Konbinatu eta Irabazi! Hitzen Semantikaren Errepresentazio Osoagoaren Bila

Abstract

Artikulu honetan hitzen semantikaren errepresentazio osoagoa lortzeko algoritmo berria aurkezten dugu.Gure proposamenak bi metodo konbinatzen ditu: ezagutza-baseetan eta corpusetan oinarritutakoak.Hasieran, konbinatu barik, bi metodo horiekin antzekotasuna deituriko prozesu kognitiboa aztertudugu, eta gero azken horiek uztartzen dituen algoritmoa enpirikoki frogatu dugu.Bi faseetan gizairizpideetan oinarritutako antzekotasun urre-patroiak hartu ditugu ebaluaziorako erreferentzia legez.Metodoak konbinatzerakoan baliabideak asko optimizatu ditugu, konbinatu barik lortutako emaitzakberdinduz edota gaindituz. Ondorioz, teknika horiek informazio semantiko osagarria dutela frogatu dugu.In the article we present a novel algorithm which achieves a more integral representation of wordsemantics. Our proposal combines two methods: the ones based in knoledge-bases and in corpuses. Inthe beggining, we analize the cognitive process called similarity with those two methods, not combiningthem, and, afterwards, we empirically test the algorithm that yokes those methods. In both phases wehave taken similarity gold-standards based on human criteria as a reference for evaluation. We haveoptimized the resources dramatically when combining methods, obtaining equal or better results thanusing them separatedly. Thus, we have proved those methods have complementary semantic information.