博文
ALPAC 黑皮书 9/9： 附录 16-20
|||
Appendix16
GovernmentSupportofMachine-TranslationResearch
NATIONALSCIENCEFOUNDATION
OfficeofScienceInformationServices,InformationSystemsProgram
1. CambridgeLanguageResearch Unit
GrantNumber | Date | NSF | Transferred | Total |
GN3398 | 3-29-57 | $7,100 | $20,000(RADC) | $27,100 |
GN4788 | 12-31-57 | 13,000 | 20,000(RADC) | 33,000 |
GN8212 | 4-3-59 | 15,650 | 20,000(RADC) | 35,650 |
GN8212.1 | 5-6-60 | – | 5,500(RADC) | 5,500 |
|
| $35,750 | $65,500 | $101,250 |
2. GeorgetownUniversity
GrantNumber | Date | NSF | Transferred | Total |
G2723 | 6-29-56 | $35,000 | $65,000(CIA) | $100,000 |
G3867 | 6-6-57 | 35,000 | 90,000(CIA) | 125,000 |
G5513 | 6-6-58 | 36,600 | 150,000(CIA) | 186,600 |
|
| $106,600 | $305,000 | $411,600 |
3. Harvard University
GrantNumber | Date | NSF | Transferred | Total |
GN4982 | 1-31-58 | $14,150 | $15,000(RADC) | $29,150 |
G5514 | 6-6-58 | 26,200 |
| 26,200 |
G6400 | 9-23-58 | 150,000 | 70,000(RADC) | 220,000 |
G10636 | 12-11-59 | 100,000 | 100,000(RADC) | 200,000 |
G15924 | 12-29-60 | 128,500 | 21,500(RADC) | 150,000 |
G24833 | 6-30-62 | 160,160 |
| 160,160 |
GN162 | 6-29-63 | 235,450 |
| 235,450 |
GN329 | 6-25-64 | 240,500 |
| 240,500 |
|
| $1,054,960 | $206,500 | $1,261,460 |
1. MassachusettsInstituteofTechnology
GrantNumber | Date | NSF | Total |
G1210 | 10-28-54 | $18,700 | $18,700 |
G2044 | 10-25-55 | 24,800 | 24,800 |
G3031 | 10-23-56 | 35,200 | 35,200 |
G4378 | 9-30-57 | 41,400 | 41,400 |
G6537 | 11-3-58 | 90,600 | 90,600 |
G10130 | 10-26-59 | 126,000 | 126,000 |
G16843 | 3-3-61 | 150,000 | 150,000 |
G24047 | 6-6-62 | 225,000 | 225,000 |
GN244 | 1-22-64 | 200,000 | 200,000 |
|
| $911,700 | $911,700 |
2. UniversityofCalifornia,Berkeley
GrantNumber | Date | NSF | Total |
G6399 | 9-30-58 | $40,500 | $40,500 |
G8737 | 6-12-59 | 57,600 | 57,600 |
G14147 | 8-15-60 | 208,000 | 208,000 |
GN92 | 2-1-63 | 249,000 | 249,000 |
GN306 | 6-8-64 | 167,300 | 167,300 |
|
| $722,400 | $722,400 |
3. OhioState University
GrantNumber | Date | NSF | Total |
G18609 | 6-16-61 | $14,700 | $14,700 |
G25055 | 6-30-62 | 40,000 | 40,000 |
GN174 | 6-24-63 | 100,000 | 100,000 |
|
| $154,700 | $154,700 |
4. Wayne StateUniversity
GrantNumber | Date | NSF | Total |
GN159 | 6-15-63 | $200,000 | $200,000 |
GN430 | 6-11-65 | 244,000 | 244,000 |
|
| $444,000 | $444,000 |
5. Ramo-Wooldridge
ContractNumber | Date | NSF | Total |
C233 | 10-2-61 | $119,477 | $119,477 |
ThompsonRamo-Wooldridge
C233(Amend) | 3-1-63 | 152,084 | 152,084 |
C320 | 8-20-63 | 50,223 | 50,223 |
Bunker-RamoCorp.
