資源描述:
《英漢機器翻譯中的介詞短語消歧》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在學術(shù)論文-天天文庫。
1、對外經(jīng)濟貿(mào)易大學碩士學位論文英漢機器翻譯中的介詞短語消歧姓名:胡克力申請學位級別:碩士專業(yè):外國語言學與應(yīng)用語言學指導教師:陳蘇東;武姜生20040401中文摘要機器翻譯是自然語言處理中的一個極為重要的領(lǐng)域和試驗田。自然語言的多種因素也阻礙了機器翻譯的發(fā)展,比如一詞多義,句法歧義,復雜的多重語法結(jié)構(gòu)等等。在這些困難當中,歧義是最主要的因素。本文集中討論了英漢機器翻譯中,對介詞短語的句法歧義現(xiàn)象的消歧方法。在對主要機器翻譯系統(tǒng)的分類進行了簡要介紹之后,本文回顧了目前用于句法分析的最重要的一種形式語法:短語結(jié)構(gòu)語法,并演示了使用短語結(jié)構(gòu)語法分析句子結(jié)構(gòu)從而
2、導致介詞短語歧義結(jié)構(gòu)的過程。此后,本文主要分析了基于優(yōu)選的介詞短語消歧策略,并提出了一個實驗性的介詞短語消歧模型。針對模型中自動獲取短語搭配的需要,結(jié)合Brill錯誤驅(qū)動算法,通過重復該算法的實驗和數(shù)據(jù),驗證了運用該算法進行介詞短語消歧的有效性。最后,針對處理介詞短語歧義時遇到的數(shù)據(jù)稀疏這一所有統(tǒng)計語言模型都面臨的難題,分析介紹了基于N—Gram的回退法介詞短語消歧及平滑技術(shù)。關(guān)鍵詞:機器翻譯,介詞短語岐義,短語結(jié)構(gòu)語法,錯誤套《的機器學習,數(shù)據(jù)稀疏AbstractMachinetranslation(MT)isanextremelyimportant
3、researchfieldandtest-inggroundofnaturallangllageprocessing(NLP).Asaninterdisciplinaryresearchfield,MTrequirestheknowledgeofLinguistics,ComputerSdence,Mathmaticsetc.,whichremainsachallenginga,rea.ManyfactorshinderthedevelopmentofMT,suchaslexiealambiguity,syntacticambiguity,multip
4、legrammaticalstructuresetc.,amongwhichvariouskindsofambiguityarethemostdifficultpart.Thisthesisfocusesondlsambiguationofprepositionalphrase(PP)at-tachment.AfterabriefintroductiontotheclassificationofmajorMTsys—tems,thisthesisreviewsoilthemostimportantformalgrammarinMT:PSGlanddem
5、onstratedhowprepositionalphraseattachmentambiguityisderivedwhenparsingbythisgrammar.Afterthat,theauthormainlyann-lyzespreference—baseddisambiguationapproachesandproposesatentativemodelforresolvingPPambi£;uity.Facingthemajorproblemoflackingofrules,thisthesisverifiesBrill’stransfo
6、rm—basederror-drivenalgorithmandexperimentforautomaticretrievaloftransformationrules.Finally,incon-nectionwiththesparsedataproblemintheprocessingofPPdisambiguationaudcurrentstatisticallanguagemodels,theauthorintroducesthebacked-offN—Grambasedalgorithmandsimilarity-basedsmoothing
7、.KeyWords:MachineTranslation,PrepositionalPhraseAttachmentProblemlPhraseStructureGraalamar】Error-drivenLearning,SparseDataAcknowledgements1wouldliketoexpressmygratitudetothefollowingpeoplefortheirsupportandassistallceinthewritingofthisthesis:MygratitudegoestoprofessorChenSudong,
8、thebestacademicadvisorandteacherIcouldhavewishe