30 lines
833 B
Plaintext
30 lines
833 B
Plaintext
|
==========
|
||
|
|
BLEU tests
|
||
|
|
==========
|
||
|
|
|
||
|
|
>>> from nltk.translate import bleu
|
||
|
|
|
||
|
|
If the candidate has no alignment to any of the references, the BLEU score is 0.
|
||
|
|
|
||
|
|
>>> bleu(
|
||
|
|
... ['The candidate has no alignment to any of the references'.split()],
|
||
|
|
... 'John loves Mary'.split(),
|
||
|
|
... (1,),
|
||
|
|
... )
|
||
|
|
0
|
||
|
|
|
||
|
|
This is an implementation of the smoothing techniques
|
||
|
|
for segment-level BLEU scores that was presented in
|
||
|
|
Boxing Chen and Collin Cherry (2014) A Systematic Comparison of
|
||
|
|
Smoothing Techniques for Sentence-Level BLEU. In WMT14.
|
||
|
|
http://acl2014.org/acl2014/W14-33/pdf/W14-3346.pdf
|
||
|
|
>>> from nltk.translate.bleu_score import sentence_bleu,SmoothingFunction
|
||
|
|
|
||
|
|
|
||
|
|
>>> sentence_bleu(
|
||
|
|
... ['It is a place of quiet contemplation .'.split()],
|
||
|
|
... 'It is .'.split(),
|
||
|
|
... smoothing_function=SmoothingFunction().method4,
|
||
|
|
... )*100
|
||
|
|
4.4267...
|