Translingual Topic Tracking Applying Lessons from the MEI Project

The University of Maryland participated in the topic tracking task submitting four runs for the required conditions ba sic and challenge In this working notes paper we present preliminary results based on those runs and six additional contrastive runs that explored translation selection post translation resegmentation post transcription document ex pansion and source dependent normalization One approach that yielded substantial improvements was post translation resegmentation into overlapping character bigrams We also implemented a new variant of balanced query translation that produced modest improvements in e ectiveness and substan tial improvements in e ciency over our the balanced docu ment translation technique that we used last year