Jure Zupan: Semantika in matematika
Povzetek. Semantika je posebna veja jezikoslovja, ki se osredotoča na pomen besed. Vprašanje je kaj beseda 'pomen' v tej zvezi pomeni? Kaj lahko posamezna beseda 'pomeni' v različnih kontekstih, v različnih okoliščinah ali izrečena od različnih govorcev? Veliko takih vprašanj, ki jih rešuje semantika, še ni rešenih. Zaradi tega bi se v tem predavanju radi vprašali ali lahko matematika s svojimi rigoroznimi postopki in koncepti pomaga reševati semantične probleme pomenov posameznih besed? Na predavanju bo najprej prikazan primer Kohonenove nevronske mreže, ki lahko loči pesmi glede na avtorstvo pesnikov. Glavni del predavanja pa je namenjen prikazu konstrukcije in generiranja mreže 60.000 pomenov slovenskih besed (44.000 samostalnikov in 16.000 glagolov), ki rezultira v enovitem usmerjenem grafu s skoraj 90.000 vozli. Prikazano bo nekaj implikacij takega grafa na področju socialne lingvistike. Nakazan bo tudi problem grafovsko-teoretičnih invariant, s katerimi bi lahko enolično označevali skupine besd, stavkov ali celih besedil.
Abstract. Semantics as a special branch of Linguistics is focused on the meanings of words, but what do we mean by 'meaning'? What the words can mean in different contexts in different situations and spoken by different speakers? Such questions are not yet resolved entirely, so the question we would like to address is: Can Mathematics help by its armory of procedures and concepts to clarify some of the problems of word meanings? In the lecture first an example how a Kohonen neural network can distinguish poems by different poets will be given. In the main part the lecture the composition and the generation of a network connecting 60,000 Slovenian word meanings (~44,000 nouns and ~16,000 verbs) in a single directed graph of almost 90,000 nodes, will be discussed and explained. Several implications of such huge structure to the social linguistics will be shown. The problem of using the graph-theoretical invariants for the assignation of groups of word, entire sentences or larger parts of text will be addressed.