C372 6-30-64 $240,000 $240,000
$561,784
1. UniversityofTexas
GrantNumber | Date | NSF | Total |
G19277 | 8-18-61 | $95,000 | $95,000 |
GN54 | 9-27-62 | 200,000 | 200,000 |
GN208 | 10-24-63 | 150,000 | 150,000 |
GN308 | 6-18-64 | 168,200 | 168,200 |
|
| $613,200 | $613,200 |
2. UniversityofPennsylvania
GrantNumber | Date | NSF | Total |
G3027 | 10-16-56 | $1,950 | $1,950 |
G3397 | 2-1-57 | 24,300 | 24,300 |
G4981 | 2-15-58 | 42,300 | 42,300 |
G6538 | 10-24-58 | 31,450 | 31,450 |
G8217 | 6-15-59 | 321,800 | 321,800 |
G17446 | 4-28-61 | 180,400 | 180,400 |
G24340 | 6-5-62 | 346,000 | 346,000 |
GN311 | 6-11-64 | 414,000 | 414,000 |
|
| $1,362,200 | $1,362,200 |
3. National BureauofStandards
GrantNumber | Date | NSF | Total |
G17815 | 6-7-61 | $15,000 | $15,000 |
G19659 | 10-3-61 | 73,000 | 73,000 |
GN107 | 3-26-63 | 75,000 | 75,000 |
GN320 | 6-29-64 | 58,200 | 58,200 |
|
| $221,200 | $221,200 |
4. UniversityofChicago(Yngve)
GrantNumber | Date | NSF | Total |
GN412 | 5-22-65 | $294,000 | $294,000 |
5. National Academy of Sciences, Automatic Language Processing AdvisoryCommittee
ContractNumber | Date | NSF | Transferred | Total |
C310 T.O.80 | 4-20-64 | $19,000
$19,000 | $20,000(CIA) 20,000(RADC) $40,000 | $59,000
$59,000 |
1. Linguistic SocietyofAmerica,MIT (Conference)
GrantNumber | Date | NSF | Total |
G11302 | 2-8-60 | $15,000 | $15,000 |
2. Wayne StateUniversity(Conference)
GrantNumber | Date | NSF | Transferred | Total |
G12887 | 5-12-60 | $3,938 | $1,000(ONR) | $4,938 |
G15859 | 12-16-60 | 3,328 |
| 3,328 |
G22890 | 3-27-62 | 357 | 5,000(RADC) | 5,357 |
|
| $7,623 | $6,000 | $13,623 |
3. MassachusettsInstitute ofTechnology (Conference)
GrantNumber | Date | NSF | Total |
G2337 | 5-1956 | $1,059 | $1,059 |
G2888 | 10-1956 | 5,351 | 5,351 |
|
| $6,410 | $6,410 |
4. UniversityofWashington
GrantNumber | Date | NSF | Total |
G13579.1 | FY-62 | $1,000 | $1,000 |
G13579 | FY-60 | 53,700 | 53,700 |
|
| $54,700 | $54,700 |
TOTALNSFSUPPORT: $6,585,227
TOTALTRANSFERREDFUNDS: $623,000
CENTRALINTELLIGENCEAGENCY
GeorgetownUniversity
GrantNumber | Date | Total |
NSFG5513Supplement | 6-6-58 | $9,890 |
XG2230 | 7-1-59 | 24,979 |
XG2239 | 7-16-59 | 153,000 |
XG2312 | 7-1-60 | 439,000 |
XG2427 | 9-1-61 | 438,000 |
Supplementto3-31-63 |
| 250,000 |
|
| $1,314,869 |
Note:OtherCIAfundsinsupportoftheGeorgetownmachine-translationproject(amountingto$205,000)weretransferredtoNSF.Seeabove.
