Combining AI to “teach” Vietnamese
The leader of a foreign direct investment (FDI) enterprise in Vietnam wants to learn Vietnamese quickly and read 90% of the content of common documents. The problem is, he is too busy and only has about 1 hour (from 12-13h every day) to study. So, how should technology be applied to language learning software to help him learn foreign languages quickly?
Above is the problem of the FDI company leadership position posed to Associate Professor, Dr. Dinh Dien, Director of the Center for Computational Linguistics, University of Science, Ho Chi Minh City National University.
As someone with many scientific research topics and international publications on the application of artificial intelligence (AI) in machine translation, critical linguistics and teaching Vietnamese to foreigners, Associate Professor Dien believes that the application of AI is essential to solve problems in linguistics.
Specifically, the first step in learning any language is to teach the language sounds. The barrier here is that Vietnamese has tones, rhythms, when teaching to learners of non-tone languages such as English, French, etc., it will be very difficult. For example, instead of asking: "Have you gone to bed yet?", learners will say "Have you gone to sleep yet?", because they cannot distinguish the tones. It is necessary to teach them where to put the game when pronouncing, what the mouth shape is, and how correct or incorrect pronunciation is different.
At this time, AI application software in foreign language teaching can simulate the mouth shape of brushing teeth, play pre-made sounds for learners to imitate. Then, learners play back the sounds, record them in the software, use technology to compare the learner's pronunciation with the standard pronunciation from the software, improve pronunciation quickly. All of the above steps must apply AI.
Another example, according to the Vietnamese Dictionary of the Institute of Linguistics edited by the late Professor Hoang Phe, the original Vietnamese vocabulary has about 34,000 words. Calculations show that it is necessary to teach about 10% of the words to the machine, equivalent to 3,400 words of application information, so that the machine can read about 90% of common Vietnamese texts. To get this data table, Associate Professor Dien had to use AI, labeling the vocabulary system in the Vietnamese corpus.
It can be said that AI has changed the way of teaching and learning in the education sector. In fact, many artificial intelligence applications have been born to support the teaching and learning process to become faster and more effective.
The interesting story of combining computer science and linguistics above shows that the process of training and applying AI in practice is very necessary, but not easy. The data needs to be separated into many layers of identification, at each layer, each variable must be processed with different specific identifiers.
When machines learn language...
Not only does artificial intelligence help humans learn languages, it also helps systems that support language intelligence better. Machines are trained and improved every day.
Similar to the story of Associate Professor Dien, below is another vivid example of how an intelligent assistant understands human language.
It is the process of researching and developing the Vietnamese voice assistant Kiki in cars, to recognize voices well with many different regional accents. In computer science, voice recognition is an important branch of artificial intelligence (AI), converting human voices into a useful and understandable format by computer applications. This technology is a bridge of interaction between machines and humans. Voice assistants have become indispensable applications around the world. The most popular are: Apple's Siri, Google Assistant, Amazon Alexa, or Kiki in Vietnam.
Mr. Nguyen Hoang Khanh Duy, who wrote the first lines of code for Kiki, shared that to train an AI model smart enough to recognize voices and respond to users with the right information, language data plays a key role.
For example, a very important function for users of the Vietnamese assistant Kiki in cars is navigation. Therefore, the product development team must prepare data and vocabulary to "smoothly" support commands from users. After the process of collecting data and training the model, the index showing the quality of voice recognition in the later version has improved by 40% compared to the original.
Voice recognition in cars is not only limited to navigation and location problems, but also many other issues.
For example, the specific use of Kiki in cars is very noisy due to the engine, wind or noise from traffic equipment on the road, which directly affects the quality of Kiki's voice recognition in the car. Therefore, the Kiki team is required to try to solve the noisy conditions by enhancing the data by speaking in noisy conditions to best match real life.
In addition, with new techniques in the world such as self-supervised learning, Kiki is trying to "learn" from even unlabeled data, to further improve the model. The stability of this Vietnamese voice assistant is improving with continuous training and product upgrades.
Obviously, technological progress is happening every day, every hour. ChatGPT, launched at the end of 2022, has partly answered the question of how big data works. Technology is "stepping" into the middle of life, especially in education, language, areas that previously depended heavily on humans. AI redefines the way we learn, work, live... as the specific examples above.
Source
Comment (0)