You can find here the data that we will use for the joint paper on neural network models for SMT

explanation of the data

Moses models [26GB]

   moses.ini moses.ini (V2)

bitexts (after selection)

LM data (after selection)

LM data (original)

dev+test data

Update May 5th 2014

New LM with a 226k vocabulary [3.0GB]

226k vocabulary itself

bitexts (2, 4, ..10%) [1.0GB]

Moses models trained on bitexts.pc2 [173MB]

Moses models trained on bitexts.pc4 [340MB]

Moses models trained on bitexts.pc6 [483MB]

Moses models trained on bitexts.pc8 [607MB]

Moses models trained on bitexts.pc10 [720MB]


distinct 1000-bests, generated with model.380M

ntst1213 (dev)

ntst14 (test)