TDT-2002 Topic Tracking at Maryland: First Experiments with the Lemur Toolkit

The University of Maryland submitted six topic tracking runs for the 2002 Topic Detection and Tracking evaluation. Two runs were produced using the Lemur language modeling toolkit, the remaining four were produced using an separate system coded in Perl. The Lemur runs outperformed the Perl runs on the required condition because term frequency information was better handled. Two of the Perl runs used native Arabic orthography with two-best translation based on a statistical lexicon, obtaining similar results to those ob- tained with the Arabic-to-English translations provided with the collection.