merge-all for relations from/to items with stacked annotations during curation?

[Markus Egg]

Hi everyone,

we are in the process of curating an annotated corpus, in which the version of one annotator is supposed to be the decisive one. However, the process of automatically accepting the annotations of this version systematically does not include relations that start or end at items with stacked annotations (see screenshots), and the Merge-all command does not work, returning only “Annotations skipped due to conflicts: [Number]”. I wonder whether there is a way to merge these relations automatically for the curated version of the annotation.

Best, Markus

[Richard Eckart de Castilho]

Hi Markus,

> On 24. Jun 2026, at 10:07, Markus Egg <kama...@gmail.com> wrote:
>
> we are in the process of curating an annotated corpus, in which the version of one annotator is supposed to be the decisive one. However, the process of automatically accepting the annotations of this version systematically does not include relations that start or end at items with stacked annotations (see screenshots), and the Merge-all command does not work, returning only “Annotations skipped due to conflicts: [Number]”. I wonder whether there is a way to merge these relations automatically for the curated version of the annotation.

So when merging the relation is on the table, the current rationale here is:

* There are multiple span annotations in the curation document that could serve as the source endpoint of the relation.
* Apparently, the user is not sure what the label of the source endpoint span should be
* Since the user cannot decide which of the labels is the correct one, I do not know which of them to choose as the relation source.
* So, I skip this relation.

Under this rationale, relations that have an endpoint with other annotations from the same layer stacked on it cannot be merged, neither automatically nor manually.

You may ask why this rationale.

Consider the case that ...

user A has annotated: [A] -<rel>-> [B]
user B has annotated: [C] -<rel>-> [B]

with [A] and [C] being at the same location.

Now the curator wants to say that C is the correct span label and merges it.
Then the curator merges the relation.

If we would consider the endpoint labels during the relation merging, we would not be able to merge it, right?

It might be possible to come up with a better set of merging rules though.

Cheers,

-- Richard