引言分析.docx

引言分析.docx

《引言分析.docx》由会员分享，可在线阅读，更多相关《引言分析.docx（7页珍藏版）》请在冰豆网上搜索。

引言分析

APASTYLE

TEXT1

TEXT2

TEXT3

JianpengCheng,MirellaLapata（2016）.NeuralSummarizationbyExtractingSentencesandWord《IpsjSigNotes》, 2016 , 2016 :

31-36

O.Adams,G.Neubig,T.Cohn,S.Bird,Q.T.Do,andS.Nakamura（2016）.

LearningaLexiconandTranslationModelsfromphonemeLatticesSubmittedon23Mar2016（v1）,lastrevised1Jul2016thisversion,v3.

ConferenceonEmpiricalMethodsinNaturalLang... , 2016 :

2377-2382

S.Cadoni,E.Chouzenoux,J.C.Pesquet,andCarolineChaux（2016）.

Ablockparallelmajorize-minimizememorygradientalgorithm

IEEEInternationalConferenceonImageProcessing , 2016 :

3194-3198

Whyitwaschosen?

Whyitwaschosen?

Whyitwaschosen?

1.Recentpublication（2016-5）

2.PublishedinascientificjournalofACL

3.Itisrelatedtothesubjectswearelearningabout.Basedonthedifferentclassesofsummarizationmodels.

4.Themethodofthemodelisclassicandhasachiveddesiredeffect.

1.Recentpublication（2016-5）

2.PublishedinascientificjournalofACL

3.Experimentsdemonstratephonemeerrorrateimprovementsagainsttwobaselinesandthemodel’sabilitytolearnusefulbilinguallexical.

4.Itisrelatedtothesubjectswearelearningabout.

1.Recentpublication（2016-3）

2.IEEEInternationalConferenceonImageProcessing 

3.Thisisanapplicationofmathematicalknowleagetosolvetheproblemof3Dimage.

4.TheBlockParallelMajorize-MinimizeMemoryGradient（BP3MG）algorithmproposedinthispapersolvestheoptimizationproblemeffectively.

Abstract

1.Classification

（1）Reportabstract:

Thistapeofabstractneedsreflectpurposes,methods,importantresultsandconclusions.

报道摘要：

这一类摘要反映了文章的目的、方法、重要结果和结论。

（2）Indicativeabstract:

Itisanabstractofthisthesisandleveloftheresultobtained.

指示性摘要：

是描述论文的主题，所得结果的水平。

（3）Report-indicativeabstract:

Intheformofareportedabstract,themostvaluablepartofthethesisisexpressed,andtheremainderisexpressedinanindicativeabstract.

报道-指示性摘要：

以报道性摘要的形式表述论文中价值最高的那部分内容，其余部分则以指示性摘要形式表达。

2.Basicelements

Abstractsshouldstatetheobjectivesoftheproject,describethemethodsused,summarizethesignificantfindingsandstatetheimplicationsofthefindings.

ElementsofAbstract；

a.Purpose;（目的）

b.Methods;（方法）

c.Results;（结果）

d.Conclusion.（结论）

3.Commontenses

Throughthestudyoftendocuments,wefindthatthetensesusedintheabstractarethepresenttenseandthepasttense.Thepresentperfecttenseisoccasionallyused.

（1）Simplepresenttense（一般现在时）

Usedtoshowageneraltruth,ortoindicateastate,orregularactionsorprocess,itismostcommonlyusedinthesepapers.

Examples:

1.Languagedocumentationbeginsbygather-ingspeech.

2.Thisarchitectureallowsustodevelopdifferentclassesofsummarizationmodelswhichcanextractsentencesorwords.Wetrainourmodelsonlargescalecorporacontaininghundredsofthousandsofdocument-summarypairs.

3.Experimentalresultsontwosummarizationdatasetsdemonstratethatourmodelsobtainresultscomparabletothestateoftheartwithoutanyaccesstolinguisticannotation.

（2）Simplepasttense（一般过去时）

Usedtodescribethediscoveryorprocessofacertainmomentinthepast.

Examples:

Weusedlessthan10hoursofEnglish–JapanesedatafromtheBTECcorpus（Takezawaetal.,2002）,com-prisedofspokenutterancespairedwithtextualtrans-lations.

（3）presentperfecttense（现在完成时）

Thepresentperfecttenseisthelinkbetweenthepastandthepresent.

Example:

Theneedtoaccessanddigestlargeamountsoftextualdatahasprovidedstrongimpetustode-velopautomaticsummarizationsystemsaimingtocreateshorterversionsofoneormoredocuments,whilstpreservingtheirinformationcontent.

4.Sentencepatterns

1）Itis…..

Smartphoneappsforrapidcollectionofbilin-gualdatahavebeenincreasinglyinvestigated（DeVriesetal.,2011;DeVriesetal.,2014;Reiman,2010;Birdetal.,2014;Blachonetal.,2016）.Itiscommonfortheseappstocollectspeechsegmentspairedwithspokentranslationsinanotherlanguage,makingspokentranslationsquickertoobtainthanphonemictranscriptions.

2）Therebe…..

Inthisworkweproposeadata-drivenapproachtosummarizationbasedonneuralnetworksandcontinuoussentencefeatures.TherehasbeenasurgeofinterestrecentlyinrepurposingsequencetransductionneuralnetworkarchitecturesforNLPtaskssuchasmachinetranslation（Sutskeveretal.,2014）,questionanswering（Hermannetal.,2015）,andsentencecompression（Rushetal.,2015）.

3）Experimentalresults…..

