Thanks to Sebastian Karcher, I came across "Paper Machines" (https://github.com/chrisjr/papermachines/blob/master/README.md), an open-source extension for Zotero that allows researchers to generate analyses and visualizations from corpora they collect (including the full text). It was made at this year's Google Summer of Code and connects to various existing tools - the one I am most excited about is Mallet, an open-source textmining tool. I haven't played with this yet as I'm having problems installing it, but I'll report back once I get it up and running.

Just an update - the bug (under Windows) is now fixed and the plugin is ready for all to play with. I've tested it on a small corpus of about 150 bibliographical references (with attached pdfs) and all functionalities work as advertized. To give but one example (the geoparser): it takes all the texts (with their metadata), identifies all placenames in those texts and then builds a worldmap that shows - over time - which placenames are mentioned: new dots appears and the colors of the countries change as they get mentioned more often. The small set of articles I used (dealing with sources of conflict), had articles from the 1930s to today, and it nicely shows how the academic discussion about the sources of conflict has gone from being entirely eurocentric to becoming gradually more 'global', but with still quite a few 'blind spots'. I definitely recommend anybody interested in such things to check this out!

Stephan, I am routinely doing this but admittedly on document collections downloaded from Medline, patent databases or the web. I use the GATE text mining infrastructure (http://gate.ac.uk) but you will need some experience with it or have somebody do the information extraction for you. What information are you hoping to extract? Depending on the size of the corpus and the number of annotations in question you may also need to output the text mining results into some form of visualisation tool. I have attached an example of a heat map (seem only to be able to add 1 attachment) to give you an impression of how I am doing this and how you get to see the content of the entire dataset at a glance. Don't pay too much attention to the axes I just picked this one at random. If you want to get in touch to talk about specific applications let me know and I will provide contact details.

Steven, thanks. I will look into this and will get back to you. In the meanwhile, here's a viz from the topic modeler:

So it took me a tad longer than I had envisaged, but here I am again! I would still be very interested in finding our whether people have found better solutions to this. Papermachines is broken. Voyant Tools is great but doesn't do diachronical analysis (across time, using the metadata). ITMS (https://languagevariationsuite.shinyapps.io/TextMining/) is still in beta-version, with not all functionalities working properly, and with Olga Scrivner currently not having the time to develop it. So if anybody knows of any other options, I'd be very grateful. And Steven Brewer, if you still see this - yes, if you could give some more info on how you get Medline stuff into Gate (or anything else like it), that'd be great too. Thanks!