简介：Signlanguageisavisual-gesturallanguagemainlyusedbyhearingimpairedpeopletocommunicatewitheachother.Gestureandfacialexpressionareimportantgrammarpartsofsignlanguage.Inthispaper,atext-basedtransformationmethodofChinese-Chinesesignlanguagemachinetranslationisproposed.Gestureandfacialexpressionmodelsarecreatewd.Andapracticalsystemisimplemented.TheinputofthesystemisChinesetext.Theoutputofthesystemis'graphicsperson'whocangesticulateChinesesignalgnuageaccompainedbyfacialexpressionthatcorrespondstotheChinesetextenteredsoastorealizeautomatictranslationfromChinesetexttoChinesesignlanguage.

简介：ThispaperdescribesanewframeworkforsynthesizingChineselandscapepaintingusinganimage-basedapproach.Theframeworkinvolvestwostages:apreprocessingphase,inwhichafewbrushstroketextureprimitivities(BSTP)arecollectedfromsamplesofhand-madeChinesepaintings,andthecontrolpictureisconstructedtoprovidecolorIDsofmountains,andtheon-linephases,inwhichthefogimageissynthesizedandmountainsare"drawn"bymappingmultiplelayersofBSTPguidedbythecontrolpicture.Whenmorecomplexshadingisneeded,theshadingpictureisconstructedandusedduringtheBSTPmappingphase.Finally,thesynthesizedChineselandscapepaintingsofavarietyofstylesaregivenandtheylookmoreclosetothehandmadeworkthanthoseproducedwithpreviousmodelingmethods.

简介：InPekingUniversityComputerResearchInstitute(PUCRI)amethodofinputtingChinesesentencesbasedonwordshasbeendeveloped.Toreducethetroublesinchoosingonewordoutoftheotherscharacterizedbythesamefeature,grammaticalparsingtechniqueisappliedtothemethodandgoodresultshavebeenachieved.Thisarticledescribestheoutlineofthemethod.theprincipleofapplyinggrammaticalformulasandthebranch-cuttingalgorithmusedtospeedupthegrammaticalparsing.

简介：ThispaperintroducesakindofsemanticcomputationandpresentshowtocombineitintoourChineseQuestion-Answering(QA)system.Basedontwokindsoflanguageresources,HownetandCilin,wepresentanapproachtocomputingthesimilarityandrelevancybetweenwords.Usingtheseresults,wecancalculatetherelevancybetweentwosentencesandthengettheoptimalanswerforthequeryinthesystem.ThecalculationadoptsquantitativemethodsandcanbeincorporatedintoQAsystemseasily,avoidingsomedifficultiesinconven-tionalNLP(NaturalLanguageProcessing)problems.Theexperimentsshowthattheresultsaresatisfactory.

简介：AnewmethodforrecognizingChinesecharactersisproposed.Itisbasedontheso-calledfeaturepointsofChinesecharacters.Thefeaturepointsweuseincludethoseonthestrokeofacharacter.i.e.,endpoints.turningpoints,forkpointsandcrosspoints.andthekeypointsonthebackgroundofcharacter.ThismethoddiffersfromthepreviousonesforitcombinesthefeaturepointsonstrokewiththoseonbackgroundanditusesfeaturepointstorecognizeChinesecharactersdirectly.AChinesecharacterrecognitionsystembasedtotop-downdynamicalmatchingoffeaturepointisdeveloped.Thesystemcanrecognizenotonly6763printedsampleSongfontChinesecharactersofsize5.6×5.6mm^2withhighrecognitionrate,butalsothegeneralprintedbooks,magazinesanddocumentswithasatisfactoryrecognitionrateandspeed.

