资源描述:
《Learning to Map Dependency Parses to Abstract Meaning Representations》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、LearningtoMapDependencyParsestoAbstractMeaningRepresentationsWei-TeChenDepartmentofComputerScienceUniversityofColoradoatBoulderWeite.Chen@colorado.eduAbstractAbstractMeaningRepresentation(AMR)isasemanticrepresentationlanguageusedtocapturethemeaningofEn
2、glishsen-tences.Inthiswork,weproposeanAMRparserbasedondependencyparserewriteFigure1:TheAMRannotationofsentence“Pierrerules.ThisapproachtransfersdependencyVinken,61yearsold,willjointheboardasaparsesintoAMRsbyintegratingthesyn-nonexecutivedirectorNov.29.
3、”tacticdependencies,semanticarguments,namedentityandco-referenceinforma-tion.AdependencyparsetoAMRgraphtheAMRparserismythesistopic,inthisproposalalignerisalsointroducedasapreliminarywewillpaymoreattentiontopreliminarywork-stepfordesigningtheparser.theA
4、MR-DependencyParsealigner.Toextracttherewriterulesandthestatistical1Introductionmodel,weneedthelinksbetweenAMRcon-AbstractMeaningRepresentation(AMR)(Ba-ceptsandthewordnodeswithinthedependencynarescuetal.,2013)isasemanticformalismthatparse.Anexamplealig
5、nmentisshowninFig.expressesthelogicalmeaningsofEnglishsen-2.AlignmentbetweenanAMRconceptandde-tencesintheformofadirected,acyclicgraph.pendencynodeisneededbecause1)itrepresentsAMRfocusesonthesemanticconcepts(nodesthemeaningofthesub-graphoftheconceptando
6、nthegraph),andrelations(labelededgesontheitschildconceptscorrespondingtothephraseofgraph)betweenthoseconcepts.AMRreliesheav-theheadwordnode,and2)thedependencynodeilyonpredicate-argumentstructuresdefinedinthecontainssufficientinformationfortheextractionPr
7、opBank(PB)(Palmeretal.,2005).Therepre-ofrewriterules.Forexample,thewordnodesentationencodesrichinformation,includingse-“Vinken”onthedependencyparsesideinFig.2manticroles,namedentities,andco-referencein-linkstothelexicalconcept“Vinken”and,further-format
8、ion.Fig.1showsanexampleAMR.more,linkstothe“p2/name”andthe“p/person”Inthisproposal,wefocusonthedesignofanconceptssince“Vinken”istheheadofthenamedautomaticAMRparserinasupervisedfashionentity(NE)“PierreVinken”andtheheadofthefromdependencyp