CORPUS-BASED SYNTACTIC-SEMANTIC GRAPH ANALYSIS: SEMANTIC DOMAINS OF THE CONCEPT FEELING

This research' exemplifies the corpus-based graph approach to the syntactic-semantic analysis of a concept feeling using the Construction Grammar Conceptual Network methodology. By constructing a lexical network from grammatically tagged collocations of the English and the Croatian web corpora, the structure of the semantic domains is revealed as a set of sub-graphs derived from the source lexeme's friend-of-a-friend graph. The subgraph structures, calculated with the community detection algorithm, are interpreted as the semantic domains associated with the source lexeme's conceptual matrix. Lexical structures are analyzed using a centrality algorithm that determines the overall rank of the salience and semantic relatedness to the source concept feeling. This empirical approach can be used for developing NLP methods and tasks, such as computing semantic similarity, sense disambiguation, sense structuring, as well as for comparative corpus and cross-cultural studies. ConGraCNet has a web application on the page http://emocnet.uniri.hr/congracnet.