Inparticulartheactivityimmediatelyprecedingthetestwaskeptasconstantincharacteraswaspossible。Sincethementalaswellasthephysicalconditionofmanissubjecttoanevidentperiodicityof24hours，itwastakenforgrantedthatlikeexperimentalconditionsareobtainableonlyatliketimesofday。However，inordertocarryoutmorethanonetestinagivenday，differentexperimentswereoccasionallycarriedontogetheratdifferenttimesofday。Whentoogreatchangesintheouterandinnerlifeoccurred，thetestswerediscontinuedforalengthoftime。Theirresumptionwasprecededbysomedaysofrenewedtrainingvaryingaccordingtothelengthoftheinterruption。Section14。SourcesofErrorTheguidingpointofviewintheselectionofmaterialandindeterminingtherulesforitsemploymentwas，asisevident，theattempttosimplifyasfaraspossible，andtokeepasconstantaspossible，theconditionsunderwhichtheactivitytobeobserved，thatofmemory，cameintoplay。
　　Naturallythebetteronesucceedsinthisattemptthemoredoeshewithdrawfromthecomplicatedandchangingconditionsunderwhichthisactivitytakesplaceinordinarylifeandunderwhichitisofimportancetous。
　　Butthatisnoobjectiontothemethod。Thefreelyfallingbodyandthefrictionlessmachine，etc。，withwhichphysicsdeals，arealsoonlyabstractionswhencomparedwiththeactualhappeningsinnaturewhichareofimporttous。Wecanalmostnowheregetadirectknowledgeofthecomplicatedandthereal，butmustgetattheminroundaboutwaysbysuccessivecombinationsofexperiences，eachofwhichisobtainedinartificial，experimentalcases，rarelyorneverfurnishedinthisformbynature。
　　Meanwhilethefactthattheconnectionwiththeactivityofmemoryinordinarylifeisforthemomentlostisoflessimportancethanthereverse，namely，thatthisconnectionwiththecomplicationsandfluctuationsoflifeisnecessarilystillatoocloseone。Thestruggletoattainthemostsimpleanduniformconditionspossibleatnumerouspointsnaturallyencountersobstaclesthatarerootedinthenatureofthecaseandwhichthwarttheattempt。Theunavoidabledissimilarityofthematerialandtheequallyunavoidableirregularityoftheexternalconditionshavealreadybeentouchedupon。Ipassnexttotwootherunsurmountable[sic]sourcesofdifficulty。
　　Bymeansofthesuccessiverepetitionstheseriesare，sotospeak，raisedtoeverhigherlevels。Thenaturalassumptionwouldbethatatthemomentwhentheycouldforthefirsttimebereproducedbyheartthelevelthusattainedwouldalwaysbethesame。Ifthiswerethecase，i。e。，ifthischaracteristicfirstreproductionwereeverywhereaninvariableobjectivesignofanequallyinvariablefixednessoftheseries，itwouldbeofrealvaluetous。This，however，isnotactuallythecase。Theinnerconditionsoftheseparateseriesatthemomentofthefirstpossiblereproductionarenotalwaysthesame，andthemostthatcanbeassumedisthatinthecaseofthesedifferentseriestheseconditionsalwaysoscillateaboutthesamedegreeofinnersurety。Thisisclearlyseenifthelearningandrepeatingoftheseriesiscontinuedafterthatfirstspontaneousreproductionoftheserieshasbeenattained。Asageneralthingthecapacityforvoluntaryreproductionpersistsafterithasoncebeenreached。Innumerouscases，however，itdisappearsimmediatelyafteritsfirstappearance，andisregainedonlyafterseveralfurtherrepetitions。Thisprovesthatthepredispositionformemorisingtheseries，irrespectiveoftheirdifferencesofalargersortaccordingtothetimeofday，totheobjectiveandsubjectiveconditions，etc。，issubjecttosmallvariationsofshortduration，whethertheybecalledoscillationsofattentionorsomethingelse。If，attheveryinstantwhenthematerialtobememorisedhasalmostreachedthedesireddegreeofsurety，achancemomentofespecialmentalclearnessoccurs，thentheseriesiscaughtonthewingasitwere，oftentothelearner’ssurprise;
　　buttheseriescannotlongberetained。Bytheoccurrenceofamomentofspecialdullness，ontheotherhand，thefirsterrorlessreproductionispostponedforawhile，althoughthelearnerfeelsthathereallyismasterofthethingandwondersattheconstantlyrecurringhesitations。Intheformercase，inspiteofthehomogeneityoftheexternalconditions，thefirsterrorlessreproductionisreachedatapointalittlebelowthelevelofretentionnormallyconnectedwithit。Inthelattercaseitisreachedatapointalittleabovethatlevel。Aswassaidbefore，themostplausibleconjecturetomakeinthisconnectionisthatthesedeviationswillcompensateeachotherinthecaseoflargegroups。
　　Oftheothersourceoferror，Icanonlysaythatitmayoccurandthat，whenitdoes，itisasourceofgreatdanger。Imeanthesecretinfluenceoftheoriesandopinionswhichareintheprocessofformation。Aninvestigationusuallystartsoutwithdefinitepresuppositionsastowhattheresultswillbe。Butifthisisnotthecaseatthestart，suchpresuppositionsformgraduallyincasetheexperimenterisobligedtoworkalone。Foritisimpossibletocarryontheinvestigationsforanylengthoftimewithouttakingnoticeoftheresults。Theexperimentermustknowwhethertheproblemhasbeenproperlyformulatedorwhetheritneedscompletionorcorrection。
　　Thefluctuationsoftheresultsmustbecontrolledinorderthattheseparateobservationsmaybecontinuedlongenoughtogivetothemeanvaluethecertaintynecessaryforthepurposeinhand。Consequentlyitisunavoidablethat，aftertheobservationofthenumericalresults，suppositionsshouldariseastogeneralprincipleswhichareconcealedinthemandwhichoccasionallygivehintsastotheirpresence。Astheinvestigationsarecarriedfurther，thesesuppositions，aswellasthosepresentatthebeginning，constituteacomplicatingfactorwhichprobablyhasadefiniteinfluenceuponthesubsequentresults。ItgoeswithoutsayingthatwhatIhaveinmindisnotanyconsciouslyrecognisedinfluencebutsomethingsimilartothatwhichtakesplacewhenonetriestobeveryunprejudicedortoridone’sselfofathoughtandbythatveryattemptfostersthatthoughtorprejudice。
