Urdatorn/sphragis-metre

Sphragis Metre

Sphragis Metre is the scanned-line companion to Urdatorn/sphragis. It supports Ancient Greek authorship attribution from exact 1-, 5-, 10-, and 50-line units combining human metrical annotation with uniform automatic dependency annotation. Every curated Hypotactic passage is parsed with the pinned Ericu950/Stoicheia-tagger-parser checkpoint.

Configurations

Configuration	Authors	Row unit	Train rows	Validation rows	Test rows
verse_1	17	line	69,450	8,300	8,300
verse_5	17	5-line chunk	13,890	1,660	1,660
verse_10	17	10-line chunk	6,945	830	830
verse_50	17	50-line chunk	1,389	166	166

The public configuration prefix is verse. Because every row belongs to this metrical-verse publication, the redundant constant genre column is omitted.

The 15 retained labels are Aeschylus, Apollonius Rhodius, Aratus, Callimachus, Hesiod, Homeric-Iliad, Homeric-Odyssey, Lycophron, Nicander, Nonnus, Oppian, Pindar, Quintus Smyrnaeus, Theocritus, and Theognis. All splits are fixed at the configuration suffix. _1, _10, _50, and _100 contain exactly the same source lines in total; seeded remainder removal is determined by the 50-line task and reused across task sizes.

Lines per poet

These are the exact retained atomic lines in verse_1. The same lines are represented in the 5-, 10-, and 50-line configurations.

Poet label	Train	Validation	Test	Total
Aeschylus	1,600	200	200	2,000 (2.3%)
Apollonius Rhodius	4,650	550	550	5,750 (6.7%)
Aratus	900	100	100	1,100 (1.3%)
Callimachus	1,100	100	100	1,300 (1.5%)
Hesiod	1,450	150	150	1,750 (2.0%)
Homeric-Iliad	12,550	1,550	1,550	15,650 (18.2%)
Homeric-Odyssey	9,650	1,200	1,200	12,050 (14.0%)
Lycophron	1,150	100	100	1,350 (1.6%)
Nicander	1,250	150	150	1,550 (1.8%)
Nonnus	17,100	2,100	2,100	21,300 (24.8%)
Oppian	2,800	350	350	3,500 (4.1%)
Pindar	2,750	300	300	3,350 (3.9%)
Pseudo-Oppian	1,700	200	200	2,100 (2.4%)
Quintus Smyrnaeus	7,000	850	850	8,700 (10.1%)
Theocritus	2,150	250	250	2,650 (3.1%)
Theognis	1,100	100	100	1,300 (1.5%)
Tryphiodorus	550	50	50	650 (0.8%)
All labels	69,450	8,300	8,300	86,050 (100.0%)

from datasets import load_dataset

metre = load_dataset("Urdatorn/sphragis-metre", "verse_10")

The Cynegetica is transmitted under Oppian's name but is not by the poet of the Halieutica. The two works previously shared one label, of which 38% was written by somebody else; Oppian is now the Halieutica alone and the Cynegetica is Pseudo-Oppian.

Pseudo-Oppian is retained rather than excluded, unlike other pseudonymous labels. The Cynegetica is the work of one Syrian poet, demonstrably neither the Cilician poet of the Halieutica nor anyone else published here, so the label denotes a single author and carries no risk of being another label's hand under a second name. Only the poet's name is unknown.

Representation

For convenient human inspection in the Hugging Face dataset viewer, syllables and conllu are the first and second columns in every configuration.

text, conllu, metre and syllables are real nested Arrow columns with one element per constituent line, not JSON packed into strings, so load_dataset returns Python lists and dictionaries with nothing left to parse. Each syllables line object contains its features mapping and its ordered syllables list. Atomic _1 rows therefore contain one line object, while _5, _10 and _50 retain 10, 50, or 100 explicitly distinguishable lines instead of flattening their annotations.