DEPARTMENT OFDEFENSE
1. UnitedStatesAirForce
FiscalYear
1956 |
| $400,000 |
1957 |
| 700,000 |
1958 |
| 800,000 |
1959 |
| 1,500,000 |
1960 |
| 1,400,000 |
1961 |
| 927,000 |
1962 |
| 561,000 |
1963 |
| 600,000 |
1964 |
| 2,045,000 |
1965 |
| 680,000 |
Total |
| $9,613,000 |
2. | UnitedStatesNavy |
|
FiscalYear |
| |
1953-1960 | $416,600 | |
1961 | 50,000 | |
1962 | 75,000 | |
1963 | 130,000 | |
1964 | 150,000 | |
1965 | 150,000 | |
Total | $971,600 | |
3. UnitedStatesArmy
FiscalYear |
|
1958-1959 | $184,000 |
1960 | 223,000 |
1961 | 225,000 |
1962 | 110,000 |
1963 | 175,000 |
1964 | 230,000 |
1965 | 175,000 |
Total | $1,322,000 |
TOTALDEPARTMENTOFDEFENSESUPPORT: 11,906,600 | |
DOD | $11,906,600 |
CIA | 1,314,869 |
NSF | 6,585,227 |
GRANDTOTAL | $19,806,696 |
TheCommitteefeelsthatthesedataformthebestestimatenowavailableofgovernmentexpendituresinsupportofmachine-translationresearch.Other estimatescouldbeobtained,however,dependingontheextenttowhichonewouldincludeorexcludefundsforthesupportofworkinrelatedareasofdataprocessingandinformationtechnologyandthecostsoftheoperationoftheForeignTechnologyDivisionmechanicaltranslationfacility.Criteriaforwhatconstitutedsupportofmechanicaltranslationresearchweredeterminedbytheindividual sponsors.
Appendix17ComputerizedPublishing
Inthepast3years,sincethefirst,andunsuccessful,attempttousecomputerizedtypesettinginnewspaperproduction,theadvancesinthis technologyhavebeensuchthatabout200computersarenowinuseinoronorderbytheprintingbusinessthroughouttheworld.NearlyallthemajorU.S.computermanufacturershaveenteredthisfield,andcompetitionforthemarketiskeen.
Althoughnewspapershavebeentheprimarypractitionersofcomputerized printing,bookmanufacturersandgovernmentagencieshavealsobeguncomputerizedoperations.Initsnewspaperapplication,atypicalsystemwouldconsistofthefollowingoperations:
1. Thereportertypeshiscopyinthecustomarywayexceptthatincertainsystemstheoutputconsistsofapunchedpapertapeinadditiontotheusual hard copy.
2. Theeditorindicatesonthehardcopywhatchangeshedesirestobemade.
3. Ifthereporter'soutputwasapunchedtape,onlythenecessarycorrectionsarepunchedup.Ifonlythehardcopyexists,itispunchedupincorporating theeditor'scorrections.
4. Theeditedpunchedpapertapeisfedintothecomputer,wherewordsare hyphenatedand linesarejustifiedautomatically.
5. Thepunchedtape(sometimesmagnetictape)outputfromthecomputer isthenused to operatelinecastingorphotocompositionmachines.
6. Subsequentoperationsareessentiallynodifferentfromthoseintheconventionalprinting process.
LINE JUSTIFICATION
Thecomputeriswelladaptedforthetypeofcomputationneededforthe justificationofprintedlines.Bysimplyaddingthewidthofthecharactersandspacesineachlineandcomparingthesumwiththecolumnwidth,thecomputerisable to apply the proper spacing techniques (e.g., insertion of thin spaces, ens, ems, orhyphenation)forjustification.
WORDDIVISION
Worddivisionstillposes aprobleminthatthetwomostwidelyusedmethods(“logic”and“dictionarylook-up”)eachhavecertaindisadvantages.Thelogicalmethod,owingtothecompletelyarbitrarynatureofEnglishsyllabificationrules,cannotattain100percentaccuracy.Thedictionarylook-upmethodrequiresamuchlargercomputermemory thanthe logical method. Sinceitisunlikelythatthedisadvantagesofeithermethodcanbecompletelyovercome,anentirelydifferentapproachhasgainedthefavorofsome.Thissystem,tobeinoperationnextyearattheCIA'sPrintingServicesDivision, justifieswithoutworddivisionhyphenationbyusingaphotocomposertovarythesetsizeofthetype.Exhibit1showsan80percentreductionofthestandard GovernmentPrintingOfficeformat,whichinitsoriginalformis20picaswide andsetin10pointModernat101/2set.Itcontains15hyphens.Exhibit2isthesamejobresetusingachoiceofsetsizes.Noworddivisionhyphenationhasbeennecessary.Exhibit3isthesameasExhibit2withbulletsnexttothelineswherealternate setsizeswere used.
