Proto-Indo-European is the ancestor of most languages of Europe, Iran and northern India. It was spoken around 4000 BCE and never written down. Everything known about it is reconstructed from its descendants, and each reconstructed form is marked with an asterisk. This project builds the language as a working system: a lexicon, a grammar engine, a corpus of texts written in it, a language model that reads and writes it, and a synthetic voice.
The language was spoken on the Pontic–Caspian steppe, north of the Black Sea and the Caspian. Its speakers herded cattle and sheep, drove wagons and kept horses.
Over the next two thousand years their descendants moved west into Europe, south into the Balkans and Anatolia, and east into Central Asia and India. The language split into separate branches as they went.
Each branch appears in writing at a different time. The dates shown are the earliest texts in each, from Hittite tablets in the seventeenth century BCE to Albanian in the fifteenth century CE. No branch wrote down the parent language.
By the time the last branch was written down, the parent language had been gone for thousands of years. What it was is known only from what the branches kept.
Words inherited from the parent language survive in its descendants in changed form. The changes are regular, so they can be reversed. Sanskrit pitár, Greek patḗr, Latin pater and Gothic fadar all go back to *ph₂tḗr, “father”. The wheel shows each word with the reconstructed forms of its descendants in each branch, from the project's lexicon.
The corpus is not recovered text. It is PIE composed to follow the reconstructed grammar. Each record states its source: a published scholarly reconstruction, an attested poetic formula, a back-projection from parallel phrases in the daughter languages, or a passage generated and checked by the engine.
We tested our model against frontier models on a fixed set of questions held out of all training: inflection, full paradigms, translation in both directions, and dialogue. Each answer gets two scores. The lenient score ignores notation, so a form without accents or with ASCII laryngeals still counts. The strict score requires the standard scholarly notation. The frontier models score well on the lenient measure and lose 25 to 36 points on the strict one. Our model loses about two.
PIE score, 0–100, weighted over five segments. Frontier models answered without tools; each is shown at its best effort level. Ours is Nemotron 3.5 Super (122B), fine-tuned, stage e3.
This hymn to the dawn goddess *h₂éwsōs comes from the corpus. Her name survives as Vedic Uṣás, Greek Ēṓs and Latin Aurōra. It is read by the project's own PIE voices, Kokoro text-to-speech models we trained for the language. Their training speech began in a simulated vocal tract (VocalTractLab, an articulatory synthesizer), where each sound is a movement of tongue, lips and larynx: the uvular fricative of h₂, the palatal stops, the breathy-voiced stops, the pitch accent. Speech-to-speech conversion then carried it into natural human voices.
The translator below runs on the project's grammar engine. It parses English, picks the PIE words, inflects them, and gives a gloss and pronunciation for each. A chat interface to the model and a reader for the corpus are in progress.