Depiction behaviour, i.e. isolating images issue

[mjy]

In discussion with @dhobern I realized our depiction model allows us to abandon images if the objects they are depicting are destroyed. This doesn't seem the best behaviour. Do we remove this auto-detachment and force users to handle the image?

[dhobern]

Would it be possible to have an option to propagate some text summarising the former linkage as a Note on the detached image, then keeping the image as an object currently not bound to the graph?

[hhopkins77]

In what ways are such images possibly of use even if no longer attached to an object? I think dhobern's comment allows for such possibility.

[mjy]

? I think dhobern's comment allows for such possibility.

I like the concept, I worry about the minimum information required to remain useful.

• For object related depictions to be destroy maybe images must have one of:
  • depictions to alternate objects
  • an Attribution
  • <some annotation> with <what content>

[dhobern]

I realise from our discussions that what I have been doing (deleting all the names in the TW Global Lepidoptera Index dataset that represent hybrids or aberrations rather than code-compliant names) has left many cards orphaned. I've not been concerned, because in the context of GLI, these are noise. I had not understood that you (and perhaps Ian) had considered this the cleanest view of all the data from the NHM card catalogue. In this context, the cards have been for me something like scratched notes to help me understand what each name was intended to represent.

I'm happy to support whatever seems the best TW-compliant path forward, but the workflow for handling these aberrations is already so lengthy that I don't want to have to add manual steps for something else. I'd be happy if we can batch annotate them with what you need, but the pain involved in cleaning out 50 names that are all non-code-compliant "synonyms" for a species is already extreme. If I had to annotate each card one at a time, it would multiple the time burden even more.

[mjy]

I think the LI status/process is OK.

1. We can retrieve an original dataset, pretty much unaltered (or completely, have to check the month lag in the rotating backup).
2. We have all the images, that would still link to the names.

This means:
3) We can ultimately have Donald batch-annotate all images with no depictions with metdata along the lines of "Non compliant name images"
3a) If we OCR all images we could full-text search them as a way to get to them for unanticipated cross-reference needs.
4) We should batch annotate all images in LI with Attribution before we get to much further, in case we decide to use an alternate strategy.

I'm leaning in general to require an Attribution, so no auto-nuking without at least having that. As a class of annotations auto-nuke should be allowed in principle, the object is gone, so should its metadata.

[hhopkins77]

I was thinking could an image be of historic or other importance? Could an image be unique in some form or fashion? Could an image be of handwritten labels that help collections managers identify a collector? Could there be value to an image outside of its connection to an object? (I'm also thinking of our mandate in TW that we don't delete things, we only add information.)