I am building an NLP model which uses MEMM in order to tag parts of speech.
My model uses history of two previous words and tags, and of the next word, alongside the current word and tag. I used those values and created basics features (f100-f107, capitals letters and digits).
I am trying to figure more advanced binary features. When I tested the model, it confuses mainly between
NN. How can I think about features that could help my model detecting those POS and not confuse and switch between them?