Linguists have long thought it would be impossible to teach a machine how to analyze speech sounds and word structures in the way human investigators do. The traditional view has been that the task is just too difficult because human languages are notoriously complex. However, recent breakthroughs in artificial intelligence research have led some experts to believe that it might be possible to create algorithms that can learn to analyze language in a way that is similar to how humans do it.

Researchers at MIT, Cornell University, and McGill University have demonstrated an artificial intelligence system that can learn the rules and patterns of human languages on its own. The system, which is based on a neural network, was able to learn the structure of several different languages after being exposed to only a small amount of data. The researchers believe that this breakthrough could lead to the development of AI systems that can effectively communicate with humans. In addition, the system may be able to help linguists better understand the underlying rules that govern language.

A new machine-learning model can learn the rules of how words change to express different grammatical functions, like tense, case, or gender. The model is designed to work with languages that have highly inflective grammar, meaning that a word can change form in many different ways to express different grammatical functions. For instance, the word for “book” in Serbo-Croatian can take on different forms depending on whether it is masculine or feminine, singular or plural, and so on. The model starts with a set of words and their various forms, along with information about the grammatical function of each form. Based on this data, the model comes up with rules that explain why the forms of the words change. In addition to being able to learn the rules of inflectional grammar, the model can also learn other types of regularities in language, such as rules for compounding words. The model could be used to help develop artificial intelligence systems that are better able to understand and generate human language.

Ready to get started?

Let us help you build the business of your dreams

To train and test the model, they used problems from linguistics textbooks that featured 58 different languages. Each problem had a set of words and corresponding word-form changes. The model was able to come up with a correct set of rules to describe those word-form changes for 60 percent of the problems. While this is a promising result, the researchers note that there is still room for improvement. In particular, they would like to see the model being able to handle more complex linguistic phenomena. Nevertheless, the study provides a valuable contribution to our understanding of how language learning works at a computational level.

The researchers used a type of AI called neural networks, which can learn by example. They fed the neural network a set of data describing the sound patterns and word structures of various languages and then gave it a series of tests. The results showed that the neural network was able to learn the relevant rules and correctly solve the problems more than 80% of the time.

This research represents an important step forward in the development of AI systems that can automatically learn from data. In addition, it provides insights into how humans learn the language, which could lead to new educational methods for teaching children phonology and morphology.

The researchers used a machine-learning technique known as Bayesian Program Learning to build the model. With this technique, the model solves a problem by writing a computer program. In this case, the program is the grammar the model thinks is the most likely explanation of the words and meanings in a linguistics problem. They built the model using Sketch, a popular program synthesizer that was developed at MIT by Solar-Lezama. But Sketch can take a lot of time to reason about the most likely program. To get around this, the researchers had the model work one piece at a time, writing a small program to explain some data, then writing a larger program that modifies that small program to cover more data. This allowed the model to more quickly learn the set of rules for assembling words, which is called grammar. As a result, the model was able to accurately solve linguistics problems after being trained on only a small amount of data.

This work was funded by a variety of organizations, including the Air Force Office of Scientific Research, the Center for Brains, Minds, and Machines, the MIT-IBM Watson AI Lab, the Natural Science and Engineering Research Council of Canada, the Fonds de Recherche du Québec – Société et Culture, the Canada CIFAR AI Chairs Program, and the National Science Foundation (NSF). The NSF also provided a graduate fellowship.

Sign up to our newsletter

Get the latest articles on all things development related right in your inbox.
Loading

Ready to get started?

Let’s discuss how we can add value to your project and build your idea from the ground up!

    By clicking ‘Send’, I’m agreeing to all the privacy terms.