The Tanl Named Entity Recognizer at Evalita 2009

We describe the tagger present in the Tanl toolkit, which is a flexible and customizable tool for use in various tagging tasks, including POS tagging and SuperSense tagging. The tagger uses a variety of features, both local and global, which can be specified in a configuration file. The tagger is based on a Maximum Entropy classifier and uses dynamic programming to select accurate sequences of tags. We applied it to the NER tagging task in Evalita 2009, customizing the set of features to use and generating a set of dictionaries from the training corpus, that also provide additional features. The final accuracy is further improved by applying simple symbolic rules.