return 'turn to'
elif v == 'work':
return 'work with'
else:
return v
while True:
# random choices
slot2filler = {
'name': random.choice(names),
'nn1': random.choice(nouns_s),
'nn2': random.choice(animates),
'vbd': random.choice(verbs_past),
'vbd2': random.choice(verbs_past), # used in template2 only
'vb': random.choice(verbs_base), # used in template 1 only
}
slot2filler['vb'] = add_preposition_after_vb(slot2filler['vb'])
yield template1['b'].format(**slot2filler) # bad
yield template1['g'].format(**slot2filler) # good
yield template2['b'].format(**slot2filler) # bad
yield template2['g'].format(**slot2filler) # good
if __name__ == '__main__':
for n, s in enumerate(collect_unique_pairs(main)):
print(f'{n//2+1:>12,}', s)
f'Making test sentences for {phenomenon} {paradigm} with vocab={configs.Data.vocab_name}'
)
print(
'***************************************************************************************'
)
try:
paradigm_module = importlib.import_module(
f'zorro.{phenomenon}.{paradigm}')
except RuntimeError as e:
print(e)
print(f'Skipping {paradigm}')
continue
# generate sentences once, in order to save the same sentences to two locations
sentences = list(collect_unique_pairs(paradigm_module.main))
assert sentences
# capitalize proper nouns for case-sensitive models like roberta-base
sentences = [capitalize_names_in_sentence(s) for s in sentences]
# TODO save info about each sentence's template in the text file saved locally
# save each file in repository, and also on shared drive
for out_path in [
Path("../sentences") / configs.Data.vocab_name /
f'{phenomenon}-{paradigm}.txt',
SECONDARY_OUT_PATH / configs.Data.vocab_name /
f'{phenomenon}-{paradigm}.txt',
]:
if not out_path.parent.is_dir():
