Proceedings of The 8th European Conference on …profkeithdevlin.com/Papers/Devlin-Ketamo_ECGBL2014.pdfProceedings of The 8th European Conference on Games Based ... Harri Ketamo1 and

ProceedingsofThe8thEuropeanConferenceonGamesBasedLearning

ECGBL2014

ResearchandTrainingCenterforCultureandComputerScience(FKI)

UniversityofAppliedSciencesHTWBerlinGermany

910October2014

VolumeOne

EditedbyProfessorDr.Ing.CarstenBusch

ReplacingPISAWithGlobalGameBasedAssessment

HarriKetamo1andKeithDevlin21SatakuntaUniversityofAppliedSciences,Finland2StanfordUniversity,[email protected]: International Educational performancemeasures, like the Programme for International Student Assessment(PISA),rankcountriesbasedontheireducationalsystem'sperformance.However,theseactivitiesrequireyearsofworkforthousands of people to collect, analyze and report the data. Furthermore, the results are always two years oldwhenpublished and there are tens ofmissing countries. In this study, a global gamebased analytics was introduced andevaluated ina realworldenvironment.Thepaper continues the study reported inECGBL2013.Realtimegamebasedassessmentsdohaveseveraladvantagescomparedtotraditionalmeasures,buttheyarenotyetreadytoreplacePISA.Forexample,samplecollectionhastoberadicallyimprovedinordertominimizenoiseandbringoutonlythevalidcases.Keywords:assessment,gamesbasedlearning,learninganalytics,mathematicslearning

1. IntroductionInternationalEducationalperformancemeasures, like theProgramme for InternationalStudentAssessment(PISA),TrendsinInternationalMathematicsandScienceStudy(TIMSS),andProgressinInternationalReadingLiteracyStudy(PIRLS)rankthecountriesbasedontheireducationalsystem'sperformance.Theyalsodescribesocioeconomicbackgroundsandfactorsbehindoutcomes.Suchactivitiesarevaluableinordertounderstanddifferencesinlearningasaglobalphenomenon.However, theseactivities requireeyearsofwork for thousandsofpeople tocollect,analyzeand report thedata. Furthermore, the results are always two years old when published and there are tens of missingcountries.Finally, thebiggestmissingpart is theunderstandinghow to improve thesystematanationalorindividuallevel.Educational games and gamebased assessment can be seen partial answer to this: distribution of theassessment(i.e.games)canbedonegloballywithoutanyextraeffort,usingoutletssuchastheAppstore.Thedataisalwaysinrealtime,andiftheinstrumentsaredesignedintermsoflearningscience,theywillmeasurelearningand learningoutcomes.Arguably themost importantadvance is the capability togive suggestionshow to improve curriculum, teachingmethodology, or learningmaterials. Furthermore, realtime analyticsenablesindividuallevelrecommendationsandadaptationinlearning.Inthispaperweintroduceoneapproachforrealtimegamebasedassessment,showglobalassessmentresultsbasedongamedata,anddiscussthefuturedirectionofresearchandchallengestobeovercomebeforegamebasedassessmentcanreplacePISAorTIMSS.ThepapercontinuesthestudyreportedinproceedingsofECGBL2013(Ketamo2013).Behaviormodelinghasalongbackgroundineducationaltechnology:Neuralandsemanticnetworks,aswellasgeneticalgorithms,areutilizedtomodeluserscharacteristics,profiles,andpatternsofbehavior inordertosupportor challenge theperformanceof individuals.Behavior recordinghasbeen studied andused in thegameindustryforalongtime.Inthemajorityofstudies,thelevelofbehaviorisingenerallimitedtoobservedpatterns(e.g.Brusilovsky2001;Houllette2003;Bowling&al2006;Bra&al2013;ShyiMing&PoJui2013).In this study, user behavior, competence, and learning are seen as semantic (neural) network that isinductively taughtby theuser.According to the cognitivepsychologyof learning,our thinking isbasedonconceptualrepresentationsofourexperiencesandrelationsbetweentheseconcepts.Learningiswhatoccurswhenthementalstructurechanges.Dataminingandanalyticsarebasedonthissemanticmodel.Whenalltheskillsandknowledgearerecordedasasemanticnetwork,theminingcanbedoneintermsofnetworkanalysis.Thenoveltyof thisstudy lies in theapproach: tobuildgamebased technologies thatenableeasy inductiveconstructionofintelligentandhumanlikebehaviors,andsoenabledetailedanalysisoflearningachievements.Thesamekindofcompetencemodeling,basedondeductivelearningtheories,hasbeendonebye.g.Biswas,Leelawong,Schwartz&Vye(2005).

