Correlation of BLEU with ranking-based human evaluation reported in WMT events, for German–English and English–German MT in a news domain.
Event . | correlation type . | correlation for German–English MT . | correlation for English–German MT . |
---|---|---|---|
WMT07 | Spearman | 0.40 | 0.26 |
WMT08 | Spearman | 0.12 | 0.58 |
WMT09 | Spearman | 0.41 | −0.43 |
WMT10 | Spearman | 0.52 | 0.39 |
WMT11 | Spearman | 0.48 | 0.44 |
WMT12 | Spearman | 0.67 | 0.22 |
WMT13 | Spearman | 0.90 | 0.83 |
WMT14 | Pearson | 0.83 | 0.22 |
WMT15 | Pearson | 0.86 | 0.57 |
WMT16 | Pearson | 0.88 | 0.78 |
Event . | correlation type . | correlation for German–English MT . | correlation for English–German MT . |
---|---|---|---|
WMT07 | Spearman | 0.40 | 0.26 |
WMT08 | Spearman | 0.12 | 0.58 |
WMT09 | Spearman | 0.41 | −0.43 |
WMT10 | Spearman | 0.52 | 0.39 |
WMT11 | Spearman | 0.48 | 0.44 |
WMT12 | Spearman | 0.67 | 0.22 |
WMT13 | Spearman | 0.90 | 0.83 |
WMT14 | Pearson | 0.83 | 0.22 |
WMT15 | Pearson | 0.86 | 0.57 |
WMT16 | Pearson | 0.88 | 0.78 |