New! We are pleased to announce that the invited speakers for TextGraphs-8 will be

Dept. of Computer Science & Engineering

University of Washington

TITLE: Extracting Tractable Probabilistic Knowledge Graphs from Text

ABSTRACT: Graphs are a powerful and intuitive representation for knowledge extracted from text and the Web, but inference over them is generally intractable. We thus have to resort to approximate inference methods that are often hard to use and can give unpredictable results. Wouldn't it be wonderful if we had a class of graphs sufficiently general to represent a large fraction of the knowledge we can extract from text, but where inference is still tractable? We've found such a class: it's called tractable Markov logic, or TML for short, and this talk is about it. Similar to semantic networks, TML uses ISA and IS-PART hierarchies to organize knowledge and direct inference. TML is surprisingly powerful, having bounded-recursion PCFGs, sum-product networks and other representations as special cases. As a result, we can use it as a basis for the entire knowledge extraction process, from syntactic and semantic parsing to knowledge base population and use. I will describe TML, what makes it work, and our early experiments applying TML to biomedical information extraction and question answering. (Joint work with Chloe Kiddon.)

Executive Director

Allen Institute of AI

TITLE: Opportunities for Graph-based Methods in Machine Reading