(From Colin Batchelor, Royal Society of Chemistry)
The ease of setting up semantic annotation really depends on how the production process works for the journal.
It was relatively straightforward for us because:
- We have a workflow based on XML; many publishers only convert their articles into XML at the end of the process before it goes to print or on the web. This means that, regardless of when we do the annotation, our technical editors are familiar with the software anyway.
- We edit in-house. If a journal has its copy-editing done by freelancers (usually in Word) then they'll need to work out who's going to do the annotation and with what. If a journal relies on authors to put their manuscripts into a fit state for publication then it would be interesting to see how authors react to being asked to annotate their text.
- We were going to be annotating journal articles with chemical structures anyway, so the GO annotation piggybacked straightforwardly onto that. The software for the chemical annotation is Oscar, which is open-source and can be downloaded from http://oscar3-chem.sourceforge.net/. It doesn't support GO terms out of the box, but the setup required is pretty simple and I'll be giving a paper about it at the ACL meeting in Prague later this month. Dietrich Rebholz-Schuhmann has a system for general biomedical mining, but to the best of my knowledge nobody has integrated this into a journals workflow.
There is a time overhead in the annotation, but this varies enormously from paper to paper, up to an afternoon for an enormous review. This includes chemical structure checking as well, so it's a bit difficult to tell what the exact workload is.
If other publishers are interested, we may well offer semantic annotation as a service in the near future. It would probably involve human curation, so it's likely to be something we would charge for.
As for dissemination of the annotations as HTML and RSS, these are extensions of the systems we use for putting up unannotated articles on the web, so the code isn't really transferable to other systems, but it's not a hard thing to do in principle.
From what we've seen of other publishers' workflows, the service provider model might be the easiest thing to implement.