Recently, the Shanghai Institute of Microsystems and Information Technology of the Chinese Academy of Sciences, together with relevant teams, has made breakthroughs in the field of language brain computer interface. Researchers have developed an implantable high-throughput flexible brain computer interface system and a real-time Chinese language neural network decoding algorithm, and have achieved real-time Chinese decoding and sentence synthesis of brain computer interfaces for the first time internationally. This study not only provides hope for Chinese aphasia patients to "speak again", but also opens up new space for the application of brain computer interfaces in multiple languages and scenarios. Decoding any Chinese character and sentence language is the main form of communication between people. However, severe brain diseases such as stroke, ALS, and traumatic brain injury often cause patients to lose their ability to express themselves verbally and fall into the dilemma of being conscious but unable to speak. There are nearly 8 million patients in China who have lost their language ability due to various brain diseases, and Chinese decoding has great significance and application value for them. ”Zhou Zhitao, a researcher at the Chinese Academy of Sciences Shanghai Institute of Microsystems and Information Technology, said that in recent years, many top international scientific research teams have made many breakthroughs in the English decoding of brain computer interfaces, including bilingual decoding in English and Spanish. However, the Chinese decoding research, which has the largest number of users and is very different from English, has made relatively slow progress due to its late start. Zhou Zhitao said that the core breakthroughs of the team's research this time are mainly two-fold, "achieving real-time decoding of Chinese for the first time, and realizing full spectrum decoding covering almost all syllables of Chinese". Compared to English, Chinese has its own uniqueness. Specifically, English is a non tonal language primarily composed of multiple syllables, while Chinese is a tonal language primarily composed of single syllables. At the same time, English has a large vocabulary, with about 20000 commonly used English words, while Chinese can construct over 3500 commonly used Chinese characters covering daily needs through about 400 Chinese syllables and 4 tones. The uniqueness of Chinese language contains advantages. The research team started with these Chinese syllables and tones, using them as stable intermediate decoding units to achieve "translation" from EEG to text. English is difficult to directly decode a large number of words, while Chinese can decode these Chinese syllables and tones to cover all pronunciation combinations, thereby decoding any Chinese character and sentence. The decoding strategy of "syllable+tone" revolves around the characteristics of Chinese language, and the research team proposes an intermediate decoding unit with "syllable+tone" as the core. According to researchers, compared to phonemes, syllables are more complete and sufficient units of pronunciation, with longer duration and more stable neural representation, which is more conducive to extracting distinguishing features from EEG signals. Meanwhile, by directly decoding syllables, the complex intermediate step of combining consonants and vowels can be eliminated, greatly improving decoding efficiency. In terms of specific implementation, the research team constructed a multi-level real-time decoding data stream, extracted High - γ frequency band EEG signals from 70Hz to 170Hz with a 50 millisecond sliding window, aligned the EEG signals with the pronunciation starting point, and drove the dual stream decoder to synchronously generate the probability distribution of syllables and tones. Then, the language model was integrated to select the most suitable sentence combination and achieve real-time Chinese sentence output. This process achieves a closed loop from "EEG syllables Chinese characters sentences". Research shows that after 9 days of language decoding tasks, the average accuracy of pure neural decoding for 394 Chinese syllables (uncovered syllables are rare and unknown to the subjects) reached 71.2%, with a single syllable decoding delay of 65ms and a real-time Chinese sentence decoding rate of 49.6 words/minute. On this basis, the team further integrated language brain computer interfaces with artificial intelligence and embodied intelligence technologies, and based on a self-developed universal brain computer operating system, achieved multiple new interaction methods. For example, subjects can use EEG coding to drive digital avatars for expression; Can directly communicate with artificial intelligence models; The decoded language can also be converted into control instructions, allowing for real-time manipulation of dexterous hands to complete grasping and other actions. Expanding the boundaries of 'speaking': Language decoding in the AI era is no longer limited to simply achieving 'speaking'. ”Zhou Zhitao believes that language decoding can also enhance a person's own abilities, making them become "superhumans" and greatly expanding the imagination and possibilities of control and interaction. Language brain computer interface can serve as a control center and information hub, helping people control various advanced software and hardware and collaborate efficiently with them through the implementation of various new interaction methods mentioned above and in the future. For clinical applications, language brain computer interface still faces some challenges. For example, how to achieve long-term stable implantation and maintain high-quality brain signal acquisition for a longer period of time to reduce tissue response. In addition, the neural encoding mechanism of Chinese language itself also needs further exploration. The research team stated that in the next step, they will prioritize using the team's latest fully wireless fully implanted fully functional brain computer interface device to conduct long-term implantation experiments, achieve real-time silent decoding, and further improve real-time decoding speed and accuracy. (New Society)