Human language exhibits complex structure. To be successful, machine learning approaches to language-related problems must be able to take advantage of this structure. I will discuss several investigations into the relationship between structure and learning, which have led to some surprising conclusions about the role that structure plays in language processing. From there, I will consider the question of: where does this structure come from. By taking insights from linguistic typology, I will show that very simple typological information can lead to significant increases in system performance for some simple syntactic problems. Moreover, I will show how this typological information can be mined from raw data. (This talk includes joint work with Dan Klein, John Langford, Percy Liang, Daniel Marcu, and some of my students: Arvind Agarwal, Adam Teichert and Piyush Rai.)
Hal Daume III is an assistant professor in the School of Computing at the University of Utah. His primary research interests are in understanding how to get human knowledge into a machine learning system in the most efficient way possible. In practice, he works primarily in the areas of Bayesian learning (particularly non-parametric methods), structured prediction and domain adaptation (with a focus on problems in language and biology). He associates himself most with conferences like ACL, ICML, NIPS and EMNLP. He earned his PhD at the University of Southern Californian with a thesis on structured prediction for language (his advisor was Daniel Marcu). He spent the summer of 2003 working with Eric Brill in the machine learning and applied statistics group at Microsoft Research. Prior to that, he studied math (mostly logic) at Carnegie Mellon University. He still likes math and doesn't like to use C (instead he uses O'Caml or Haskell). He doesn't like shoes, but does like activities that are hard on your feet: skiing, badminton, Aikido and rock climbing.
This talk is part of the CLIP Colloquium Series. For the complete schedule, please visit http://www.umiacs.umd.edu/research/CLIP/colloq/.