallyduetoparametersbeingeithertooinexibleor,paradoxically,tooexible.Thismismatchbetweentheexibilityneededandtheexibilityprovidedbythesearchtoolsisinfactagoodillustrationofwhatwehaveidentiedasthemajorbottleneckinthecollaborationbetweencomputationalandpalaeographicalresearch,andthatbottleneckonceagainisdowntocommunication.Itmightseematrstthatproblemsincommunicationareeasytosolve,andthatitisjustamatteroflisteningandunderstanding,amatterofironingoutdierences.However,eveninourgroupoftwentypeopleatDagstuhlfromdierentbackgrounds,whereallwereaccustomedtocollaborativescholarship,astrikingrecurringdicultyinunderstandingeachotherwasapparentatraitthatwouldmostdenitelybeaccentuatedinalargergroupandinagroupwherecross-disciplinarycommunicationisnotacurrentpractice.Someofthesymptomsofthisproblemareenumeratedbelow,aswellassomeexamples,andpossiblemeasurestotreatthem.Unfortunately,thesehintsforsolutionswillonlytreatthesymptoms;therootsoftheproblemrunmuchdeeperandaremostlycultural,originatinginthetraditionsofeachandeverydiscipline.3.1.3InterdisciplinaryApproachestoResearchScholarlyendeavoursareallrootedintheirowntraditions.Inspiteofournon-subscriptiontoSnow'sTwoCulturestheory[46]thenextsectiononterminologywillillustratehowthebusinessofcreatingknowledgeisaMany-CulturessystemratherthanaTwo-Culturessystemoneofthehigh-levelobservationsaboutresearchmethodologiesinpalaeographyandincomputerscienceisthattheydierwidely.Asnotedabove,computerscientiststendtobeproblemsolvers.Theirapproachtoknowledgecreationistypicallytobreakdownalargetaskintosmallertasksandthentosolvethesetasks,iteratively,untilasatisfactorysolutionoftheinitiallargetaskisfound(wheresatisfactoryisoftenlefttotheirowndiscretion).Inthetraditionofcomputersciences,thereisafurtherconventionofnotderivingnaturalinterpretationfromthemethodology.Inotherwords,theoutputneedsadditionalcognitiveprocessingtobeinterpreted,andcomputersciencesdonottraditionallyhavewaysofdoingso.Incontrast,palaeographerstendtoapproachknowledgecreationinadierentway.Theirmethodistypicallytoderivequestionsfromquestions,whereanewquestionoftenhasthevalueofansweringtheprecedingquestion(seefurtherExploringandQuestioning,notAnswering,below).Communicatingbetweenthesetwoapproachescanevokesituationsinwhichwhatmayrstseemtobeamisunderstandingormisconceptionturnsout,nally,toreleasesynergies.Take,forinstance,aquestioninpalaeographyforwhichacorrespondingcomputationalsolutionhasbeendeveloped.Thediscussionbetweenpalaeographersandcomputerscientistsmightleadtoanemphasisontheweaknessesorincompletenessoftheproposedcomputationalsolution.Butitmightalsorevealtheneedtoreformulatetheoriginalquestion,ormightopenupthepotentialfornew,relatedquestions.Inanycase,thisinterdisciplinarycommunication 26ComputationandPalaeography:PotentialsandLimitsprinciplesandassumptionsbehindtheinnerworkingoftheblackboxesandnotofthedetailsofthemethodsandtheirimplementation.Establishingthattrustwillalleviatetheanxietiesthatblackboxestendtogenerate;itwilltherebyeasecommunicationandcollaboration.Two(non-exclusive)naturalsolutionstosuchbottlenecksandlackoftrustcanbesummarizedastheintroductionofanin-betweenerandcommunicationofmid-levelfeatures;bothofthesearediscussedfurtherbelow.3.2DirectionsItisworthnotingthatthetechnicallimitationsoutlinedabovearenotreviewedinmoredetailherebecause,inthelightofthepotentialproblemsincommunicationalreadydiscussed,theyseemlargelysurmountable.Infact,throughthediscussions,roundtables,andQ&A-sessionsduringtheDagstuhlworkshop,itoftenemergedthatcomputationalapproachesoeralotmorepossibilitiesthansingleexpertsmighthavepredicted.Asaresult,anyprognosisoftechnicallimitationsvoicedherewouldcarrytheinherentriskofoutliningpre-emptivedelimitations.3.2.1InterdisciplinarityandtheIn-BetweenerTheDagstuhlworkshopcanserveasabest-practiceortemplateforfutureinterdisciplinarycommunication.Furtherjointsessionsatconferencesandsimilareventsneedtobeheld.ButcommunicationbetweencomputerscientistsontheonehandandHumanitiesscholarsontheotherisonlyastartingpoint.Interdisciplinaryprojectsbetweentheeldsneedtobestrengthened,andallparticipatingdisciplineswilldrawsignicantbenetsfromthem.ExpertsinscienticcomputingshouldnotmerelyimplementrequirementsformulatedbytheHumanities,butshouldalsosuggestideasbasedontheirexcellenceandexpertise.Atthesametime,scholarsinComputerScienceshouldacknowledgetherelevanceofresearchquestionsandmethodsfromtheHumanities.Althoughthedisciplineshavedierentsemioticsandseparateproofsystems,interdisciplinarycommunicationandcooperationleadstobetterunderstandingandconsequentlytonewknowledge.Interdisciplinaryworkshopsareinvaluable,butalsonecessaryistheinterdisciplinaryindividual:thein-betweenerintroducedabove.Thisisamiddle-person,atranslator:apersonwhoisversedenoughineachofthecollaboratingeldstounderstandenoughofeachofthediscipline-speciclexicaleldstofostergoodcommunicationandfruitfulexchanges.Dedicatedspecializedin-betweenershavealreadybeenusedverysuccessfullyinsomeDigitalHumanitiescontexts,suchasthepositionsofprojectanalystsattheDepartmentofDigitalHumanitiesinKing'sCollegeLondon,andtheirapplicationtopalaeographyistobeencouraged.3.2.2Communication,IntelligibilityandtheBlackBox:Evidence-basedPalaeographyGiventhetaskofclassifyingawrittenfragment,anauthoritativepalaeographermightexaminethepageandsimplystatehisorherclassicationofit,typicallyprovidinglittleevidenceforhowthisconclusionwasreached[16,10,9].Somewhatanalogously,givenanimageofafragment,acomputerizedsystemmightoutputtheclassofscriptforwhichthefragmentscoredthehighest,alongwiththescoreitself.Themathematicalproceduresandformulasthatledtothisconclusionwouldremaininaccessibleinsidetheblackboxes.Both 28ComputationandPalaeography:PotentialsandLimitseortshavemetwithgreatchallengesandwerenotabletoachievetheirgoals.Wesuggestembracingthedierencesinterminologiesandthecomplexrelationsthatexistbetweenthemandfocussinginsteadondevelopinganontology.Anontology,inthisunderstanding,isarepresentationofaknowledgedomainwhichisbaseduponwelldenedentities,eachhavingauniquemeaning.Variousstructurallinksarethenusedtodenerelations,suchassubstof(alsoknownasisa),relatedto,andsoon.Eachtermcanalsocontainalistofsynonymsandtranslations,adenition,referencestootherterminologies,andremarks.Insteadoftraditionalclassicationsystems,ontologiesarebeingusedmoreandmorewidelyinHumanitiesscholarship,especiallyinculturalheritagedocumentation,becauseofthemuchgreaterexibilitythattheyallow.ExamplesofontologiesthatarewidelyusedintheHumanitiesincludetheEDMmodeloftheEuropeanalibraryofdigitalobjects[14];andtheCIDOCConceptualReferenceModel[20],whichhasbecomeaninternationalstandard[21].Anexampleofpartoftheenvisionedontologyforpalaeographycouldbe:[TERM]id:PCO0000345name:triangularascenderdef:triangulardecorationatthetipofanascendersynonym:wedgedascenderEXACTrelatedto:is_a:PCO0000221ascenderdecorationThetopleveloftheontologycouldcontainthetermsManuscriptStudiesandunderneathPalaeographyandCodicology.Thelattercouldbebasedonaneortrecentlyenvisioned([28],andcomparealso[31]).Wenotethattheeldofpalaeographyismuchlessorderedthancodicology,andthiscanleadtochallengesinrepresentingit.Nevertheless,thetopdistinctionsunderpalaeographycouldbeAllographsandGraphemes,orsomethingsimilar.Someofthetermswouldbedescendantsoftermsfromboththesebranches,e.g.,Carolinea(cf.