258

HarriKetamoandKeithDevlin

Weknowthatchildrenarereadytodomoreworkfortheirgamecharactersthanforthemselves(e.g.Chase&al.2009).ThatiswhySkillpixels(acompanyKetamohasfoundedin2011)hasdevelopedgameswheregamecharacters learn likehumansdo,withchildren taking the roleofa teacher.The frameworkand technologybehind thegames supportdetailed learninganalyticsandprovides real timeanalysisof learningprocesses,difficultiesinlearning,andchallengesinthecurriculum.

2. ResearchprocedureThesample(N=49080)wascollectedwithSmartKidMathsOnlineversion,between1.8.and31.12.2013.Thisonlineversionwastargetedespeciallyforschoolpiloting,whichhelpsustomakeanassumptionthatmostoftheplayerswere1stand2ndgradepupils.Onlytheplayerswhohadcompletedat least10%ofthecontent(approximately1.52hoursofgameplayonaverage)wereincludedintothissample.Thisdecisionwasmadetoensurethatalltheincludeddataiscollectedfrompupilswhoreallyplayedthegame,notjusttrieditbrieflyorhadashortperiodofuse.Because all collecteddata is completely anonymous,we canbe sureonly about the country that thedatacomes from.We cannot guarantee that all the first grade players are really first grade pupils, and so on.Furthermore,thenatureofthegameisthatwhenaplayerhascompletedkindergarten,he/shecontinuestofirst grade.Thismeans there are a lotof kindergartenpupils included in first gradedata and a lotof firstgraders in second grade data. In this paper, we focus on describing the procedure and findings, so thesampling isnot aproblem,butwhendescribing learning achievements and competences indetails, this ismaybeoneofthemostcriticalchallengestocounter.Inthispaper,onlycountrieswithmorethan1500playerswereincludedintheanalysis.Thefinalsample,1000pupilspercountry,wasrandomlyselectedfromallthedatafromthespecificcountry.Thedifferences intheresultsbetweenthreedifferentrandomsampleswerebelow2%.Furthermore,thesamplewasdividedintwosubgroups:casualplayersandhardgamers.Casualplayerswerethosewhocompletedmorethan10%but lessthan35%ofthegame. Inotherwordsthosewhoplayed26hoursonaveragewithseveralsessions.Thehardgamersgroupconsistofthosecompletedmorethan35%ofthecontent.Someof themplayedonly5hourswithextremelygoodperformance,andsomeplayersspentmorethan100hourswithSmartKidMaths,endingupwithagoodperformance.However,mostofthepupilswhoplayedmorethan35%ofthecontentwerekindergartenpupilswhoplayedfirst grade contentor firstgradepupilswhoplayed second grade content.Whenanalyzing the results,wenotedthatthemorelevelsplayeropened,thatmorehe/shefacedcompletelynewtopics.In the results sectionwe provide analysis from Finland, Sweden,United States, Canada,United Kingdom,Ireland,Malaysia,India,LatinAmerica(asagroupinordertoreachthedefinedN),andAustralia.

