I am annotating speech and co-speech gesture and I have many cases where a gesture starts a bit before the end of an utterance and continues as the speaker utters the next sentence (or the other way around: it ends during the next utterance). This means that my gesture annotation overlaps with the annotation of two utterances. When I try to export the file, all the cells concerning the speech annotations that should go with the gesture are empty as the system does not know which annotations are relevant. I have been filling them in manually but as it is time consuming, I was wondering if there is a way to solve this problem.
Yes, in the tab-delimited text export with the option “Repeat values of annotations spanning other annotations” (assuming this is what you used), annotations only end up in the same row if they either have exactly the same begin and end time or if one annotation completely spans one or more other annotations.
In case you didn’t do so already, you could try the option “Sliced annotation output etc.” to have the overlaps on the same row. But this has other disadvantages, the actual/original start and end time and duration of annotations gets a bit lost.
Alternatively you could also have a look at the multiple file export as “Annotation Overlaps Information”; most of the exported columns are probably not relevant, but some might include the information you need.
Thank you for your answer and solutions! Unfortunately, with the “sliced annotation output” option, the annotations of the concerned gestures are duplicated, which is problematic as I have to count them.
When I click on export as “Annotation Overlaps Information” and try to load the domain, nothing happens…
Yes, one should be careful with counting with specific tab-delimited export settings. Not only with the sliced output, but also with the ‘separate column’ option and combined with ‘repeat’ options. But, going back to your problem description, if you don’t want ‘empty cells’ but still want to capture that e.g. a gesture overlaps two utterance annotations, I don’t immediately see how that could be done without duplicating annotation values.
If counting is the main purpose, the tab-delimited export without the ‘separate column’ is probably best (but then any relation with annotations on other tiers is more or less lost). If the combination of utterances and gestures is the main purpose, maybe the (multiple layer) search can be useful?
Concerning the Domain problem, this is not entirely clear to me. Did you create a domain but if you select it for the mentioned function it doesn’t load? Doesn’t the ‘Export annotation overlap information’ window appear? Or is the creation of a domain the problem?
Thank you for your answer. When I export my data, I create two separate files: one for speech and one for gestures in relation to speech. This way, I am able to count both without the repeated speech values.
Yes, I had created a domain but when I tried to load it, the “export annotation overlap window” disappeared and nothing happened. It seems to work now. It was probably a small bug.
