语法分析器
1680计 算 机 学 报2004年theprobabiIitiesofsub-constituents.Highaccuracyof
90.1%precision/recaIIonaveragehasbeenreported
[4].ontheEngIishparsersevaIuatedonPennTreebank
ForChinese,afeweffortsweremade.ZhouOiangpro-
[6]posedastatistics-basedChineseparser,inwhich
differentkindsofstatisticsextractedfromtreebankwere
combined.Highaccuracywasachieved,butitwase-
vaIuatedonasmaIIcorpus.ZhouMingproposeda
[7]bIock-baseddependencyparserforunrestrictedChi-
nesetext,whichintegratedphrasestructureapproach
anddependencyparsingapproachsuccessfuIIyanda-
chievedsatisfactoryinitiaIresuIts.trainingandevaIuatingtheparserisbriefIyintroducedinsection2.ThenthestatisticaImodeIandaIgorithmaredescribedinsection3and4indetaiI.Someexper-imentaIresuItswithanaIysisaregiveninsection5,anddetaiIederroranaIysesarepresentedinsection6.Fi-naIIy,concIudingremarksandfutureresearchdirec-tionsaregiveninsection7.2 SemanticdependencynetAIargecorpusannotatedwithsemanticdepend-[9]encywasbuiIt.Thecorpusconsistedofabout000,000words(about1,500,000Chinesechar-1,
However,aIIaboveworksarefocusedonsyntax.
VeryIimitedworkhasbeendonewithsemanticparsing
fordomain-unrestrictedtextinaIIIanguages.ForEng-
Iish,aIargesemanticannotatedcorpus(FrameNet)[8]
wasbuiIt,andsomeworkonthecorpushasbeenre-
ported,whiIeforChinesefewworkswasreportedat
present.ButSemanticparsingisanimportanttaskfor
naturaIIanguageprocessing(NLP),becauseunder-
standingthemeaningofsentences,evenparagraphs
anddocumentsisoneofthemaingoaIsofNLP,and
thiscannotaccompIishwithoutthesemanticparsing.
Furthermore,ChineseisaIanguageguitedifferentfrom
EngIish,foritismeaning-combinedandword-order-
freeIanguage.Sosemanticparsingismoreimportantto
ChineseNLP.
Inthispaper,anoveIparserbasedonsemantic
dependencyispresented.Theparsercandiscoverthe
semanticdependencyreIationsinasentence,which
areusedtorepresentthemeaningandstructureofthe
sentence.AIargecorpusannotatedwithsemanticde-
pendency,SemanticDependencyNet(SDN),was
buiIt[9].TheparseristrainedandevaIuatedonit.Va-
riousexperimentsarecarriedouttoanaIyzetheper-
formanceoftheparser.AIthoughtheresearchofthe
parserisstiIIinitsearIystage,ithasshownapromis-
ingperformance.TheaccuracyreachedashighIeveI
astheChinesesyntacticparser[7].
ThisparserhasthreenotabIefeatures.First,it
parsesthesemanticstructureofsentencedirectIy,a-
voidingthestageofparsingsyntacticstructure.Sec-
ond,itisadata-drivenparser,fornohand-crafted
ruIesareneededunIikephrasestructureparsers.In
ourmodeI,everywordispermittedtobedependenton
anyotherwordinasentence.FinaIIy,theprobabiIity
ofreIationbetweenwordsistrainedonannotatedcor-
pus.AIIknowIedgeusedintheparserisautomaticaIIy
acguiredfromthecorpus.
InthefoIIowing,theannotatedcorpususedinacters).ThetextswereseIectedfromthenewsofPeo-pleDaily,anditsdomaincoversthepoIitics,econo-my,science,sports,etc.AIIwordsinthecorpusweretaggedwiththesemanticcIasses.AIIsentenceswereannotatedwithsemanticdependencyreIations,whichcouIdrepresentChinesesyntacticandsemanticstruc-turesatthesametime.2.1 SemanticclassesThedefinitionofSemanticcIassesfoIIowsDictiona-ryofSynonymouSWordS(TongYiCiCiLin)[10].AIIthesesemanticcIassesareorganizedasatree,whichhasthreeIeveIs.ThefirstIeveIcontains18cIasses,thesecondIeveIcontains101cIasses,andthethirdIeveIcontains1434cIasses.Sincetheprojectoftag-gingsemanticcIasseswascompIetedin1998,thean-notatedcorpushasbeenusedforseveraIyearsintheresearchesonautomaticsemanticcIasstagging.Now,highaccuracyof92.73%hasreached[11]ontaggingsemanticcIasses.2.2 SemanticdependencyrelationshipInourtaggingscheme,dependencygrammarisusedtorepresentthemeaningandstructureofasen-tence,andconceptionsofheadwordandIexicaIinfor-mationarestressed.LetSbethesentencecomposedofwordstaggedwithsemanticcIasses,S={〈w1,S1〉,〈w2,S2〉,…,〈wN,SN〉}.Then,aIistofsemanticde-pendencyreIations[12]isdefinedas:SRL={SR(1),SR(2),…,SR(N)},whereSR(i)=(Hi,Ri).SRstandsforsemanticreIation.SR(i)=(Hi,Ri)statesthattheHi-thwordistheheadwordtothei-thwordwithse-manticreIationRi.Iftheword isthesententiaIhead,thenSR( )=(H ,R )=(-1,“Kernelword”).ForexampIe,asentenceisshowninFigure1(a).ThesemanticdependencyreIationIistandse-manticdependencytreeareshowninFigure1(b)and(c)respectiveIy.
中文语义依存关系分析的统计模型_英文__文学_高等教育_教育专区 暂无评价0人阅读0次下载举报文档 中文语义依存关系分析的统计模型_英文__文学_高等教育_教育...
基于配价结构和语义依存关系的句法分析统计模型_天文/地理_自然科学_专业资料 暂无评价
中南大学 信息科学与工程学院,湖南 长沙,410083) 摘要:利用语义、语法等语言知识,建立一种基于依存关系的句法分析统计模型,并利用改进的句法分析模型 进行句法分析...
语共现统计的方法是一种最基本 的方 的基 本 ...斯 坦福 大学 自然语 言 句子语义依存关系分析可以...(6):531—534.(in Chinese) 模型进行 了 比较 ...
,主要存在两种方法 : ①基于向量空间模型的方法 。...的句法分析 , 找出依存关系 , 并在 依存分析结果...下文中的统计特性 ,而没有考虑词本身的语义信息 。...
汉语语义依存分析_计算机软件及应用_IT/计算机_专业资料。汉语句法分析方面的文章...汉语语义分析模型研究述... 7页 1下载券 基于树库的汉语依存句法... 5页 ...
关键词: 计算机应用; 中文信息处理; 语料库; 语义依存关系;5 知网6; 动态...类分得越细, 统计数据的稀疏问题越严重, 会对分析模型的鲁棒性造成不良 影响。...
原信息为特征,基于该特征训练得到支持 向量机(SVM)依存分析模型,从而有效识别出了术语内部的依存关系;在依存分析基础上利用统计和 规则相结合的方法进行术语语义分析...
汉语语义分析模型研究述评_文学_高等教育_教育专区。汉语语义分析模型研究述评维普...中文语义依存关系分析的... 11页 免费 基于统计学习模型的句法... 12页 免费...
http://research.microsoft.com/~jfgao/paper/ 附中文参考文献: [10] [11] [13] [14] 李明琴,李涓子,王作英,陆大纟 .中文语义依存关系分析的统计模型....
热门文档