Stabilizing Minimum Error Rate Training

The most commonly used method for training feature weights in statistical machine translation (SMT) systems is Och's minimum error rate training (MERT) procedure. A well-known problem with Och's procedure is that it tends to be sensitive to small changes in the system, particularly when the number of features is large. In this paper, we quantify the stability of Och's procedure by supplying different random seeds to a core component of the procedure (Powell's algorithm). We show that for systems with many features, there is extensive variation in outcomes, both on the development data and on the test data. We analyze the causes of this variation and propose modifications to the MERT procedure that improve stability while helping performance on test data.