Understanding how humans learn language provides a blueprint for designing other intelligent learning systems.
Jon Rawski, assistant professor in the department of linguistics and language development at San Jose State University, discusses how to do so.
Jon Rawski is an Assistant Professor in the Department of Linguistics and Language Development at San José State University in San Jose, California, where he teaches courses on general and computational linguistics.
Professor Rawski’s research aims to understand the fundamental laws of language and learning. He seeks to mathematically understand how humans, unlike other species, are capable of learning at least one or more of the 7,200 languages that exist with so little data so soon after being born. Understanding how humans learn language provides a blueprint for designing other intelligent learning systems such as modern AI systems. His work spans linguistics, cognitive science, philosophy, and theoretical computer science.
A Blueprint for Designing Intelligent Learning Systems
No other species can learn a language. Only humans can. There’s something cognitively special about this.
It’s a mystery how children gather data to begin to learn language. They are born with a blueprint to learn it. The linguist wants to know what is it about language, the fundamental structure of it, that needs to be learned from that data, learned from other people, learned from the community, and other language speakers–and, conversely, that doesn’t need to be learned or which couldn’t even be learned.
It’s part of our biology.
Every human knows, learns, and uses at least one of the 7,200 languages on the planet. Every single definition of AI involves mastering language. An overwhelming amount of human knowledge is stored in language. This means language is important to study scientifically.
Language is a system, and learning is a process, and where there are systems and processes, there is mathematics.
My work mathematically proves what kinds of languages are possible or impossible, based on what kinds of learning humans do. Studying the fundamental laws of language mathematically, like any other field, is necessary since data and experiments are hard to get with humans. This also helps translate the discoveries to other fields, such as computer science.
Mathematically studying sentence structure–syntax–led to the development of programming languages and compilers. Mathematically studying language learning kick started the field of machine learning. Both occurred in the 1960s. Mathematical results today are used to robustly test large language models.
Why is understanding all of this important? It helps us understand ourselves as a species. We’re good at understanding other species but understanding ourselves is a tricky beast.
Also, if we’re going to design systems such as AI that learn language we ought to do it in a way that is as successful as when humans learn it.
