I'm an outsider to the SemWeb community. A lot of the enthusiasm around the SemWeb reminds me of the AI hullabaloo of the 1980s. Moreover, much of the technology seems little different from the knowledge representation work invented in that era.
Over the past 20 years, AI researchers have come to appreciate the limitations of traditional knowledge representation techniques. It seems that statistical methods and machine learning have proven more productive than reasoning based on ontologies. ThereĆs even some evidence that people perform common recognition based on simple feature matching, rather than using classification-based reasoning.
In light of this, it's interesting that major web applications like Google and Flickr are increasingly relying on statistical and clustering methods, while "big S" SemWeb technologies are encountering resistance from the folksonomy communities with claims that RDF and OWL are too complex and "unnatural".
In my talk, I'll expand on some of these ideas and highlight a few challenges for Semantic Webheads.