A group of researchers and computer scientists is building an artificial intelligence system that can analyze instruction via speech recognition and natural language processing, says Sean Kelly, a sociologist at the University of Pittsburgh School of Education.
Developers want to replicate, as closely as possible, the feedback that a content expert might give to a teacher.
Recently published research focused on “authentic” questions that middle school teachers asked to spark discussion and for which there was no preset answer. Human observers reviewed hundreds of recordings of four classroom sessions and classified how the teachers presented lessons.
LINK TO MAIN CIO NEWS ARTICLE: Teachers teaching teachers
An adaptive algorithm was then created to identify speech patterns associated with authentic questions, Kelly says. For this experiment, the difference in questions deemed authentic by humans and the computer was not statistically significant, at 3.9 percent versus 3.6 percent, respectively.
However, it took programmers a long time to teach the robot the algorithms for a single experiment.
The feedback system shows promise, but Kelly emphasizes that the AI needs more accurate algorithms, along with the ability to discern components of other classroom discussions, before it can adequately analyze instruction.