ZeroDivisionError when computing bleu_score

[betterenvi]

Python version: 3.5
NLTK version: 3.2.5

  File "/usr/local/lib/python3.5/dist-packages/nltk/translate/bleu_score.py", line 89, in sentence_bleu
    emulate_multibleu)
  File "/usr/local/lib/python3.5/dist-packages/nltk/translate/bleu_score.py", line 199, in corpus_bleu
    hyp_len=hyp_len, emulate_multibleu=emulate_multibleu)
  File "/usr/local/lib/python3.5/dist-packages/nltk/translate/bleu_score.py", line 544, in method4
    incvnt = i+1 * self.k / math.log(hyp_len) # Note that this K is different from the K from NIST.
ZeroDivisionError: float division by zero

[alvations]

@betterenvi thank you for raising the issue. Could you provide the hypothesis and reference(s) input that caused the zero division error?

[truthless11]

@alvations ZeroDivisionError also happens to me, here's an example of inputs:

from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
smoothie = SmoothingFunction().method4
ref = ['a', 'quick', 'brown', 'fox', 'jumped', 'over', 'the', 'lazy', 'dog']
hyp = ['a']
sentence_bleu([ref], hyp, smoothing_function=smoothie)

[purificant]

@alvations ZeroDivisionError also happens to me, here's an example of inputs:

from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
smoothie = SmoothingFunction().method4
ref = ['a', 'quick', 'brown', 'fox', 'jumped', 'over', 'the', 'lazy', 'dog']
hyp = ['a']
sentence_bleu([ref], hyp, smoothing_function=smoothie)

len(hyp) needs to be > 1

It makes no sense to use method 4 smoothing ( see http://acl2014.org/acl2014/W14-33/pdf/W14-3346.pdf ) as it causes a 1 / ln(len(hyp)) operation, which is division by zero for len(hyp) = 1

Even if method 4 is patched to work around that and essentially do nothing, there are still issues with calculating BLUE due to ln(0) operations for the example input.

[tomaarsen]

I suspect this has been solved through #2839. Let us know if any normal uses of BLEU still throw unexpected exceptions!