An assistant professor at DePaul University has developed computer software that talks with people and responds with a variety of emotions, including joy, love, hope, and compassion.

The program, called the Affective Reasoner (AR), is a collection of artificial intelligence programs that allows a computer to engage in spoken conversations with people, perform some forms of intelligent reasoning about emotions, and respond to these emotions in kind.

Clark Elliott, associate director of DePaul's Institute for Applied Artificial Intelligence, began work on the project six years ago and sees several practical applications for his research. The AR could be used to create friendly computer tutoring systems, emotionally interactive storytellers and games, or training programs for sales people who need to know how to interact with different personality types.

The AR's programs are embodied in multimedia agents, or cartoon-like, on-screen characters. The characters have faces that morph to show up to 70 different facial expressions and have mouths that move while talking with people.

The personalities of AR's agents have two parts: a disposition which controls how they construe the world, which in turn leads to different emotional responses to situations that arise, and a temperament which controls how they express emotions

The AR is capable of projecting 24 general categories of emotions--including love, hope, fear, gloating, reproach, joy, anger and gratification--in response to emotional situations discussed by a computer user. Each category of emotion has 22 variables that control the quality and intensity of the emotion expressed and 440 channels to express emotions. Like humans, AR can have mixed emotional responses.

Speech recognition and synthesization technology allow the system to listen to spoken English and respond in English. The program also can retrieve and play music that suits the mood of the conversation.

Despite the sophisticated technology involved, the AR can run on a standard multimedia computer and is efficient enough to be accessed over the World Wide Web through a standard modem, although Elliott is still developing the web version.

While AR can reason about human emotions and respond in a variety of ways, "it doesn't understand anything but emotion," Elliott explained. "For example, if you say you are worried about your test, it might ask whether the test is important to you, if a negative outcome is possible and other questions. Then it will respond that it is sorry you are worried and offer encouragement. "The program understands worry and asks questions to determine the degree of worry, the person's goals and actions, and then responds appropriately. But it doesn't know what a test is," he said.

Elliott has already experimented with programming the agents to read and act out parts of a children's story, which is possible because the agents are able to talk and respond to each other as well as to people. He showed the program to a group of children and they interacted with AR's storytellers by making verbal requests, such as asking that a favorite part to be told again. "Even a child as young as two could interact with the computer agents," he said.

One of Elliott's goals is for AR to be used as a friendly computer tutor that can not only offer students access to a vast amount of information, but also respond in an understanding way when a student becomes frustrated, or possibly in a competitive way when a student says he is bored.

Elliott, who has been a member of the faculty of DePaul's School of Computer Science, Telecommunications and Information Systems for the last five years, said the friendly tutor idea is an extension his own teaching philosophy. He began his teaching career at 15, tutoring trumpet students, and later switched to computer science, earning a Ph.D. in 1992 from Northwestern's Institute for the Learning Sciences.

"I've been devoted to teaching for 25 years now. I feel that it's important to pay close attention to my students' goals, what's important to them and what motivates them. To be an effective teacher, you have to understand that emotions are an integral part of what people think about and how they operate," he said.

"My research is an extension of this teaching style," he said "It's about teaching computers to behave in compassionate, understanding ways in situations where they currently do not have these skills."