Re: stratml vs cl

> On 10 Jan 2020, at 04:16, Paola Di Maio <paoladimaio10@gmail.com> wrote:
> 
> Dave
> 
> Is a formal KR really needed?  There is no evidence that biological systems use formal KR as opposed to other forms of computation.
> 
> This is an important question. It would probably require an essay, for which I do not have time. 
> I ll try to be very brief
> - what doe we mean by formal?  (different levels of formalization?)
> - I think what we need is enough formality to support 
> a) logic /reasoning
> b)robustness/repeatability/reliability consistency
> c) verifiability/proof that a) is correct to some extent
> 
> I gave a talk once that was aiming to say natural language is sufficiently formal
> to enable abc, but not sure I fully managed to put my point across as crisply as i would have liked
> workshop page
> http://www.cs.stir.ac.uk/events/network-analysis/ <http://www.cs.stir.ac.uk/events/network-analysis/>  
> My slides
> http://www.cs.stir.ac.uk/events/network-analysis/slides/dimaio-analysis.pdf <http://www.cs.stir.ac.uk/events/network-analysis/slides/dimaio-analysis.pdf>  
> 
> (I am indebted to Sowa for explaining this at length on ontolog forum) 
> 
> Regarding biological systems, we really dont know enough, I d say and biological systems
> may use different forms of communication than language as we know it
> until we evolve to communicate without language, some degree of formalization may be necessary/beneficial
> 
> The crux for me is consistency. ability to express intent and to follow through and verify it ETC
> for this we normally require some degree of formalization. but if you can find a way Dave to achieve logical consistency without formalization I d be very interested
> :-)

Whilst there is general agreement on the value of graph representations, Industry is showing a lot more interest in Property Graphs than in RDF. This has two corollaries: the first is that Property Graphs are allegedly easier to work with, and the second is that formal semantics and logical deduction (at centre stage for the Semantic Web) are not important for the majority of industry use cases. 

As you hinted at, logical consistency can be considered in terms of robustness, repeatability, reliability and consistency over use cases of interest.  Learning is about adapting to new use cases which don’t quite fit the existing model.  An example is extending data types for people’s names to allow for accented characters in people’s names, or to allow for more than one family name (as is the case in Spain).  Today, adding support for such extensions involves contacting the IT department, as the semantics are implicit in the data queries embedded in application code, and hence require talking with programmers to make the changes.

Natural language semantics are established through usage by a community of language speakers. The meanings often change over time as new patterns of usage appear. Trying to formalise this would be both challenging and rather futile.  A better plan is to model how people learn new meanings from what they read and hear in conversations with other people or through listening to media. Formal languages have a role to play where the context is clearly defined and relatively static. However, for AI, those conditions typically don’t hold.

Best regards,

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