Morpheme segmentation: statistical cross-check
Learns purely from corpus frequencies where words most likely split into building blocks: glyphs are iteratively merged into the most frequent adjacent pairs (byte-pair encoding), then at every possible word boundary it measures how many different continuations occur in the corpus (branching factor). A high branching factor marks a likely morpheme boundary. The candidates found this way, without any Hebrew hypothesis, are then matched against the hand-derived prefix/suffix table: does pure statistics independently confirm the same boundaries?
Heuristic approximation via a branching-factor threshold, not a full information-theoretic cost model.