Anyone can easily translate: New AI invents its own languages

Researchers have developed an AI tool that not only translates texts, but also invents completely new languages. ConlangCrafter uses large language models to generate grammar and vocabulary. This helps linguists and the film industry.

AI creates new languages

Scientists at the Universities of Miami and UC Berkeley have unveiled a new tool called ConlangCrafter. The system, which is based on large language models, not only translates existing texts, but also generates independently constructed languages, so-called conlangs (for “constructed languages”). The team responsible wants to show how artificial intelligence can help build linguistic structures instead of just analyzing existing grammar. The aim of the project is to design linguistic structures that rarely or do not occur at all in natural languages.

To date, the team has generated over 60 different language systems. The software divides the process into subtasks. First the AI ​​defines the sounds, then the grammar and finally the vocabulary. An integrated correction mechanism continuously checks the rules for contradictions.

AI languages ​​for aliens and films

The developers also tested unusual specifications to check the flexibility of the system. As TechXplore reported, the software, for example, designed a language completely without consonants. Another experiment involved a communication system for fictional squid-like aliens. This system is based on colors and gestures instead of traditional sounds. Such constructed languages ​​are often used in the entertainment industry. Well-known examples are the languages ​​in Game of Thrones or Lord of the Rings. According to the researchers ConlangCrafter twice as diverse and almost 70 percent more consistent than traditional models such as Gemini when users simply ask them to invent a language.

Benefits for science

In addition to creative applications, the tool offers advantages for research in the area of ​​machine language processing, also known as natural language processing. Scientists can use it to study how the structure of a language affects the performance of computer models. Simulations can also be carried out to analyze the evolution of communication.

However, the system still has weaknesses. The developers admit that ConlangCrafter has its limitations when it comes to complex semantic meanings, contextual conversation flow, and creating visual writing systems. The source code is still available to the public free of charge to promote further developments.