I don't think that the OpenNLP named entity recognizer itself is able to generate more than one label. A solution for you would be to train separate classifiers, one for "person" and one for "director" and add two OpenNlpNamedEntityRecognizer components to the pipeline, each using of of your trained models.
See also: https://github.com/dkpro/dkpro-core/issues/901 Cheers, -- Richard > On 11.07.2016, at 16:53, John Bower <j...@zode64.com> wrote: > > I'm currently using OpenNLP with UIMA to label words in a sentence. It's > important that a single word can be labelled more than once. For example David > Cronenberg should be labelled as director and person. > > I know the training process is implemented correctly because I have a > custom model file and when all sentences with one of the labels is removed > from the model file the other label is detected. > > I would preferably be able to continue to use OpenNLP to double label > words. Is there a way to do this? If not is this possible with another > library such as Stanford CoreNLP. > > The code that gets the labels is below: > > List<NamedEntity> entities = JCasUtil.selectCovered( > NamedEntity.class, aConstituent ); > if ( !entities.isEmpty() ) { > // is never more than 1 > } > > And some sample training data is below (there are hundreds of lines similar > to this.) > > <START:person> David Cronenberg <END> directed <START:film> Crash > <END> .<START:director> David Cronenberg <END> directed <START:film> > Scanners <END> .
