Re: Tuple Store, Artificial Science, Cognitive Science and RDF (Re: What is a Knowledge Graph? CORRECTION)

OK, I think this discussion is going beyond what is appropriate for this forum, so if you want to continue, lets take this off-list, OK?

> On Jun 26, 2019, at 9:50 AM, Dave Raggett <dsr@w3.org> wrote:
> 
> 
> 
>> On 26 Jun 2019, at 16:24, Patrick J Hayes <phayes@ihmc.us <mailto:phayes@ihmc.us>> wrote:
>> 
>> A quick remark:
>> 
>>> On Jun 26, 2019, at 8:03 AM, Dave Raggett <dsr@w3.org <mailto:dsr@w3.org>> wrote:
>>> 
>>> I very much agree and have been arguing for a blend of symbolic and statistical techniques using insights from decades of work in Cognitive Psychology.  Rational belief is about what can be justified given prior knowledge and past experience. 
>> 
>> So far in this thread we have been talking about knowledge representation notations. You are here talking about mechanisms, not quite the same topic. I entirely agree about the need to put together symbolic and statistical, but I don’t see any reason why the use of the statistical would change the nature or the semantics of the symbolic. (Do you?) 
> 
> Good question. Statistical approaches alter the nature of reasoning, and this will influence the semantics.

I don’t see why it would alter it. Take a simple example, say a statement like ‘Capital cities are financial centers’, which might get rendered as something like (forall (x)(if (CapitalCity x) (FInancialCenter x))). Statistical tests of this assertion might inolve counting the numbers of examples of CapitalCity and working out percentages and so on. They might cause one to adjust confidence levels for such a claim, or even to modify it in some way, perhaps by inventing a new category of FinancialCapitalCity. But none of this kind of arithmetic would alter the /meaning/ of “CapitalCity” or “FinancialCenter”. In what way would you expect that /semantics/ would change because statistical methods were applied?

>>> This is not infallible, but nonetheless very useful in practice. It can support higher order reasoning, something that is essential for modelling human reasoning. 
>> 
>> What kind of higher-order reasoning are you referring to here? The term ‘higher-order’ has various meanings. If you simply mean that the logic can mention, describe and quantify over properties and relationships as first-class entities, then I would agree; but versions of FOL, even RDF, can do that. 
> 
> You will need to explain further.

If you want more details of how an essentially FO language can describe relations, an old paper gives the basic idea: https://pdfs.semanticscholar.org/d061/e6667716fec03325e586fe3020134d45a058.pdf?_ga=2.85123801.1565722984.1561617498-676702878.1561363073 <https://pdfs.semanticscholar.org/d061/e6667716fec03325e586fe3020134d45a058.pdf?_ga=2.85123801.1565722984.1561617498-676702878.1561363073>

You might also find this useful, and the links in there to other expositions:
http://www.jfsowa.com/talks/clintro.pdf <http://www.jfsowa.com/talks/clintro.pdf>

> Reification allows RDF to describe relationships, e.g. the time interval that a given relationship holds, 

This is a common claim, but it is false. Reification allows RDF to describe /RDF triples/, not relationships. RDF can already describe relationships (‘properties’ in RDF-jargon) without using reification: there are many examples in the RDFS axioms.

Relationships in RDF (and logics in general) do not hold for times: they are timeless. To describe temporally limited relations, one adds the time as an extra parameter or argument ot the relation. (Or at any rate that is one way to do it; there are others. There is a survey in section 2 of http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.529.5189&rep=rep1&type=pdf <http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.529.5189&rep=rep1&type=pdf>, which covers some of the common methods.)

> but reification is painful in practice.

No argument from me there. RDF reification should have been strangled at birth. 

> Extensions to Turtle and to N3 have been proposed that makes this less painful to express and process. We also want to explore ideas for fresh representations of declarative knowledge and procedural rules that are easier for the average developer.
> 
> We would like to be able to model the meaning of natural language in a way that mimics what we know about human reasoning

Well, that is a hugely ambitious goal. I wonder if you have any idea of the scope of this ambition and how much work has already been done towards it in AI, linguistics and cognitive science. (By the way, what do you think we DO know about human reasoning?)

> without reducing everything to logic.

? Why do you say /reducing/? Do you know of any other notation which is richer or more expressive than FOL?  Have you looked at how logics have actually been used to express complex knowledge about real topics, such as the OBO foundry or the Cyc knowledge base, or  ISO 15926-2 <http://15926.org/topics/data-model/index.htm> and ISO 15926-4 <http://15926.org/topics/reference-data/index.htm>? Have you read Carnap’s “Logical Structure of the World”, or studied any of the many published upper-level ontologies? Have you looked at Montague semantics for natural language, or any of the other linguistic ideas along these lines? Have you looked at AI work on NL comprehension and how systems such as Allen’s TRIPS represent meanings? I do not know of any serious work on linguistic meaning which does not use something at least as expressive as FO logic as its meaning representation. 

> The semantics are defined operationally in terms of the application of rules on graphs, including the means to compile graphs to rules.

With respect, this is not a semantics. Nor will any such approach ever succeed in capturing more than a tiny fragment of natural meaning. But good luck trying.

> Natural language is very flexible in its expressivity. 

Again, no argument from me there.

Best wishes

Pat Hayes
> 
> Dave Raggett <dsr@w3.org <mailto:dsr@w3.org>> http://www.w3.org/People/Raggett <http://www.w3.org/People/Raggett>
> W3C Data Activity Lead & W3C champion for the Web of things