BLEU, WER, and KSMR results for the Europarl test corpora using conventional (batch learning without retraining) and online SMT systems. Log-linear weights were adjusted via MERT. The average online learning time (LT) in seconds is shown for the online system.
| Corpus . | SMT system . | BLEU . | WER . | KSMR . | LT (sec) . |
|---|---|---|---|---|---|
| Eng-Spa | conventional | 21.0±0.5 | 65.4±0.7 | 45.9±0.4 | - |
| online | 22.5±0.6 | 63.4±0.7 | 44.7±0.4 | 0.2 | |
| Eng-Fre | conventional | 21.2±0.6 | 64.4±0.7 | 44.4±0.5 | - |
| online | 22.6±0.6 | 63.2±0.7 | 43.2±0.5 | 0.2 | |
| Eng-Ger | conventional | 13.1±0.5 | 73.8±0.7 | 49.2±0.4 | - |
| online | 14.1±0.5 | 72.8±0.7 | 48.0±0.4 | 0.2 |
| Corpus . | SMT system . | BLEU . | WER . | KSMR . | LT (sec) . |
|---|---|---|---|---|---|
| Eng-Spa | conventional | 21.0±0.5 | 65.4±0.7 | 45.9±0.4 | - |
| online | 22.5±0.6 | 63.4±0.7 | 44.7±0.4 | 0.2 | |
| Eng-Fre | conventional | 21.2±0.6 | 64.4±0.7 | 44.4±0.5 | - |
| online | 22.6±0.6 | 63.2±0.7 | 43.2±0.5 | 0.2 | |
| Eng-Ger | conventional | 13.1±0.5 | 73.8±0.7 | 49.2±0.4 | - |
| online | 14.1±0.5 | 72.8±0.7 | 48.0±0.4 | 0.2 |