perceptron parallelization discriminative SMT BLEU WER lexicalized disfluent pointwise listwise Viterbi derivations parametrized commonly iteratively SGD computable alia lowercased tokenized binarized backoff Hiero cdec hypernode hypergraph summands symmetrized arity SCFG bigrams bitext Mert dtrain hyperparameters hypergraphs initializations optimizers synthetical multipartite interpretable formalization evaluable TER Levenshtein iff precisions MBR datasets hyperparameter infeasibly suboptimal rescoring partite runtime linearithmic differentiable SVMs regularizer infeasible pre-defined pre tradeoff sharding parallelized overfitting MapReduce unregularized gzip hadoop IPC resharding segmenter ALPAC gisting transduction HAMT IAMT termbases belonging reranking belonging optimality additivity backends TransType termbase employing translatables trigram tokenization parallelizable evaluators passively assessment HTER keystrokes KSR WSR KSMR repeatability verifiably NMT PWR ANOVAs MLE anonymized tf idf truecasing subsample trigrams Adadelta HBLEU mini subwords reachability adaptive RNN embeddings Europarl subword