[51]).Ontologiesallowforuniedtreatmentofmetadataassociatedwithdocumentsaswellastominingofsuchresources.Forexample,projectsliketheMedievalElectronicScholarlyAlliance[34]andtheManuscriptsOnlineproject[32]aimtoprovidefederatedsearcheswhichspanmultipleresources.However,eachresourcemightuseadierentnameforexactlythesameterm,orthesamenamefordierentthings.Forexample,EnglishvernacularscriptoftheeleventhcenturyhasbeenlabelledCarolineminuscule[27],Anglo-SaxonRoundminuscule[6],Anglo-SaxonVernacularminuscule[12],andEnglishVernacularminuscule[52].Givenanontology,itisfairlystraightforwardtocomeupwithreasonablemethodstoexpandthesearchtoincludealltheseterms,andthenrankthecombinedresultstogether,andthisisanapproachwhichthoseprojectswilluse,althoughitsapplicationinpracticeisfarfromtrivial.Inthepast,someprojectsonbuildingontologies(ingeneral)werelesssuccessfulthanothers.Oneoftheauthorsofthismanifesto(TH)hasparticipated,asastudent,intheconstructionofanontologyforrepresenting3Dshapes.Hedescribesaveryfrustratingprocessinwhichcommitteemembersdebatedmanyminutedetailsandwhichultimatelyledtoatoolthatnoresearcheruses.Incontrast,anotherauthor(LW)isanaviduseroftheHuman-Phenotype-Ontology[19],whichisusedbyclinicalgeneticiststodescribephenotypes,manyofwhicharevisual.Hehaswitnessedthepoweroftheontologyinfacilitatingthe T.Hassner,M.Rehbein,P.A.Stokes,andL.Wolf29mergingofdisperseterminologiesandtheextremelyusefuldataminingandclassicationtoolsthatthatithasentailed.Interestingly,thisontologyhasoriginatedfromacollectionofmedicaldatacalledOMIM,whichwaswrittenbythousandsofauthors,eachusingtheirownterminology[36].Basedondiscussingthesecaseswecametotheconclusionthatontologiesshouldrelyatrstontheexpertiseofspecicauthoritativepalaeographersratherthanonthecommunityatlarge,andthattheymustbeaccompaniedbydatasetsandcomputationaltoolsthatemploytheontologyinquestion.3.2.4ExploringandQuestioning,notAnsweringItisincreasinglybeingrecognisedinrelatedeldsofDigitalHumanitiesthattheblackandwhiteresponseoftengivenbycomputationalmethodsisincompatiblewiththeapproachesandinterestsofHumanitiesscholars.Furthermore,asalreadydiscussedhere,itisverydiculttomovefromcomputationalresultstorealworldmeaning,butformostHumanitiesscholarsthisrealworldmeaningistheprincipaloronlypointofinterest.Itisthereforebecomingincreasinglyevidentthatpalaeographersprefertoharnesscomputationalmethodsnottoprovideanswerstoreal-worldquestions,butrathertomanagelargeamountsofdatainwaysthatallowthemtodrawtheirownconclusions.Furthermore,ithasbeenobservedthatcognitiveprocessesinacademicresearchcanbeenhancedthroughvisualisation,particularlywhenappliedtomaterialwhichisinherentlyvisualsuchaspalaeographyandmanuscriptstudies[22,50].SomeresearchquestionsofhistoricalcontenthaveresultswhichHumanitiesscholarscanverifywitharelativelyhighdegreeofcondence.Oneexampleisjoins,thatis,identifyingpagesorfragmentsofpagesfromnowdismemberedbooks.Forproblemslikethese,com-putationalmethodscanusefullyproposereal-worldanswers,forexamplebyprovidingasetofimagesofpageswhicharelikelytobefromthesamebook,andwhichthescholarcanthencheck.Incontrast,otherproblemsaredicultorimpossibletoverifyagainstthehistoricaltruth,andcomputationalmethodswhichattempttoanswerthesehavenotbeenacceptedbecauseofthisdicultyofverication.Here,Humanitiesscholarsneedtobeabletocross-examinetheresults,includingalsothemethodandtheassumptionswhichunderliethem[9,43]:iftheycannotverifythemthentheycannothaveanycondenceintheresults.Thisiscloselyrelatedtotheblackboxproblemdiscussedaboveand,asalreadynoted,itisasignicantchallengeforfuturework.However,analternativeapproachisrathertodevelopcomputationalmethodsthatallowresearcherstomanipulateandvisualisethecontentontheirownterms,andtocommunicatethisdataasevidencetoabroaderaudience.ScholarsinDigitalHumanitieshavereferredtothevirtueofautomatedanalysis...not[as]thereadydeliveryofobjectivetruth,butinsteadthemoreprofoundvirtueofbringingusupshort,ofdisturbingusinourpreconceptions[44].Therefore,onceacriticalmassofdatahasbeenreached,problemsofthissortshouldbetreatedwithcomputationalmethodsdesignedtoaiddiscovery,exchange,interpretation,andpresentationofknowledge,notprovidinganswerstohistoricalorotherrealworldquestions([22];cf.also[8]).Thishasimportantimplicationsforcollaborativeworkwithcomputerscientists,sinceitisfundamentallydierentfromthealgorithmstosolveproblemsapproachwhichismoretypicalofthelatter'seld(seeInterdisciplinaryApproachestoResearch,above). 12382 T.Hassner,M.Rehbein,P.A.Stokes,andL.Wolf31funds,andonlineresourcescanhelptoovercomethis.Indeed,thisincreasedaccessanddemocratisationisafrequentpromiseofDigitalHumanities,althoughithasnotnecessarilybeenfullledinpractice[41].Theintroductionnotonlyofdigitisedimagesbutalsoofcomputerizedtechniquesopensupnewwaysofsharingthisinformationwiththebroadpopulation.OneparticularlyeectiveexampleofthisistheWaltersArtMuseum,whosepolicyofreleasingdigitalimagesofmanuscriptsusingCreativeCommonslicensing,andofdistributingtheseimagesthrougharangeofsocialandothermedia,hasleaddirectlytoverywidepublicrecognitionoftheirholdings,somuchsothatasearchforkoraninGoogleImagesreturnsahighlydisproportionatenumberofresultsfromthatmuseumfarmorethanfrommuchlargerandbetter-knowninstitutionssuchastheBritishLibraryortheBibliothèquenationaledeFrance[35].Evenmoreexcitingistheprospectofpeopleconductingtheirownresearch,ortappingintonon-experttraditionsasawayofenrichingscholarlyknowledge.Althoughtheprocessofopeningupvirtualmanuscriptarchivestothepublichasalreadybegun,theseprojectsarestillintheirinfancy.Reachingout,collecting,andprocessingtheknowledgethatmaybeavailableinregionaltraditions,ontheotherhand,hasnotbeensucientlyexplored.Doingsobyusingcrowdsourcingtechniquesisanexcitingnewresearchdirectionandhasalreadybeenappliedtotranscriptionandidenticationofmanuscriptsandmusicalscores,amongothers(e.g.[57,60,1]).Inordertorealisethispotentialfullycertainrequirementsremain.AstheDigitalWaltersprojectclearlydemonstrates,onerequirementisagainthatofsucientlypermissivecopyrightandlicensingconditions:ifpeoplearenotallowedtouseimagesinwaysthattheywish,orifitisunclearwhethertheymaysousethemornot,thentheytypicallywillnotusethematall[35].Thematerialmustalsobefreenotonlyoflicensingrestrictionsbutalsooftechnicalones:again,iftheimagesareavailableonlyinproprietaryviewersorotherlimitingformatsthenaccesstothemdiminishesaccordingly.Furthermore,thedicultiesincommunicationwhichhavebeendiscussedbetweenpalaeographersandcomputerscientistsbecomeevenmorepronouncedwhenmovingbeyondtheprofessionalresearchertothewiderpublic.However,thesameprinciplesadvocatedhere,suchasmid-levelfeaturesandin-betweenerspecialists,arealsorelevanttothisbroaderchallenge.Theseprinciplesneedtobeextendedtootherareasbothofacademicbutalsoofpublicinterestsuchaslocalhistory,genealogy,arthistory,language(includingregionaldialects),namestudies,calligraphy,artsandcrafts,andsoon.Asresearchersareincreasinglypressuredtodemonstratetheimpactandvaluetosocietyoftheirwork,andastheydiscusshowbesttomeasureandachieveit[55],digitalpalaeographyisalreadyaddressingtheseconcernsandalsohasanidealscopeofstudywhichalreadyhasdemonstrablepublicinterest.Extendingtheseconcernsandcombiningthepre-existinginterestpresentsanoutstandingopportunityfortakingthisnewandrelativelymarginalisedeldofstudyandbringingittotheforefrontofpublicandacademicawareness. AcknowledgementsTheauthorswishtothankSchlossDagstuhlfortheirsupportofthisworkshop.PASalsothankstheEuropeanResearchCouncil(ERC):someoftheresultspresentedherearefundedbytheEuropeanUnionSeventhFrameworkProgramme(FP7)undergrantagreementno263751.ThistexthasreceivedsubstantialcontributionsfromallmembersoftheWorkshop,forwhichseeParticipantslistbelow. 12382 34ComputationandPalaeography:PotentialsandLimits26JISC.DevelopmentofPersonalisationfortheInformationEnvironment1(DPIE1),2008.Availableonline:http://www.jisc.ac.uk/whatwedo/programmes/resourcediscovery/dpie1.aspx.27NeilR.Ker.CatalogueofManuscriptscontainingAnglo-Saxon.ClarendonPress,Oxford,1957.28RobertKummer.Semantictechnologiesformanuscriptdescriptions-conceptsandvisions.InFischeretal.[15],pp.133154,2011.Availableonline:http://kups.ub.uni-koeln.de/4347/.29N.Levy,L.Wolf,N.Dershowitz,andP.A.Stokes.Estimatingthedistinctivenessofgraph-emesandallographsinpalaeographicclassication.InDH2012BookofAbstracts,ed.byJanChristophMeisteretal.,pp.264267.UniversityofHamburg,2012.Availableonline:http://www.dh2012.uni-hamburg.de/conference/programme/abstracts/estimating-the-distinctiveness-of-graphemes-and-allographs-in-palaeographic-classication/.30London:King'sCollege.DigiPal:DigitalDatabaseandResourceofPalaeography,Manu-scriptsandDiplomatic.Availableonline:http://www.digipal.eu/.31London:UniversityoftheArts.Ligatus:AnEnglish/GreekTerminologyfortheStructuresandMaterialsofByzantineandGreekBookbinding.Availableonline:http://www.ligatus.org.uk/glossary/.32ManuscriptsOnline.WrittenCulture1000to1500.Availableonline:http://www.manuscriptsonline.org/.33MaxPlanckInstitute.BestPracticesforAccesstoImages:RecommendationsforScholarlyUseandPublishing.Berlin,2009.Availableonline:http://www.mpiwg-berlin.mpg.de/PDF/MPIWGBestPracticesRecommendations.pdf.34MESA.MedievalElectronicScholarlyAlliance.Availableonline:http://mesa.performantsoftware.com/.35W.Noel.TheCommonsandDigitalHumanitiesinMuseums.CUNY,NewYork,2012.Availableonline:http://www.youtube.com/watch?v=XPJ_kciC15I.36OMIM.OnlineMendelianInheritanceinMan.Availableonline:http://www.ncbi.nlm.nih.gov/omim/.37Pareto.ParetoPrinciple.Wikipedia.Availableonline:http://en.wikipedia.org/wiki/Pareto_principle.38SusannahB.F.Paletz,ChristianD.Schunn,andKevinH.Kim.Theinterplayofconictandanalogyinmultidisciplinaryteams.Cognition,126(1):119,2013.doi:10.1016/j.cognition.2012.07.020.39E.PierazzoandJ.André.Autourd'uneséquenceetdesnotesduCahier46:enjeuducodagedanslesbrouillonsdeProustAroundaSequenceandsomeNotesofNotebook46:EncodingIssuesaboutProust'sDraftsKing'sCollege,London,2012.Availableonline:http://research.cch.kcl.ac.uk/proust_prototype/about.html.40MalteRehbein,PatrickSahle,andTorstenSchaÿan,editors.KodikologieundPaläographieimDigitalenZeitalter-CodicologyandPalaeographyintheDigitalAge.SchriftendesInstitutsfürDokumentologieundEditorik.BooksonDemand,Norderstedt,2009.Availableonline:http://kups.ub.uni-koeln.de/volltexte/2009/2939/.41M.Reisz.Surfdom.TimesHigherEducationSupplement,2028:3439,8December2011.Availableonline:http://www.timeshighereducation.co.uk/story.asp?sectioncode=26&storycode=418343&c=2.42RightsReserved.