PEOPLE who send cellphone text messages to Internet chat rooms will soon be able to rest their aching thumbs. Michael Kearns at the University of Pennsylvania has developed an intelligent system that lets them join in simply by speaking into their phones.
Chat rooms are online “virtual” environments with different rooms in which people can socialise, move around and even perform basic actions and gestures, such as laughing, hugging and nodding. This is all done by typing dialogue and commands into a keyboard.
But to enter this sort of world via a phone requires more than just a system that recognises speech and which converts text to speech in return. Just as in a real room, multiple conversations are going on at the same time between different people who pop in and out of the chat room at will. To make it possible for someone on the phone to make sense of it all Kearns uses a “software agent”, called Cobot, to act as their proxy presence in the chat room.
Advertisement
Cobot allows users to enter and modify environments. They can create new rooms, for example. “Cobot allows users, wherever they are, to call in and get detailed information about that environment and get confirmation that their actions in that environment take place,” says Kearns. Working with colleagues at MIT, Georgia Tech and the University of Pittsburgh, Kearns developed an interface so someone on the phone can talk to Cobot and get it to tell them what is happening in the chat room.
Most speech recognition systems aren’t that good, says Kearns. To improve recognition they use subject-specific “grammars” to reduce the risk of ambiguity. And Cobot’s dialogue system exploits these grammars to maintain the vernacular of the chat rooms.
“This is a very pop-culture environment and so a lot of people have slang phrases that they frequently use,” says Kearns. So phone users can specify a particular grammar which contains a list of a few hundred phrases that they know they will repeatedly use, such as “Chillin’ like a villain” or “Eat my shorts”.l
