Next: Summary
Up: Estimating Model Parameters
Previous: Estimating Model Parameters
In a binary choice between English and Spanish strings drawn from a
bilingual corpus, an accuracy of 92%
can be got from 20 bytes of test data and 50Kbytes of training data,
improving to about 99.9%
when 500 bytes of test data are allowed.
If you have very small amounts of training or test data it may be
better to stick with low-order models.
Chris Brew
8/7/1998