Currenlty, HMEG uses mlconjug3 in order to get conjugations of words.
The library contains pre-trained conjugation models that maps input word with one of the predefined templates, that guide conjugation. Although this is a nice approach that can be used even for the new (previously unseen) words, it makes mistakes, as any classifier would do facing 171 imbalanced classes (for the English conjugations case).
Apart from occasional mistakes some other reasons to try replacing the mlconjug3:
For the HMEG purposes the mlconjug3 seems like an overkill. The vocabularies are fixed and are much smaller and their scope is ok to make conjugations manually.
(minor) mlconjug3 has scikit-learn as a dependency, which is quite heavy.
Simpler candidate approach(es):
Conjugate verbs manually by:
first checking whether verb is regular or not (eg using this)
get conjugation based on the outcomes of the check.
Currenlty, HMEG uses
mlconjug3in order to get conjugations of words.The library contains pre-trained conjugation models that maps input word with one of the predefined templates, that guide conjugation. Although this is a nice approach that can be used even for the new (previously unseen) words, it makes mistakes, as any classifier would do facing 171 imbalanced classes (for the English conjugations case).
Apart from occasional mistakes some other reasons to try replacing the
mlconjug3:mlconjug3seems like an overkill. The vocabularies are fixed and are much smaller and their scope is ok to make conjugations manually.mlconjug3hasscikit-learnas a dependency, which is quite heavy.Simpler candidate approach(es):