Research Class: : Identifying Semantic Domains of Lexical Concepts Using Syntactic Logical Connectives: From Corpus to Conceptualization

Datum održavanja: petak, 30.11.2018. u 11:00 sati, prostorija O-357
Predavač: doc. dr. sc. Bendikt Perak, Filozofski fakultet, Sveučilište u Rijeci
Naziv predavanja: Identifying Semantic Domains of Lexical Concepts Using Syntactic Logical Connectives: From Corpus to Conceptualization


Abstract:


The talk will present methods for identification polysemous and meronymic features of lexical concepts using corpus methods of extracting collocates of syntactic-semantic constructions. The idea is based on the application of graph theory and community detection algorithms that can be used to identify lexical near synonyms from purely syntactic-semantic features of friend-of-a-friend extended coordinated constructions, such as [noun1|adjective1|verb1 +and+ (noun2|adjective2|verb2]  +and+ noun3|adjective3|verb3). The graph communities are interpreted as conceptual networks that activate various facets of meaning within a latent matrix of the interconnected lexical meaning(s). The output of the networks can be used to produce measures of the conceptualization distance between concepts and domains yielding significantly richer structures than top-down methods such as WordNet, and more cognitively grounded ontological explanations of causal word correlation measures than standard computational methods,k such as word embeddings.

The talk will present the linguistic theory, logical principles, as well as the NLP pipeline developed by the researchers in the EmocNet project http://emocnet.uniri.hr/ and FORMALS.hr for corpus data extraction, data modelling, graph database storing, data enrichment, querying and visualization.