Training a Constraint Grammar Tagger

In (Lager 1999a) it was shown how transformation-based learning can be used to induce a Constraint Grammar (Karlsson et al. 1995) from tagged corpora. In the following, we perform a small experiment based on this idea.

Train and test, and see what happens! From the OS prompt, run:

> ./mutbl -f examples/cg_pos_tagging.script

Inspect the script for information about where templates and training and test data are located.

For your report: Note the setting of the accuracy threshold to 1.0. Why is this important when learning CG rules? (You will find the answer in the relevant paper if you cannot figure it out by yourself.) What happens if you lower the accuracy threshold? Try with (say) 0.98 and 0.90. Report the results (both precision and recall) and explain (if you can) the tradeoffs involved..