 Human beings are the only species that have invented spoken language. We speak to communicate and we speak to survive. Even people who cannot talk have invented sign languages. So I think therefore I am, really is, I speak therefore I am. And the evolution of human language parallels that of the evolution of our civilization and it's a very, the right reflection of our society and the right representation of human knowledge. We talk to our infants as soon as they are born. We teach them to speak. We transmit knowledge, our love, our hope to the infants. A child listens to about from several thousand to tens of thousands of words a day. By the age of two they can generate about two to three hundred words. A machine can recognize 20 names 10 years ago. We even talk to animals just like when we talk to babies. It's how we speak the tone of our voice, the melody that the animal can understand or what we mean even without understanding the actual words. So the emotion in our speech is also a very important component of semantic representation. The cognitive, neurological and muscular chain of activities that constitutes the speech chain. We formulate our thoughts, we choose the words we use and we speak in complete sentences and then when the sentences goes to the listener the process is reversed. So the listener then decode the message. And the speech chain between human and machine goes like this. The speech waves are converted into digital form and then into parameters. And then the speech recognition that transforms these parameters into words and then the semantic decoder transforms words into meaning. So, and then the dialogue system controls the interaction. We are now very familiar with using spoken language systems to look for information, to ask for, to call somebody and a command such as open the door, it should be clear enough to machines, that's how we feel. Unfortunately, the machine response is not always what we expected. This is due to many multiple challenges. One major challenge is when we do speech recognition of signal to words we can make mistakes, we meaning machines and the engineers. The semantic decoder can also make mistakes from words to the meaning. Machines do not always understand the intention of the user. We're used to now using spoken language systems to help us. All right, we are, for 10 years now there has been commercial systems and we know we're talking to machines most of the time. But increasingly people ask questions like, Siri, do you love me? Siri, are you a man? Siri, do you want to marry me? So, in the future there's going to be a blurring of machine and humans. For example, in last year, 2014, there's a chatbot that has fooled 30% of the scientists into believing that it is not a machine, it is indeed a 13-year-old Ukrainian boy. So this is the end game of a Turing test. And the machines will become empathetic. All right, we are training machines to understand human emotions and to have emotions. Machine intelligence is not at the stage where they can be intentionally malicious. So don't worry about that. We are trying very hard to teach them to be nice. All right, but machines which can really understand human emotions can be of great help to the society. They can be companions and caretakers of the elderly, for example. China as well in many other societies have fewer and fewer young people to take care of the elderly. So empathetic machines can really be an asset to the aging society. And empathetic machines that can learn will also be a great asset to education. They can teach children through playing new knowledge. And at the same time, machines will also learn from their interaction with humans. They will learn from us and become more intelligent with use over time. For example, in the future, there will be a professor who looks like me, talks like me, but who is not me. This humanoid professor will be able to search for information on the Internet to prepare a lecture, to formulate PowerPoint slides automatically and to speak with the synthesized voice that sounds very much human-like. And even within 10 years, maybe there will be ideas lab of humanoid professors. And we are also training machines to understand humor, sarcasm, irony. People will consider to be the mystical parts of human intelligence. To do that, we have to build better understanding, better models of our own human intelligence. Today, we have about 6,000 languages in the world. Most of them are fast-disappearing. Machines that can learn a lot of these languages can help to translate from different languages to other languages. When we build machines, it's no longer just for human-machine communication. It will help human-to-human communication. It will help to build bridges between civilizations and improve and enhance part of our human society. Thank you very much.