An implementation of BLEU evaluation metric.
$ python calculatebleu.py /path/to/candidate /path/to/reference
The first parameter will be the path to the candidate translation (a single file), and the second parameter will be a path to the reference translations (either a single file, or a directory if there are multiple reference translations).
The program will write an output file called bleu_out.txt which contains a single floating point number, representing the BLEU score of the candidate translation relative to the set of reference translations.
Language | Candidate | Reference | BLEU score |
---|---|---|---|
German | candidate-1.txt | reference-1.txt | 0.151184476557 |
Greek | candidate-2.txt | reference-2.txt | 0.0976570839819 |
Portuguese | candidate-3.txt | reference-3.txt | 0.227803041867 |
####Notes
- The candidate and reference files will be in UTF-8 encoding.
- You may assume a line-by-line correspondence of sentences between the candidate and reference translations.