　　Theresultsaremethalfwaywithananticipatoryknowledge，withakindofexpectation。Simplyfortheexperimentertosaytohimselfthatsuchanticipationsmustnotbeallowedtoaltertheimpartialcharacteroftheinvestigationwillnotbyitselfbringaboutthatresult。Onthecontrary，theydoremainandplayarôleindeterminingthewholeinnerattitude。
　　Accordingasthesubjectnoticesthattheseanticipationsareconfirmedornotconfirmed（andingeneralhenoticesthisduringthelearning），hewillfeel，ifonlyinaslightdegree，asortofpleasureorsurprise。
　　Andwouldyounotexpectthat，inspiteofthegreatestconscientiousness，thesurprisefeltbythesubjectoverespeciallystartlingdeviations，whetherpositiveornegative，wouldresult，withoutanyvolitiononhispart，inaslightchangeinattitude？Wouldhenotbelikelytoexerthimselfalittlemorehereandtorelaxalittlemoretherethanwouldhavebeenthecasehadhehadnoknowledgeorpresuppositionconcerningtheprobablenumericalvalueoftheresults？Icannotassertthatthisisalwaysorevenfrequentlythecase，sincewearenothereconcernedwiththingsthatcanbedirectlyobserved，andsincenumerousresultsinwhichsuchsecretwarpingofthetruthmightbeexpectedshowevidentindependenceofit。
　　AllIcansayis，wemustexpectsomethingofthesortfromourgeneralknowledgeofhumannature，andinanyinvestigationsinwhichtheinnerattitudeisofverygreatimportance，asforexampleinexperimentsonsenseperception，wemustgivespecialheedtoitsmisleadinginfluence。
　　Itisevidenthowthisinfluenceingeneralmakesitselffelt。Withaveragevaluesitwouldtendtoleveltheextremes;whereespeciallylargeorsmallnumbersareexpecteditwouldtendtofurtherincreaseordecreasethevalues。Thisinfluencecanonlybeavoidedwithcertaintywhenthetestsaremadebytwopersonsworkingtogether，oneofwhomactsassubjectforacertaintimewithoutraisinganyquestionsconcerningthepurposeortheresultoftheinvestigations。Otherwisehelpcanbeobtainedonlybyroundaboutmethods，andthen，probably，onlytoalimitedextent。Thesubject，asImyselfalwaysdid，canconcealfromhimselfaslongaspossibletheexactresults。Theinvestigationcanbeextendedinsuchawaythattheupperlimitsofthevariablesinquestionareattained。Inthisway，whateverwarpingofthetruthtakesplacebecomesrelativelymoredifficultandunimportant。Finally，thesubjectcanproposemanyproblemswhichwillappeartobeindependentofeachotherinthehopethat，asaresult，thetruerelationoftheinterconnectedmentalprocesseswillbreakitswaythrough。
　　Towhatextentthesourcesoferrormentionedhaveaffectedtheresultsgivenbelownaturallycannotbeexactlydetermined。Theabsolutevalueofthenumberswilldoubtlessbefrequentlyinfluencedbythem，butasthepurposeofthetestscouldneverhavebeentheprecisedeterminationofabsolutevalues，butrathertheattainmentofcomparativeresults（especiallyinthenumericalsense）andrelativelystillmoregeneralresults，thereisnoreasonfortoogreatanxiety。Inoneimportantcase（sec。38）Icoulddirectlyconvincemyselfthattheexclusionofallknowledgeconcerningthecharacteroftheresultsbroughtaboutnochange;inanothercasewhereImyselfcouldnoteliminateadoubtIcalledespecialattentiontoit。
　　Inanycasehewhoisinclinedaprioritoestimateveryhighlytheunconsciousinfluenceofsecretwishesonthetotalmentalattitudewillalsohavetotakeintoconsiderationthatthesecretwishtofindobjectivetruthandnotwithdisproportionatetoiltoplacethecreationofhisownfancyuponthefeetofclay——thatthiswish，Isay，mayalsoclaimaplaceinthecomplicatedmechanismofthesepossibleinfluences。Section15。MeasurementofWorkRequiredThenumberofrepetitionswhichwerenecessaryformemorisingaseriesuptothefirstpossiblereproductionwasnotoriginallydeterminedbycounting，butindirectlybymeasuringinsecondsthetimethatwasrequiredtomemoriseit。Mypurposewasinthiswaytoavoidthedistractionnecessarilyconnectedwithcounting;andIcouldassumethattherewasaproportionalrelationexistingbetweenthetimesandthenumberofrepetitionsoccurringatanytimeinadefiniterhythm。Wecouldscarcelyexpectthisproportionalitytobeperfect，since，whenonlythetimeismeasured，themomentsofhesitationandreflectionareincluded，whichisnottruewhentherepetitionsarecounted。Difficultseriesinwhichhesitationwilloccurrelativelymorefrequently，will，bythemethodoftimemeasurement，getcomparativelygreaternumbers，theeasierserieswillgetcomparativelysmallernumbersthanwhentherepetitionsarecounted。Butwithlargergroupsofseriesatolerablyequaldistributionofdifficultandequalseriesmaybetakenforgranted。Consequentlythedeviationsfromproportionalitywillcompensatethemselvesinasimilarmannerinthecaseofeachgroup。
　　When，forcertaintests，thedirectcountingoftherepetitionsbecamenecessary，Iproceededinthefollowingmanner。Littlewoodenbuttonsmeasuringabout14mms。indiameterand4mms。attheirgreatestthicknesswerestrungonacordwhichwouldpermitofeasydisplacementandyetheavyenoughtopreventaccidentalslipping。Eachtenthpiecewasblack;theothershadtheirnaturalcolor。Duringthememorisationthecordwasheldinthehandandateachnewrepetitionapiecewasdisplacedsomecentimetersfromlefttoright。Whentheseriescouldberecited，aglanceatthecord，sinceitwasdividedintotens，wasenoughtoascertainthenumberofrepetitionsthathadbeennecessary。Themanipulationrequiredsolittleattentionthatinthemeanvaluesofthetimeused（whichwasalwaystabulatedatthesametime）nolengtheningcouldbenotedascomparedwithearliertests。
