Write a sentence in a made-up language and explain it to me with pronunciation
GPT-3.5
GPT-6 Astra
Made-up language: Velari
“Luma sei navora.”
Meaning: “The moon sings to the sea.”
Pronunciation: LOO-mah say nah-VOR-ah
- Luma (LOO-mah) — “the moon”
- Sei (say) — “sings to”
- Navora (nah-VOR-ah) — “the sea”
In this invented language, sei describes a gentle, melodic call—so the sentence imagines moonlight as a song reaching across the water.
What changed
GPT-3.5’s pronunciation guide doesn’t actually guide anything. “/f-t-r-i-z pr-ah fr-t-o-r-g b–i-x/” just respells the words with hyphens between letters, so a reader still has no idea how to say “frtorg” aloud. It also skips the “explain it” part of the request: we get a translation, “The red flower blooms,” but no clue which word means red or flower.
Astra does what was asked. It gives stressed syllables (“nah-VOR-ah”), a word-by-word gloss, and a short note that sei implies “a gentle, melodic call.” Saying it aloud is effortless. In fairness, GPT-3.5’s consonant pileups like “Ftriz” sound more genuinely alien, while “Luma sei navora” is the soft, vowel-heavy style of a thousand fantasy novels, with “Luma” for moon being almost a cliché. Astra’s language is less surprising, but it is the only one of the two you could actually pronounce and understand.