Translations of
Translations of
- Translations are sorted by increasing edit distance (ED) between the Statistical
Machine Translation (SMT) and the reference translation (REF)
- PER means Position Independent Error Rate between SMT and REF
- REP is the number of times the same source sentence occurred in the test corpus
- Translations differing from the reference (edit distance >= 1)
have been evaluated by hand (SMT or REF)
- they are both shown on a light gray background when they are judged equivalent
- otherwise the worse translation is shown on a dark gray background
an the better translation is given a lighter gray backgroung
- sentences in red (SMT or REF) are rated as bad translations that
should be corrected manually by reading the source sentence
- sentences in blue (SMT or REF) are rated as bad formulations that
could be corrected manually by only looking at the translation (no need to look at the source)
- The classical Sentence Error Rate (SER) for this corpus is:
(
when weighted by the number of repetitions).
It is
obtained by dividing the number of translations Worse than reference by the sum of the
number of Evaluated translations and the number of Identical with reference.
Statistics on the SMT and reference translations |
| raw scores | weighted by REP |
Rating of SMT | NB | % | NB | % |
Rating of sentences | |
|
30%
top
eqv
worseTrans
betterTrans
worseTrans
betterTrans
badForm
badTrans
|
|
|
|
|