Hi Jan,
We’re working on a new experimental stack for parsing Word or PDF documents into fully-structured JATS XML; you can try out a standalone version here http://pkp-udev.lib.sfu.ca/, look at the code here https://github.com/pkp/xmlps, obtain an (early) plugin for integrating that stack in OJS here GitHub - pkp/ojs-markup: A Public Knowledge Project Open Journal Systems (OJS) plugin for converting various document types to xml, pdf and html, or, if you want to just dump out a JATS <front> stub (i.e. just article metadata) from the articles in an OJS site, I have some code that can be easily rigged up to do that here GitHub - axfelix/metadump: dumps out OJS article metadata from DB and puts it all in JATS stubs.
Hope that helps! All of our new XML functionality is under active development.