Researchers of the Human Communication Research Centre at the University of Edinburgh are developing a computer system that learns how to communicate. Oliver Lemon and his team are working on a tourist information system that aims at assisting tourists coming into a foreign city. Their Talk City Dialogue System allows tourists to find hotels or restaurants by simply asking their computer for help. The System applies machine learning techniques to understand the great range of communication possibilities. While talking to the person, the system tries out different possibilities of having a conversation. In exchange it receives rewards or punishments. For example, if the user is angry and hangs up the phone, the computer knows something has gone wrong and that it may not repeat this way of communicating. The difficulty is that the computer needs more information than just the language to be able to communicate. It has to know its exact tasks to be able to provide the correct information.
Another example where computer interfaces can be helpful is when travelling into a country where a foreign language is spoken. At IBM in Prague, Jan Kleindienst is developing a real-time speech-to-speech translation system. The system translates from the tourist’s native language into the target language. The computer proceeds in three steps: firstly, it transcripts the spoken words, then it translates them and finally synthesises the sentence. The researchers at IBM have developed a method of breaking down sounds to phonemes which are very basic sounds. Additionally, they are in possession of a phonetic alphabet for every language. Based on these phonetic alphabets and with the help of phonemes their speech to speech translation system is able to recognise every word in every language.
The systems invented in Edinburgh and Prague are still in development process and their biggest challenge is to consider all aspects of conversation. The computer systems are often compared to Hal, the artificial intelligence of the Discovery spaceship from the novel “2001: A Space Odyssey” by Arthur C. Clarke. The difference between this fictional computer and the latest research results from Edinburgh and Prague is that Hal did not have to cope with natural language. Herein lays the biggest challenge: the lack of understanding. As soon as the computers can understand human speech, it will mean a dramatic step forward for the researchers. Nevertheless, it might take many more years until “human” computers will become a reality.
