At present, the relevant research has only realized the simple information transmission between humans and animals to a certain extent. I'm afraid there is still a long way to go before the real cross species communication can be realized. In the short term, it is still difficult to achieve cross species communication, but more and more research is undoubtedly opening a door for it. —— Tan Mingzhou, director of Artificial Intelligence Division of Yuanwang think tank and Chief Strategic Officer of Turing robot In Andersen's fairy tales and other literary works, as well as many excellent film and television works, the same theme - the communication and interaction between man and animals is reflected. Nowadays, scholars all over the world try to break the language barrier between human and animals through artificial intelligence and other ways to truly realize cross species communication and even emotional communication. Recently, an international research team composed of researchers from the University of Copenhagen in Denmark, the Federal Institute of technology in Zurich, Switzerland, and the French National Institute of agriculture, food and environment has developed an artificial intelligence product that can translate the sounds of pigs in various scenes and successfully decode the "joys, sorrows and joys" conveyed in their cries. The research results were published in the latest issue of the journal Science report. So, can artificial intelligence realize the communication between human and animals? How does the algorithm distinguish animal emotions? At present, what difficulties do humans need to overcome to understand animal language through AI? Many animal languages have been studied through algorithms Animals, like people, have their own emotions. They will be happy, sad, afraid and angry, but limited by language and expression, animal emotions are difficult to know. In fact, we may hide their different emotions in the sounds of animals that are similar to each other. The above paper shows that in order to train AI to translate pig language, researchers specially recorded more than 7000 calls of 411 pigs in 19 different scenes. The results show that the call sound of positive emotion is shorter and the amplitude is lower than that of negative emotion. The researchers said that the accuracy of this algorithm is as high as 92%, which can basically accurately distinguish the emotion from the pig's cry. Coincidentally, a research team at Cambridge University had asked AI to identify whether a sheep is in trouble only based on its facial expression. The AI system first lists several "facial action units" (AUS) related to different degrees of pain according to the facial expression of sheep pain, and then marks these aus in 480 sheep Photos - deformation of nostrils, rotation of each ear and narrowing of eyes, so as to judge the situation of sheep. "In fact, there is a precedent for using algorithms to study animal language and communication between humans and animals. There have been projects to study pet dogs and cats before. The purpose of these studies is to make it easier for humans to get along with them." On May 4, Tan Mingzhou, director of Artificial Intelligence Division of Yuanwang think tank and Chief Strategic Officer of Turing robot, pointed out in an interview with science and technology daily. For example, in order to realize the communication between people and pets, Dr. Matsumi Suzuki, a famous Japanese acoustics expert, once invented a "pet dog translator" by using the animal translation technology based on machine learning. Pet owners only need to pin a mini microphone on their collar, and the collected pet calls will be transmitted to the translator for speech recognition and conversion, so as to convey the meaning of pets to their owners and understand their emotions. "In addition, there are researchers who study the swarm intelligence of simple animals, such as the communication mode of bees and ants. These studies have certain bionic reference significance for military tactics and equipment; there are interdisciplinary studies on the organizational ability of animals such as dolphins and whales, which are very valuable for exploring the history of biological evolution." Tan Mingzhou further explained. You can gain translation ability even if you don't understand language Do animals have their own language? If so, what would they talk about? Understanding animals can be said to be a long-term research topic for humans. At present, AI is helping us find the answer. "Animals don't have the unique language system of humans, so researchers can analyze their demands by combining their calls, behaviors and habits in order to better understand them," Tan Mingzhou said A researcher once watched the intermittent "dialogue" between two sperm whales in a relatively static position for up to 40 minutes. Almost every sentence of their "dialogue" was not duplicate, and was accompanied by various actions. This makes one wonder: are the two female whales "pulling the family" and sharing their parenting experience? The research on the content of sperm whale "dialogue" is one of the research contents in the "whale language translation project" (project CETI) sponsored by the international team of scientists and supported by National Geography in recent years. It is reported that researchers are using natural language processing system (NLP) to analyze 4 billion communication codes of sperm whales. NLP is a sub field of artificial intelligence, which focuses on dealing with human written and oral