Mining Linguistic Patterns from Historical Educational Resources Using Hybrid AI-Blockchain

Historical educational resources-digitised textbooks, manuscript pedagogical texts, and archival instructional materials-represent an underexploited source of linguistic pattern data for researchers studying language change, educational discourse evolution, and shifting pedagogical vocabulary, but their value is limited by two challenges: NLP models that underperform on archaic text, and provenance uncertainty about whether a digitised document accurately represents its physical original. This paper proposes a hybrid AI-Blockchain architecture combining a BERT-based linguistic pattern extractor (fine-tuned for archaic language) with a Hyperledger Fabric provenance chain and IPFS content-addressed storage, applied to 24,000 historical educational documents (18,000 training / 6,000 test). The hybrid achieves$\mathbf{F} \mathbf{1}=\mathbf{0. 8 6 4}$for linguistic pattern extraction, outperforming AIonly (F1 0.764; +13.1%) and traditional NLP (0.584; +47.9%). The blockchain layer verifies provenance for 96.0% of documents (23,040/24,000) and detects tampering in 97.8% of manipulated test documents (47/48), with IPFS retrieval at 284ms and Hyperledger Fabric consensus at 4.8 seconds. OCR noise degrades F1 by 16.2% and cross-lingual documents by 9.3%, each with a documented mitigation.

Hali tarjima qilinmagan