UPGRADE YOUR BROWSER! If you see this message, it means your browser, which is CCBot/1.0 (+http://www.commoncrawl.org/bot.html), does not support current webstandards. Please, see the webstandards project.

This web page contains the solution to an enhanced version of the fourth assignment assignment of the second part (transformation based learning) of the course on machine learning taught at the graduate school of language technology, fall 2003. Authors of this page are Fredrik Olsson and Magnus Sahlgren.

Noun phrase chunking

Results from initial run:

DATA STATISTICS:

            Corpus Size: 9597
         Number of Tags: 9597
 Number of Correct Tags: 8978
       Number of Errors: 619
                 Recall: 93.6% 
              Precision: 93.6%
                F-Score: 93.6%
Number of Tags per Word: 1.000

Applied 26 rule(s) for feature(s) [tag] in 1.800 seconds

DATA STATISTICS:

            Corpus Size: 9597
         Number of Tags: 9597
 Number of Correct Tags: 9194
       Number of Errors: 403
                 Recall: 95.8% 
              Precision: 95.8%
                F-Score: 95.8%
Number of Tags per Word: 1.000

The speed of the learner incresed dramatically when we constrained the template file to only include the templates that were in fact used to generate rules. The reason for this is that the TBL algorithm searches through every possible rule instantiation, which means that the more templates you have, the longer time it will take for the learner to find the highest scoring rule.