　　Bymeansofthissimultaneousmeasurementoftimeandrepetitionsincidentalopportunitywasaffordedforverifyingandmoreaccuratelydefiningthatwhichhadbeenforeseenandwhichhasjustbeenexplainedwithregardtotheirinterrelation。Whentheprescribedrhythmof150strokesperminutewaspreciselymaintained，eachsyllablewouldtake0。4second;andwhenthesimplereadingoftheserieswasinterruptedbyattemptstoreciteitbyheart，theunavoidablehesitationswouldlengthenthetimebysmallbutfairlyuniformamounts。This，however，didnotholdtruewithanyexactness;
　　onthecontrary，thefollowingmodificationsappeared。
　　Whenthedirectreadingoftheseriespredominated，acertainforcing，anaccelerationoftherhythm，occurredwhich，withoutcomingtoconsciousness，onthewholeloweredthetimeforeachsyllablebelowthestandardof0。4
　　sec。
　　Whentherewasinterchangebetweenreadingandreciting，however，thelengtheningofthetimewasnotingeneralconstant，butwasgreaterwiththelongerseries。Inthiscase，sincethedifficultyincreasesveryrapidlywithincreasinglengthoftheseries，thereoccursaslowingofthetempo，againinvoluntaryandnotdirectlynoticeable。Bothareillustratedbythefollowingtable。
　　Assoonasthisdirectionofdeviationfromexactproportionalitywasnoticedthereappearedinthelearningacertainconsciousreactionagainstit。
　　Finally，itappearedthattheprobableerrorofthetimemeasurementswassomewhatlargerthanthatoftherepetitions。Thisrelationisquiteintelligibleinthelightoftheexplanationsgivenabove。Inthecaseofthetimemeasurementsthelargervalues，whichnaturallyoccurredwiththemoredifficultseries，wererelativelysomewhatgreaterthaninthecaseofthenumberofrepetitions，becauserelativelytheywereforthemostpartlengthenedbythehesitations;conversely，thesmallertimeswerenecessarilysomewhatsmallerrelativelythanthenumberofrepetitions，becauseingeneraltheycorrespondedtotheeasierseries。Thedistributionofthevaluesinthecaseofthetimesisthereforegreaterthanthatofthevaluesinthecaseoftherepetitions。
　　Thedifferencesbetweenthetwomethodsofreckoningare，asisreadilyseen，sufficientlylargetoleadtodifferentresultsinthecaseofinvestigationsseekingahighdegreeofexactness。Thatisnotthecasewiththeresultsasyetobtained;itisthereforeimmaterialwhetherthenumberofsecondsisusedorthatoftherepetitions。
　　Decisioncannotbegivenaprioriastowhichmethodofmeasurementismorecorrect——i。e。，isthemoreadequatemeasureofthementalworkexpended。Itcanbesaidthattheimpressionsaredueentirelytotherepetitions，theyarethethingthatcounts;itcanbesaidthatahesitatingrepetitionisjustasgoodasasimplefluentreproductionoftheline，andthatbotharetobecountedequally。Butontheotherhanditmaybedoubtedthatthemomentsofrecollectionaremerelyaloss，In，anycaseacertaindisplayofenergytakesplaceinthem：ontheonehand，averyrapidadditionalrecollectionoftheimmediatelyprecedingwordsoccurs，anewstart，sotospeak，togetovertheperiodofhesitation;
　　ontheotherhand，thereisheightenedattentiontothepassagesfollowing。
　　Ifwiththis，asisprobable，afirmermemorisationoftheseriestakesplace，thenthesemomentshaveaclaimuponconsiderationwhichcanonlybegiventothemthroughthemeasurementofthetimes。
　　Onlywhenaconsiderabledifferenceintheresultsofthetwokindsoftabulationappearswillitbepossibletogiveonethepreferenceovertheother。Thatonewillthenbechosenwhichgivesthesimplerformulationoftheresultinquestion。Section16。PeriodsoftheTestsThetestsweremadeintwoperiods，intheyears1879-80and1883-84，andextendedeachovermorethanayear。Duringalongtimepreliminaryexperimentsofasimilarnaturehadprecededthedefinitetestsofthefirstperiod，sothat，forallresultscommunicated，thetimeofincreasingskillmaybeconsideredaspast。AtthebeginningofthesecondperiodIwascarefultogivemyselfrenewedtraining。Thistemporaldistributionofthetestswithaseparatingintervalofmorethanthreeyearsgivesthedesiredpossibilityofacertainmutualcontrolofmostoftheresults。
　　Frankly，thetestsofthetwoperiodsarenotstrictlycomparable。Inthecaseofthetestsofthefirstperiod，inordertolimitthesignificanceofthefirstfleetinggrasp[3]oftheseriesinmomentsofspecialconcentration，itwasdecidedtostudytheseriesuntiltwosuccessivefaultlessreproductionswerepossible。LaterIabandonedthismethod，whichonlyincompletelyaccomplisheditspurpose，andkepttothefirstfluentreproduction。Theearliermethodevidentlyinmanycasesresultedinasomewhatlongerperiodoflearning。
　　Inadditiontherewasadifferenceinthehoursofthedayappointedforthetests。Thoseofthelaterperiodalloccurredintheafternoonhoursbetweenoneandthreeo’clock;thoseoftheearlierperiodwereunequallydividedbetweenthehoursof10-11A。M。，11-12A。M。，and6-8P。M。，whichforthesakeofbrevityIshalldesignateA，B，andC。
　　Footnotes[1]Thevowelsoundsemployedwerea，e，i，o，u，ä
　　ö，ü，au，ei，eu。Forthebeginningofthesyllablesthefollowingconsonantswereemployed：b，d，fg，h，j，k，l，m，n，p，r，s，（=sz），t，wandinadditionch，sch，softs，andtheFrenchj（19altogether）;
　　fortheendofthesyllablesf，k，l，m，n，p，r，s，（=sz）t，ch，sch（11altogether）。Forthefinalsoundfewerconsonantswereemployedthanfortheinitialsound，becauseaGermantongueevenafterseveralyearspractiseinforeignlanguagesdoesnotquiteaccustomitselftothecorrectpronunciationofthemediaeattheend。ForthesamereasonIrefrainedfromtheuseofotherforeignsoundsalthoughItriedatfirsttousethemforthesakeofenrichingthematerial。
　　[2]Ishallretaininwhatfollowsthedesignationsemployedaboveandcallagroupofseveralsyllableseriesorasingleseriesa"test。"Anumberof"tests"Ishallspeakofasa"testseries"
　　ora"groupoftests。"
　　[3]Describedinsec。14。
