I am working with Gary Marcus of the Psychology Department on commonsense physical reasoning about containers in human cognition and in automated reasoning. The web page for the project is here.
Developing ontologies of matter for representing and reasoning about simple physics and chemistry experiments, such as the one depicted here, from Michael Faraday's The Chemical History of a Candle, in which hydrogen is produced by passing steam over heated iron filings. Understanding such experiments involves combining formal scientific knowledge, such as the chemical equations; commonsensical physical and spatial knowledge, such as the knowledge that the gas will be trapped in the inverted test tube; and knowledge of the perceptual and manipulative powers of the experimenters, such as knowing they can see the water level lowering in the test tube, though they cannot see the hydrogen gas directly. A particular focus of my work is on the use of partial knowledge of the geometry involved; for instance, the exact shape of the test tube is not critical, but it is critical that it does not have a hole at the top.
Gary Marcus and I have also written two papers studying the limits of simulation as a technique in commonsense physical reasoning: one analyzing the issue in AI systems, the other analyzing it in cognitive models.
The TACIT project, aims toward building a corpus of short texts that are hand-annotated with all the commonsense knowledge and reasoning needed to interpret them, and developing a systematic analysis of the issues involved. The project is currently on a back-burner, but may be resumed at some point, if I can find a meaningful way to move it forward. The project is currently on a back-burner, but may be resumed at some point, if I can find a meaningful way to move it forward.
In collaboration with Hector Levesque and Leora Morgenstern, I have assembled a collection of Winograd schemas.