ADVANTAGESOFCOMPUTERIZEDPRINTING
Someoftheadvantagesthathavebeenmentionedbytheusersofthismethod ofprintingare:
1. improvedoutputbytypistsresultingfromeliminationofthespacingand hyphenationdecisions,
2. reductionoftime neededto trainnewperforator operators,
3. more efficientuse oflinecasting machines,
4. the abilityto setcloser deadlines,and
5. increase inproduction.
PHOTOCOMPOSITION
Inthefuture,photocomposingmachineswillhavetobeusedinorderto takefulladvantageofthecomputer.Thefastestline-castingmachinesarecapableofanoutputofonly15newspaperlinesaminute,whereasthenewestphotocomposingmachines arecapableofprinting 1,000-2,000linesa minute.
Appendix18
RelationBetweenProgramming LanguagesandLinguistics
EFFECT OF LINGUISTICSONPROGRAMMING
Thiseffectvariesfromperiodtoperiodofprogramminghistory(whichis veryshort).Inpre-Fortrantimestheeffectwasalmostnil since allprogrammingwasinmachinelanguageandalmostallcomputationwas scientific.
IntheperiodfromFortrantoALGOL(1956-1960)theconnectionwas almosttotallyterminological:wordsanddefinitions,butnottheoryandtechnique,wereborrowedfromlinguistics,forexample,grammarandsyntax.Thereallinkwasbetweenprogrammingandmathematicallogic,aswitnessthedevelopmentofADESlanguage1basedonrecursivefunctionsandthedevelopmentofseveralPolishprefix-orientedlanguages.Syntaxanalysisduringthisperiodwasacollectionofadhoctechniques.ThusthepaperbySheridanonFortran2isenormouslycomplex.Descriptionsofevenmorecomplexgrammarsare much more clearlyunderstandabletoday.
TheperiodfromALGOLtothepresentshowsintenseborrowingof currentmathematicallinguistictheory,technique,andnotation.ThesourceofthisdependencycanbetracedtothedefinitionofALGOL60syntaxproduction notation.Thesimilaritybetweenthisnotationandtherewritingrulesofsomelinguisticmodelscausedthistheorytoberapidlyemployedinprogramming. Still,itisimportanttonotethatthedefinitionoftheALGOLlanguagewastotallyinspiredbyprogrammingconsiderations(Fortran,LISP),andnot linguisticones.
Theeffectofthissyntaxformalismhasbeenenormousandalltothegood. ThusALGOLsyntaxis“essentially”ofType2.Hence,parsingmechanismsforType2languagescanbeappliedintheconstructionofALGOLtranslators. Manyoftheparsingtechniquesemployedwere,however,discoveredby programmersoperatinginparallelto,butindependentof,similardevelopments in mathematicallinguistics.
Theexistenceofatheoryhasmadeitpossibletodefinevariationsonagivengrammarthatpermitthesametaskspecificationbutinagrammarmoreefficientlyparsed(onepush-downstackinsteadofmany,noretracingofpathsin atreeofsyntax choices),for example, precedencegrammars.
Certainlyitisnowthecasethatthedesignofprogramminglanguages followsamorerationalprocedurethanbeforebecauseofmathematicallinguistics,andproceedsin the followingsteps:
A. Asetoftasksisisolatedandtheirinformalalgorithmicdescriptionsarespecified.
B. Thedatastructuresinherentinthisclassofproblemsareisolatedandappropriatecomputer representationsaredefined.
C. Thenaturaloperators onthe data areisolated.
D. Agrammarofincreasinglycomplexunitsisspecified,e.g.,atoms,expressions,statements, and programs.
E. A parser-recognizer isconstructedfor thegrammar.
F. ThestepsDandEareiterateduntilareasonablemixtureofflexibilityandefficiencyisattained.
G. Asemiformalstatementoftheevaluationofalgorithmsdescribedinthis languageisgiven,whichbecomesthebasisforatranslationprocess takingthislanguageintosomeothergivenlanguage(usuallymachinecode).
