The CEDAR project produced impressive results in a short span of time. Even considering that its promoters had a good idea of what has to be done, results are here and question the state-of-the-art.
I was surprised by the way the project is separated into a language and experimental track, because in my opinion, quite some good experimental work has been performed in the language track. I am more familiar with the work on "compiling" semantic web technologies into order-sorted feature structures (OSF) that has been performed in the language track, so I will mostly comment on it.
From the results, it seems clear that the implemented reasoner shows tremendous performances on non toy ontologies with respect to state-of-the-art reasoners. Meanwhile, there remain questions to be precisely and definitely answered: what are the characteristics of the language and those of the solved problem on which it is so good? This would help putting such results in perspective and characterise the very contribution of this work. Moreover, there seems to be some room to extend it, both on the encoding side and on the language side, e.g., encoding disjointWith in ontologies. On the problem side, it would be useful to evaluate the gain beyond the Jena store and try with more optimised triple stores because there is a possibility that the gain provided by the CEDAR Reasoner is erased by optimisations of such stores.
Now, this work cannot stay the way it is: it has to gain more visibility. I see two particular actions to take in order to help this.
Implement standard APIs. This can be used for evaluating on "equal" grounds. But this can also be a way to promote the CEDAR Reasoner as an actual tool. If it implemented the classical owl:Reaseoner interface, I could try it (and not only me), in my implementation instead of HermiT in the same painless way that I used HermiT instead of Pellet.
This API aspect is also useful for places where there is no standard API. Proposing one is something worthwhile. I can think about the hierarchy encoding that has been used. Hassan Aït-Kaci mentioned the eventuallity to use a recursive modular encoding and Peter van Roy suggested giving a try to BDDs (for which there readily exist usable Java packages). Designing such an abstract interface could be useful for experimenting by yourself the various alternatives, it could also help you promote the encodings which have been designed in the project. Another area in which I think that could be worth doing is that the presented work can be packaged as an ontology compiler. Such a tool may be useful within your competitor tools even if they still use tableau methods.
The other avenue that I see is to publish about the techniques that have been used. Given that the web ontology languages are defined, it is very important to characterise the flavour of these languages that is covered. This remark applies to the ontology language as well as the query language (SPARQL is not just Datalog conjunctive queries, so comparing the later to the former is prone to lead to mistakes: the SPARQL predicate may be a variable, which, as far as I know is not covered by OSF). It is important to characterise sharply which fragments are covered. There is no problem to cover a fragment: it may be sufficient for large classes of applications.
Finally, talking about "niches" it seems to me that there is a place where this work could have an easier impact: querying SPARQL modulo ontologies. This is typically what is demonstrated by the prototype and typically what neither OWL reasoners, nor SPARQL query engines are very good at. Yet, this is a tremendous need for the community. Moreover, there is a hiatus between description logics with their object-orientation and SPARQL with its relationnal orientation. OSF terms, given their origin, do not suffer from this: an OSF term is already a restricted SPARQL query (and subsumption test, query entailment, etc.). So I think that targetting that particular area should be a good way to show-case the work done.
Together with colleagues in the Tyrex team, we worked on query containment in this context. That would be interesting to compare the OSF approach to what we did [Chekol 2012] and to understand the benefits.
Melisachew Wudage Chekol, Static analysis of semantic web queries, Thèse d'informatique, Université de Grenoble, Grenoble (FR), December 2012 ftp://ftp.inrialpes.fr/pub/exmo/thesis/thesis-chekol.pdf