Monday, May 03, 2010
10:00 AM
Computer Studies Bldg. Room 703
Ph.D. Thesis Proposal
Jonathan Gordon
University of Rochester
Learning World Knowledge Suitable for Inference
To enable human-level artificial intelligence, machines must have access to the same kind of commonsense knowledge about the world that people have. It must be represented in a way that allows it to be used for reasoning, and it must be available in high-volume. We argue that the best source for such knowledge is text -- learning by reading. Our work so far has focused on expanding the scale of knowledge extraction to the vast -- and noisy -- sources of writing found on the Web and the use of crowds of non-experts to evaluate such results. Now we propose addressing the issue of acquiring knowledge suitable for reasoning by considering several issues in determining the meaning of conditional knowledge, both when directly interpreting generic statements and when abstracting specific instances.