Is machine learning “too stupid” for language learning? Learn more from mom and dad

New inspiration for NLP?

Is machine learning "too stupid" for language learning? Learn more from mom and dad

Source: foreign media

[Guide] Researchers from the University of Chicago and Carnegie Mellon University have designed a game to understand how parents can accurately adjust the language they use when communicating with their children based on their children’s acquired language knowledge. The research results also contribute to the development of machine learning.

Is machine learning "too stupid" for language learning? Learn more from mom and dad

Do you still remember how our parents taught us to speak when we were babbling?

“Call Mom, Mom—Mom—” “Dinner”

This way of prolonging pronunciation and repeating words is the most basic way for every child to learn to say a word.

And this way of talking between parents and children is different from that between adults in many ways. In addition to the above method of prolonging pronunciation and repeating words, parents will continue to teach us other content based on the words we have learned.

So, how does a machine learn a language?

Researchers from the University of Chicago and Carnegie Mellon University have discovered a new method that uses experiments to evaluate how parents change their language according to the language they already master when talking with their children. And this way of educating children to speak is also suitable for machine learning.

This is the first language model study of this type, and related papers have been published on Psychological Science.

Choose the right language model for your child

“We have always known that the way parents and children speak is different from the way other adults speak. For example, simplifying the language, repeating words, or prolonging the pronunciation can help children better master the language. “

“However, we don’t know whether parents change the way they speak while their children are learning languages, so that the children can learn the next content “just right”.” said Daniel Yurovsky, associate professor of psychology at Carnegie Mellon University .

Is machine learning "too stupid" for language learning? Learn more from mom and dad

Daniel Yurovsky

Researchers have found that adults will slow down and speak with children in a higher pitch. They also use exaggerated pronunciation, repeating or simplifying their own language. In addition, adults will also ask children questions to confirm their understanding. The entire communication model will continue to change as the child’s language fluency improves.

Professor Yurovsky said that this is like the way students learn mathematics in school.

Students start with algebra, then plane geometry, and then calculus. Professor Yurovsky said, “People use the same structure to communicate with children without thinking, in order to grasp the children’s understanding of the language, and modify their way of speaking, so that the children can better understand them.”

How to adjust the language mode?

As children grow up, parents will also change the way they interact with them. To better understand this process, the research team designed a game that allows parents to assist their children in choosing the correct animal among the three animals. Among these options, half of the animals are known to the child before the age of 2, while the rest are animals that the child only recognizes after the age of 2.

Forty-one pairs of parents and children were invited to participate in the study, and the researchers measured how parents communicate when they think their children should know certain animals.

Is machine learning "too stupid" for language learning? Learn more from mom and dad

Parents describe animals to guide children to choose the correct pattern

The results of the study found that parents understand their children’s language very well because they have been witnessing their children’s growth and learning. This shows that parents use their knowledge of their children’s language development to adjust their language.

Is machine learning "too stupid" for language learning? Learn more from mom and dad

Use longer expressions to describe animals that children don’t know

Research has also found that parents use various methods to express “unknown” animals to their children. For example, when describing animals, they use other words that children are familiar with to describe animals.

Is machine learning "too stupid" for language learning? Learn more from mom and dad

In Figure b, after the child makes the right choice, the parents will continue to use a longer expression to describe the animals they think the child does not know. However, if the parent thinks that the child knows the animal, but the child chooses the wrong option, a longer expression will be used the next time the same animal appears.

The experiment validated the team’s previous ideas and also found that parents not only used the language knowledge their children had mastered, but also changed their way of expression when they realized that their children did not know a certain animal.

Application in the field of machine learning

Natural language processing (NLP) is an artificial intelligence that specializes in analyzing human language. Modern NLP is a mixed subject that combines linguistics, computer science and machine learning.

Everyone is familiar with voice assistants, and all kinds of voice assistants must be trained before they go online. The first step of training is text preprocessing. Simply put, it is to enter text for subsequent analysis.

The language model can learn rich semantic knowledge from unrestricted large-scale monolingual corpus. This is how we now train machine learning languages: provide all our language data.

Professor Yurovsky believes that their research results can help people understand the language system of machine learning.

In other words, if people no longer train all the language data at once, but input the appropriate language data into the machine at the appropriate time, the machine can learn the language at the appropriate level of complexity.

Machines are to researchers what children are to parents. The latter have a very clear grasp of the learning process and learning ability of the former.

If researchers can adjust the progress and methods of machine learning language like parents, it will be a matter of opinion whether they can better help machine learning.

Reference materials:

Posted by:CoinYuppie,Reprinted with attribution to:
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.

Like (0)
Donate Buy me a coffee Buy me a coffee
Previous 2021-07-08 14:33
Next 2021-07-08 14:34

Related articles