Each conllu unit is a list of sentences, each sentence a list of token objects keyed by the CoNLL-U column names. xpos is not among them: the nine-position tag is folded into feats, which carries all seven Greek tenses as Tense plus Aspect — see "Column structure" in the Sphragis README, whose representation this publication shares. scripts.conllu_units.render_conllu returns canonical CoNLL-U text for a unit.

The line-level features object always contains hiatus, longa, brevia, morae and caesurae. Long syllables contribute 1 mora and short syllables 0.5; transcribed elided syllables contribute no metrical time. longa and brevia count long and short syllables, while hiatus counts syllables carrying that feature.

caesurae holds one object per caesura, giving its position measured two ways: isochronic is the metrical time before it and isosyllabic is the syllable count before it. Earlier releases spread the same information over sparse keys with positional names (isochronic-caesura-3, isosyllabic-caesura-4), which no Arrow type the Hugging Face datasets library understands can express.

[
  {
    "features": {
      "hiatus": 1,
      "longa": 2,
      "brevia": 2,
      "morae": 3,
      "caesurae": [
        {"isochronic": 2, "isosyllabic": 3}
      ]
    },
    "syllables": [
      {"text": "μῆ", "quantity": "long", "features": []},
      {"text": "νι", "quantity": "short", "features": []},
      {"text": "ν", "quantity": "short", "features": ["caesura", "hiatus"]},
      {"text": "ἄει", "quantity": "long", "features": []}
    ]
  }
]

Punctuation-delimited contexts are first reconstructed across consecutive metrical lines. All editorial punctuation is then removed before inference, including punctuation attached to words; a malformed Hypotactic word span that contains internal whitespace is split into separate parser tokens. The cleaned context is parsed in full with Stoicheia and every dependency relation is normalized to the Universal Dependencies v2 inventory (AuxP, for example, becomes case; unknown residual labels fall back to dep). Each sentence tree is then cropped back to its constituent line boundaries. Thus a line break never becomes an artificial parser sentence boundary. If one line contains material from more than one sentence, its conllu unit holds those sentences in order as separate token lists; 2,092 units do. Context-window fallback splits, if required, occur only between lines. Every result is validated with the official CoNLL 2018 loader. syllables contains the human syllable-level metrical annotation plus two uniformly derived word-boundary features. caesura is added to the final syllable of every non-line-final Hypotactic span.word, independently of incidental HTML formatting. hiatus is likewise derived uniformly: upstream hiatus classes are removed, and the feature is added to the first word's final syllable exactly when grc_utils.vowel returns true for both that word's final character and the next word's initial character. Line-final syllables never receive either derived boundary feature. The redundant scansion field is not published.

Source syllable transcriptions are preserved rather than silently corrected. Fourteen retained lines contain an upstream text–syllable transcription mismatch; this audit count is recorded in metadata/build_report.json.

syntax_annotation is uniformly predicted, and treebank_source is uniformly stoicheia_tagger_parser, in atomic rows and chunks. source_records pins Stoicheia revision cd8ae1658c364874c3b6f4df37bd1d46313e3cea. The prior gold trees are not published as model input. Where exact alignment exists, they are used only as a diagnostic reference for token-boundary, lemma, UPOS, dependency-relation, UAS, and LAS measurements recorded in metadata/build_report.json.

Sphragis Metre publishes the same model-facing representation as Urdatorn/sphragis, so one model can read both without changing its feature extraction. CoNLL-U documents contain token rows only; the text column carries the text. MISC is empty, and any head repair is recorded per token in native_syntax. Word forms, lemmas and text are lowercased with grc_utils.lower_grc, elision is written the same way everywhere, and UPOS, XPOS, FEATS and DEPREL use the shared inventories documented in the Sphragis README. Full audit provenance remains available in dedicated metadata columns.

This publication is annotated by one parser throughout, so it does not carry the cross-treebank confound that motivates the shared inventories; they are applied here for comparability, not for leakage control.

The source revisions, split manifest, exclusions, and validation results are recorded under metadata/.