[Ml-stat-talks] Wed: Kuzman Ganchev on rich prior knowledge in NLP

David Mimno mimno at CS.Princeton.EDU
Mon May 7 11:32:31 EDT 2012

Can we make computers understand written language given effectively
unlimited data, massive amounts of processor time, and the best stats
talent available? I'm guessing this week's Machine Learning lunch
speaker knows the answer. If you want to find out how much he can tell
us about it, come to this talk!

Kuzman Ganchev, Google

CS402, Wed May 9, 12:30

Rich Prior Knowledge in Learning for Natural Language Processing

We possess a wealth of prior knowledge about most prediction problems,
and particularly so for many of the fundamental tasks in natural
language processing.  Unfortunately, it is often difficult to make use
of this type of information during learning, as it typically does not
come in the form of labeled examples, may be difficult to encode as a
prior on parameters in a Bayesian setting, and may be impossible to
incorporate into a tractable model.  Instead, we usually have prior
knowledge about the values of output variables.  For example,
linguistic knowledge or an out-of-domain parser may provide the
locations of likely syntactic dependencies for grammar induction.
Motivated by the prospect of being able to naturally leverage such
knowledge, no less than five different groups have recently developed
similar, general frameworks for expressing and learning with side
information about output variables.  This talk will provide a brief
survey of these frameworks as well as some examples of how they have
been applied.

More information about the Ml-stat-talks mailing list