　　ClassicsintheHistoryofPsychology——Ebbinghaus（1885）Chapter4Memory：AContributiontoExperimentalPsychologyHermannEbbinghaus（1885）TranslatedbyHenryA。Ruger&ClaraE。Bussenius（1913）CHAPTERIVTHEUTILITYOFTHEAVERAGESOBTAINEDSection17。GroupingoftheResultsoftheTestsThefirstquestionwhichawaitsananswerfromtheinvestigationscarriedoutinthemannerdescribedis，asexplainedinsections7and8，thatofthenatureoftheaveragesobtained。Arethelengthsoftimerequiredformemorisingseriesofacertainlength，underconditionsasnearlyidenticalaspossible，groupedinsuchawaythatwemaybejustifiedinconsideringtheiraveragevaluesasmeasuresinthesenseofphysicalscience，oraretheynot？
　　Ifthetestsaremadeinthewaydescribedabove，namely，sothatseveralseriesarealwaysmemorisedinimmediatesuccession，suchatypeofgroupingofthetimerecordscouldscarcelybeexpected。For，asthetimedevotedtolearningatagivensittingbecomesextended，certainvariableconditionsintheseparateseriescomeintoplay，thefluctuationsofwhichwecouldnotverywellexpect，fromwhatweknowoftheirnature，tobedistributedsymmetricallyaroundameanvalue。Accordinglythegroupingoftheresultsmustbeanasymmetricaloneandcannotcorrespondtothe"lawoferror。"
　　Suchconditionsarethefluctuationsofattentionandthedecreasingmentalfreshness，which，atfirstveryquicklyandthenmoreandmoreslowly，giveswaytoacertainmentalfatigue。Therearenolimits，sotospeak，totheslowingdownofthelearningprocessescausedbyunusualdistractions;
　　asaresultofthesethetimeforlearningaseriesmayoccasionallybeincreasedtodoublethatofitsaveragevalueormore。Theoppositeeffect，thatofanunusualexertion，cannotintheverynatureofthecaseoverstepacertainlimit。Itcanneverreducethelearningtimetozero。
　　If，however，groupsofseriesequalinnumberandlearnedinimmediatesuccessionaretaken，thesedisturbinginfluencesmaybeconsideredtohavedisappearedorpracticallyso。Thedecreaseinmentalvigourinonegroupwillbepracticallythesameasthatinanother。Thepositiveandnegativefluctuationsofattentionwhichunderlikeconditionsoccurduringaquarterorhalfhourareapproximatelythesamefromdaytoday。Allthatisnecessarytoask，then，is：Dothetimesnecessaryforlearningequalgroupsofseriesexhibitthedesireddistribution？
　　Icananswerthisquestionintheaffirmativewithsufficientcertainty。
　　Thetwolongestseries，obtainedunderconditionssimilartoeachother，whichIpossess，are，tobesure，notlargeintheabove-mentionedtheoreticalsense;theysuffer，moreover，fromthedisadvantagethattheyoriginatedattimesseparatedbycomparativelylongintervalsduringwhichtherewerenecessarilymanychangesintheconditions。Inspiteofthis，theirgroupingcomesasnearascouldbeexpectedtotheonedemandedbythetheory。
　　Thefirsttestseriestakenduringtheyears1879-80comprises92tests。
　　Eachtestconsistedinmemorisingeightseriesof13syllableseach，whichprocessoflearningwascontinueduntiltworeproductionsofeachserieswerepossible。Thetimerequiredforalleightseriestakentogetherincludingthetimeforthetworeproductions（butofcoursenotforthepauses，seep。25，4）amountedtoanaverageof1，112secondswithaprobableerrorofobservationof±76。Thefluctuationsoftheresultswere，therefore，verysignificant：onlyhalfofthenumbersobtainedfellbetweenthelimits1，036and1，188，theotherhalfwasdistributedaboveandbelowtheselimits。Indetailthegroupingofthenumbersisasfollows：
　　Intheinterval，1/4P。E。to1/2P。E。，thereoccursaslightpilingupofvalueswhichiscompensatedforbyagreaterlackinthesucceedinginterval，1/2P。E。toP。E。Apartfromthis，thecorrespondencebetweenthecalculatedandtheactualresultsissatisfactory。Thesymmetryofthedistributionleavessomethingtobedesired。Thevaluesbelowtheaveragepreponderatealittleinnumber，thoseabovepreponderatealittleinamountofdeviation：onlytwoofthelargesteightdeviationsarebelowthemeanvalue。Theinfluenceofattentionreferredtoabove，thefluctuationsofwhichintheseparateseriesshowgreaterdeviationstowardtheupperlimitthantowardthelower，hasnot，therefore，beenquitecompensatedbythecombinationofseveralseries。
　　Thecorrectnessoftheobservationsandthecorrespondenceoftheirdistributionswiththeonetheoreticallydemandedaregreatlyimprovedinthesecondlargeseriesoftests。Thelattercomprisestheresultsof84seriesofteststakenduringtheyears1883-84。Eachtestconsistedinmemorisingsixseriesof16syllableseach，carriedonineachcasetothefirsterrorlessreproduction。Thewholetimenecessaryforthisamountedto1，261secondswiththeprobableerrorofobservationof±48。4——i。e。，halfofallthe84numbersfellwithinthelimits1，213-1，309。Theexactnessofthsobservationsthushadgreatlyincreasedascomparedwiththeformerseriesoftests：[1]
　　Theintervalincludedbytheprobableerroramountstoonly71/2percentofthemeanvaluesasagainst14percentintheearliertests。Indetailthenumbersaredistributedasfollows：
　　Thesymmetryofdistributionisheresatisfactorilymaintainedapartfromthenumbers，whichareunimportantonaccountoftheirsmallness。
　　Thedeviationwhichisgreatestabsolutelyistowardthelowerlimit。
　　Ifseveralofourseriesofsyllableswerecombinedintogroupsandthenmemorisedseparately，thelengthoftimenecessarytomemoriseawholegroupvariedgreatly，tobesure，whenrepeatedtestsweretaken;but，inspiteofthis，whentakenasawholetheyvariedinamannersimilartothatofthemeasuresoftheideallyhomogeneousprocessesofnaturalscience，whichalsovaryfromeachother。So，atleastinexperimentalfashion，itisallowabletousethemeanvaluesobtainedfromthenumericalresultsforthevarioustestsinordertoestablishtheexistenceofcausalrelationsjustasnaturalsciencedoesthatbymeansofitsconstants。
