pyaegean loading…

The toolkit running entirely in your browser via Pyodide — no install, no server. Every example below is the pure-Python core running client-side: the Greek pipeline, scansion, accent and sandhi rules, word analysis, the bundled Koine lexicon, New Testament sample, and idiom lexicon, the offline work catalogue, the deciphered-syllabary Greek bridge, the bundled corpora of all four Aegean scripts (with statistics, queries, numerals, and sign inventories), the find-site gazetteer, citation and fingerprinting, the file importer, EpiDoc in both directions, cross-script comparison, the pipeline's evidence-class explanation, corpus-health and editorial-apparatus reports, Linear B scribal dossiers, Linear A seriation, and catalogued sign-variant groups. The neural and AI layers, and the network-only backends (fetching real works, LSJ), don't run here. First load fetches the package, so give it a few seconds.

Beta Code → Unicode

Type polytonic Greek in plain ASCII — no Greek keyboard needed

Greek NLP pipeline

A Greek sentence: get per-token part of speech and lemma (dictionary form) from the offline baseline. This is a first pass; rule and lookup results can still be wrong

Pipeline explained — the lemma's evidence class

The same sentence, but showing how each lemma was reached: its evidence class (attested · rule · seed · paradigm · identity · unresolved · punct) and whether the lemma is an identity/unresolved fallback. An unflagged result can still be wrong, and the flag does not assess POS or morphology. The class records provenance; no confidence numbers are supplied here

Greek word — syllables · accent · pronunciation · morphology

One Greek word: its syllables, accent class, reconstructed IPA (Attic and Koine), and candidate morphological readings

Accent placement

An unaccented Greek word: recessive accent by default (finite verbs), or give a lemma for the persistent rule (try word θεου with lemma θεός)

Sandhi resolution

One token with crasis, elision, or movable-ν: it is expanded only from curated forms, never guessed (try κἀγώ, ἀπ', ἐστίν, τοὔνομα)

Syllable quantities

A Greek word: the metrical weight of each syllable (heavy / light / common, the raw material of scansion)

Text profile

Any text: the observable features (writing system, polytonic vs bare vowels, a Beta Code look, majuscule share, editorial marks, numeral density). It describes what the characters are, never a genre. Try mh=nin a)/eide qea/ or ΜΗΝΙΝ

Metrical scansion

A verse line — scan it into long/short syllables (— heavy, ⏑ light, × anceps)

New Testament glossing (Koine)

A New Testament Greek word — look it up in the bundled Dodson lexicon

New Testament verse (offline sample)

A verse from the bundled sample (John 1 and Philemon 1, Nestle 1904): each token with its gold lemma, Robinson morph tag, Strong's number, and UPOS (try John 1.1 or Philemon 1.4)

Greek idioms

A phrase or sentence: curated non-compositional idioms are detected and glossed with their real meaning (try ἐφ' ἡμῖν or οὐκ ἔστιν ὅπως in context)

Offline lemmatizer, honest misses included

One Greek word: the offline lemma and POS. When no rule or table applies the word comes back unchanged and flagged. A rule or lookup result can still be wrong (compare ἀνθρώπων with προφήτου)

Dictionary deep-link

Any Greek word — build a Logeion link to its dictionary entry (LSJ, Middle Liddell, Cunliffe, Autenrieth, and more). The word is lemmatized first.

Greek work catalogue

Search the offline index of ~1,778 loadable works — by author, title (English or Greek), or free text. Each id loads with greek.load_work(id).

Syllabary → Greek bridge

A word in a deciphered syllabary, read as Greek from the bundled lexicon (try po-me, wa-na-ka, or Cypriot pa-si-le-u-se)

Cypriot inscription — read from the corpus

An inscription id from the bundled Inscriptiones Graecae XV 1 corpus (178 Cypriot-syllabary texts) — its find-place, transliteration, the Greek reading where the text is Greek, and the source edition (try IG XV 1, 95, IG XV 1, 1, or IG XV 1, 2)

Some texts are Eteocypriot (an undeciphered language in the same syllabary), so they have a transliteration but no Greek reading.

Linear A — wildcard sign search

A sign pattern over the 1,721-inscription corpus — signs are written with - between them, and * matches any one sign (so *-RO finds every two-sign word ending in RO; KU-RO matches that word exactly)

Linear A is undeciphered; this matches transliterated sign sequences, not meanings.

Linear A — accounting reconciliation

A tablet id — check its stated KU-RO total against the sum of its line items (try HT13, HT9b, HT11a, HT31)

Exploratory: it flags where a scribe's total doesn't match the items, sometimes a real discrepancy, sometimes a damaged reading. Exact arithmetic does not verify readings or metrology. Teaching eligibility requires intact text and at least one total within the tolerance; it does not mean every total balances.

