Learning Recursive Phrase Structure Combining the Strengths of PDP and X-Bar Syntax

《Learning Recursive Phrase Structure Combining the Strengths of PDP and X-Bar Syntax》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、DepartmentofComputerScienceUniversityatAlbanyStateUniversityofNewYorkLearningRecursivePhraseStructure:CombiningtheStrengthsofPDPandX-BarSyntaxTR91-5GeorgeBergHHUNIVERSITYATALBANYSTATEUNIVERSITYOFNEWYORKLearningRecursivePhraseStructure:CombiningtheStrengthsof3PDP

2、andX-BarSyntaxGeorgeBergComputerScienceDepartment,andDepartmentofLinguisticsandCognitiveScienceStateUniversityofNewYorkatAlbanyAlbany,NYUSAberg@cs.albany.edu1IntroductionOneofthebiggestchallengesfacingproponentsofconnectionistmodelsofnaturallanguageprocessing(NL

3、P)istherichstructureoflanguage.Asconstruedinmostlinguistictheories,asentenceconsists(inpart)ofvarioustypesofphraseswhicharecomposedofotherphrasesand/orthewordsofthesentence.Thisstructurealsohelpsconstrainthesemanticsormeaningofthesentence(e.g.whatelementslltheA

4、gent,Patient,Instrument,etc.semanticcaseroles;towhatotherelement(s)inthesentencemayapronounlegitimatelyrefer).TobetakenseriouslyasmodelsofNLP,connectionistsystemsmusteithershowhowtheycanaccommodatestructureandstructurally-sensitiveprocessing,orproposeandjustifys

5、omealternativewhichexplainsthesamephenomena.Inthepastfewyears,systemshaveemergedwhichaddresssomeoftheseissues.However,thesesystemshaveapriorilimitsonthesentencestructurestheysupport.Inthispaper,weintroducetheXERICParser,aconnectionistsyntacticanalyzerwithaxedne

6、tworkarchitecturewhichplacesnopriorlimitontheextentofthestructureofasentence.Inpracticethelimitsre ecttheactualstructuresofthetrainingdataandthelimitationsofthenetwork'strainingregimenandrepresentation.2StructureinConnectionistNLPOneoftheunderlyingtenetsofthemod

7、erncognitivesciencesisthathumanlanguageshaveaformandconstraintsontheirmeaningswhicharewell-accountedforbyassumingthatsyntacticallyandsemanticallytheyhaveacompositionalstructureandthatthisstructureaectsthemeaningofthesentencesinthelanguage.And,asputforthmostforc

8、efullybyFodorandPylyshyn(1988),itisdiculttocapturethisnotionofstructureandstructure-sensitivityinconnectionistmodelsofNLP.Despitetheiruseofstrawmentomaketheircase,Fo