3. ResearchmaterialsSmartKidMathsisagamebased,virtualmathematicsschoolformobile,tabletandonlineusers(Figure1).ThebiggestdifferencefromothermathematicsgamesisthatinSmartKidMathstheplayerteachesthecomputer,notviceversa.Whentheplayerisresponsibleforcharactersmentaldevelopment,he/sherecordsalsohis/hermentalconceptualstructureduringthegameplay.

Figure1:CollectionofscreenshotsfromSmartKidMaths

SmartKid Maths include all mathematics in the US common core K2 curriculum, which representsapproximately300400pagesofexercisesintraditionalbooks.Inthegame,thereare100levelsandonelevel

259


represents approximatelyone schoolweek,35pagesof the exercisebook, in traditional school. SmartKidMathscoversthefollowingCommonCoretopics:GradeK:CountingandCardinality

Knownumbernamesandthecountsequence.

Counttotellthenumberofobjects.

Comparenumbers.

Grade1:OperationsandAlgebraicThinking

Understandandapplypropertiesofoperationsandtherelationshipbetweenadditionandsubtraction.

Addandsubtractwithin20.

Workwithadditionandsubtractionequations.

Grade2:OperationsandAlgebraicThinking

Representandsolveproblemsinvolvingadditionandsubtraction.

Representandsolveproblemsinvolvingmultiplicationanddivision.

Addandsubtractwithin100.

Multiplyanddividewithin100.

Workwithequalgroupsofobjectstogainfoundationsformultiplication.

Oneofourspecialfocihasbeenthescientificproofofconcept:Wehaveshowneducationaloutcomesaswellasmotivation towards teachingvirtualpets:Under strict laboratoryexperiment settings,more than60%ofplayersincreasedtheirskillsremarkablyduringthegameplay.Theoutcomeinanaturallearningenvironment,withthepossibilityoflongergameplay,isevengreater:thebestoutcomeisachievedwhentherearebreaksandinformaldiscussionsbetweengameplay.Whentheplayerisresponsiblefortheircharactersmentaldevelopment,he/sherecordsalsohis/hermentalconceptual structure during the gameplay. The most important finding is that assessment carried outaccording to learningdatacollectedduring thegameplaycorrelateswithassessmentdonewith traditionalpaper tests: Taught conceptual structure is strongly related to paper tests scores received after gameplay(0.4


times.ThegameAIusesallofthetaught informationtosupport itsdecisions,andtherefore ittakestimetooverrideanywrongrelationsintheagentsconceptualstructure.

4. ResultsAnalytics outcomes of the learning process can be used when developing national curricula or learningmaterials.When summarizing the individualgameachievements (taught conceptualnetworks), schoolsandnational levelpolicymakerscanreceiveanalysisabout learningachievementsandcompetencesatagenerallevel.Theycanapplythistodeveloptheirteachinginstructionsorformalcurriculum.Ourgoalinthispaperisnottorankcountries,buttoprovideinformationthatmightbeusefulfordevelopingcurriculumorclassroompractices.Inthisanalysiswecomparedifferencesinunderstandingandapplyingthecoreconceptsofnumeracy:

countofobjects

numbersymbol

placeinnumberline

Thegamerecordsalltheeventsduringthegameplay.Firstweidentifythenumberofdifficultiesorrecordedmisconceptions encountered during the game play. After that, identify playerswho have overcome theirpreviousdifficultieswhenthelevelorgameiscompleted.OntheleftsideofTables16arethepercentagesharesofpupilswhomasteredthecontentallthetimeduringthegameplay(below5%misconceptionsatanystageofthegameplay)andontherightsidethepercentagesharesofpupilswhomasteredthecontentafterthelevelorgameiscompleted.Thedifferencebetweenthesesharesshowstheeffectofthegame:themorepupilsmastersthecontentafterthegameplay,thatbetterthegameperformsasalearninggame.InTables13,wefocusoncasualplayersgroup(completedmorethan10%butlessthan35%ofthegame).Focusingonlyonaconceptuallevel,wecanseethattheconceptofnumbersymbol is theeasiestone. Inotherwords,pupils recognize thewrittennumbersandonlya relatively smallnumberofpupilshadproblemsapplyingnumber symbols.Thenumber linewas themostdifficultconcept.Surprisingly,applyingcountswasmorechallengingthanthenumbersymbols.Thisphenomenawasobservedinallthecountries.

