Structured multilingual corpora for
enterprise AI.
Chedai builds parallel translation datasets from scholarly editions of out-of-copyright texts, with token-level metadata and documented edition, provenance, and rights status.
Ancient Greek, Latin, Biblical Hebrew, Aramaic, and Classical Chinese into English are in production. The pipeline is parameterized by source language; additional source and target languages are in development.
Each passage is rendered twice under one editorial standard: once in a close register that preserves the source structure token for token, and once as fluent English. Every token carries 24 fields of linguistic metadata. Edition, provenance, and rights status are documented per text for every text we release.
The Poem of Mulan · 木蘭詩
唧唧復唧唧，木蘭當戶織。
不聞機杼聲，唯聞女歎息。
問女何所思，問女何所憶，
女亦無所思，女亦無所憶。
昨夜見軍帖，可汗大點兵。
軍書十二卷，卷卷有爺名。
阿爺無大兒，木蘭無長兄。
願為市鞍馬，從此替爺征。
Close register
Sigh upon sigh, and sighing again. Mulan weaves facing the door. One does not hear the loom and shuttle; one hears only the girl sighing. Ask the girl what she thinks of; ask the girl what she longs for. The girl thinks of nothing; the girl longs for nothing. Last night she saw the muster notice: the khagan calls a great levy. Twelve scrolls of the army register; scroll upon scroll bears father’s name. Father has no eldest son; Mulan has no elder brother. I would buy saddle and horse, and from this day take father’s place in the campaign.
Fluent
Sigh after sigh, and sighing again — Mulan sits weaving at the door. You cannot hear the loom; you hear only the girl’s sighing. What are you thinking of? What do you long for? I think of nothing, she says. I long for nothing. Last night I saw the muster rolls: the khagan is raising a great levy, twelve scrolls of the register, and my father’s name on every one. Father has no grown son. Mulan has no elder brother. I will buy a saddle and a horse, and from this day I take my father’s place in the campaign.
A delivery contains
- Parallel source and target text, in both registers
- Token-level records against a versioned schema
- Edition and provenance documentation
- Rights-status documentation and license terms
- Release notes, in JSONL and Parquet