　　Thenumberofseriesofsyllableswhichistobecombinedintoasinglegroup，ortest，isnaturallyindeterminate。Itmustbeexpected，however，thatasthenumberincreases，thecorrespondencebetweenthedistributionofthetimesactuallyfoundandthosecalculatedinaccordancewiththelawoferrorswillbegreater。Inpracticetheattemptwillbemadetoincreasethenumbertosuchapointthatfurtherincreaseandtheclosercorrespondenceresultingwillnolongercompensateforthetimerequired。
　　Ifthenumberoftheseriesinagiventestislessened，thedesiredcorrespondencewillalsopresumablydecrease。However，itisdesirablethateventhentheapproximationtothetheoreticallydemandeddistributionremainperceptible。
　　Eventhisrequirementisfulfilledbythenumericalvaluesobtained。
　　Inthetwolargestseriesoftestsjustdescribed，Ihaveexaminedthevaryinglengthoftimenecessaryforthememorisationofthefirsthalfofeachtest。Intheolderseries，thesearetheperiodsrequiredbyeach4seriesofsyllables，inthemorerecentseriestheperiodsrequiredbyeach3ofthemtakentogether。Theresultsareasfollows：
　　1。Intheformerseries：meanvalue（m）=533（P。E。o）
　　=±51。
　　2。Inthelaterseries：m=620，P。E。o=±44。
　　Bybothtablesthesuppositionmentionedaboveoftheexistenceofalessperfectbutstillperceptiblecorrespondencebetweentheobservedandcalculateddistributionofthenumbersiswellconfirmed。
　　Exactlythesameapproximatecorrespondencemustbepresupposedif，insteadofdecreasingthenumberofseriescombinedintoatest，thetotalnumberoftestsismadesmaller。InthiscasealsoIwilladdsomeconfirmatorysummaries。
　　Ipossesstwolongtestseries，madeatthetimeoftheearliertests，whichwereobtainedunderthesameconditionsastheabovementionedseriesbutatthelatertimesoftheday，B。andC。
　　Oneofthese，B，comprised39testsof6serieseach，theother，C，38testsof8serieseach，eachseriescontaining13syllables。Theresultsobtainedwereasfollows：ForthetestsattimeB：m=871，P。E。o=±63。ForthetestoftimeC：m=1，258，P。E。o=±60。
　　InadditionImentionaseriesofonlytwentytests，withwhichIshallconcludethissummary。Eachtestconsistedofthelearningofeightseparateseriesofthirteensyllableseach，whichhadbeenmemorisedonceonemonthbefore。Theaveragewasinthiscase892secondswithaprobableerrorofobservationof54。Thesinglevaluesweregroupedasfollows：
　　Althoughthenumberofthetestswassosmall，theaccordancebetweenthecalculationbytheoryandtheactualcountofdeviationsisinallthesecasessoclosethattheusefulnessofthemeanvalueswillbeadmitted，thewidelimitsoferrorbeing，ofcourse，takenintoconsideration。Section18。GroupingoftheResultsoftheSeparateSeriesThepreviouslymentionedhypothesesconcerningthegroupingofthetimesnecessaryforlearningtheseparateserieswerenaturallynotmerelytheoreticalsuppositions，buthadalreadybeenconfirmedbythegroupingsactuallyfound。Thetwolargeseriesoftestsmentionedabove，oneconsistingof92testsofeightsingleserieseach，andtheotherof84
　　testsof6singleserieseach，thusgiving736and504separatevaluesrespectively，affordasufficientlybroadbasisforjudgment。Bothgroupsofnumbers，andbothinthesameway，showthefollowingpeculiarities：
　　1。Thedistributionofthearithmeticalvaluesabovetheirmeanisconsiderablylooserandextendsfartherthanbelowthemean。Themostextremevaluesabovelie2timesand1。8times，respectively，asfarfromthemeanasthemostdistantofthosebelow。
　　2。Asaresultofthisdominancebythehighernumbersthemeanisdisplacedupwardfromtheregionofthedensestdistribution，andasaresultthedeviationsbelowgetthepreponderanceinnumber。Thereoccurrespectively404and266deviationsbelowasagainst329and230above。
　　3。Thenumberofdeviationsfromtheregionofdensestdistributiontowardsbothlimitsdoesnotdecreaseuniformly——asonewouldbeverymuchinclinedtoexpectfromtherelativelylargenumberscombined——butseveralmaximaandminimaofdensityaredistinctlynoticeable。Thereforeconstantsourcesoferrorwereatworkintheproductionoftheseparatevalues——i。e。，inthememorisationoftheseparateseries。Theseresultedontheonehandinanunsymmetricaldistributionofthenumbers，andontheotherhandinanaccumulationofthemincertainregions。Inaccordancewiththeinvestigationsalreadypresentedinthischapter，itcanonlybesupposedthattheseinfluencescompensatedeachotherwhenthevaluesofseveralserieslearnedinsuccessionwerecombined。
　　Ihavealreadymentionedastheprobablecauseofthisunsymmetricaldistributionthepeculiarvariationsintheeffectofhighdegreesofconcentrationofattentionanddistraction。Itwouldnaturallybesupposedthatthepositionoftheseparateserieswithineachtestisthecauseoftherepeatedpilingupofvaluesoneachsideoftheaverage。If，inthecaseofalargetest-series，thevaluesaresummedupforthefirst，thesecond，andthirdseries，etc。，andtheaverageofeachistaken，theseaveragevaluesvarygreatly，asmightbeexpected。Theseparatevaluesaregroupedabouttheirmeanwithonlytolerableapproximationtothelawoferror，butyettheyare，onthewhole，distributedmostdenselyinitsregion，andtheseseparateregionsofdensedistributionmustofcourseappearinthetotalresult。
　　Thefollowingmaybeaddedbywayofsupplement：onaccountofthementalfatiguewhichincreasesgraduallyduringthecourseofatest-seriesthemeanvaluesoughttoincreasewiththenumberoftheseries;butthisdoesnotprovetobethecase。
　　OnlyinonecasehaveIbeenabletonoticeanythingcorrespondingtothishypothesis，namely，inthelargeandthereforeimportantseriesof92testsconsistingofeightseriesof13syllableseach。Inthiscasethemeanvaluesforthelearningofthe92firstseries，the92secondseries，etc。，werefoundtobe105，140，142，146，148，140seconds，therelativelengthsofwhichFig2exhibits。ForalltherestofthecaseswhichIinvestigatedthetypicalfactis，onthecontrary，rathersuchacourseofthenumbersaswastrueinthecaseoftheseriesof84testsofsixseriesof16syllableseachandasisshowninFig。3。
