Swedish AI decodes 17th-century Borg Cipher

According to AzerNEWS, researchers in Sweden developed an AI pipeline that substantially shortens the time needed to decode historical ciphers. AzerNEWS reports that, per project leader Professor Beata Megyesi, the system deciphered the Borg Cipher - a roughly 400-page 17th-century manuscript that uses 34 distinct symbols and records medical and pharmaceutical knowledge - in about 28 minutes, compared with prior manual efforts measured in weeks. Stockholm University's pages document the longer-running, interdisciplinary DECRYPT project led by Megyesi and the data and tools associated with it.

Per AzerNEWS, the system combines character recognition, pattern reconstruction, and language modeling to reconstruct plaintext from cipher symbols. Stockholm University materials describe TranscripTool for turning cipher images into machine-readable transcriptions and CrypTool for assisting decipherment workflows, alongside language resources and diplomatic transcriptions used for evaluation. The project collections DECODE and HistCorp provide training and benchmark data.

the project exemplifies applying computer vision, computational linguistics, and statistical or neural sequence models to archival and humanities problems. Comparable efforts typically pair OCR-like symbol recognition with probabilistic or neural language models and human-in-the-loop validation to handle noise, sparse training data, and diverse scripts. Recurring challenges for practitioners include domain-specific symbol inventories, limited parallel corpora, and error propagation between transcription and decipherment stages.

For digital-humanities and NLP practitioners, reproducible pipelines and shared datasets matter more than single-case runtimes. Observers should look for releases of transcription and alignment data, code and model checkpoints for TranscripTool and CrypTool, benchmark evaluations across multiple ciphers and languages, and documentation of human-in-the-loop steps and uncertainty estimates. Note that source figures differ - Stockholm University reports 7,000+ encrypted items while AzerNEWS cites a 20,000+-document database - a discrepancy worth flagging.