Table1:Mastering the conceptofNumber (as symbol)duringandafter thegameplay.Playerswhoplayedmorethan10%butlessthan35%ofcontentareincludedinfigure

Table2:MasteringtheconceptofCountduringandafterthegameplay.Playerswhoplayedmorethan10%butlessthan35%ofcontentareincludedinfigure

261


Table3:MasteringtheconceptofNumberLineduringandafterthegameplay.Playerswhoplayedmorethan10%butlessthan35%ofcontentareincludedinfigure

Whencomparingcountries, thereare relativelybigdifferences inpercentagesharesof thosemastering thecontent.Wewould expect to find statistically significant differences, but because our data collection andsampledefinitionswererelativelysuperficial,wedidnotrunanystatisticalanalysisonthissample.Instead,wechosetodescribethemethodandreportitsstrengthsandweaknesses.Mostofthedifferencescanbeexplainedbyfollowingtwofacts:1)Kindergarten isstartedbetween3and6yeasofage,dependingoncountry.Thiskindofdifferenceinageobviouslyaffectstoperformanceinthegame.2)Weknowmanykindergartenpupilsplayedthefirstgradegame,butwealsoknowthat insomecountrieswehadremarkablybigsample from firstandsecondgrade.Whendoingarandomsample,thebigshareofsecondgradersaffectsthesample. InTables46wefocusonthehardgamersgroup (completedmorethan35%ofthecontent).Thisgroupisverydifferentfromthecasualplayersgroup:Mostoftheplayerscompletedcontent(levels)thatgoesfarbeyondtheirschoolliteracy.Inotherwords,kindergartenpupilwhohaveplayedhalfof the first grade contentbelongs into this group. Furthermore, is seems that thosepupilswho facedchallengesduring thegameplayandovercame those challengesexperiencedkindofaFlowexperience,andthereforeplayedfor longerthanthosewhodidnot learnasmuchduringthegameplay.FromTables46weseethatthistimetheconceptofcountwastheeasiestone.Thisismostlybecauseofthetasksinthegame:Theconceptofcountwasexcluded from themostdifficult tasksandhence thepupilsdidnot face thatbigchallengewithcounts.

Table4:MasteringtheconceptofNumberduringandafterthegameplay.Playerswhoplayedmorethan35%ofcontentareincludedinfigure

Table5:MasteringtheconceptofCountduringandafterthegameplay.Playerswhoplayedmorethan35%of

contentareincludedinfigure

262


Table6:MasteringtheconceptofNumberLineduringandafterthegameplay.Playerswhoplayedmorethan35%ofcontentareincludedinfigure

Learning is about changes in our understanding. Educational games in general aremeant to provide goodlearningoutcomes.Thishasbeen themost importantgoal forSmartKidMaths.When focusingon the rightsideofTables16wecanseethatalmostalltheofpupilsmasteredthecontentafterthegameplay,nomatterhowmanydidnothavemasterybeforethegameplay.Thisinnotonlybecauseofthemethodandcontent.Itisalsobecauseofgametypestorytelling:Whenpayinggames,childrenareusedtoworkinghardinordertoaccesstothenextlevel.Beforeaplayercanaccessthenext level,he/shemusthaveenough skill fromprevious levels.The same isnot thecase for school,wherepupilscancontinuewiththenexttopicnomatterhowmanybasicskillstheyaremissing.BecauseSmartKidMathsisdesignedtoprovideadaptiveandpersonalizedlearning,itensuresallthebasicsarelearnedbeforepupilareallowedtocontinue.Inotherword,whenanalyzingthelatestperformance,almostallpupilshadachievedthebasicnumeracyskillsduringthegameplay.Thisgivesastrongbackgroundtocontinueaddition,subtraction,multiplication,anddivision.

