Artificial Versifying (2019) (jeffreymbinder.net)

John Peter’s 1677 Artificial Versifying is essentially a hand-era algorithm: a fully specified, deterministic procedure that assembles grammatically and metrically correct Latin verses by drawing letters from indexed tables according to a pattern. By choosing sequences of digits (five for pentameter, six for hexameter; no zeroes) a user can reproducibly generate one of roughly 600,000 distinct lines, without needing to read or understand Latin. The original method — a fixed lookup-and-combine rule set that guarantees syntactic and metrical constraints — is a clear ancestor of template- and rule-based text generation. For the AI/ML community this is historically and technically instructive. It demonstrates how formal constraints and combinatorial mechanisms can produce valid output without semantic modeling, highlighting a long-standing split between syntactic correctness and semantic understanding. Peter’s system resembles finite-state or template systems (deterministic, fully interpretable, reproducible), contrasting with modern probabilistic models that learn distributions and produce fluent but less explainable outcomes. The piece is a useful reminder that generative creativity can be achieved by simple, auditable algorithms and that evaluation should separate form (grammar, meter) from meaning—an enduring design and evaluation lesson for current generative systems.