Next: A first language model
Up: Introduction
Previous: Chapters
Our approach differs
significantly from that of McEnery and Wilson's textbook, as it does from
Charniak's. The big difference is that we take a thoroughly
Bayesian approach to all the necessary statistics, with several
major consequences:
- In general we aim for a higher level of mathematical rigour than
the other sources.
- We need to introduce some basic probability theory early on,
in order to motivate the Bayesian approach.
- We don't present algorithms and statistical tests in the usual
cookbook style. We feel that creative data-intensive linguistics demands a
flexible approach to statistics,
and we hope that a systematically Bayesian
approach will give students a secure
basis from which to apply old principles to
new problems.
Chris Brew
8/7/1998