5. ConclusionsAccording to our studies, users can relatively quickly and easily teach their game character. The behaviormodelingmakes itpossibletoemulateconceptual learningandthusuncoverthefrequencies,dependencies,and patterns behind conceptual change and learning transfer. From the data we can determine theperformanceindicatorsforconceptuallevelandcancomparetheperformancebetweencountries.Ingeneral,the results in this study differs from e.g. PISA results, which is both an expected and an interestingobservation.Realtimeglobalgamebasedassessment,described inthisstudy, isnotyetreadytoreplacePISA2015.Wehavetoradicallyimproveourdatacollectionandsampledesigninordertobringoutreallyvalidresults.Thismightbedoneviaschools:whenschoolscananonymouslyidentifythepopulation,wellbeabletoeliminateunknownplayersandfilteroutconceptualchangesresultingfromlongplayingtime.One future researchactivity todecrease this samplingchallengecouldbebasedonestimatingwhenpupilsplayata level that is completelyoutof their competence.Wehaveobserved thatpupilswhohaveplayedthrough a lot of content hadmajor conceptualmisunderstandings during the gameplay,while thosewhoplayedlessdidnotdemonstratesuchmisunderstandings.Iftheonsetofcriticalconceptualmisunderstandingcouldbe identified,we could separate firstgraders from secondgradersandkindergartenpupils from firstgraders.Ingeneral,gamesprovideadifferentviewoflearningbecauseofthetimeframe.Learningisaboutchangeandchangerequirestime.Whenwecanprovidevalidanalysisaboutlearningprocessandlearningassessment,wecanprovidesomethingnewinadditiontoPISA,TIMSSandPIRLS.

References

Biswas,G.,Leelawong,K.,Schwartz,D.,Vye,N.(2005).LearningbyTeaching:ANewAgentParadigmforEducationalSoftware.AppliedArtificialIntelligence,19(34),363392.

263


Bowling,M.,Furnkranz,J.,Graepel,T.&Musick,R.(2006).Machinelearningandgames.MachineLearning,vol63,pp.211215.

Bra,P.,Smits,D.,Sluijs,K.,Cristea,A.,Foss,J.,Glahn,C.&Steiner,C.M.(2013).GRAPPLE:LearningManagementSystemsMeetAdaptiveLearningEnvironments.InPeaAyala,A.(ed.)IntelligentandAdaptiveEducationalLearningSystems:SmartInnovation,SystemsandTechnologies.Volume17,2013,pp133160.

Brusilovsky,P.(2001).AdaptiveHypermedia.UserModelingandUserAdaptedInteraction,vol11,p.87110.Chase,C,Chin,D,Oppezzo,M,Schwartz,DL.(2009).Teachableagentsandtheprotegeeffect:Increasingtheeffort

towardslearning.JournalofScienceEducationandTechnology,18,334352.Houlette,R.(2003)PlayerModelingforAdaptiveGames.InRabin,S.(ed.)AIGameProgrammingWisdomII.

Massachusetts:CharlesRiverMedia,Inc.Ketamo,H.(2013).LearningAnalyticswithGamesBasedLearning.Inproceedingsofthe7thEuropeanConferenceon

GamesBasedLearning.34October,Porto,Portugal,pp.284289.Ketamo,H.(2011).SharingBehaviorsinGamesandSocialMedia.InternationalJournalofAppliedMathematicsand

Informatics,vol.5(1),pp.224232.Ketamo,H.&Kiili,K.(2010).Conceptualchangetakestime:Gamebasedlearningcannotbeonlysupplementary

amusement.JournalofEducationalMultimediaandHypermedia,vol.19(4),pp.399419.Ketamo,H.(2009).Semanticnetworksbasedteachableagentsinaneducationalgame.TransactionsonComputers,vol

8(4),pp.641650.ShyiMing,C.&PoJui,S.(2013).Constructingconceptmapsforadaptivelearningsystemsbasedondatamining

techniques.ExpertSystemswithApplications.Volume40(7),pp.27462755.

