Generated: 2026-07-12 20:46:34 UTC
This site provides comprehensive statistical analysis of Stephanos of Byzantium's Ethnika. Select a section below to explore different aspects of the text.
Explore word count distributions by entry type and starting letter. Includes normalized histograms with KDE curves and statistical tests comparing different entry types.
Identify Greek and English words associated with longer or shorter English translations after controlling for Greek source length.
Compare AI prompt versions against approved human translations with BLEU-4, chrF++, METEOR, ROUGE-L, BERTScore, COMET, BLEURT, length regression, and residual analysis.
Predict which approved-human passages are likely to score poorly under ordinary v3 translation runs using Greek vocabulary, current recogniser-rule matches, and a combined feature model.
Discover what the original Stephanos emphasized versus what the Byzantine epitomizer emphasized. Interactive visualizations reveal what was lost in the epitome and what was added or expanded.
Examine the distribution of etymology categories across the corpus, with comparisons between Delta and Non-Delta entries.
Statistical comparison of word counts between entries from the original Stephanos (Delta) and the Byzantine epitome (Non-Delta).
Detailed analysis of how different categories of proper nouns correlate with entry length. Explore which authors, historical figures, places, ethnic groups, and deities Stephanos emphasized.
Analysis of Stephanos's citations of Pausanias the Periegete. Did Stephanos have access to the complete text of Pausanias, or only certain portions? Statistical analysis of citation distribution with links to the cited passages.
Daily statistics for translation-guidance rules: discovery estimates, Zipf-like rank frequency, and top-rule headword coverage.
Exploratory fingerprinting work for formula, gloss, and grammar recogniser feature vectors, including UMAP clustering, Kappa/non-Kappa checks, and non-epitomised control coverage.
DB-backed vocabulary profiles, Zipf-style segment summaries, printed-edition control tests, and unsupervised sliding-window clustering over the Meineke word-lemma index.