Available from the web science workshop report.
If the SW is, or depends on, the traditional AI knowledge representation task, then there is no particular reason to expect progress in this new form of presentation as all the problems and challenges of logic reappear and it will be no more successful outside the narrow scientific domains where KR seems to work and the formal ontology movement has brought some benefits.
This is true. Semantic Web practice is different from the traditional AI knowledge representation task, which still have many problems after so many years of research. The most important difference, as I believe, is that traditional AI practices are more or less limited on a small group of people because of its methodology. Semantic Web practices, however, can employ the huge man power through collaborative web applications (as what Web 2.0 is practising now). It is this difference that may determine the varied fate of these two practices. We must try to use the cheap and powerful human resources (regular web users) to realize the Semantic Web. Otherwise, just as what we quoted, the current status of traditional AI practices will be the future of Semantic Web practices.
Alternatively, if the SW is the WWW with its constitute documents annotated so as to yield their content or meaning structure, then NLP/IE will be central as the procedural bridge from texts to KR.
NLP/IE has already been critical technology. It is possible that NLP/IE be not only the procedural bridge but also the universal identifier. As a fundamental question: why must URIs be identifiers? Can procedures be identifiers?
In the discussion, Berners-Lee argued that the SW rests not on NLP but logic. Logic and ontologies will suffice to extract much of the value from the data held in structured relational databases.
NLP-based logic or logic-based NLP, which one is more preferable? Or even Logic NLP, if we simply put them side-by-side.
Wilks responded that the unstructured (legacy) part of the Web needs NLP for annotation, even in other media, at least until visual recognition systems become more reliable.
NLP again. NLP is important and valuable. But have we over-emphasized it?