Artificial Brain Learns to Use Human Language

Robots That Talk

Humanity has long dreamed of a robot (or any kind of artificial intelligence, really) that is able to actually talk to you, to carry on a conversation in a way that is natural. We started working on this technology in earnest over half a century ago, when scientists began to develop something that is called "natural language processing" (NLP) in the 1940s. This technology is meant to fool individuals into believing that they are communicating with a computer when, in fact, the intelligence they are conversing with is entirely artificial.

But as anyone who has ever used CleverBot can tell you, the technology has been somewhat lackluster. At least, it was until now.

A cognitive model composed of two million interconnected artificial neurons has been developed by researchers from the University of Sassari and the University of Plymouth. The system is known as ANNABELL (Artificial Neural Network with Adaptive Behavior Exploited for Language Learning), and it is able to communicate through human language from a starting point that is “tabula rasa,” which translates from Latin to "blank slate."

To break this down a bit, ANNABELL is still a computer, and computers work through programs with coded rules that they must follow in order to perform a given task. However, ANNABELL is unique because it doesn’t have pre-coded language knowledge. It learns only through communication with a human interlocutor. To achieve this, the researchers made use of two mechanisms that are also present in biological brains. Namely, they used synaptic plasticity and neural gating.

Like a Biological Brain

Synaptic plasticity is the ability of two neurons to increase their connection efficiency when they are active simultaneously, or nearly simultaneously. This is essential for learning and for long-term memory. Meanwhile, neural gating is based on the properties of certain neurons (called bistable neurons) to switch themselves "on" or "off" due to a control signal coming from other neurons. When on, the bistable neurons transmit the signal from a part of the brain to another, otherwise they block it.

As ANNABELL has a mimicked version of synaptic plasticity, it is able to learn to control the signals that open and close the neural gates, so as to control the flow of information among different areas. ANNABELL has been validated using a database of about 1500 input sentences, based on literature on early language development. It has responded by producing a total of about 500 sentences in output, containing nouns, verbs, adjectives, pronouns, and other word classes, demonstrating the ability to express a wide range of capabilities in human language processing.