'Mutation of AA-[NN]'

[abojchevski]

• Include it as a feature
• If not good enough add as post-processing rule

[juanmirocks]

@abojchevski didn't you try this already?

[abojchevski]

No.

I am not sure to what extent it actually happens. My observations that we are not recognizing such mentions was based on just what I saw during annotating which is not super reliable.

Therefore we have to first see how often it happen, and only then I can create a feature/pp rule and evaluate if it is worth adding it.

I propose:

1. Wait until we have all new iterations (without test-set of-course)
2. Split data into train/test (stratified)
3. Count how many FN are of the above mentioned type
4. Add feature and evaluate if it helps

[juanmirocks]

We will not try this before paper publication. Not a priority.