资源描述:
《Improvement of Translation Quality of English》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、MachineTranslation16:233250,2001.233©2003KluwerAcademicPublishers.PrintedintheNetherlands.ImprovementofTranslationQualityofEnglishNewspaperHeadlinesbyAutomaticPre-editingTAKEHIKOYOSHIMIInformationSystemsGroup,SharpCorporation,3-2-20IshiyamaderaOtsuShiga,520
2、-0861,JapanE-mail:yoshimi@mx.scn.tvAbstract.SincetheheadlinesofEnglishnewsarticleshaveacharacteristicstyle,differentfromthestyleswhichprevailinordinarysentences,itisdifficultforMTsystemstogeneratehigh-qualitytranslationforheadlines.Wetrytosolvethisproblembya
3、ddingtoanexistingsystemapre-editingmodulewhichrewritesheadlinesasordinaryexpressions.Rewritingofheadlinesmakesitpossibletogeneratebettertranslationswhichwouldnototherwisebegenerated,withlittleornochangestotheexistingpartsofthesystem.Focusingontheabsenceofaf
4、ormoftheverbbeasamissingpartofnormalEnglish,wehavedescribedrewritingrulesforputtingproperlytheverbbeintoheadlines,basedoninformationobtainedbymorpho-lexicalandroughsyntacticanalysis.WehaveincorporatedtheproposedmethodintoourEnglishJapaneseMTsystem,andcarrie
5、doutanexperimentwith312headlinesasunknowndata.Ourmethodhassatisfactorilymarked81.2%recalland92.0%precision.Keywords:automaticpre-editing,headline,newsarticle,rewriting,rule-basedMT1.IntroductionAsJapanesepeoplehavemoreopportunitiestoseeEnglishnewspaperartic
6、lesthroughtheInternet,itbecomesmoreimportanttotranslatethemcorrectlyintoJapanese.Sincethemostessentialinformationofarticlesisexpressedbyheadlines,theirsuccessfultranslationisoftheessence.Headlinesmustconveymuchinformation,expressingitasconciselyaspos-sible,
7、sotheyhavecharacteristicstylesdifferentfromthoseofordinarysentences.Thecharacteristicstylesarelikelytodamagethequalityoftranslationbyprevent-ingasyntacticparserfromgeneratingappropriateparsetreesforheadlines,thusdecreasingthequalityoftranslation.Thereareatl
8、easttwopossiblesolutionstothiskindofproblem.Oneistoextendthesyntacticrulessothataparsermayanalyzecharacteristicexpressions.TheotheristoaddtotheexistingMTsystemapre-editingmodulewhichrewritescharacteris