　　Themeanvaluesherewere191，224，206，218，210，213seconds。Theystartin，asmaybeseen，considerablybelowtheaverage，butriseimmediatelytoaheightwhichisnotagainreachedinthefurthercourseofthetest，andtheythenoscillateratherdecidedly。Ananalogouscourseisshownbythenumbersinthe7testsofnine12-syllableseries，namely：71，90，98，87，98，90，101，86，69（Fig。4）。
　　Furthermorethevaluesfor39testsofsixseriesof13syllableseachobtainedintimeBwereasfollows：118，150，158，147，155，144（Fig。
　　5lowercurve）。
　　Thosefor38testswitheight13syllableseriesoftimeCwere139，159，167，168，160，150，162，153（Fig。5uppercurve）。
　　FinallythenumbersobtainedfromseventestswithsixstanzasofByron’s"DonJuan"were：189，219，171，204183，229。
　　Eveninthecaseofthefirstmentionedcontradictorygroupoftestsagroupingoftheseparatemeanvaluesharmonisingwiththenormaloneoccursif，insteadofallthe92testsbeingtakenintoconsiderationatonce，theyaredividedintoseveralparts——i。e。，iftestsarecombinedwhichweretakenataboutthesametimeandunderaboutthesameconditions。
　　Theconclusioncannotbedrawnfromthesenumericalresultsthatthementalfatiguewhichgraduallyincreasedduringthetwentyminutedurationofthetestsdidnotexertanyinfluence。
　　Itcanonlybesaidthatthesupposedinfluenceofthelatteruponthenumbersisfaroutweighedbyanothertendencywhichwouldnotaprioribesoreadilysuspected，namelythetendencyofcomparativelylowvaluestobefollowedbycomparativelyhighonesandviceversa。Thereseemstoexistakindofperiodicaloscillationofmentalreceptivityorattentioninconnectionwithwhichtheincreasingfatigueexpressesitselfbyfluctuationsaroundamedianpositionwhichisgraduallydisplaced。[2]
　　Afterorientingourselvesthusconcerningthenatureandvalueofthenumericalresultsgainedfromthecompletememorisations，weshallnowturntotherealpurposeoftheinvestigation，namelythenumericaldescriptionofcausalrelations。
　　Footnotes[1]Ofcourse，theexactnessobtainedherecannotstandcomparisonwithphysicalmeasurements，butitcanverywellbecomparedwithphysiologicalones，whichwouldnaturallybethefirsttobethoughtofinthisconnection。TothemostexactofphysiologicalmeasurementsbelongthelastdeterminationsofthespeedofnervoustransmissionmadebyHelmholtzandBaxt。Onerecordoftheseresearchespublishedasanillustrationoftheiraccuracy（Mon。Ber，d。Berl。Akad。1870，S。191）afterpropercalculationgivesameanvalueof4。268withtheprobableerrorofobservation，0。101。Theintervalitincludesamounts，therefore，to5percentofthemeanvalue。Allformerdeterminationsaremuchmoreinaccurate。Inthecaseofthemostaccuratetest-seriesofthefirstmeasurementsmadebyHelmholtz，thatintervalamountstoabout50percentofthemeanvalue（Arch。f。Anat。u……Physiol。1850，S。340）。EvenPhysics，inthecaseofitspioneerinvestigations，hasoftenbeenobligedtoputupwithalessdegreeofaccuracyinitsnumericalresults。InthecaseofhisfirstdeterminationsofthemechanicalequivalentofheatJoulefoundthenumber838，withaprobableerrorofobservationof97。（Phil。Mag。，1843，p。435ff。）
　　[2]Ifitshouldeverbecomeamatterofinteresttheattemptmightbemadetodefinenumericallythedifferenteffectsofthattendencyindifferentcases。Fortheprobableerrorsofobservationforthenumericalvaluesofseries-groupsaffordameasurefortheinfluenceofaccidentaldisturbancestowhichthememorisationisdailyexposed。
　　Ifnowthelearningoftheseparateseriesingeneralwereexposedtothesameofsimilarvariationsofconditionasoccurfromtesttotest，thenaccordingtothefundamentalprinciplesofthetheoryoferrors，aprobableerrorofobservationcalculateddirectlyfromthespearatevalueswouldrelateitselftotheonejustmentionedas1to[thesquarerootof]n，where"n"denotesthenumberofseparateseriescombinedintoatest。Ifhowever，asisthecasehere，specialinfluencesassertthemselvesduringthememorisationoftheseseparateseries，andifsuchinfluencestendtoseparatethevaluesfurtherthanothervariationsofconditionswoulddo，the"P。E。0"calculatedfromtheseparatevaluesmustturnoutsomewhattoogreat，andthejustmentionedproportionconsequentlytoosmall，andthestrongertheinfluencesare，themoremustthisbethecase。
　　Anexaminationoftheactualrelationsis，tobesure，alittledifficult，butfullyconfirmsthestatements。Inthe84tests，consistingofsixseriesof16syllableseach，the[squarerootof]n=2。45。Wefound48。4tobetheprobableerrorofobservationofthe84tests。Theprobableerrorofthe504separatevaluesis31。6。Thequotient31。6：48。4is1。53;thereforenotquite2/3ofthevalueof[thesquarerootof]n。
　　ClassicsintheHistoryofPsychology——Ebbinghaus（1885）Chapter5Memory：AContributiontoExperimentalPsychologyHermannEbbinghaus（1885）TranslatedbyHenryA。Ruger&ClaraE。Bussenius（1913）CHAPTERVRAPIDITYOFLEARNINGSERIESOFSYLLABLESASAFUNCTIONOFTHEIR
　　LENGTHSection19。TestsBelongingtotheLaterPeriodItissufficientlywellknownthatthememorisationofaseriesofideasthatistobereproducedatalatertimeismoredifficult，thelongertheseriesis。Thatis，thememorisationnotonlyrequiresmoretimetakenbyitself，becauseeachrepetitionlastslonger，butitalsorequiresmoretimerelativelybecauseanincreasednumberofrepetitionsbecomesnecessary。
　　Sixversesofapoemrequireforlearningnotonlythreetimesasmuchtimeastwobutconsiderablymorethanthat。
　　Ididnotinvestigateespeciallythisrelationofdependence，whichofcoursebecomesevidentalsointhefirstpossiblereproductionofseriesofnonsensesyllables，butincidentallyIobtainedafewnumericalvaluesforitwhichareworthputtingdown，althoughtheydonotshowparticularlyinterestingrelations。
