Robustness on short food phrases

[g-simmons]

Hi @chambliss, thanks for this cool project. Not sure what your plans are for maintenance, but just wanted to put this experience out there.

I have a use case where it would be nice to identify food names in short phrases. It looks like the model struggles with food names in isolation (I consistently encounter it identifying only the latter half or two-thirds of single-token food names when they appear in short blurbs rather than full sentences).

As a hack, prepending a "some" before the short food phrase seems to help.