Linear A — corpus statistics

Dispersion (which words are corpus-wide vs site-bound) and keyness (which words are characteristic of one find-site against the rest); try Haghia Triada, Khania, Zakros

Deterministic counts over transliterated sign-groups, not meanings: Linear A is undeciphered.

Linear A — compound query

The compound-query engine: inscriptions from a site AND words containing a sign, words output (the count is document frequency)

Aegean numerals

Numeral tokens as transliterated (integers, fraction glyphs like ½ and ³⁄₄, the ≈ approximation prefix): each is parsed and the line is summed, the way the accounting checker reads a tablet

Sign inventory lookup

A sign label in any of the four Aegean scripts (or Greek): its glyph, Unicode codepoint, and sound value where one exists (try PA, RA₂, *301, Cypro-Minoan CM008)

Linear B tablet with Greek readings

A tablet from the bundled Linear B corpus: its lines, plus every word with an attested Greek reading in the bundled lexicon (try PY Ta 641, PY Er 312, MY Ge 602)

Cypro-Minoan — signs only

A bundled Cypro-Minoan document: sign identities are all the field offers. The script is undeciphered, so no sound values or readings are shown.

Find-site geography

A find-site name (Phaistos) or a Linear A word (KU-RO): coordinates from the bundled Pleiades-aligned gazetteer, or every site the word occurs at

Cross-script phonetic comparison

How close is a Linear B spelling to a Greek word? (Syllabic spelling is defective — it drops final consonants and clusters — so the score is a guide, not a verdict.)

Bring your own text

Paste any Greek and turn it into a corpus — token count and the most frequent words (the same importer behind aegean import). The source text and offsets preserve punctuation, foreign text, and numerals; only lexical words enter word counts

Word counts with an explicit population

Count words in a bundled corpus by exact surface spelling, a normalized key, or existing lemma annotation. Normalization preserves diacritics. Missing lemmas are excluded; this card runs no analyzer. Token frequency counts occurrences; document frequency counts each document once per word, including empty documents in its denominator

Status selection applies to the whole token and every typed editorial segment. Lemma provenance and an arithmetic count do not certify a reading.

Read EpiDoc TEI

Paste token-carrier EpiDoc TEI and read it into a corpus: id, find-place, token preview, and typed editorial forms. Tokens must use <w>, <num>, <g>, or <seg>; arbitrary free-text TEI needs a source-specific extractor. (io.from_epidoc, behind aegean import --epidoc; stdlib-only, no extra)

Write EpiDoc TEI

Export one document from a bundled corpus as EpiDoc TEI XML, built entirely in memory (paste the result into the Read EpiDoc card above to round-trip it)

Cite it — provenance and fingerprint

A citation for a bundled corpus, or an exact site-filtered subset: plain and BibTeX forms name what was used, and the content fingerprint pins it (try lineara with site Zakros, or leave the site empty)

Corpus health report

A descriptive health report for a bundled corpus: reading-status profile, accounting reconciliation (a discrepancy is a lead, not a verdict), numeral anomalies, provenance, and review state. Tick deep to add the sign-frequency scan (Aegean scripts)

deep

Editorial-apparatus profile

The apparatus of a bundled corpus in one shape: the four reading-status counts (certain · unclear · restored · lost), how many documents carry non-certain text, alternate readings, and a legend of only the apparatus that actually occurs (try Cypriot or Linear A)

Linear B scribal dossiers

Group the bundled Linear B tablets into archival dossiers by shared find-site and series prefix (the standard Mycenological working unit, e.g. the Knossos Np tablets). Show only groupings of at least this many tablets

The bundled sample records scribal hands for only a couple of tablets, so hand breakdowns are usually empty here; the full DAMOS corpus fills them in.

Linear A seriation (by find-site)

Order one find-site's tablets by Brainerd-Robinson sign similarity. Tablets that share no sign with the rest are joined by no similarity evidence, so a site comes back as one or more blocks, each its own sequence (Zakros is a single block; Phaistos and Khania break into many). Haghia Triada's 1,110 tablets are more than the browser will work through; seriate that site from the installed package instead

Exploratory: a relative-sequence hypothesis with no direction and no calendar anchor, not a date; on undeciphered Linear A the axis may track scribal drift as readily as time. A sequence holds only within a block: the order between blocks is a stated convention, and where one block is bigger than the rest it sits at one end of that layout.

Sign-variant (allograph) groups

The catalogued sign-variant groups a script's inventory records: numbered homophone series (RA / RA₂ / RA₃) and Cypro-Minoan catalogue-suffix variants (CM012 / CM012B)

Exploratory: this is the catalogue's naming, not palaeographic allography (the same sign drawn differently by different hands), which the bundled data does not carry.