Some NER tools such as the CoreNlpNamedEntityRecognizer mark every token individually as a NE instead of creating a multi-token NE.

IMHO the default behavior should be that NEs with the same label are joined unless the model uses a BIO-like encoding in which case the BIO markers should be respected.

Also the unit tests for the NER tools should be changed to include a multi-word NE, e.g. change John from the current unit tests into John Smith.

• CoGrOO Named Entity Recognizer
• CoreNLP Named Entity Recogizer (old API)
• CoreNLP Named Entity Recognizer
• Illinois CCG Named Entity Recognizer
• LingPipe Named Entity Recognizer
• NLP4J Named Entity Recognizer
• OpenNLP Named Entity Recognizer