engleski

In search of the beauty and the beast: antonymy in syntactic constructions

The article describes syntactic dependency and sentiment potential graph methods for identifying antonymous lexical relations and lexical clusters in a corpus. Antonymy, understood as a cognitive mechanism of knowledge categorization (Lakoff 1987, Murphy 2006, Paradis 2011, Jones et al. 2012), provides access into both semantic and syntactic patterns of language (English and Croatian) that enable the conceptualization and expression of intricate emotions and sentiment. The ConGraNet graph method, developed at the EmocNet project (http://emocnet.uniri.hr/congracnet/), yields lexical communities of collocated lexemes that represent the sense structure of a seed word based on a syntactic dependency. By projecting the semantic value to the coordinated syntactical relation, we are able to filter out lexical collocations with high conceptual similarity, and construct labelled clustered lexical networks, which reveal polysemous and ambiguous senses of a source lexeme, as well as lexemes with antonym potential. In addition to the semantic potential of syntactic constructions (Goldberg 2006), the procedure uses the WordNet antonymy relation dictionary to filter out typical antonymic lexemes and sentiment analysis graph-based algorithms to assign psychological hedonic valency values to lexemes in a sense cluster. This, in turn, provides access to possible overarching patterns of emotion and sentiment conceptualization (Lüdtke 2015). The procedure of antonym potential identification includes the semantic analysis of syntactic relation enrichment and sentiment value distribution within a lexical dependency graph. We exemplify the application of the procedure on several lexemes in different languages and corpora. This graph approach (Perak and Ban Kirigin 2020) relies on the NLP processing of syntactic relations in natural languages, and can be used as a complementary method to other contemporary NLP resources to enrich semantic tasks, including word disambiguation, domain relatedness, sense structure, synonymy, metonymy, and metaphoricity, as well as to establish cross-/intra-cultural discourse variations of antonymical prototypical conceptualization patterns and knowledge representations. Insights procured by this methodology could deepen our understanding of conceptual-lexical relations as mirror images of our knowledge structuring processes.

lexical constructions, antonymy, sentiment analysis, graph-based algorithms

nije evidentirano