EMNLP 2015 TENTH WORKSHOP ON STATISTICAL MACHINE TRANSLATION Tuning Task September 2015, in conjunction with EMNLP 2015 in Lisbon, Portugal
http://www.statmt.org/wmt15/tuning-task/ As you may have noticed, the EMNLP WMT15 workshop includes a shared task on *tuning a given MT system*. The actual English-to-Czech and Czech-to-English hierarchical systems (Moses) have been released. Your task is to find the best settings of weights in the moses.ini, possibly adding sparse features as available in Moses Release 3.0. If you wish to make use of our *complimentary manual evaluation* for translations into Czech during your development, please register by February 22, 2015. In the task, you are most welcome to experiment with: - various optimizers - various MT evaluation metrics - various sparse features - additional development data (non-constraint track) or any other changes in the tuning process. Any source-code change is permitted for the tuning phase. A submission to the tuning task consist of the weight settings in the moses.ini (and optionally a weights file for sparse features). We will run the fixed translation system with your settings on the WMT15 test sentences and the outputs will be evaluated in the same style of manual evaluation as the main WMT15 translation task. More details: http://www.statmt.org/wmt15/tuning-task/ Looking forward to your submissions! Tuning Task Organizers Ondrej Bojar (Charles University in Prague) Milos Stanojevic (University of Amsterdam) Amir Kamran (University of Amsterdam) -- Ondrej Bojar (mailto:[email protected] / [email protected]) http://www.cuni.cz/~obo _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
