Artificial intelligence isn’t just capable of translating between existing languages—it can also create entirely new ones.
That’s according to a recent paper published in the Proceedings of the Association for Computational Linguistics on ConlangCrafter, a tool the authors developed that leverages large language models (LLMs) to create novel languages with their own grammar rules and vocabulary.
“The idea of ConlangCrafter is: ‘How can we make new languages that have different linguistic features than what we normally see in natural languages?’” explained Morris Alper, the paper’s first author and an incoming assistant professor in the Department of Computer Science at the University of Miami College of Arts and Sciences.
Morris and co-authors Moran Yanuka, Raja Giryes, and Gašper Beguš designed ConlangCrafter to build new languages step by step, generating a system of sounds as well as grammar rules and vocabulary. They have so far used the tool to create more than 60 different languages and have shared their code on the ConlangCrafter website so others can generate their own.
Users can give ConlangCrafter specific parameters for the constructed languages. The researchers asked the tool to create a language with no consonant sounds, for example, as well as a language for an alien cephalopod species that uses colors and gestures to communicate.
Once it creates a new language, ConlangCrafter translates sentences from natural languages into the constructed one and reviews and revises its work, identifying inconsistencies and fixing them. It continually updates a “language sketch,” akin to a design document that keeps track of the new language’s rules.
The resulting languages are more diverse and internally consistent than those generated by simply asking a general-purpose LLM, like Gemini, to create its own language from scratch.
“Imagine if you just say, ‘Make me a language.’ It will give you something that doesn’t make sense,” Alper said. “What we did is build this pipeline where you say, ‘Okay, what are the sounds? And then let’s check them. And then, what are the rules for building words? What about syntax?’ We split the problem apart and have the LLMs solve each sub-problem and combine them together.”
There are numerous potential applications for ConlangCrafter, including helping human language designers create fictional languages for video games, movies, books, and TV shows (think “Game of Thrones” and “The Lord of the Rings,” both of which feature constructed languages).
Beyond creative applications, the authors foresee potential uses in linguistics and computer science research. The tool could potentially help researchers develop technologies for poorly documented languages, for example, for which detailed descriptions exist but not large collections of text. It could also be used to study how languages evolve over time and how AI agents can use constructed languages to communicate with each other.
The most challenging aspect of the research was figuring out how best to evaluate the new languages, Alper said.
“The hardest part of the work was defining an objective measure to give us numbers that say, ‘How well is the model performing at this task?’” he explained. “That’s really hard to do for creative things.”
The research team created a framework to test how consistently the translations followed each language’s rules and how diverse the new languages are in terms of linguistic features such as the presence of unique sounds and different sentence structures.
Although ConlangCrafter was only recently released, it has already garnered attention in the technology sector. It was recently highlighted in articles in Science and IEEE Spectrum.
Alper developed ConlangCrafter while working as a postdoctoral researcher at Carnegie Mellon University’s Language Technologies Institute. He completed his Ph.D. in multimodal machine learning at Tel Aviv University and his bachelor’s degree in mathematics and linguistics at the Massachusetts Institute of Technology.
As a machine learning researcher studying the intersection between language and multimodal AI, Alper has collaborated with faculty members from other disciplines, including archaeologists studying ancient inscriptions. He is a co-first author of a paper on leveraging artificial intelligence to help read cuneiform, an ancient writing system used across Mesopotamia.
Alper said he was drawn to the University of Miami in large part by its focus on interdisciplinary research.
“It felt to me like the University really encourages collaboration between fields, and I am a very interdisciplinary person,” he said. “My research is on the seam between computer science, linguistics, and digital humanities, and I found that the University of Miami really appreciates that.”
