Short syllables, hiatus and function words in the Demosthenic corpus: an automatic check of Blass's law and three independent signals on disputed speeches

Abstract

Note: a newer version of this record is available (version 1.1), with the preprint as a PDF that can be read in the browser. The content is unchanged. Please cite and use the latest version: https://doi.org/10.5281/zenodo.23442934 Blass’s law states that Demosthenes avoids sequences of three or more short syllables. This preprint describes a rule-based scansion of Attic prose, validated on verse whose meter fixes syllable quantity (0.5% and 1.0% error on Sophoclean trimeters and on the Odyssey), and a quantity lexicon derived from the spoken trimeters of Aeschylus, Sophocles and Euripides (0.12% error, one play held out at a time). Applied to 130 speeches of eight Attic orators and of the Demosthenic corpus, the measures recover Blass’s law: Demosthenes has 3.5 runs of short syllables per 1,000 syllables, against 7.3–10.3 in the other orators. Shuffling the words within each sentence raises his rate by 54% (71% with the lexicon) and leaves the others nearly unchanged. The effect survives a stricter shuffle that keeps word groups intact (+61%) and a comparison with his contemporaries Aeschines and Dinarchus alone. The avoidance therefore lies in the order of words, not in vocabulary. Hiatus places Demosthenes second only to Isocrates. Together with a function-word measure (Burrows’s Delta), these signals bear on disputed speeches: speech 45 is Demosthenic in all measures; speech 47 goes with the Apollodoran group; on speech 51 the measures divide two against two; 33, 34 and 48 are not Demosthenic; for 46 and 59, vocabulary agrees with the morphosyntactic results of Gorman and Gorman (2024). In Against Timocrates, sections 111–150 lack both rhythmic habits, as Blass had already observed. The automatic measures recover that observation without prior knowledge of it. All tests were pre-specified and dated before they were run. Failed hypotheses are reported alongside the others. The results are offered as converging evidence, not proof of authorship. The author is not a classicist; the study was carried out with the assistance of an AI system (Claude, by Anthropic), as stated in the Preface. Corrections and collaboration from specialists are welcome, in particular a manual annotation of vowel quantities in Attic prose. Files: the preprint (PDF); per-speech values for all 130 speeches (CSV) with a description of the columns (README).