Experimentalresultsontwosummarizationdatasetsdemonstratethatourmodelsobtainresultscomparabletothestateoftheartwithoutanyaccesstolinguisticannotation.

4）Wepresentamethodto......

5）Thisarchitectureallowsusto......

LearningaLexiconandTranslationModelfromPhonemeLattices

Title

Languagedocumentationbeginsbygatheringspeech.

Providesbackground

Manualorautomatictranscriptionatthewordlevelistypicallynotpossiblebecauseoftheabsenceofanorthographyorpriorlexicon,andthoughmanualphonemictranscriptionispossible,itisprohibitivelyslow.Ontheotherhand,translationsoftheminoritylanguageintoamajorlanguagearemoreeasilyacquired.

Problemdescription

Weproposeamethodtoharnesssuchtranslationstoimproveautomaticphonemerecognition.

Designthinking

Themethodassumesnopriorlexiconortranslationmodel,insteadlearningthemfromphonemelatticesandtranslationsofthespeechbeingtranscribed.

Designinnovation

Themethodassumesnopriorlexiconortranslationmodel,insteadlearningthemfromphonemelatticesandtranslationsofthespeechbeingtranscribed.

Presentsthesignificanceandachievementofthestudy

NeuralSummarizationbyExtractingSentencesandWords

Title

Traditionalapproachestoextractivesummarizationrelyheavilyonhuman-engineeredfeatures.

Providesbackgroundandproblemdescription

Inthisworkweproposeadata-drivenapproachbasedonneuralnetworksandcontinuoussentencefeatures.Wedevelopageneralframe-workforsingle-documentsummarizationcomposedofahierarchicaldocumentencoderandanattention-basedextractor.

Designtechniques

Thisarchitectureallowsustodevelopdifferentclassesofsummarizationmodelswhichcanextractsentencesorwords.

Advantagesofthe

method

Wetrainourmodelsonlargescalecorporacontaininghundredsofthousandsofdocument-summarypairs.

Designmethod

Experimentalresultsontwosummarizationdatasetsdemonstratethatourmodelsobtainresultscomparabletothestateoftheartwithoutanyaccesstolinguisticannotation.

Designresults

ABlockParallelMajorize-minimizeMemoryGradientAlgorithm

Title

Inthefieldof3Dimagerecovery,hugeamountsofdataneedtobeprocessed.

Providesbackgroundandproblemdescription

Paralleloptimizationmethodsarethenofmaininterestsincetheyallowtoovercomememorylimitationissues,whilebenefitingfromtheintrinsicaccelerationprovidedbyrecentmulticorecomputingarchitectures.Inthiscontext,weproposeaBlockParallelMajorizeMinimizeMemoryGradient（BP3MG）algorithmforsolvinglargescaleoptimizationproblems.

Designtechniques

Thisalgorithmcombinesablockcoordinatestrategywithanefficientparallelupdate.

Designinnovation

Theproposedmethodisappliedtoa3Dmicroscopyimagerestorationprobleminvolvingadepth-variantblur,whereitisshown

toleadtosignificantcomputationaltimesavingswithrespecttoasequentialapproach.

Advantagesofthe

Methodandapplication

LearningaLexiconandTranslationModelfromPhonemeLattices

Introduction

Analysis

⑴Mostoftheworld’slanguagesaredyingoutandhavelittlerecordeddataorlinguisticdocumentation（AustinandSallabank,2011）.⑵Itisimportanttoad-equatelydocumentlanguageswhiletheyarealivesothattheymaybeinvestigatedinthefuture.⑶Languagedocumentationtraditionallyinvolvesone-on-oneelicitationofspeechfromnativespeakersinor-dertoproducelexiconsandgrammarsthatdescribethelanguage.However,thisdoesnotscale:

lin-guistsmustfirsttranscribethespeechphonemicallyasmostoftheselanguageshavenostandardizedorthography.⑷Thisisacriticalbottlenecksinceittakesatrainedlinguistabout1hourtotranscribethephonemesof1minuteofspeech（Doetal.,2014）.

（1）Thissentenceexplainsthesocialcontextoftherecordedlanguageandthecurrentresearchsituation

（2）Thesentencereflectstheimportanceofrecordingthelanguage

（3）Attributiveclause:

That………

（4）Causeadverbialclause:

Since……..

⑸Smartphoneappsforrapidcollectionofbilin-gualdatahavebeenincreasinglyinvestigated（DeVriesetal.,2011;DeVriesetal.,2014;Reiman,2010;Birdetal.,2014;Blachonetal.,2016）.⑹Itiscommonfortheseappstocollectspeechsegmentspairedwithspokentranslationsinanotherlanguage,makingspokentranslationsquickertoobtainthanphonemictranscriptions.

（5）Passivesentence:

Havebeendone……….

（6）Formalsubject:

Itis……….

⑺Wepresentamethodtoimproveautomaticphonemetranscriptionbyharnessingsuchbilingualdatatolearnalexiconandtranslationmodeldirectlyfromsourcephonemelatticesandtheirwrittentar-gettranslations,assumingthatthetargetsideisamajorlanguagethatcanbeefficientlytranscribed.ABayesiannon-parametricmodelexpressedwithaweightedfinite-statetransducer（WFST）frameworkrepresentsthejointdistributionofsourceacousticfeatures,phonemesandlatentsourcewordsgiventhetargetwords.⑻Samplingofalignmentsisusedtolearnsourcewordsandtheirtargettranslations,whicharethenusedtoimprovetranscriptionofthesourceaudiotheywerelearntfrom.Importantly,themodelassumesnop