Johns Hopkins Computer
Science

Finite-State Methods In Natural Language Processing

Offered as a pair of short courses: 600.405 (Fall 2000) and its sequel 600.406 (Spring 2001).

Catalog description: Finite-state transducers, a generalization of finite-state automata, can efficiently compute many useful functions and weighted (probabilistic) relations on strings. We'll cover theory and practice, including algebraic operations for building transducers, software tools, and a range of applications to natural language.

Announcements

Assignments 4 and 5 due by May 12, please. I'm not on campus so you'll need to use email/fax/snailmail.

Where to find things for spring semester

Policies


Syllabus

Finite-state methods are enjoying a lot of new attention these days. The NLP community has been inventing new algorithms and applications, as well as unearthing old results. A sampling of topics is below. We will not be able to cover more than a fraction of them, so rather than setting a fixed schedule in advance, I'll try to take your interests and background into account as the course progresses.

Really this class is about constructing useful functions. We'll focus on what can be done theoretically; how to do it elegantly and efficiently; and how to do it in practice with existing software toolkits.

Resources

Overview readings that will reinforce this material:


Some possible topics for future lectures in 600.405 and 600.406

Send me mail if you see something you're particularly interested in, or if you'd like to propose another topic.

Jason Eisner - jason@cs.jhu.edu - $Date: 2002/12/22 18:32:40 $