Computational Scansion Across Early Modern English Drama: LLM-Derived Stress and Metrical Profiling with Shakespeare as a Benchmark Corpus

Manual scansion of Early Modern English drama remains indispensable for interpretation, yet it is neither scalable nor stable as a measuring instrument: competent readers disagree over stress realization, and such disagreements multiply in a corpus whose spelling, contraction, and rhythmic freedoms resist uniform treatment. This paper presents an LLM-based pipeline that annotates dramatic verse at scale, producing syllabified line strings, position-indexed prominence (0/1) profiles, descriptive meter labels, candidate caesura locations, and feminine-ending flags from minimally normalized texts. Instead of treating line-level agreement with human scansions as the primary criterion—an approach that presupposes a single "correct" reading where none is guaranteed—I validate the method distributionally, asking whether aggregated stress-position profiles recover the strong/weak asymmetries and cross-play stability associated with iambic pentameter. Using Shakespeare as a benchmark-rich case study, the analysis frames systematic divergences from hand-derived results as acceptable when they preserve the macroscopic shape of the profile and concentrate in predictably ambiguous environments (e.g., function-word stress, emphasis, and boundary effects), where human practice itself is least convergent. The principal advantage of computational scansion is therefore not interpretive authority but procedural consistency and scale: once prosodic structure is enumerated across tens of thousands of lines, metrical evidence becomes a reusable quantitative feature for authorship attribution, collaboration studies, and models of periodization and chronology across the wider Early Modern dramatic archive.