Thanks for taking an interest in our work (SAOR). I would also like to point you to the ‘final’ extended version of our SAOR paper which may be of interest. The message is very much the same, but we offer more careful treatment of the subject as well as an alternative implementation approach for more complex rules, etc.

Indeed, we have been in discussion with Renaud about combining reasoning approaches with the Sindice group. However, it is not clear how best to approach this. Our “one conservative T-Box model applies to everyone” approach stems from the underlying SWSE architecture of batch processing techniques, sorts, scans, etc. Renaud’s approach on “one model per document” stems very much from their document centric view of the Web. We are more restrictive in that we disallow all non-authoritative inferences and only have a global scope. However, Renaud’s approach misses some inferences from not considering the merge of all documents (one may question whether such inferences are, in any case, a good idea or not).

We could perhaps incorporate some “non-authoritative” inferences within a local context for SAOR, but the cost/benefit ratio is unclear. The main question here is how much web publishers abide by the understood semantics of published T-Box data and how much people need to “customise” the semantics of popular classes and properties for their local documents.

One other important issue we have yet to properly address is that of equality reasoning and A-Box-level authority. The question is how people should contribute on an assertional level. Should it be completely open? Can anyone contribute to my foaf:Person description, or should I be allowed import contributions through use of :sameAs from my authoritative document? For example, naïve equality reasoning on IFPs is very, very messy.

Sorry it took me a bit to read the paper. You are correct; this is a very apropos approach that should have been included in my initial roundup. The contextual reasoning approach is different but quite applicable. I look forward to what the Sindice and SAOR teams migh suggest as a combined approach.

There is another reasoning methodology for Linked Data that could be of interest. Within the sindice project, we apply a context-dependent reasoning approach for linked data [1]. The methodology is different than the SAOR approach and is based on contextual reasoning (Guha), but the goal is similar: avoid “non-authoritative” inference and scalable web reasoning. We are currently discussing with Axel Polleres and Aidan Hogan how to combine this approach with SOAR.

[1] http://axel.deri.ie/~axepol/publications/delb-etal-2008.pdf