　　Theseriesinquestioncomprised（inthecaseofthetestsoftheyear1883-84），12，16，24，or36syllableseach，and9，6，3，or2serieswereeachtimecombinedintoatest。
　　Forthenumberofrepetitionsnecessaryinthesecasestomemorisetheseriesuptothefirsterrorlessreproduction（andincludingit）thefollowingnumericalresultswerefound：
　　Inordertomakethenumberofrepetitionscomparableitisnecessary，sotospeak，toreducethemtoacommondenominatorandtodividethemeachtimebythenumberoftheseries。Inthiswayitisfoundouthowmanyrepetitionsrelativelywerenecessarytolearnbyheartthesingleseries，whichdifferfromeachotheronlyinthenumberofsyllables，andwhicheachtimehadbeentakentogetherwithasmanyothersofthesamekindaswouldmakethedurationofthewholetestfromfifteentothirtyminutes。[1]
　　However，aconclusioncanbedrawnfromthefiguresfromthestandpointofdecreaseinnumberofsyllablesThequestioncanheasked：Whatnumberofsyllablescanbecorrectlyrecitedafteronlyonereading？Formethenumberisusuallyseven。IndeedIhaveoftensucceededinreproducingeightsyllables，butthishashappenedonlyatthebeginningofthetestsandinadecidedminorityofthecases。Inthecaseofsixsyllablesontheotherhandamistakealmostneveroccurs;withthem，therefore，asingleattentivereadinginvolvesanunnecessarilylargeexpenditureofenergyforanimmediatelyfollowingreproduction。
　　Ifthislatterpairofvaluesisadded，therequireddivisionmade，andthelastfaultlessreproductionsubtractedasnotnecessaryforthelearning，thenthefollowingtableresults。
　　NumberofsyllablesinaseriesNumberofrepetitionsnecessaryforfirsterrorlessreproduction（exclusiveofit）Probableerror71
　　1216。6±1。1
　　1630。0±0。4
　　2444。0±1。7
　　3955。0±2。8ThelongerofthetwoadjoiningcurvesofFig。6illustratestheregularcourseofthesenumberswithapproximateaccuracyforsuchasmallnumberoftests。AsFig。6shows，inthecasesexamined，thenumberofrepetitionsnecessaryforthememorisationofseriesinwhichthenumberofsyllablesprogressivelyincreased，itselfincreaseswithextraordinaryrapiditywiththeincreaseinnumberofthesyllables。
　　Atfirsttheascentofthecurveisverysteep，butlateronitappearstograduallyflattenout。Forthemasteryoffivetimesthenumberofsyllablesthatcanbereproducedafterbutonereading——i。e。，afterabout3secondsover50repetitionswerenecessary，requiringanuninterruptedandconcentratedeffortforfifteenminutes。
　　Thecurvehasitsnaturalstartingpointinthezeropointoftheco-ordinates。
　　Theshortinitialstretchuptothepoint，x=7，y=1，canbeexplainedthus：
　　inordertorecitebyheartseriesof6，5，4，etc。，syllablesonereading，ofcourse，isallthatisnecessary。Inmyeasethisreadingdoesnotrequireasmuchattentionasdoesthe7-syllableone，butcanbecomemoreandmoresuperficialasthenumberofsyllablesdecreases。Section20。TestsBelongingtotheEarlierPeriodItgoeswithoutsayingthatsincetheresultsreportedwereobtainedfromonlyonepersontheyhavemeaningonlyasrelatedtohim。Thequestionariseswhethertheyareforthisindividualofageneralsignificance——i。e。，whether，byrepetitionofthetestsatanothertime，theycouldbeexpectedtoshowapproximatelythesameamountandgrouping。
　　Aseriesofresultsfromtheearlierperiodfurnishesthedesiredpossibilityofacontrolinthisdirection。They，again，havebeenobtainedincidentally（consequentlyuninfluencedbyexpectationsandsuppositions）andfromtestsmadeunderdifferentconditionsthanthosementioned。Theseearliertestsoccurredatanearlierhourofthedayandthelearningwascontinueduntiltheseparateseriescouldberecitedtwiceinsuccessionwithoutmistake。
　　Atestcomprised15seriesof10syllableseach，or8""13""
　　or6""16""
　　or4""19""So，again，fourdifferentlengthsofserieshavebeentakenintoaccount，buttheirseparatevaluesliemuchclosertogether。
　　Sincetherepetitions——whichareinquestionhere——werenotcountedatallintheearlierperiod，theirnumberhadtobecalculatedfromthetimes。Forthispurposethetableonp。31hasbeenusedaftercorrespondinginterpolation。Ifthenumbersfoundareimmediatelyreducedtooneserieseach，andifalongwithitthetworepetitionsrepresentingtherecitationaresubtractedasabove，weobtain：
　　ThesmallercurveofFig。6exhibitsgraphicallythearrangementofthesenumbers。Asmaybeseen，thenumberofrepetitionsnecessaryforlearningequallylongserieswasalittlelargerintheearlierperiodthaninthelaterone。Becauseofitsuniformitythisrelationistobeattributedtodifferencesintheexperimentalconditions，toinaccuraciesinthecalculations，andperhapsalsototheincreasedtrainingofthelaterperiod。Theoldernumbersfallveryclosetothepositionofthelaterones，and——whatisofchiefimportance——thetwocurveslieascloselytogetherthroughouttheshortextentoftheircommoncourseascouldbedesiredfortestsseparatedby31/2yearsandunaffectedbyanypresuppositions。Thereisahighdegreeofprobability，then，infavorofthesuppositionthattherelationsofdependencepresentedinthosecurves，sincetheyremainedconstantoveralongintervaloftime，aretobeconsideredascharacteristicforthepersonconcerned，althoughtheyare，tobesure，onlyindividual。Section21。IncreaseinRapidityofLearningintheCaseofMeaningfulMaterialInordertokeepinmindthesimilaritiesanddifferencesbetweensenseandnonsensematerial，IoccasionallymadetestswiththeEnglishoriginalofByron’s"DonJuan。"TheseresultsdonotproperlybelongheresinceIdidnotvarythelengthoftheamounttobelearnedeachtimebutmemorisedoneachoccasiononlyseparatestanzas。Nevertheless，itisinterestingtomentionthenumberofrepetitionsnecessarybecauseoftheircontrastwiththenumericalresultsjustgiven。
