"Other" Class in Stanford NLP Classifier for lines that are not related to ANY of the Trained Classes

Question

I'm using the Stanford NLP just fine. I made a train file with all my classes. and it identifies the test lines just fine. BUT what if I have an Other line (that is not of any of the Classes I've trained it). Can I ask the algorithm to return null etc. when the line is not recognized with any of the Classes? If not, what/How do you recommend that I should create an "Other" Class with "other" lines.. but that could be infinite. Thanks, Aryeh.

Answer 1

If I understand your question correctly, yes, you should create an "Other" / O class to capture all "null" labels. This is the standard in e.g. NER, where the majority of the tokens in the corpus receive an O label (indicating no named entity label).