Itisnowpossibletoteachsyntaxanalysisofprogramminglanguages,i.e., the basic knowledgeisnow availablein anorganized form.
Itisnowpossibletoconstructprogramsthataregeneral-purposesyntax analysers inthesensethat theyparse any programminglanguage ofa given type.
EFFECT OFPROGRAMMING ONLINGUISTICS
Sinceprogrammingisan“applied”activityandlinguisticsamoreabstractone,programminghasprovidedlinguisticswith“real”modelsthataresufficientlycomplicatedto permitthe developmentofdiverse theories.
Programminghasalsoledtothedefinitionoflinguisticmodelspossessingatheoryoftheirown3andspecificallytailoredforuseasprogramminglanguages.4
Theexistenceofabodyoftechniqueinprogramminghasmadeitpossibletodevelopspecialprogramminglanguagesforsolvingcertainlinguistic problems,e.g.,SNOBOL5andCOMIT.6
Similarly,programming,beingconcernedwithagrowingsetofdemands,providesapressureonlinguistictheorydirectingit toward problemsparticularlyrelevanttocomputation,e.g.,problemsofefficiencyofrepresentationandspeedofcomputation.
FUTURERELATIONSHIPBETWEENPROGRAMMINGAND LINGUISTICS
Inprogrammingtherewillbeconcentrationondevelopingtheoriesof evaluation,i.e.,whatismeantbytheexecutionofaprogramwritteninlanguage`?Wemaycallthisthesematicsof`.Suchstudieswillreplacethe presentadhocdevelopmentofcompiler-compilerswithatheoryoftheirpropertiesandmoreinsightintothedesignofcomputingmachines.Thisisthe translation problemforcomputerlanguages.
Theselanguageswillbecomesufficientlycomplexsothatatheoryoftheirsemanticsorevaluationwillbeasufficientlyinterestingmodelforthe equivalentproblemsarisinginnaturallanguagetranslation.
Similarly,therewillbeareverseflowfromthedevelopmentofsemantictheorieswithinnaturallinguisticsintomathematicallinguisticmodels,which,inturn, willinfluence programming.
REFERENCES
1. E.K. Blum,“Automatic Digital Encoding System,”NAVORD Rep.4411 (1956).
2. P.B.Sheridan,“TheArithmeticTranslator-CompileroftheIBMFortranAutomaticCoding System,” Commun.Assoc.ComputingMach.2 ,9 (1959).
3. S.GinsburgandR.H.Gordon,“TwoFamiliesofLanguagesRelatedtoALGOL,”J.Assoc.ComputingMach. 9,350(1962).
4. R.W.Floyd,“SyntacticAnalysisandOperatorPrecedence,”J.Assoc.ComputingMach.10,316(1964).
5. D.J.Farber,R.E.Griswold,andI.P.Polonsky,“SNOBOL,AStringManipulationLanguage,” J.Assoc.ComputingMach.11,21 (1964).
6. The ResearchLaboratoryofElectronicsandtheComputationCenter,COMIT Programmers Reference Manual,2nded.,TheM.I.T.Press,Cambridge,Mass.(1962).
BIBLIOGRAPHY
1R.W.Floyd,IEEETrans.Electron.Computers13,346(1964).Thisbibliographyincludessubjectsrelatedtothesyntaxofprogramminglanguagesinsofarastheyilluminatetheproblemsofanalysisandsynthesisofformally defined programminglanguages.
