A new study using artificial intelligence demonstrates that language actively evolves over generations to become more structured — making it easier for both humans and machines to learn. The research, published in the Proceedings of the National Academy of Sciences, bridges cognitive linguistics and deep learning to reveal the architectural principles behind language acquisition.
The Research
Dr. Devon Jarvis, a Lecturer in Computer Science at the University of the Witwatersrand (South Africa) and Fellow at the Wits Machine Intelligence and Neural Discovery (MIND) Institute, led the study. The team built a deep linear neural network modeled after a child’s progressive learning stages. These networks mimic how children learn in hierarchies — starting with basic concepts (e.g., plants vs. animals) before understanding finer distinctions (e.g., types of birds).
Over successive generations, the network was exposed to data with properties similar to human language. The key finding: iterated learning — the process of transmitting language across generations — naturally filters out unstructured elements while retaining structured, easily learnable patterns. The network’s depth proved critical: only deep networks with multiple processing layers succeeded in capturing these regularities; shallow networks failed entirely.
“We built a computer brain with similar characteristics to a child’s, and compared it to behaviors we see in children’s brains,” Dr. Jarvis explained. “Computer brains find structure in data the same way children favor certain language properties. Language becomes more structured over generations because it makes learning easier.”
Why It Matters
This study reveals that the evolution of language is not random but driven by cognitive constraints. For anyone curious about their own learning, it highlights that our brains are wired to prefer order and repetition. The same principles that make language learnable apply to how we absorb any new skill — from music to math. If you’ve ever struggled with a confusing textbook, this research suggests the problem may not be you, but the material’s lack of hierarchical structure.
What You Can Do
- Seek structure: When learning something new, break it into hierarchical categories (big ideas first, then details).
- Practice iteration: Revisit material over multiple sessions; each pass reinforces structured patterns.
- Test your cognitive strengths: Understanding your own learning style can make study time more effective.
Source: Neuroscience News
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
