Learning to Resolve Natural Language Ambiguities: A Unified Approach

Weanalyzea fewofthecommonlyusedstatisticsbased andmachinelearningalgorithmsfornaturallanguage disambiguationtasksandobservethattheycanbcrecastaslearninglinearseparatorsinthefeaturespace. Eachofthemethodsmakesa prioriassumptions,which itemploys,giventhedata,whensearchingforitshypothesis.Nevertheless,asweshow,itsearchesaspace thatisasrichasthespaceofalllinearseparators. Weusethistobuildan argumentfora datadriven approachwhichmerelysearchesfora goodlinearseparatorinthefeaturespace,withoutfurtherassumptions onthedomainora specificproblem. Wepresentsuchanapproach- a sparsenetworkof linearseparators,utilizingtheWinnowlearningaigorlthrn-andshowhowtouseitinavarietyofambiguity resolutionproblems.Thelearningapproachpresented isattribute-efficientand,therefore,appropriatefordomainshavingverylargenumberofattributes. In particular, wepresent an extensive experimental comparisonof our approach with other methodson several well studied lexical disambiguationtasks such as context-sensltlvespelling co...

• Type: Conference
• Year: 1998
• Where: AAAI
• Authors: Dan Roth
• Keywords: AAAI 1998, Argumentfora Datadriven Approachwhichmerelysearchesfora, Intelligent Agents, Prepositional Phrase Attachmentand, Weanalyzea Fewofthecommonlyusedstatisticsbased Andmachinelearningalgorithmsfornaturallanguage