Appendix19
MachineTranslationandLinguistics
Theadventofcomputationallinguisticspromisestoworkarevolutioninthestudyofnaturallanguages.Hockettisfondoftheappellation“computer revolution”or“thirdhumanrevolution”fortheeventsthatareengulfingus[see
C.F.HockettandR.Ascher,“TheHumanRevolution,”CurrentAnthropol.5,135(1964)].Therewasspeech,makingtheaggregateofcodwellinganimalsaconglomeratetribe.Therewasthetool,theleverwithwhichmankindmoved theworld.Andnowthereisthecomputer,thefirstpowerfulmanipulatorof symbolsoutsidethehumanhead.Whetherthecomputerisasgreataninvention asthefirstartefact,oronlythefirstintellectualtool,itspotentialforlinguistics isalreadyprofound.Itcanchangethelevelofanalysisofnaturallanguages,asthemicroscopechangedbiology.Itfacilitatesmathematizationasithasaided physics.Andithaslinkedtheory,empiricalstudies,and,perhaps,practical application.Mel'chuksaysthatcomputationallinguisticsisnotafieldof linguistics,asubspecialtyforthosewholikecomputation;itisatechniqueinescapableforanylinguistwhohonorshisdiscipline.InO.S.Akhmanova,I.
A.Mel'chuk,R.M.Frumkina,andE.V.Paducheva,ExactMethodsinLinguisticResearch,UniversityofCaliforniaPress,Berkeley(1963),p.46we read,“MTissimultaneouslybothaworkshop,wherethemethodsofpreciselinguisticresearchareperfectedindependentlyoftheconcretesphereofapplicationofthesemethods,andanexperimentalfield,wheretheresultsareverifiedbyexperience.”
Muchoftherecentchangeinlinguisticshascomefromclarificationgainedthroughformalizingdisciplines,andthesechangesaresurelyconnectedwiththedevelopmentsunderlyingcomputerstudies,aswellaswithtrendsinthegrowthofcontemporarylogicandphilosophy.Thoughitseemsclearthatthecomputerwasnotatthecenterofmostofthisinadirectcausalfashion,ithas surelyplayed asignificantrole,both ofinterplayand asa toolforvalidation.
SurelythemostdramaticrecentchangeshavebeencausedbyChomsky [see,forexample,Proc.9thInternatl.Cong.ofLinguistics,Cambridge,Mass.,1962,MoutonandCompany,TheHagueNetherlands(1964)]andsimilar thinkers,andtheyhaveexplicitlyhadlittletododirectlywithcomputers(seepage922oftheabove-mentionedProceedings).Thefundamentalchangesthat theyhavebroughttolinguisticsinhereratherinanalteredviewtakenbylinguisticsofthenatureofscience,ofascientifictheory,andoftherelationofempiricismtoscience.Butthesechangeshavebeenbroughtaboutandspurredonnotbyscholarswholiveandworkinvacuo,butwithagooddealofcross-fertilizationfromareasinclosetouchwithcomputationalactivities,andevenwith machinetranslation.
Moreover,thedepthofsyntacticanalysishaschanged.Adecadeago,mostlinguistsbelievedthatsyntaxhadtodowithwordorder,inflection,functionwords(e.g.,prepositionsandconjunctions),andintonationorpunctuation.Theyalsobelievedthatmostsentencesutteredbynativespeakersinordinarycontextsweresyntactically,evenifnotsemantically,unambiguous.Theimportantdifferenceintheirbeliefofthattimewasthattheythoughtsyntax relatedonlytothesurfacestructure,thevisibleoraudibleconfigurationsofthe output,andtheydeniedbyandlargethatprocess-typestatementsrelatingto rulesthatworkedonunderlyingabstractexpressionswereproperlyapartofgrammar.Therecanbenodoubtthatexperimentsincomputerparsingofordinarysentences,usingreasonablegrammarsashithertoconceivedandprogramsthatexposeallambiguities,havegreatlyhelpedmanylinguiststoabandontheirearlierinadequatesyntacticviews.ArecentandaccessibleaccountoftheseambiguitiesisthatofR.A.LangevinandM.F.Owens[“ComputerAnalysisoftheNuclearTestBanTreaty,”Science146,1186(1964)]. They usethe Kuno-Oettingerparser.
Whileitistruethataverynewviewofsyntaxhasgrownup,theinterestingresulthasbeenthatwithinthelast3yearsorso,interestamong generativegrammarianshasbeenperhapsaslivelyonquestionsofphonologyasithascometobeonsyntax.Infact,thisisanaturalconsequenceifoneviews agrammarasatotalsetoforderedrules,withcomponents(e.g.,phrase-structureandtransformational)simplydifferentiatedbytypeofrule,ratherthanasetoflevelsdifferentiatedbythephenomenatowhichtheyseverallyapply, andfromwhichonecanthenmakeachoicefortheapplicationofone'sanalyticeffortsbased on taste.
