A UM professor built AI software that invented a language with no consonants, and another for octopuses. The same tool could help save languages at risk of disappearing entirely.
Morris Alper is an assistant professor of computer science at the University of Miami. He has been interested in inventing languages since he was young. Recently, he got curious whether AI could actually do it.
Where Off-the-Shelf AI Fell Short
Alper first tried using tools like ChatGPT directly to generate constructed languages, the kind familiar from Elvish or Dothraki.
"I saw sparks of it being able to do this but not completely," Alper said.
So he built his own framework instead, called Conlang Crafter. It's designed to work in tandem with AI systems rather than ask them to invent a language from scratch. He broke the problem into three parts: sounds, grammar and words.
"There are languages in Southern Africa that use click sounds, which we don't have in English," Alper explained. Word order varies just as much. "In English you would say, 'I eat cake.' In Japanese you would basically say 'I cake eat.'"
A Tool Built for More Than Fiction
Conlang Crafter isn't only useful for building fantasy languages. Alper sees a more practical application in endangered language preservation.
More than 3,000 endangered languages exist worldwide, according to Ethnologue, a research center for language intelligence. Many have no written rules, dictionary or framework that AI systems could otherwise learn from.
"People are trying to study how to use things like LLMs with an actual description of the language that maybe a linguist wrote, so it can follow rules to translate into it, rather than just learning from a lot of data," Alper said.
The "Impossible" Languages the Software Already Built
Alper's framework doesn't stop at realistic languages. One output has no consonants at all, communicating only through vowel sounds.
Another was designed for octopuses. It doesn't use sound at all, instead relying on color and movement to communicate.
Both follow internally consistent rules that could, in theory, be taught and learned like any real language.
What This Means for Miami's Research Community
Conlang Crafter sits at an unusual intersection for a computer science department. It's part linguistics tool, part AI research project, part answer to a childhood fascination.
For a field usually measured in benchmarks and accuracy scores, Alper's work points to a different kind of AI contribution. Preserving what a community says before there's no one left who remembers how to say it.
UM Study Uses AI to Predict Parkinson's Decline Early
