资源描述:
《Building an Annotated English-Vietnamese Parallel Corpus for Training Vietnamese-related NL》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、BuildinganAnnotatedEnglish-VietnameseParallelCorpusforTrainingVietnamese-relatedNLPs***DinhDienandHoangKiem*FacultyofInformationTechnology,UniversityofNaturalSciences,VNU-HCMCTel:+84-8-8497-181E-mail:ddien@saigonnet.vn**CenterofInformationTechnologyDevelopment,VietnamNati
2、onalUniversityofHCMCTel:+84-8-9317-442E-mail:hkiem@citd.edu.vnAbstract:InNLP(NaturalLanguageProcessing)tasks,thehighestdifficultywhichcomputershadtofacewith,isthebuilt-inambiguityofNaturalLanguages.Todisambiguateit,formerly,theybasedonhuman-devisedrules.Buildingsuchacompl
3、eterule-setistime-consumingandlabor-intensivetaskwhilstitdoesn’tcoverallthecases.Besides,whenthescaleofsystemincreases,itisverydifficulttocontrolthatrule-set.So,recently,manyNLPtaskshavechangedfromrule-basedapproachesintocorpus-basedapproacheswithlargeannotatedcorpora.Cor
4、pus-basedNLPtasksforsuchpopularlanguagesasEnglish,French,etc.havebeenwellstudiedwithsatisfactoryachievements.Incontrast,corpus-basedNLPtasksforVietnameseareatadeadlockduetoabsenceofannotatedtrainingdata.Furthermore,hand-annotationofevenreasonablywell-determinedfeaturessuc
5、haspart-of-speech(POS)tagshasprovedtobelaborintensiveandcostly.Inthispaper,wepresentourbuildinganannotatedEnglish-VietnameseparallelalignedcorpusnamedEVCtotrainforVietnamese-relatedNLPtaskssuchasWordSegmentation,POS-tagger,WordOrdertransfer,WordSenseDisambiguation,English
6、-to-VietnameseMachineTranslation,etc.Keywords:Corpora,bilingualcorpus,corpusannotation,machinelearning,machinetranslations.formats,spelling,etc.intouniqueones.Sentence1.INTRODUCTIONalignmentofEVC.Nowadaysmoreandmorepeopleareinterestedin•WordalignmentofEVC:toautomaticallye
7、xtractinginformationaboutlanguagefromverylargeword-alignEVCbySemantic-Classapproach.annotatedcorpora.SuchannotatedcorporahavebeenManuallycorrectingword-alignments.Problemsofbuiltforpopularlanguages(e.g.PennTreeBankforVietnameseWordSegmentations.English,French,Japanese,etc
8、.)andthesecorpora•AnnotatingEVCwithPOS-tagsandSEM-tags:havebeenusedtoeffectivelyservesuchwell-kn