You can see that the tagger basically works on this specific case by looking at the distributional similarity of the tagged tokens, as well: https://sense2vec.

You can see that the tagger basically works on this specific case by looking at the distributional similarity of the tagged tokens, as well: https://sense2vec.spacy.io/?Jordan%7CGPE https://sense2vec.spacy.io/?Jordan%7CGPE https://sense2vec.spacy.io/?Jordan%7CPERSON https://sense2vec.spacy.io/?Jordan%7CPERSON (This specific word works well because there are plenty of examples of both entities in the training data, which has a lot of text from the 1980s Wall Street Journal. Other examples are much worse. I have high hopes that distributional similarity can help surface these errors, and fix them.)