Liviu Ciortuz, Vlad Saveluc

This report outlines a couple of lattice-based (un)learningstrategies proposed in a recent development ofunification-based grammars, namely the Fluid ConstructionGrammar (FCG) setup.

These (un)learning strategies are inspired by two linguisticphenomena occurring in a dialect spoken in the Banatarea of Romania. Children from that region – whereinfluences produced over centuries by Serbian, a Slaviclanguage, are obvious – learn in school the modernRomanian language, which is a Romance language.

This particular setup offers us the possibility to modelin FCG a two-step learning process: the first step isthat of learning a (perfective) verbal aspect similar tothe one already presented by Kateryna Gerasymova inher MSc thesis, while the second one is concerned withun-learning (or, learning another linguistic “construction”over) this verbal aspect. Thus, the interesting issue here ishow learning could continue beyond learning the verbalaspects. We will first give linguistic facts, after which wewill outline the way in which FCG could model such alinguistic process.

From the computational point of view, we show that theheuristics used in this grammar repairing process can beautomatically derived since the meanings associated towords and phrases are organized in a lattice of featurestructures, according to the underlying constraint logics.

We will later discuss the case of another verbal markerin the dialect spoken in Banat. It will lead us to sketch acomposite, quite elaborated (un)learning strategy.

Full Document (PDF)


author = "Liviu Ciortuz and Vlad Saveluc",
title = "{Learning to unlearn in lattices of concepts: A case study in Fluid Construction Grammars}",
institution = "``Al.I.Cuza'' University of Ia{c s}i, 
                 Faculty of Computer Science",
year = "2011",
number = "TR 11-01",
note = "URL:"