n.05_September_October.2009.Englisch._

• lANGUAGEMACHINETRANSLATION

"Improvements in machine translation

come gradually"

JohnHutchins,nowretired,workedfor
manyyearsasa librarianattheUniver-
sityofEastAngliainEngland,andhas
written extensively about linguistics. information retrieval,
andmachinetranslation(seetheboxonpage17).
Whyisonlinetranslationstillsopoer?
Weil,onereasonisthatonlinesystemsaresupposedtobeable
totranslateanything,creverything,thatisthrownatthem.
Machine-translationsystemshavebeenprimarilydevelopedfor
technicaldocumentation.Therefore,onceyouusesystemsfor
somethinglikeaWebpage,whichmaybemoresensitiveto
socialorculturalmatters,theresultsaremuchpoorer.It'sthe
natureofthelanguage,andthestyleofthelanguage.

15 machinetranslationbettetinsomelanguagesthanothers?
FrenchandEnglishgoprettyweil,SpanishandEngtishgoptet-
tyweil,butyougetsomeverypeculiarresuttsifyoutranslate

tromGermanintoEnglish,orEnglishintoGerman.InJapan,

theyusemachinetranslationforJapaneseintoEnglisha great

deal,notsomuchforEnglishintoJapanese.Butformanyoth-

erAsianorAfricanlanguages,thesystemsareverypoorindeed.

So,irscommercialandpolltlealimportaneethatdetermines

whethera systemisdevelopedornot.

Willtherebeabreakthroughsooo?

Rightthroughthehistoryofmachinetranslation,researchers

havebeencJaiming:"We'reimprovingeveryyearandwe'rego-

ingtohavea greatbreakthroughinthenearfuture,andthe

improvementwill bedramatic!"- butit'sneverhappened.

Improvementseomegradually.Lookingforward.Idon'tseeany

reasonwhyitshouldgoanyfasterinthefuture.

Willmachinetranslationeverreplacehumantranslation?

No,therewillalwaysbea demandforreliablehumantransla-

tion.Peoplewillnevercompletelytrustmachines.

listen 10 lohnHutchinsonBusinessSpotlightAudio

assemblymanualswillbemorequicklyavailablerhan

everbeforeinnumerouslallguagcs.

EUCAM,whichsrandsforMultilingualCommunica-

tioninEuropeanCarManufacruring,isa learningpro-

gramthatimmediarclyrranslatl"'sworkflowchangesin

automobileproduction,mainlyatDaimlerfacrories.

"So,rheworkerinJapanbenefirsframthedarabasein

exacrly[hesameway ashiseolleaguein Derroitor

Mannheim,theminutea ehangeis made,"saysDr.

AdrianeRinsehe,directorofeheLanguageTechnolog)'

Centre(LTC),whiehdevelopedthclangl!age-rranslation

elementofehcsystem.

AnotherexampleofatranslationsolutionmadetO

mcasure,inpanicularforSAPappliea[ionsinforeign

counrries,canbefoundinLucySoftware.Ircanhelp

Aworldwidenetwork?

ThefocusonEnglishinglobalbusinessmeansthatoth-

erlanguagesareneglected,includingthose spokenin

AsiaandtheMiddleEast.Suchlanguagcshavediffer-

entalphabets,andtheyf10wfromrÜ~httolefr,ormn

[0bonom. Localizationisslow.""InArahcountries,

we'reabout 15 to 20 yearsbehindrheWestin[heavail-

abilityandusabilityofsoftware,"saysEgyprianhusi·

nesswomanManalAmin,headofArahi7.e,a loealiza-

tioneompanyinCairo.

Parrieularlybetweenlanguageswhichfollowver)'dif-

ferentrules,statistiealmachinetranslation(seethebox

onpage12)is deliveringpromisingrcsults,sayirssup-

porrers.Thesearch-engjnegiantGoogJehasinvesred

millionsindcvelopingthistechnologyforitsWeb-site

translarionservice.Thecompanyhasfeditsstatistical

systemveryhigh-qualitytextsrhathavebeenrranslat·

edbyhumans.~orexample,thcUnitedNationsprovid·

ed 200 billionwordsiniessixofficiallanguages.

Ina 2007 nldiointerviewwichtheS(/lIloseMercury

News,~ranzOehofGoogle'smachine-translarioneffoTt

said: "'\Y/chaveseendramaticqualityimprovemenes

for very hard language pairs, Arabic-English or

ChineSl..'-English. Thismethod issignificamly better

thanwhattraditionalsysremshavcprodueed."

Translationis also the key ro workingwith col-

leaguesinothercountries.Thecrisisintheautomobile

indusrryhasproduccdnewcross-bordercompaniesand

languagceombinations.Whetherit'sfor~iar-Chrysler,

Opcl-Magna,orChinese-madeHummers,thanksro

machineandcompurer-aidcdtranslation,thousandsof

pagesofpartdeseriptions,produetionworkflows,and

16 BusintSSSpodigb

aswrnbly!;l'scmbli]

comparativelyLt.;)(ß'p;n;lllvlij

cross-borderI,rn:s'b:l;rWrI

flowfromrilhttoreff

Lfluofr;>m,rallt:;>'lefll

~Iygrall[,houli'grel'lJ

informationN!itrieval

L,mf'r'meII'nri,lri:v·l]

hbrarianllal'brcri:;>n]

lrterall'llI~

madetomeasureL,meldIU'me3'T]

neglectsth.lm'glct.l]

peeuliarIpl'li:jU:lj..-]

resean::hef(ri'sa;II..-]

sean::henglner'>u:if,cnd;>m]

sensitiw:bemore-to...I'""ns:ltlv]

toptabot10m:flowfrom-

1,10:pt:l'bu:l"ml'~~__

lIsabllity[j..'!:z;l'blbli]

verse[v~:~

workflow['w3:kFloul

Montage

heiligerGlal

Informalions(wieder)-

gewinnung

Bibliothekar(in)

wÖrtlich{zunehmen}

maBge.schneidert

etw.vemachl1s~gen

eigentOmIich

F~her{in}

SuchlTlilSChine

eherum...gehen

senkrechtverlaufen

. Brauchbarkeit,Verwendbarkeit
v"
Albeitsablauf

512009