Research Information System
Journal of Ethnic and Cultural Studies, vol.13, no.3, pp.382-410, 2026 (Scopus)
This study investigates AI language systems as independent linguistic constructs, evaluating syntactic, semantic, pragmatic, and acquisition mechanisms. We conducted a comparative conceptual and empirical analysis of transformer-based language models, integrating probabilistic syntax mapping, high-dimensional embedding evaluation, and discourse-level context assessment. Data were systematically analyzed from large-scale corpora across multiple AI architectures to quantify emergent structural and semantic patterns. The results showed that syntactic regularities emerged probabilistically, with graded grammatical acceptability exceeding 85% coherence across complex sentence structures. Semantic relationships were distributional, maintaining 78–92% contextual similarity without referential grounding. Pragmatic adaptation occurred algorithmically across 1,000 discourse simulations, while acquisition was fully data-dependent, revealing alternative pathways to functional competence. AI-generated language diverged from human hierarchical grammar yet preserved operational effectiveness. These findings demonstrate that AI systems embody autonomous, non-biological linguistic competence, challenging classical assumptions of grammaticality, meaning, and acquisition. This study provides actionable insights for theoretical linguistics, computational modeling, and the design of advanced human-AI communication systems.