n.05_September_October.2009.Englisch._

(Ben Green) #1

  • LANGUAGEMACHINETRANSLATION


TEACHING EACH OTHER


InderglobalisiertenWeltgibtesmehrTextezuübersetzendennje.Die Übersetzungssoftwaredazu
hatFortschrittegemachtundwirdauchweiterhinmitjedemTagbesser.CAROLSCHEUNEMANNberichtet,
wiederÜbersetzerundseinComputereinerfolgreichesTeambildenkönnen. "Rt1tl

...Machine rranslation was one of the first non-
numeriealapplic:ltionsofthecomputer,"writesJohn
HurchinsinMachineTrallslation:Past,preseut,(/ftllre
(seetheboxonpage17).
Today,wehavccotranslatemorethanever.Compa-
niescommunicateacrossborders,andglobalbusiness
depcndsonhavingmulrjJjn~ualdocuments,Websires,
andproducrmaterial.Inrernationalorganizarions,such
astheUnited ationsandtheEuropeanUnion,inform
worldeiti7.cnsoftheirservicesinmanylanguages.
Andtheinrernerhascreatedoceansofwords:Weh
siresfromUkraine,blogsfromlraq,newsfromChina,
chats,forums,andTwieterareavailnbletunearlyevery-
oneontheplanet.Withourthehelpofrnaehinetrans-
lation,fewofuswouldeverheablereadthern.

TImeismoney
Theterm"rnachinetranslation"originallyreferredro
Handalonesoftwareprograms,burhascornecoinclude
anysysremrhatlIsesa computerteltransforma text
&omonelanguageintoanotherIanguage.Inmosrcases,
theresultisfar from perfeet. Nevertheless, machine
translationallowspeopletogetthettillofa text- in
orherwords,cogetehebasicme:ll1ing..
Thismaybcgoodenoughforinternalcompan)'mes-
sagesandiocQmiogcorrespondenee.Ir al10wsYOllco
sec,forexample,wherherthare-mailfromTokyoisa
rcquestforproductinformationorjusrspam.Sending
suchtextstoa translararwouldcosrarleaStreneuro
centSperword,or€SOperpage,andisjusttOOexpen-
sive.Moreimponant,irrakestoomucbtime.
largeorganizationsusehighlydevclopedtranslation
systems,called"enrerprisesoluriQns,"thatcanprocess
tensofthOLisandsofwordsanhour,inconrrasttoabout~

,


'I


n thebeginning,Godcreatedthchcavensand
theearrh.Theearthwasformlessandempty,
anddarkncsscoveredthedeepwaters."M..!:ill:
llL..k.Jis, Thewordsyoureaddepcndonthe
versionafrhe Bihleyou'reholdingand,in
parricular, on its rr:mslation. Ofthe world's 6,
known I3nguages,acampleteversionoftheßibleis
availableinonlyaoout 440 ofthem,saysJonRiding,a
specialistinlinguistjccomputingatthcBritisbandFor-
eign BibleSociery, anorganization with theaimof
translatingtheBibleintoeverylanguage.Partsofthe
ßibleareavailableinaround2,000languages,farfew-
erthanhalf."TbemosteommonqueseionIhearfrom
peopleis: 'Hasn'trhatall heendone?',"saysRiding.
Some 550 yearsafterGurcnbergprintedehefirsrversion
oftheSeripcureinLatin,computertechnologiesmay
helpcospeedupBibletranslationprojeets,roo.
Theideaofusingmachinesrotransferwordsfrom
onelanguageintoanotheris widelysaidtocomefrom
FrenehphilosopherandmarhemaricianReneDescartes.
Intheearly17thcentury,hesuggestedthathurnansare
rnachincs,andthatsymbolscouldbeusedtotransfer
ideasinstcadoflanguagc.Threehundrcdyearslater,se-
rious acremptsco use computerstorranslarebegan.

Howmachinetranslation works


T


heU.$.militaryledthewayinusmgcomputerstodecipher
Russianinthe1950s.Theearliestmachine-translationsys-
temsmatchedword-for·worddirectly,muchlikea dietionary.
Eachdecadesincethenhasseennewmethodsforteaching
computershowtobuildsentences.Manysystemsarerule-based,
sothecomputereananalyzesentencesbasedongrammatical
rules,wardorder,andstyle.Knowledge-basedorexample-based
systemsareindireetjtheyconeentrateonsemanties,themean-
ingofwords.Tilisrequiresreal-worldknowledge- forexample,
whethera "bank"isaninstitutionthatholdsmoney,ora place
at ariver'sedge.
Sinceabout1990,manycomputerlinguistshavefocusedon
statisticalmachinetranslation.Thisisa~-basedmethod,
anditeomparessentencesin!wolanguages.Usingmillionsof
sentencestandthroughthehelpofstatistieallearningmethods
andmathematicalalgorithrns,thesoftwareworkstofindtheeor-
reetwordsincontext.Someofthelatestsystemsuseaeombi-
nationofmethods,or"hybrid"technology.


12 BusiMSSSp>IJighi

corpus(pi.corpora)['b:rp;)S
decipherslh.ldi'~d>(]
enlerprisesolutionl'enl"l"pratzg,I",J.....]
gist[d31stJ
incomllll['m,kAmu;lJ
moreorless[,nn;r>('k~l
mullililliual['m"lti'lr!lg~IJ
non-numerlcalLn,,:nnu'menk-II
processI'pru:scs]
Scripturel'sknpl~"-l
standaloner'sl",no.l.~,lounJ
wordorderl'w"J:d,:>:rd'r]

Korpus(Sammlungvon
Te~tmaterialien,Ö';;-:;;:__
etw.dechiffrieren,etw.
entziffern
Unternehmenslllsung
dasWesentliclK!
eingehend
(so)un~fahr
mehrsprac.hi~

512009
Free download pdf