BPE VMML Cross-Text Instability in Tagalog: Implications for Voynich Manuscript Discrimination

Paper 7 (Silva 2026, DOI: 10.5281/zenodo.20633920) found that Tagalog is the only natural language whose BPE morpheme structure enters the Voynich discriminant zone (VMML = 5.914, CI [5.77–6.05]) among 71 tested corpora. The obvious next question: is the elevated VMML a property of Tagalog, or of one text? I tested this by running the same full-corpus computation on a second Rizal novel, El filibusterismo (n = 114,777 tokens; same protocol as Paper 7: max_merges=200, min_freq=3, no token cap). The result was 5.578 — below the alphabetic ceiling (5.748). Cross-text Δ = 0.336 units, larger than the Voynich zone width (0.28 units). Tagalog's elevated VMML is text-specific, not a language property. A preliminary Austronesian expansion (6 languages) finds Ilocano tentatively above the ceiling (VMML = 5.785, n = 12,807; preliminary). All Malayo-Polynesian languages remain well below. A corpus-size matched character shuffle permutation test (n=50, all corpora at n=19,968) reveals that despite similar VMML values, Voynich and Tagalog reach them through structurally different mechanisms: Voynich drops 22.0% (CI95: [5.07, 5.82]) vs Tagalog 12–14% (CI95: [4.77, 5.04]); CI ranges do not overlap. The Voynich discriminant zone remains unoccupied by any known natural language among the 71+ tested.