Results on (Hierarchical Sample Clauses

Results on (Hierarchical. Phrase-based Translation Table 2 shows the results on phrase-based and hierarchical phrase-based translation systems. We find that our approach systematically outperforms GIZA++ and BERKELEY on all NIST datasets. In particular, generalizing the agreement to model the discrepancy between word alignment and phrase segmentation is consistently beneficial for improving translation quality, suggesting that it is important to introduce structural constraints into word alignment to increase the correlation between alignment and translation. While “SM+SV” improves over “SM” signifi- cantly on phrase-based translation, the margins on the hierarchical phrase-based system are relative- ly smaller. One possible reason is that the “SV” system loss |AC→E| |AE→C| |AC→E ∩ AE→C | F1 GIZA++ N/A 29.39M 27.64M 17.07M 59.86 BERKELEY HM 29.12M 28.09M 21.30M 74.46 this work SM 29.84M 29.31M 20.24M 68.42 SM+SV 30.04M 29.50M 20.54M 69.00 Table 3: Agreement evaluation of GIZA++, BERKELEY and our approach. The F1 score reflects how well two asymmetric alignments agree with each other. loss function can better account for phrase-based rather than hierarchical phrase-based translation. It is possible to design new loss functions tailored to hierarchical phrase-based translation. We also find that the BLEU scores of XXXXX- XXX on hierarchical phrase-based translation are much lower than those on phrase-based transla- tion. This might result from the fact that XXXXX- XXX is prone to produce one-to-one alignments, which are not optimal for hierarchical phrase- based translation.