Towards a Pattern Science for the Semantic Web

Paper Title: 
Towards a Pattern Science for the Semantic Web
Aldo Gangemi and Valentina Presutti
With the the web of data, the semantic web can be an empirical science. Two problems have to be dealt with. The knowledge soup problem is about semantic heterogeneity, and can be considered a difficult technical issue, which needs appropriate transformation and inferential pipelines that can help making sense of the different knowledge contexts. The knowledge boundary problem is at the core of empirical investigation over the semantic web: what are the meaningful units that constitute the research objects for the semantic web? This question touches many aspects of semantic web studies: data, schemata, representation and reasoning, interaction, linguistic grounding, etc.
Full PDF Version: 
Submission type: 
Responsible editor: 

Review 1 by Pascal Hitzler:
The authors emphasize the need for context-driven knowledge management for the Semantic Web. Starting from a linked open data perspective, they identify two conceptual difficulties in current approaches, called the "knowledge soup" and the "knowledge boundary" problem. These are discussed in historic context, and the authors then argue strongly that these are of central importance for the semantic web vision, and that, in order to address these, "knowledge patterns" should be studied as primary research objects.

I find the paper very inspiring. It aligns nicely with other articles in this special issue (Raubal, Kuhn, Hitzler et al., Polleres et al.).

Can you comment on scalability issues in the context of the approach you propose?

Perhaps you could also add a sentence, how frames connect to mirror neurons (page 4, left, bottom).

Minor requests:

- please do a careful proofreading

- improve the figures so that they are more readable

- something wrong with the author list of [IB82]

- the third author of [JHY+10] has surname Yeh, not Yehy

Review 2 by Ban Adams:
This paper takes a very interesting tack on the knowledge soup and boundary problems by proposing frame based knowledge patterns. My understanding is that use of frames as described in the paper is not necessarily to replace the description logic formalism for concepts as that is still one facade (fig. 5), but rather acts as a meta-metadata for organizing knowledge into patterns for different contexts. Also, I absolutely agree with the notion presented here that the process of ontology design has (lamentably) been relegated to the sidelines in the semantic web in favor of computational issues. In my opinion, the semantic web has suffered from theory driving application rather than vice versa. Overall, a very thought provoking paper and the following comments/questions come in no particular order:

The primary argument for frames as the "unit of meaning" for the semantic web seems to be from the perspective that frames are a more natural (i.e., easier) model for people to use, which, while possible, is not corroborated by any evidence in the paper and seems subjective. I'm not convinced that they are more expressive semantically other than that some slots seem to be filled with natural language text and others with DL and rules.

There are a few times in the paper when I am unsure what the intended agent is that is interpreting the knowledge represented in a frame. It certainly makes sense to me that the description of the concepts involved in ontology design need not be formal, since we are not asking artificial agents to design ontologies, yet, only infer meaning from them. As described, knowledge patterns seem more of use to human designers of ontologies. However, the authors say "Agents and reasoners must be able to recognize such frames and reason on them...", so to what degree are these frames for synthesis of ontologies vs. analysis of (i.e., reasoning over) ontologies? The only facades that seem possible for agent-based reasoning are "formal representation" and "inferential structure", which is already covered by OWL/SWRL but how is an agent to reason over the facades without formal representation, such as "use case", "vocabulary", etc. (without NLP)? Also, for example, on page 6: "To make sense of these data, formal semantics alone is not enough." Is that the human ontology designer / programmer that needs to makes sense of the data or an artificial agent? If the latter how does an agent make "sense" of data that is not formally defined?

INTRODUCTION -- I would present a clear definition of what is meant by "research object". There are some typos (didn't realize that the web of data was male ;-) in the introduction. What is meant by the canonical data sets (do you mean data models?) for geographical and biological data, musical, etc? In general, the introduction reads less well than the rest of the paper and I would suggest proofreading once more.

Since aligning to a specific upper ontology like DOLCE does not seem like a very feasible solution to the knowledge soup problem, the authors present the notion of representing knowledge contexts in the ontology. Context is, of course, everything but formalizing it can be tricky. A bit more on what the mechanism is for how the contexts that are listed (descriptive, informational, situational, etc.) practically help in terms of interpretation of meaning would be nice, though perhaps beyond the scope of the paper.

A few additional questions/comments:

In what sense are knowledge patterns like many other frame-based systems in that they make a closed world assumption (in contrast to description logics' open world assumption)? Are knowledge contexts ways of designing if not closed worlds, then "fenced worlds"? Perhaps these assumptions are something to revisit in the semantic web, since for many research objects there is an assumed universe of discourse.

Why is the inferential structure (which appears to be rules) a separate facade from the formal representation? Also, the definition of the data facade is unclear to me: "Data that can be used to populate an ontology whose schema is a formal representation for the KP". Can you clarify the wording: are you saying the data schema is defined in the formal representation facade?

A connection might be made to design patterns in programming languages given their popularity and success [e.g., cite Gamma, et al. 1995].

Section 2 could be trimmed down a bit.

More of a philosophical question, but at what point does a knowledge pattern become too restrictive in terms of constraining the kinds of research activities that can be performed with the research object?

At what point is the knowledge boundary defined in the knowledge base vs being defined by the application using the knowledge? E.g., I am not sure what the interaction structure has to do with the "research object". That seems like a different question for every application that uses the knowledge.

Maybe want to reference the Manchester syntax for OWL 2, which is designed to be frame-based: , and show what is different/same in your designation of a frame-based system.