　　Thereareonlyseventests（1884）tobeconsidered，eachofwhichcomprisedsixstanzas。Whenthelatter，eachbyitself，werelearnedtothepointofthefirstpossiblereproduction，anaverageof52repetitions（P。E。m=±0。6）wasnecessaryforallsixtakentogether。Thus，eachstanzarequiredhardlyninerepetitions;or，iftheerrorlessreproductionisabstracted，scarcelyeightrepetitions。[2]
　　Ifitisborninmindthateachstanzacontains80syllables（eachsyllable，however，consistingontheaverageoflessthanthreeletters）andifthenumberofrepetitionsherefoundiscomparedwiththeresultspresentedabove，thereisobtainedanapproximatenumericalexpressionfortheextraordinaryadvantagewhichthecombinedtiesofmeaning，rhythm，rhyme，andacommonlanguagegivetomaterialtobememorised。Iftheabovecurveisprojectedinimaginationstillfurtheralongitspresentcourse，thenitmustbesupposedthatIwouldhaverequired70to80repetitionsforthememorisationofaseriesof80to90nonsensesyllables。Whenthesyllableswereobjectivelyandsubjectivelyunitedbythetiesjustmentionedthisrequirementwasinmycasereducedtoaboutone-tenthofthatamount。
　　Footnotes[1]Theobjectionmightbemadethat，bymeansofthisdivision，recoursemadedirectlytotheaveragesforthememorisingofthesingleseries，andthatinthiswaytheresultoftheFourthChapterisdisregarded。For，accordingtothat，theaveragesofthenumbersobtainedfromgroupsofseriescouldindeedbeusedforinvestigationintorelationsofdependence，buttheaveragesobtainedfromseparateseriescouldnotbesoused。Idonotclaim，however，thattheabovenumbers，thusobtainedbydivision，formthecorrectaverageforthenumbersbelongingtotheseparateseries，i。e。，thatthelattergroupthemselvesaccordingtothelawoferrors。Butthenumbersaretobeconsideredasaveragesforgroupsofseries，and，forthesakeofabettercomparisonwithothers——aconditionwhichinthenatureofthecasecouldnotbeeverywherethesame——ismadethesamebydivision。Theprobableerror，themeasureoftheiraccuracy，hasnotbeencalculatedfromthenumbersfortheseparateseriesbutfromthoseforthegroupsofseries。
　　[2]Forthesakeofcorrectevaluationofthenumbersandcorrectconnectionwithpossibleindividualobservationspleasenotep。24，1。Inordertoprocureuniformityofmethodthestanzaswerealwaysreadthroughfrombeginningtoend;moredifficultpassageswerenotlearnedseparatelyandtheninserted。Ifthathadbeendone，thetimeswouldhavebeenmuchshorterandnothingcouldhavebeensaidaboutthenumberofrepetitions。Ofcoursethereadingwasdoneatauniformrateofspeedasfaraspossible，butnotintheslowandmechanicallyregulatedtimethatwasemployedfortheseriesofsyllables。Theregulationofspeedwaslefttofreeestimation。Asinglereadingofonestanzarequired20
　　to23seconds。
　　ClassicsintheHistoryofPsychology——Ebbinghaus（1885）Chapter7Memory：AContributiontoExperimentalPsychologyHermannEbbinghaus（1885）TranslatedbyHenryA。Ruger&ClaraE。Bussenius（1913）CHAPTERVIIRETENTTONANDOBLIVISCENCEASAFUNCTIONOFTHETIMESection26。Exp1anationsofRetentionandObliviscenceAllsortsofideas，iflefttothemselves，aregraduallyforgotten。
　　Thisfactisgenerallyknown。Groupsorseriesofideaswhichatfirstwecouldeasilyrecollectorwhichrecurredfrequentlyoftheirownaccordandinlivelycolors，graduallyreturnmorerarelyandinpalercolors，andcanbereproducedbyvoluntaryeffortonlywithdifficultyandinpart。
　　Afteralongerperiodeventhisfails，except，tobesure，inrareinstances。
　　Names，faces，bitsofknowledgeandexperiencethathadseemedlostforyearssuddenlyappearbeforethemind，especiallyindreams，witheverydetailpresentandingreatvividness;anditishardtoseewhencetheycameandhowtheymanagedtokeephiddensowellinthemeantime。Psychologists——eachinaccordancewithhisgeneralstandpoint——interpretthesefactsfromdifferentpointsofview，whichdonotexcludeeachotherentirelybutstilldonotquiteharmonise。Oneset，itseems，laysmostimportanceontheremarkablerecurrenceofvividimagesevenafterlongperiods。Theysupposethatoftheperceptionscausedbyexternalimpressionsthereremainpaleimages，"traces，"which，althoughineveryrespectweakerandmoreflightythantheoriginalperceptions，continuetoexistunchangedintheintensitypossessedatpresent。Thesementalimagescannotcompetewiththemuchmoreintenseandcompactperceptionsofreallife;butwherethelatteraremissingentirelyorpartly，theformerdomineerallthemoreunrestrainedly。Itisalsotruethattheearlierimagesaremoreandmoreoverlaid，sotospeak，andcoveredbythelaterones。Therefore，inthecaseoftheearlierimages，thepossibilityofrecurrenceoffersitselfmorerarelyandwithgreaterdifficulty。Butif，byanaccidentalandfavorablegroupingofcircumstances，theaccumulatedlayersarepushedtooneside，then，ofcourse，thatwhichwashiddenbeneathmustalppear，afterwhateverlapseoftime，initsoriginalandstillexistentvividness。[1]
　　Forothers[2]theideas，thepersistingimages，sufferchangesmoreandmoreaffecttheirnature;theconceptofobscurationcomesinhere。
　　Olderideasarerepressedandforcedtosinkdown，sotospeak，bythemorerecentones。Astimepassesoneofthesegeneralqualities，innerclearnessandintensityofconsciousness，suffersdamage。Connectionsofideasandseriesofideasaresubjecttothesameprocessofprogressiveweakening;itisfurtheredbyaresolutionoftheideasintotheircomponents，asaresultofwhichthenowbutlooselyconnectedmembersareeventuallyunitedinnewcombinations。Thecompletedisappearanceofthemoreandmorerepressedideasoccursonlyafteralongtime。Butoneshouldnotimaginetherepressedideasintheirtimeofobscurationtobepaleimages，butrathertobetendencies，"dispositions，"torecreatetheimagecontentsforcedtosinkdown。Ifthesedispositionsaresomehowsupportedandstrengthened，itmayhappenatanyonemomentthattherepressingandhinderingideasbecomedepressedthemselves，andthattheapparentlyforgottenideaarisesagaininperfectclearness。