Pausanias Analysis - Stylometry

This section compares passage-level units using the current gpt-5.4-mini LLM grammar parses. It publishes the morphosyntactic feature model alongside traditional word-frequency and character n-gram baselines.

Interactive Projection

Mouse over passages in each feature family to inspect labels, token counts, and nearest-neighbor context.

Open Projection

Statistical Outputs

Distance summaries, outlier lists, nearest-neighbor tables, and target-vs-rest feature deltas.

Open Statistics

Feature Families

Morphosyntactic Grammar

UPOS tags, dependency labels, morphological features, and head-dependent POS relations from the LLM grammar parser.

180 selected features; projection method: UMAP.

Feature Number=SingFeature Gender=MascFeature Number=PlurFeature Case=AccUPOS NOUNFeature Gender=FemUPOS VERBFeature Case=GenUPOS DETDepRel detDepRel/UPOS det:DETHead direction det:left

Traditional Word MFW

Most frequent Greek surface forms after casefolding.

120 selected features; projection method: UMAP.

word καὶword δὲword ἐσword τοῦword τὸword ἐνword τὴνword τεword μὲνword τὸνword τῶνword τῆσ

Traditional Character 4-Grams

Most frequent normalized Greek character 4-grams, a language-agnostic stylometry baseline.

160 selected features; projection method: UMAP.

char 4-gram καὶ char 4-gram καὶchar 4-gram δὲ char 4-gram ἐσ char 4-gram τοῦchar 4-gram σιν char 4-gram τοῦ char 4-gram αὐτchar 4-gram ν καchar 4-gram τὴν char 4-gram σ καchar 4-gram τὸ