A scholarly paper is not a PDF. A PDF is merely one view of a scholarly paper. To push ‘beyond the PDF’, we need design patterns that allow us to segregate the user interface of the paper (whether it is displayed as an aggregation of triples, a list of assertions, a PDF, an ePub, HTML, …) from the thing itself.
I think a lot of the disagreement about the role of the PDF can be put down to trying to overload its function: to try to imbue it with the qualities of both ‘model’ and ‘view’. … One of the things that software architects (and I suspect designers in general) have learned over the years is that if you try to give something functions that it shouldn’t have, you end up with a mess; if you can separate out the concerns, you get a much more elegant and robust solution.
My personal take on this is that we should keep these things very separate, and that if we do this, then many of the problems we’ve been discussing become more clearly defined (and I hope, many of the apparent contradictions, resolved).
So… a PDF (or come to that, an e-book version or a html page) is merely a *view* of an article. The article itself (the ‘model’) is a completely different (and perhaps more abstract) thing. Views can be tailored for a particular purpose, whether that’s for machine processing, human reading, human browsing, etc etc.
[paragraph break inserted]
The relationship between the views and their underlying model is managed by the concept of a ‘controller’. For example, if we represent an article’s model in XML or RDF (its text, illustrations, association nanopublications, annotations and whatever else we like), then that model can be transformed in to any number of views. In the case of converting XML into human-readable XHTML, there are many stable and mature technologies (XSLT etc). In the case of doing the same with PDF, the traditional controller is something that generates PDFs.
[paragraph break inserted]
The thing that’s been (somewhat) lacking so far is the two-way communication between view and model (via controller) that’s necessary to prevent the views from ossifying and becoming out of date (i.e. there’s no easy way to see that comments have been added to the HTML version of an article’s view if you happen to be reading the PDF version, so the view here can rapidly diverge from its underlying model).
[paragraph break inserted, link added]
Our Utopia software is an attempt to provide this two-way controller for PDFs. I believe that once you have this bidirectional relationship between view and model, then the actual detailed affordances of the individual views (i.e. what can a PDF do well / badly, what can HTML do well / badly) become less important. They are all merely means to channeling the content of an article to its destination (whether that’s human or machine).
The good thing about having this ‘model view controller’ take on the problem is that only the model needs to be pinned down completely …
Perhaps separating out our concerns in this way — that is, treating the PDF as one possible representation of an article — might help focus our criticisms of the current state of affairs? I fear at the moment we are conflating the issues to some degree.
I’m particularly interested in hearing if this perspective, using the MVC model, makes sense to others.