文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 幼儿教育 > 少儿英语 > a hierarchical phrase-based Model for SMT
侵权投诉

a hierarchical phrase-based Model for SMT

AHierarchicalPhrase-BasedModelforStatisticalMachineTranslation

DavidChiang

InstituteforAdvancedComputerStudies(UMIACS)UniversityofMaryland,CollegePark,MD20742,USA

dchiang@umiacs.umd.edu

Abstract

Wepresentastatisticalphrase-basedtransla-tionmodelthatuseshierarchicalphrases—phrasesthatcontainsubphrases.Themodelisformallyasynchronouscontext-freegram-marbutislearnedfromabitextwithoutanysyntacticinformation.Thusitcanbeseenasashifttotheformalmachineryofsyntax-basedtranslationsystemswithoutanylin-guisticcommitment.Inourexperimentsus-ingBLEUasametric,thehierarchicalphrase-basedmodelachievesarelativeimprove-mentof7.5%overPharaoh,astate-of-the-artphrase-basedsystem.

Englishsentenceeismodeledas:(1)(2)

argmaxP(e|f)=argmaxP(e,f)

e

e

=argmax(P(e)×P(f|e))

e

ThetranslationmodelP(f|e)“encodes”eintofbythefollowingsteps:

1.segmenteintophrasese¯1···e¯I,typicallywithauniformdistributionoversegmentations;2.reorderthee¯iaccordingtosomedistortionmodel;3.translateeachofthee¯iintoFrenchphrasesac-cordingtoamodelP(f¯|e¯)estimatedfromthetrainingdata.Otherphrase-basedmodelsmodelthejointdistribu-tionP(e,f)(MarcuandWong,2002)ormadeP(e)andP(f|e)intofeaturesofalog-linearmodel(OchandNey,2002).Butthebasicarchitectureofphrasesegmentation(orgeneration),phrasereordering,andphrasetranslationremainsthesame.

Phrase-basedmodelscanrobustlyperformtrans-lationsthatarelocalizedtosubstringsthatarecom-monenoughtohavebeenobservedintraining.ButKoehnetal.(2003) ndthatphraseslongerthanthreewordsimproveperformancelittle,suggestingthatdatasparsenesstakesoverforlongerphrases.Abovethephraselevel,thesemodelstypicallyhaveasimpledistortionmodelthatreordersphrasesin-dependentlyoftheircontent(OchandNey,2004;Koehnetal.,2003),ornotatall(ZensandNey,2004;Kumaretal.,2005).

Butitisoftendesirabletocapturetranslationswhosescopeislargerthanafewconsecutivewords.

1Introduction

Thealignmenttemplatetranslationmodel(OchandNey,2004)andrelatedphrase-basedmodelsad-vancedthepreviousstateoftheartbymovingfromwordstophrasesasthebasicunitoftransla-tion.Phrases,whichcanbeanysubstringandnotnecessarilyphrasesinanysyntactictheory,allowthesemodelstolearnlocalreorderings,translationofshortidioms,orinsertionsanddeletionsthataresensitivetolocalcontext.Theyarethusasimpleandpowerfulmechanismformachinetranslation.

Thebasicphrase-basedmodelisaninstanceofthenoisy-channelapproach(Brownetal.,1993),1inwhichthetranslationofaFrenchsentencefintoan

Throughoutthispaper,wefollowtheconventionofBrownetal.ofdesignatingthesourceandtargetlanguagesas“French”and“English,”respectively.Thevariablesfandestandforsourceandtargetsentences;fijstandsforthesubstringofffrompositionitopositionjinclusive,andsimilarlyforeij.

1

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email