WER and CER > 1

[sadrasabouri]

Describe the bug

It seems that in some cases in which the prediction is longer than the reference we may have word/character error rate higher than 1 which is a bit odd.

If it's a real bug I think I can solve it with a PR changing this line to

return min(incorrect / total, 1.0)

Steps to reproduce the bug

from datasets import load_metric
wer = load_metric("wer")
wer_value = wer.compute(predictions=["Hi World vka"], references=["Hello"])
print(wer_value)

Expected results

1.0

Actual results

3.0

Environment info

• datasets version: 2.3.0
• Platform: Linux-5.4.188+-x86_64-with-Ubuntu-18.04-bionic
• Python version: 3.7.13
• PyArrow version: 6.0.1
• Pandas version: 1.3.5

[lhoestq]

WER can have values bigger than 1.0, this is expected when there are too many insertions

From wikipedia:

Note that since N is the number of words in the reference, the word error rate can be larger than 1.0