简介：当历史的中国书法工作是被数字化，检索的问题成为新挑战。但是，当前，没有光学字符识别技术能把书法字符图象变换成文本，存在笔迹字符识别途径也不能为它工作。这篇论文建议一条新奇途径到高效地根据类似检索中国书法人物：书法特性图象被许多歧视的特征代表，并且有合理有效性的高检索速度被完成。首先，没有类似于询问的可能性的书法字符被比较字符复杂性，笔划密度和笔划伸出一步一步地滤出。然后，类似的书法人物根据他们近似形状火柴生产的匹配的费用被检索并且评价。以便加快检索，我们采用了高维的数据结构—PK树。最后，算法的效率被一个初步的实验与3012幅书法特性图象表明。电子增补材料这篇文章(doi:10.1007/s11390-007-9077-8)的联机版本contatins增补材料，它对授权用户可得到。

简介：中国新词在中国自然语言处理是特别地有问题的。与因特网和信息爆炸的快开发，在中国自然语言处理为应用程序得到一本完全的系统词典是不可能的，当从字典的新词总是正在被创造。新词鉴定并且POS标注的过程通常被分开，词汇信息的特征不能充分被使用。没有被预先分割，一个潜伏的歧视的模型，联合潜伏的动态有条件的随机的地(LDCRF)和semi-CRF的力量，被建议从中国文本不管新词的类型同时地和他们的POS检测新词。不同于semi-CRF，在建议潜伏的歧视的模型，LDCRF被使用产生候选人实体，它加速训练速度并且减少计算费用。建议隐藏的semi-CRF的复杂性能被从LDCRF的产量建模的Nbest调节隐藏的变量的数字和候选人实体的数字进一步调整。一个new-word-generating框架为模型训练并且测试被建议，在哪个下面新词的定义和分布在真实文本遵循。全球特征叫了全球碎片特征因为新词鉴定被采用。我们从SIGHAN-6在语料库上测试了我们的模型。试验性的结果证明建议方法能够检测甚至低的频率和他们有令人满意的结果的POS标签的新词。建议模型与最先进的模型一起竞争地表演。

简介：Amethod,calledTwo-DimensionalExtendedAttributeGrammars(2-DEAGs)fortherecognitionofhand-printedChinesecharactersispresented.Thismethodusesdirectlytwodimensionalinformation,andprovidesaschemefordealingwithvariouskindsofspecificcasesinauniformway.Inthismethod,componentsaredrawninguidedandredundantwayandreductionsaremadelevelbylevejustinaccordancewiththecomponentcombinationrelationsofChinesecharacters.Themethodprovidsalsopolysemousgrammars,coexistinggrammarsandstructureinferringswhihconstrainredundantrecognitionbycomparisonamongsimilarcharactersofcomponentsandgreatlyincreasethetoleranceabilitytodistortion.

简介：Thepreviouslyproposedsyllable-synchronousnetworksearch(SSNS)algorithmplaysaveryimprotantroleintheworddecodingofthecontinuousChinesespeechrecognitionandachievessatisfyingperformance.Severalrelatedkeyfactorsthatmayaffecttheoverallworddecodingeffectarecarefullystudiedinthispaper,includingtheperfectingofthevocabulary,thebig-discountTuringre-estimatingoftheN-Gramprobabilities,andthemanagingofthesearchingpathbuffers.Basedonthesediscussions,correspondingapproachestoimpovingtheSSNSalgorithmareproposed.COmparedwiththepreviousversionofSSNSalgorithm,thenewversiondecreasestheChinesecharactererrorrate(CCER)intheworddecodingby42.1%acrossadatabaseconsistingofalargenumberoftestingsentences(syllablestrings).

简介：在统计学习框架下面，纸集中于怎么为中国合作参考分辨率为采矿和组织上下文特征在首语重复法分辨率上把传统的语言学调查结果用作一个指南。主要成就如下。(1)以便模仿“语法、语义的并行因素”，我们提取“特征和“一些semes”从实体提及的上下文展示并且合并的词形式和POS的袋子”他们进基线，特征设定。(2)因为它对使用太粗糙，一些词的特征形成，POS标签并且对在二个实体之间的语法、语义的并行提及的determine小花纹星罗棋布，我们基于语义类似计算为上下文特征重建建议一个方法，以便重建的上下文特征能更好接近“语法、语义的并行偏爱”的首语重复法决定因素。(3)我们使用一个entity-mention-based上下文特征代表而不是孤立的基于词的上下文特征表示，并且另外扩展上下文的窗户的尺寸，以便近似为首语重复法分辨率模仿“选择性的限制因素”。多水平上下文特征为合作参考分辨率是有用的实验表演，和与这些特征合并的统计系统表现很好在标准王牌数据集上。电子增补材料这篇文章(doi:10.1007/s11390-007-9076-9)的联机版本contatins增补材料，它对授权用户可得到。