In our early education day, while learning English one of the most important tasks was to fill the blanks.
Two types of question were provided:
- Cloze test with clues
- Cloze test without clues
It was totally difficult for non-native English learners to fill the blanks. And cloze test without cloze was the most tricky.
But nowadays the machine can fill the blanks smartly. Mask Language Modeling(MLM) based language model give machine such power.
FillBlank is a BERT based package, can fill cloze test without clues type text. This package mostly builds using huggingface transformers.
pip install fillblank
- pytorch 1.6+
keep in mind, one sentence can have not more than one .
from fillblank.fillblank import FillBlank
text = """Man is a rational being <blank> wisdom, intellect and sense of self-respect. He had immense <blank> in himself.
It keeps him aloof from all sorts of evil <blank>. To become an ideal man he should <blank> the feelings of others."""
filltheblank = FillBlank()
output, output_dictionary = filltheblank.fill(text)
print(output)
print(output_dictionary['predict_words'])
# man is a rational being <with> wisdom, intellect and sense of self - respect. he had immense <faith> in himself. it keeps him aloof from all sorts of evil <things>. to become # an ideal man he should <respect> the feelings of others.
# ['with', 'faith', 'things', 'respect']