You can call external annotation tools directly from within CorA as long as they’ve been set up accordingly by an administrator. This feature can be used, for example, to pre-annotate a text and then correct these annotations manually. If the annotation tool supports it, you can also retrain it with data you’ve already annotated in CorA in order to improve the results.
To call an external annotator, click on “Automatic annotation” in the toolbar of the “Edit” tab while you’ve opened a document in CorA. A dialog window will open where you can choose from the available annotators.
Using this feature requires that an external annotator is set up by an administrator first. If you cannot see an annotation tool in the list that you expect to be there, please consult a local administrator. Admins should make sure that annotators are actually linked to a tagset that the document uses.
Performing an automatic annotation
To perform an automatic annotation, open the automatic annotation dialog, then select one of the available annotators from the list and click “Annotate”. While the annotation process is running, the interface will be blocked and a waiting screen will be shown. The total duration of this process depends on the annotation tool that is being used.
Calling an automatic annotator will overwrite existing annotations and currently cannot be undone. You can, however, control what will get overwritten. Read on for more details.
Automatic annotators will always respect the green progress bar in the editor: tokens in the range of the green progress bar will not be affected by the automatic annotation, while all tokens with a gray progress bar will have their annotations overwritten by the external tool.
Note that this only affects the annotations of the type the external annotator uses: a pure part-of-speech annotator, for example, will not overwrite annotations in the normalization column.
Retraining an annotator
To retrain an automatic annotator on data in CorA, click the yellow “Retrain” button in the automatic annotation dialog. If the button is not available (i.e. grayed out), the selected annotator cannot be retrained.
Retraining will cause the annotator’s training functionality to be called, with data from CorA as its input. Typically, this data will not only be taken from the currently opened document, but from all documents within the same project. The green progress bar is again used as the criterion for this: only text passages marked in green will be added to the training data. This way, you can make sure that only manually verified annotations will be used.
If you are using a retrainable annotator for the first time, it is likely that you need to train it once before you can use it for annotation. If you get an error message during annotation with a retrainable annotator, try performing the retraining first.