Closing the Loop between Knowledge Patterns in Cognition and the Semantic Web

Tracking #: 2207-3420

Authors: 
Aldo Gangemi

Responsible editor: 
Guest Editor 10-years SWJ

Submission type: 
Other
Abstract: 
This article discusses currently open issues in the discovery and representation of knowledge patterns in computational processing of meaning. Starting from a formal semantics of knowledge patterns as discussed in cognitive science and knowledge representation, some measures are suggested to improve interoperability and validity of web-based semantics.
Full PDF Version: 
Tags: 
Reviewed

Decision/Status: 
Minor Revision

Solicited Reviews:
Click to Expand/Collapse
Review #1
By Frank van Harmelen submitted on 15/Jul/2019
Suggestion:
Major Revision
Review Comment:

This paper sets itself an ambitious goal, namely to review the current
state of play for knowledge patterns (KP's) and to identify the new
directions that this work should take.

The paper is full of interesting analyses and observations, but also
full of sentences for which I have no idea what they might mean, and
an entire section (section 2) that I can not make head or tail of.

After reading the paper multiple times, I feel each time that I have
read a bunch of interesting stuff, that I was totally confused in
other places, but each time I fail in any attempt to summarise to
myself the main content or message of the paper.

Also, the paper is stronger in its analysis of why things haven't
worked out until now than in its proposal on how to fix things. As
mentioned, I find the formal proposal in section 2 rather
impenetrable, and oerall it is not so clear why any newly proposed
approach to KP's in the paper would make such a radical change. In
particular: the discussion about compositionality is philosophically
well-informed, but gives no concrete technical/computational
proposal.

All this is the global justification for my "major revision"
suggestion: there's good stuff here, but it needs to be clarified and
sorted out.

More specifically:

Section 1 starts with an interesting analysis including some
historical context. No further comments.

Section 2 is the most problematic of the entire paper. Despite
multiple attempts, I can still not make sense of the formalisation.
(there are also mysterious sentences such as "The orderly view of
relations corresponds to the cognitively reified view of knowledge
patterns, and vice versa". I have no idea what this sentence could
possibly mean.
- I fail to understand what "an n-ary predicate with multiple arity"
is.
- I would really need an example earlier on, and when I get the
example on p4 I cannot parse it, let alone understand.
- I do not succeed in interpreting the axiom schema's in (1)-(9) (at
least, I assume they are axiom schema's). For example:
- is any \phi a KP? Surely not?
- (6) imposes an antencedent condition on \i, but \i not longer
appears in the consequent
- etc
- my problem with the example is that I don't understand if this is an
instantiated KP or a generic KP ("pattern", after all). It seems to
be the former, but then it doesn't teach me how a generic knowledge
*pattern* would look like.
As is clear from the above, I'm entirely in the dark on what section 2
tries to do and how it tries to do it. (could be just me, of course..)

Fortunately, coherence is restored in section 3 with an interesting
wide-ranging analysis of the notion of knowledge patterns. I'm
actually rather surprised that this section doesn't immediately follow
on from section 1? Some minor comments here, but see my comments in
the PDF at
https://www.dropbox.com/s/e874gadx2wmvax8/swj2207.pdf?dl=0
My only concern in this section is the claim that these insights "are
backed by neurological results". I don't think the quoted neurological
results back anything (such as the specific/generic trade-off of
KP's). Aren't the neurological results simply inconclusive?

Section 4 is again interesting, although I am left with the question
why predicates like Over(o1,o2) or Contact(o1,o2) are _frames_. Are
they not simply single predicates? So this begs the question when a
relation is either an atomic predicate or a frame / patterns. Some
discussion on this would be welcome.

Secondly, the "5 classes of asymmetry" appear rather unmotivated. Why
these 5? Why only these 5? More motivation for this would be welcome.

Section 5 concludes, and leaves me, as stated, wondering why we should
now be more hopeful of any future success of KP's establishing
distributed computational semantic interoperability then before.

Finally, in the annotated PDF at
https://www.dropbox.com/s/e874gadx2wmvax8/swj2207.pdf?dl=0
I have underlined all the sentences and phrases of which I honestly cannot make
sense. I would suggest for each of them to either clarify them (in
case they are really needed) or remove them otherwise.

Small details:
- there is a missing reference on p2.
- the phrase "such as http:///www.ontologydesignpatterns.org" suggests
there are other repositories. Are there?

Review #2
Anonymous submitted on 15/Aug/2019
Suggestion:
Minor Revision
Review Comment:

This "blended" essay-position style paper aims to clarify the rationale, brief history, and research issues related to knowledge patterns in cognition and the context of semantic web. The author did a great job covering the scope from philosophical, logical, web, and cognitive aspects around knowledge patterns. It would be most helpful if the author can succinctly describe a desirable state of affairs as an overarching vision or goal at the beginning - a high-level specification of the requirements for the fields to become.

A few suggestions for improvement:

* Citations. A few additional prior work should be cited. Wrt multigrade predicate, "Worlds as a Unifying Element of Knowledge Representation" by
J. R. Scally, Nicholas L. Cassimatis, and Hiroyuki Uchida proposed a similar syntactic style of logical expression. Since there has not been mathematical formal semantics established in either approaches, the fundamental distinction, if any, is not apparent (perhaps one is more general than the other). Wrt categorical/mathematical approach to modeling situations and conceptual blending, the works of Jon Barwise (situation) and Joseph Goguen (blending) should be mentioned. Mention of FAIR and deep learning should either be omitted or expanded a bit with references.

* Style. Very long sentences could be broken into several shorter ones for clarity, readability (and "compositionally"). Single sentence paragraphs should be avoided.

* Typesetting
Page 3: \ [2] add space before [
Page 3: Right column - should be single line spacing
Page 5: Fix .". - one . only
Page 8: bottom left - reference to "next section" - there is no next section
Page 9: reference [23] - page numbers gabbled