language. The research team plans to let AI link each sound to a specific background, a process that is expected to take at least five years. If the team achieves these goals, the next step will be to develop and deploy an interactive chat robot to talk to sperm whales living in the wild. Scientific research published in the journal Nature has proved that artificial intelligence is very effective in deciphering ancient human languages. This opens up the possibility of using AI to explore animal language. According to the study, machine learning technology provides new tools that can help archaeologists understand the past faster, especially when deciphering ancient texts. The AI system uses the ancient Greek language and the inscriptions of the whole ancient Mediterranean world for training. The training data comes from the largest digital data set of Greek inscriptions provided by the relevant humanities colleges, and each of these inscriptions is marked with metadata, which describes the writing place and time of the inscriptions investigated by historians. With these data, AI can find patterns and laws in these information, use complex mathematical models to encode these information, and then further use these inferred information to infer the content, compilation place and age of other inscriptions. Research shows that the AI achieves 62% accuracy in repairing damaged words. This also provides inspiration for translating animal language. Artificial intelligence often follows the same methods and guidelines in cracking ancient characters and translating animal languages. Tan Mingzhou said: "in the classic task of translation, machines do not need to understand the language, but only rely on the corpus of a single language to master the syntax, grammar and other key elements of the language. In other words, deep learning does not understand English and Chinese, but through a large number of learning materials, they can obtain the ability of Chinese-English translation." "In the final analysis, AI can translate and understand animal language, which is still based on human's effective interpretation of language." Tan Mingzhou said. There is still a long way to go to realize cross species communication "Human language has rules to follow, so languages in different countries can follow the rules to learn. But there are unknown barriers to animal language rules. Therefore, AI still has a long way to go and needs to overcome some difficulties to realize cross species language translation." Tan Mingzhou said. First of all, it has been proved that the algorithm trained with data affected by human bias can easily lead the results to the "wrong way". For example, a dog will make a rapid cry, which may be because it wants to beg from its owner, remind its owner of strangers, or blame its owner for not playing with itself. However, if researchers only based on their own cognition, when marking the call data, they think that the call only expresses the pet's demand for food from the owner, so as to mark the data, artificial intelligence will have great limitations in learning data and translation. This kind of translation can easily lead to communication barriers between people and pets, thus losing the meaning of pet language translation. "In research, researchers need to go to the 'human center', that is, cross species communication with the help of algorithms needs algorithms to avoid some human biases." Tan Mingzhou pointed out. Secondly, matching human language with animal language through AI algorithm requires a large number of extensive and perfect data acquisition and scene training to complete the interpretation of animal language and summarize the "rules". This requires extensive simultaneous collection and comparison of animal cry and EEG data, and then put them into the database. However, the vocal cord characteristics of different dog breeds are different, and the vocal performance in the face of the same scene is also different. There are countless combinations of such scenes and barks, which brings great challenges to the data acquisition work. Tan Mingzhou said that in terms of technology, to achieve accurate translation of an AI translation product, at least several problems need to be solved: on the form side, if the form of shooting translation is used, the problems related to image recognition should be solved, and the form of simultaneous translation should be solved; On the content side, AI translation products also need to overcome problems such as text language analysis and big data collection. Due to AI's lack of common sense judgment on visual scene, auditory scene and natural language processing, AI needs to develop to the stage where it can deal with these problems very accurately. In addition, some scholars pointed out that the generation gap between animal language and human language exists objectively. What AI can do is to continuously improve its own functions and improve the database, content, corpus and scene by scientific means; Only in this way can people and animals effectively communicate on a solid foundation. "Although some studies have made great progress, the problems can not be ignored. At present, relevant studies only realize the simple information transmission between humans and animals to a certain extent, and there is still a long way to go before the real cross species communication is realized. In the short term, it is still difficult to realize cross species communication, but more and more studies are undoubtedly opening a door for it 。” Tan Mingzhou said. (Xinhua News Agency)