Allow passing an outcome converter to TcAnnotator

[reckart]

Currently, to use TcAnnotator in a production environment, one has to wrap it in another UIMA component which delegates to the TcAnnotator in its process method and after running it picks up the TextClassificationOutcome annotations, creates the actually desired outcome annotations and then removes the outcomes.

In order to avoid the need for creating a wrapper UIMA component all the time, it might be a nice idea to pass the name of a converter class to TcAnnotator which would be called with the JCas as an argument after TcAnnotator completes its process method - thus the conversion can happen within the scope of the TcAnnotator.

[Horsmann]

@reckart One could also pass the name of the annotation directly with the field-name? I am not sure how the API looks for this but this would probably most straight forward? It was possible to reference the field of an annotation by a string-name, right?

[reckart]

I think the conversion process of outcome to actual annotation may be more complicated than the case that simply creating an instance of AnnotationFS and setting a field on it:

• the annotation you want to add the outcome too might already exist from a previous annotation step - e.g. if you have a TcAnnotator which does performs entity linking on NamedEntity annotations that have been created by say StanfordNamedEntityRecognizer.
• there may be multiple fields to set - obvious cases could be the label and the confidence score; but the label itself might be split up into multiple fields, e.g. in case of filling in the various features of MorphologicalFeatures
• the label may require post-processing - e.g. in the case of BIO-encoded labels which a converter could merge

[Horsmann]

Ok. I uploaded a change. There is a test case in the example project in the package org.dkpro.tc.tcAnnotator that demonstrates / test using such a conversion annotator.

[Horsmann]

@reckart Is the current implementation what you had in mind?