He asked Dinh Dien, director of the Center for Computational Linguistics at the University of Natural Sciences, a member of the Vietnam National University, HCM City, what software should he use to learn Vietnamese.

Dien, who has internationally published works on artificial intelligence (AI) in machine translation and teaching of Vietnamese to foreigners, said that AI application can help solve linguistic problems.

The first step in teaching a language are speaking and pronunciation. The barrier is that Vietnamese language has tones, which makes it difficult for people who speak languages with no tones such as English and French to pronounce Vietnamese.

In this case, software using AI to teach languages simulates the movements of teeth, and the pronunciation of sounds for learners to imitate. After that, learners pronounce the sound again, record it on the software, and use technology to compare the sounds pronounced by the learner and the standard pronunciation on the software. 

This method allows learners to improve their pronunciation quickly. All of these stages need to use AI, he said.

According to the Vietnamese dictionary from the Institute of Linguistics (the chief editor is late Professor Hoang Phe), there are 34,000 Vietnamese words. So, it is necessary to teach 10 percent of the words, or 3,400 words to the machine so that the machine can read 90 percent of general documents in Vietnamese. 

To retrieve the data, Dien has to use AI and label the vocabulary in the Vietnamese linguistic database.

Experts agree that AI has changed the way of teaching and learning in the educational sector. Many AI apps have been created, making teaching more effective.

When machines learns languages

AI also helps systems better support linguistic intelligence. Machines are trained and improve day by day.

For example, an intelligent assistant such as Kiki can understand human language.

Kiki, the Vietnamese voice assistant used in a car can recognize voices with different regional intonations.

In computing science, voice recognition is an important branch of AI, converting the human voice into a useful format and understandable to computer apps. 

The technology is a bridge for communicating between machines and humans. Voice assistants have become an indispensable app. They include Apple’s Siri, Google Assistant, Amazon Alexa and Vietnam’s Kiki.

Nguyen Hoang Khanh Duy, who wrote the first code lines of Kiki, said to train the AI model and make it intelligent enough when recognizing voice and replying with true information to users, language data plays the key role.

Navigation, or showing directions, is an important function for the Vietnamese assistant Kiki users when driving cars. 

The product development team must prepare data for users to command. After data collection and model training, the voice recognition quality index in the later version improved by 40 percent compared to the original.

Kiki still needs solutions in order to improve. For example, the noise caused by the engine, or wind or noise from traffic on the road, affects the quality of Kiki's capability of recognizing voices. The Kiki team is solving the problem by augmenting data of speech in noisy conditions that suit real life.

In addition, with new techniques, such as self-supervision, Kiki is trying to learn unlabeled data to further improve the model. The stability of Kiki has gradually improved thanks to continuous training and upgrading.

Trong Dat