264

Al-Wahmi Blanchfield et al 049Al-Washmi Blanchfield et al 054Amresh Clarke and BeckwithBeg Looy and AllBratitsis and PapachristouBusch Steinicke et alCaponetto Earp and OttChaudy Connelly and HaineyCrockettCronkDijk et alEjsing-Duun and KaroffFabricatore and Lopez 025Fabricatore and Lopez 026Farrell and MoffatForlizzi et alFranken-Wendelstorf Konrad and SchuchardtGerodetti and NixonGjeddeHainey et al 075Hainey et al 096Hammais Ketamo and KoivistoHanghoj et alHautopp and HanghojHawlitschek and KoppenHussaan and SehabaJohansson et alJuzeleniene Baranauskiene and SarkauskieneKalmpourtzisKandroudi Bratitsis and LambropoulosKarakostas Tsiastsos et alKetamo and DevlinKetelhut et alKing and CazessusKlonari and GousiouKonig Pfeiffer WernbacherKordaki and GousiouKwan et alLockley and BoyleLynch Mallon and ConnollyMaciuszek and MartensMagnussenMalliarakis Satratzemi and XinogalosMarchetti and ValenteMarfisi George and TarpinMavridis Tsiatsos et alMehm et alMisfeldtMohtasham GilaniMoreno Cavazotte and AzevedoMozeliusMozgaleva Gulyaeva and ZamyatinaNaik 022Naik 023O Broin and PowerPlohn and AalbergPlohn Louchart and AalbergRetalis et alReuter et alRiedel and AzadeganRos-McDonnell Bajor and FejesSanchez and MartinezSchoenau-FogSelmbacherova Sisler and BromShabalina et alSillaotsSobke and londongTorrao Carvalho and RestivoTriantafyllou Timcenko and TriantafyllidisTveitVahed Singh and McKennaVate-U-LanWeitzeWeizmanWendel et alWestera Slootmaker and KurversWiklund Mozelius et alXiang et alYilmaz Saran and O'ConnorZamyatina et alZapusek and RugeljZX-All Castellar and LooeyZX-Djeli et alZX-FialovaZX-Koivisto Merilampi and SirkkaZX-Linssen and TheuneZX-Lopez-Arcos et alZX-Meesters Olthof and VandewalleZX-Nguyen et alZX-Szander Strommer and BajorZX-Titus and Ng'ambiZX-Vasileva-Stojanovska et alZX-WilkinsonZXX-Burke O'Broin and Mc EvoyZXX-Sideris et alZZ-BarrettZZ-Friess Kolas and KnochZZ-Gouveia -de Carvalho et alZZ-Hansen and HansenZZ-Janssen et alZZ-KalendaZZ-Korkut et alZZ-Lane-KelsoZZ-Marques da Silva and da SilvaZZ-Nogueira Vaz de Carvalho et alZZ-StanfordZZ-Tan et alZZ-Tegos et alZZ-Terracina and MecellaECGBL 2014 front-dropbox vol 1-front.pdfSlide Number 1Slide Number 2Slide Number 3

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 300 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 1200 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile () /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False

/CreateJDFFile false /Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles false /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /DocumentCMYK /PreserveEditing true /UntaggedCMYKHandling /LeaveUntagged /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> ]>> setdistillerparams> setpagedevice
































































/CreateJDFFile false /Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles false /MultimediaHandling /UseObjectSettings /Name

Proceedings of The 8th European Conference on …profkeithdevlin.com/Papers/Devlin-Ketamo_ECGBL2014.pdfProceedings of The 8th European Conference on Games Based ... Harri Ketamo1 and

Documents