I've been reading Seb Paquet's thesis, now that something in the new Adobe Reader made the text visible, and one section is particularly interesting (though it's all good). If I may quote:
In many cases it is quite possible that similar problems exist in other areas, that they have been worked on, and that the relevant material is only keystrokes away, thanks to the internet; but it is equally probable that these problems have been described using a different vocabulary, making it very hard for the researcher to find.
A Socio-Technological Approach to Sharing Knowledge Across Disciplines (Sebastien Paquet, 2003)
The ontological variation between different, but related, areas of work is obvious when researching a new area using PubMed. From an initial article, the Related Articles link brings up a list of articles that share a high frequency of the same keywords. Repeating with the most relevant of these eventually produces just about all the articles that could be said to cluster around a particular topic (which you can easily visualise with TouchGraph).
This is fine, but then you try a different initial search (still looking for more papers on the same subject) and find a new bunch of articles that are all related to each and cluster around a particular topic, but never showed up in the first cluster, as they used different keywords.
There are obviously some keywords, or semantics, or something, shared between the two clusters, but these must be shared with many hundreds of other papers. The problem is how to recognise that the two clusters are related. The NCBI's MESH headings attempt to categorise each paper into a number of hierarchical categories, but they don't really cover the vague, intercategory topics that are the most interesting for research. Maybe it's time to put together a user-defined Navigable Sythesis Ontology for MEDLINE.
The closest thing available so far is Make-A-List in HubMed: I can make a list containing all the papers I found on the topic described above and submit it to biologging, which will send a Trackback to all the papers included in the list. Then anyone looking at one of the papers will be able to look at my list and easily find all the others. I could even give the list a name, to define that topic area, then link the list to a wiki page for that topic. That would be good.