Mathematicallinguisticswouldhavehadnosignificancein1686,if Newtonhadinventedit.Thesliderulewastheperfectmathematicalmachinefor mechanicsandmanyotherbranchesof physics;withpencilandpaperandasliderule,generaltheoriescouldbesolvedabstractlyforspecialcases,andspecificexamplesworkedoutforobservedorproposedparameters.Ofcourse,otherbranchesofphysicscouldnotprogressfarwithoutmassivedigitalcalculations:thestudyofnuclearreactions,for example,orofcrystalstructure.Alloflinguisticsfallsinthelattercategory.Whenamathematicalstructureispromulgatedasalinguisticmodel,itsspecific correspondencewithanyonenaturallanguagecanbetested,inaseriousway,onlybytheexaminationofmanystringsthatitgeneratesassentences[severaltransformationalistshavetriedthistechnique,buttheonlypublicationsknown tousearebyV.H.Yngveandhisstudents;e.g.,hisRandomGenerationofEnglishSentences,”in1961InternationalConferenceonMachineTranslationofLanguagesandAppliedLanguageAnalysis,H.M.StationeryOffice,London (1962),pp.65-82],or,conversely,bythestudyofthestructuresthatitassignstonaturallyoccurringsentences.Thisplanhasbeentriedmanytimes.ThesituationisreviewedbyD.G.Bobrow,inhispaper“SyntacticAnalysisofEnglishbyComputer–ASurvey,”inAFIPSConferenceProceedings,SpartanBooks,Baltimore,Md.(1963),Vol.24.Onlyahigh-speedautomaticcomputer(i.e.,symbolmanipulator)canserveadequatelyinempiricaltestsofsuch theories.
Eventodaytherearelinguistictheoreticianswhotakenointerestin empiricalstudiesorincomputation.Therearealsoempiricallinguistswhoare notexcitedbythetheoreticaladvancesofthedecade–orbycomputers.Butmorelinguiststhaneverbeforeareattemptingtobringsubtlertheoriesinto confrontationwithricherbodiesofdata,andvirtuallyallofthem,inevery country,are eagerforcomputationalsupport.
Ifeveramachine-aidedsimulationoftotallinguisticanalysissynthesis(orvoice-to-ear-to-voicetranslation)becomespossible,itwillnotbebecauseof adherence tothe typeoflinguistictheory widelycurrent around1950.
Therecanbenodoubtthatthedisappointinglyslendercomputerresultsrealizedonthebasisofsuchtheorymusthavebeenimportantinshakingatleastsomeinquisitivelinguistsoutoftheircontentment.Ifmachinetranslationhadvariousnegativeresults,thiswasonethat waspotentinasingularlyfruitful way.
Appendix20
PersonsWhoAppearedBeforetheCommittee
June2-3, 1964
EdmundGlenn,DepartmentofState JulesMersel,Bunker-RamoCorporationSeptember30 - October1,1964
FranklinClark, President,Language ServiceBureau, Inc. TheodoreSchaeffer,Free-lancetranslator
KurtGingold,President, AmericanTranslators Association HowardSteensen, TranslationDirector,F. W.DodgeCompany ThomasMiller, Director,JointPublications ResearchService CharlesZalar, NationalScience Foundation
December9-10,1964
VincentGiuliano, Arthur D.Little,Inc. StephenPollock,ArthurD. Little,Inc.
ErnestR.Sohns,NationalScience Foundation March17-18, 1965
Paul L.Garvin,Bunker-RamoCorporation Gilbert King,TheItekCorporation
J. C.R.Licklider,TheIBMCorporation DavidLieberman, TheIBMCorporation WarrenStrohm,The IBMCorporation WinfredP. Lehmann,TheUniversity ofTexas
https://blog.sciencenet.cn/blog-362400-730694.html
上一篇：ALPAC 黑皮书 8/9： 附录 11-15
下一篇：设计上需要多方面的思考
