Machine translation will be the most difficult task in the field of artificial intelligence

Machine translation (MT) involves mathematics, linguistics, cognitive science, computer science and other disciplines, and has been recognized by the scientific community as one of the most difficult topics in the field of artificial intelligence. It has evolved from rule-based MT technology to example-based MT technology, and then to statistics-based MT technology, and finally to the current mainstream neural network MT technology.

Neural network machine translation (NMT) is a machine translation method proposed in recent years. Compared with traditional machine translation methods, neural network machine translation can train a neural network that can map from one sequence to another, and the output can be a variable-length sequence, which can achieve better performance in translation.

Currently, the most mature machine translation technology on the market is Google Translate. The technology behind Google Translate is a statistical machine translation method. The basic operating principle is to search a large amount of bilingual web content as a corpus, and then the computer automatically selects the most common word-to-word correspondence and finally gives the translation result. Now Google Translate uses neural networks and deep learning methods, which has greatly improved machine translation technology.

However, no matter which machine translation method is used, the biggest factor affecting the development of machine translation is the quality of the translation. Judging from the existing achievements, the quality of the translation is still far from the ideal goal (reaching the translation level of senior translators). As we all know, the process of manual translation (abbreviated as "human translation") is a comprehensive process of understanding, analysis, selection and re-creation for the human translator, which is the process of brain thinking activities. Therefore, if the quality of machine translation is to reach the level of human translation, it is necessary to solve the mystery of how the brain processes language information.

In fact, more than 20 years ago, Chinese scientist and futurist Professor Zhou Haizhong predicted that it would be impossible for machine translation to achieve the level of "faithfulness, fluency and elegance" when humans still did not understand how the brain performs fuzzy recognition and logical judgment of language. This prediction has been confirmed today.

The European Brain Project, the American Brain Project, and the Chinese Brain Project implemented in recent years all aim to use computers to simulate the human brain, with neuroinformatics as their core content. The research results of this emerging frontier discipline will help people understand how the brain processes natural language and are expected to pave the way for the advancement of artificial intelligence, especially machine translation technology.

There is still a long and difficult way for machine translation to reach the level of human translation. This requires the development of future science and technology, especially major breakthroughs in brain science. At present, people can only combine machine translation with human translation to complement each other, which can not only save translation time but also produce high-quality translations.

(The author is a postdoctoral fellow at the School of Electronics at the Technical University of Berlin, Germany)