BibTeX
@MISC{_towardsfullautomationoflexiconconstruction,
author = {},
title = {TowardsFullAutomationofLexiconConstruction},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Wedescribeworkinprogressaimedatdevelopingmethodsforautomaticallyconstructinga lexiconusingonlystatisticaldataderivedfrom analysisofcorpora,aproblemwecalllexical optimization. Specifically,weusestatistical methodsalonetoobtaininformationequivalent tosyntacticcategories,andtodiscoverthesemanticallymeaningfulunitsoftext,whichmaybemulti-wordunitsorpolysemousterms-incontext. Ourguidingprincipleistoemploya notionof“meaningfulness”thatcanbequantifiedinformation-theoretically,sothatplausible variantsofalexiconcanbejudgedrelativeto eachother.Wedescribeatechniqueofthisnaturecalledinformationtheoreticco-clustering andgiveresultsofaseriesofexperimentsbuilt arounditthatdemonstratethemainingredientsoflexicaloptimization. Weconcludeby describingourplansforfurtherimprovements, andforapplyingthesamemathematicalprinciplestootherproblemsinnaturallanguageprocessing. 1
