I want the extractor to be able to extract unseen entities in particular names of people. I have a general dataset where the name entity is annotated in each senstence but the intent isn’t (the intent is irrelevant). I can’t get repetable results from examples that have relevant intents. Unfortunately there are no pretrained extractors that support my language. I think, setting all those examples to an intent could worsen the intent prediction. Do you have any tips on how I could achieve a more general extractor?
Name entities are notoriously hard to extract. Best advice is to use a lookup-table with all or most common names for your application.