הטעמה חישובית Computational Cantillation

We present ongoing work on computational approaches to biblical cantillation. We are interested in questions like: To what extent can explicit, explainable rules predict te,amim? Can AI be trained to correctly predict them, not necessarily explainably? How much variation is there between cantillation styles in different books/parts of the Bible?

AI models are used as powerful exploratory aids, especially for enumerating possible configurations, such as which "disjunctive signs" can separate between the words of a construct form. (Answer: almost all.)

Initial experiments using the Tiberian Masoretic text as structured data include rule-based tests of short-verse configurations, phonotactic constraints, and the munah/pashta alternation before zaqef. In the latter case, preliminary rules based only on syllable and vowel counting reach about 97.5% precision, suggesting that some accentual choices may be more formally constrained than traditional descriptions imply.

We will briefly demonstrate etnachta.com, an interactive tool for cantillation search.