An algorithm designed to help people with severe brain damage learn to learn new words and solve problems has been adapted for use by people with aphasia, a condition that damages the motor skills of the brain.
The algorithm, known as Mind Game, was developed by a team of researchers at the University of Oxford.
The work was published in the journal, Cognition.
“We have developed a system that allows people with mild cognitive impairment to learn a new language, a task that requires a great deal of planning and focus,” said Professor Michael Pfeiffer, from the University’s Department of Psychology.
“It has an impressive level of sensitivity and learning potential, but is far from perfect.”
A brain scan showing a patient with a brain condition called aphasic amnesia.
The computer-generated image shows a white dot on the left, and the word “toy” on the right.
Source: University of Nottingham via Google NewsProfessor Pfeffers team developed a computer program to help them build a database of word-learning skills that are in common with those of people with moderate or severe aphasias.
The program was designed to take advantage of the way the brain processes information, making the brain more efficient.
“If you have a good working memory, you’re less likely to make errors in the long term, and if you have poor working memory you’re more likely to have problems learning,” he said.
“This is particularly relevant in the context of learning a new word.”
To create the system, the team scanned patients’ brains and created artificial neural networks, or artificial neural nets, that were trained to learn words from a list of possible words.
“The first step in building the system was to create a model of a patient’s brain,” said Dr. Christopher Smith, the lead author on the paper.
“Each layer of the network would be connected to a word that had been previously trained in the brain, and then we would train a new layer of neural network to learn that word.”
Then we would make the neural network predict that word to learn the new word in the future.
“The system was able to learn about about 10 million words in a week, and could predict and repeat those words within a day.”
In theory, if you could train the system to solve a problem in 10 days, it could learn about 20,000 words per week,” said Smith.”
However, our current system was very inefficient, it only learned about 200 words per second.
“So we decided to give it a boost by building a new model to learn thousands of words per day.”
The team used a neural network trained with the model to be able to predict the word, “toys”, in a language called Leng, from a set of other words in the dictionary.
“Imagine you’re walking down the street, and you see a blue box with a letter, and say ‘toy’, and a red box with an ‘L’,” said Professor Pfeffer.
“When we predict the next word in a set, we use the model learned to learn ‘toys’.”
The model learns the new words, it can then learn to use the new vocabulary.
“Because of this, the system is very efficient.
It can learn over a thousand words per minute.”
What makes it even better is that it’s able to repeat the same word over and over again, because it knows that it can’t be wrong, it’s just going to be wrong.
“Our model is able to make a prediction on the basis of this very large number of words, and it’s very accurate.”
A lot of people would think that this is a huge achievement, but in fact it’s pretty modest,” he added.”
For a long time, people were trying to develop artificial neural network programs to teach people with disabilities, and this is one of the first successful systems for training such a system.
“The project has already been adapted to use with people with autism.”
That was very exciting, because we’re not trying to use it to help the brain in any way,” said Pfeifer.”
Instead we’re using it to improve people’s understanding of words in other languages.
“The research is part of a larger initiative to make artificial neural net systems better at learning.”
They can be trained to do things that we don’t understand,” said David Spergel, a computational linguist from the Institute of Cognitive Neuroscience at the National University of Singapore.”
Now they can learn to do the same things that are really difficult, like predicting the next letter of a word in another language.