-FreeAccess.NutzungsrechtefürhistorischeQuellenundHandschriftenimInternet.Availableonline:http://www.infoclio.ch/de/node/26076.43LambertSchomaker.Advancesinwriteridenticationandverication.Proc.of9thInt.Conf.onDocumentAnalysisandRecognition(ICDAR2007),26September,Curit-iba,Brazil,2:769773,2007. T.Hassner,M.Rehbein,P.A.Stokes,andL.Wolf3544D.SculleyandBradleyM.Pasanek.Meaningandmining:Theimpactofimplicitassump-tionsindataminingforthehumanities.LiteraryandLinguisticComputing,23(4):409424,2008.doi:10.1093/llc/fqn019.45Y.Choueka,L.Wolf,R.Shweka,andN.Dershowitz.Automaticextractionofcatalogdatafromdigitalimagesofhistoricalmanuscripts.Forthcoming.46CharlesPercySnow.TheTwoCultures.CambridgeUniversityPress,2012.47F.M.Stenton.ThesupremacyoftheMerciankings.EnglishHistoricalReview,33(132):433452,1918.doi:10.1093/ehr/XXXIII.CXXXII.433.48PeterA.Stokes.Palaeographyandimageprocessing:Somesolutionsandproblems.Di-gitalMedievalist,3,2007/8.Availableonline:http://www.digitalmedievalist.org/journal/3/stokes/.49PeterA.Stokes.Computer-aidedpalaeography:Presentandfuture.InRehbeinetal.[40],pp.309338,2009.Availableonline:http://kups.ub.uni-koeln.de/volltexte/2009/2978/pdf/KPDZ_I_Stokes.pdf.50PeterA.Stokes.Palaeographyandthevirtuallibraryofmanuscripts.InB.NelsonandM.Terras,DigitisingMedievalandEarlyModernCulture,pp.137169.MedievalandRenaissanceTextsandStudies,426.ArizonaCenterforMedievalandRenaissanceStudies,TempeAZ,2012.51PeterA.Stokes.Modelingmedievalhandwriting:Anewapproachtodigitalpa-laeography.InDH2012BookofAbstracts,ed.byJanChristophMeisteretal.,pp.382385.UniversityofHamburg,2012.Availableonline:http://www.dh2012.uni-hamburg.de/conference/programme/abstracts/modeling-medieval-handwriting-a-new-approach-to-digital-palaeography.52PeterA.Stokes.EnglishVernacularScriptfromÆthelredtoCnut,ca990ca1035.Boydell,Woodbridge,Forthcoming.53D.Stutzman.AAA�A
Alphabet,AmbiguitéetActualité(paléographique):l'ontologiedesformesalphabétiques.PaléographieMédiévale,2012.Availableon-line:http://ephepaleographie.wordpress.com/2012/01/22/aaa-%CE%B1%CE%B4%CE%BB-alphabet-ambiguite-actualites-paleographique-ontologie-formes-alphabetiques/.54Tagung.CodicologyandPalaeographyintheDigitalAge.Conferencereport,July2009.Availableonline:http://www.i-d-e.de/events-des-ide/2009-tagung-kpdz/report.55SimonTanner.MeasuringtheImpactofDigitalResources:TheBalancedValueIm-pactModel.King'sCollege,London,2012.Availableonline:http://www.kdcs.kcl.ac.uk/innovation/impact.html.56M.Terras.DigitalImagesfortheInformationProfessional.Ashgate,Aldershot,England,2008.57TranscribeBentham.Availableonline:http://www.transcribe-bentham.da.ulcc.ac.uk/.58GeorgVogeler.KodikologieundPaläographieimdigitalenZeitalter/Codicologyandpalaeographyinthedigitalage.AHF-Information,(206),2009.Availableonline:http://www.ahf-muenchen.de/Tagungsberichte/Berichte/pdf/2009/206-09.pdf.59C.VonOertzenandK.Wilder.ScholarlyPublishingandtheIssuesofCulturalHeritage,FairUse,ReproductionFeesandCopyrights.Berlin,MaxPlanckInstitute,2008.Availableonline:http://www.mpiwg-berlin.mpg.de/PDF/MPIWGWorkshop1-2008Report.pdf.60What'stheScore.Availableonline:https://www.zooniverse.org/lab/wts/.61LiorWolf,TalHassner,andYanivTaigman.Eectiveunconstrainedfacerecognitionbycombiningmultipledescriptorsandlearnedbackgroundstatistics.PatternAnalysisandMachineIntelligence,IEEETransactionson,33(10):19781990,2011.62EuropeanScienceFoundation.ExploratoryWorkshoponDigitalPalaeography.Conferencereport,2011.Availableonline:http://hsozkult.geschichte.hu-berlin.de/tagungsberichte/id=3968. 12382