TRAINING AND VALIDATION OF ARTIFICIAL NEURAL NETWORK MODELS WITH THE CATEGORIES OF SENSE AND ABSURDITY

This paper argues for the originality of working with the categories of sense and absurdity, given the complexity of their processing—complexity that stems from the fact that such analysis extends beyond the purely computational aspects of textual data. Conventional approaches typically focus on syntactic or statistical regularities. In contrast, the categories in question demand cognitive interpretation of contextual cues, background knowledge, linguistic and national culture, world knowledge, implicit assumptions, and more. Our findings indicate that:1. While the current architectural sophistication of artificial neural networks allows for highly effective recognition of syntactic, statistical, and other patterns or anomalies, these models remain limited when processing textual absurdity that stems from violations of logic, continuity, referential coherence, or cultural context (including discourse-related factors).2. Such models perform better in tasks involving sense classification when context is enriched (e.g., through annotations) or when training data includes manually labeled examples of the target constructions.3. The problem of distinguishing between sense and absurdity, however, remains far from resolved. A comprehensive solution requires an integrated, interdisciplinary approach–combining mathematics, applied and computational linguistics, cognitive semantics, philosophy of language, natural language processing, and sentiment analysis. Future research should focus on evaluating the effectiveness of neural models integrated into systems for monitoring political online discourse. Specifically: a) developing specialized text corpora annotated for ontological, contextual, and other forms of absurdity; b) implementing linguistic models in logical and ontological knowledge representations informed by advances in the humanities (e.g., psychology, sociology); c) conducting a comprehensive analysis of how pragmatic and stylistic factors influence the classification of sense and absurdity. These directions may enhance existing models for machine-based textual analysis and generation, while also offering deeper insights into the nature, boundaries, and uniqueness of human interpretation.