When an IBM supercomputer named Watson takes on human contestants on the quiz show Jeopardy! this week, it will represent the latest triumph of advances that could one day enable people to interact much more naturally with computers, according to a UT Dallas expert in human-machine interaction.

“This is a milestone because Watson will compete directly against humans, and moreover against the champions of Jeopardy!” said Dr. Sanda Harabagiu, a leader in the field of human-computer interaction for more than a decade. “This is important, and it’s a more realistic gauge than simply comparing a system to search engines.”

It’s hard enough to create a system that can interact via spoken language fast enough to compete on Jeopardy!, but Harabagiu and her collaborators in the University’s Human Language Technology Research Institute are already working on even more complex human-computer interactions.

“We have focused on several classes of complex questions, not only on the kind that occur on Jeopardy!” said Harabagiu, who is an associate professor of computer science and director of the research institute. “We have tackled temporal questions and causal questions, and we have also developed methods that capture complex information about events and actions, thus enabling interaction about complex questions that imply elaborate scenarios.”

For example, consider the question, “What were the effects of the winter storm in Dallas preceding Super Bowl XLV?” Understanding and responding conversationally to such a question is much more complicated than, say, simply identifying who played Frau Blücher in the movie Young Frankenstein. (Cloris Leachman, of course.)

“We have to find new ways for combining the semantic processing of questions with many aspects of the pragmatics encountered in possible answers,” said Harabagiu, who is also the University’s Jonsson School Research Initiation Professor. “The problem of question-answering is not a solved one, but IBM’s Watson will bring attention to the level of effort required to be able to answer even trivia or general-knowledge questions. When we imagine systems capable of common-sense reasoning, and the capability to process causality and procedural questions, we need to think in terms of increasing their capabilities by several orders of magnitude.”

Harabagiu and her colleagues recently collaborated with researchers at Stanford University and the International Computer Science Institute in their continuing efforts to lay the groundwork for a day when computers could understand and reply to virtually any question, regardless of complexity.

IBM’s Watson will face Ken Jennings and Brad Rutter, the two most successful players in Jeopardy! history in episodes broadcast the week of Feb. 14. At stake will be $1 million, which will go to charity if the computer wins. Producers said Watson qualified for the show by passing the same test human contestants take.