- From: John McClure <jmcclure@hypergrove.com>
- Date: Mon, 05 Jan 2015 12:44:07 -0800
- To: public-owl-dev@w3.org
Hi Dan I agree much with your observation that 'good' models are prerequisite to solving these problems long-term. I've looked closely on schema.org, concluding that its use cases are a bit too contextual for my taste. Thus in my schema, I do try to sidestep some of the 'wooliness' you mention. Overall, its predicators implement a "verb:connector" construct. A "connector" indicates whether a triple's object is in a nominative, genitive, dative, accusative, indicative or other relation with respect to its subject. Samples of these constructs: a. Author: Moby Dick -- is:this -- Person: Herman Melville b. Author: Moby Dick -- is:that -- Author: Omoo c. Author: Herman Melville -- is:of -- Book: Moby Dick d. Person: Herman Melville -- is:of -- City: New York e. Book: Moby Dick -- is:for -- Group: General Audiences f. Person: Herman Melville -- is:a -- Nationality: American A 'predicate namespace' reflects past, present, future, conditional, existential &or other tenses of relations; a triple's nouns are therefore strictly quarantined to the context of the subject or object of the triple. a. Ticket: X12345 -- is:for -- Person: ABCD b. Ticket: X12345 -- was:for -- Person: ABCD c. Ticket: X12345 -- will-be:for -- Person: ABCD d. Ticket: X12345 -- is:for -- Organization: ABCD e. Ticket: X12345 -- had:by -- Organization: ABCD f. Ticket: X12345 -- had:by -- Person: ABCD g. Person: ABCD -- has:this -- Ticket: X12345 OK now to address the question of a triple's object that in some contexts is a text-string, and others a resource, I note the obvious: text-strings are resources as well "within" which one finds multiple translations (as a for instance, but also it's the container for lots&lots of metadata too!) a. Book: Moby Dick -- has:this -- Title: Moby Dick b. Title: Moby Dick -- is:a -- Type: Title (i.e., Title: Moby Dick -- rdf:type -- Class: Title) c. Title: Moby Dick -- is:of -- Book: Moby Dick d. Title: Moby Dick -- iso:en -- "Moby Dick" e. Title: Moby Dick -- iso:ar -- "موبي ديك" So bottom line I am trying to say that, while understandable I suppose from a desire for a rough equivalence of Assembler data structures with semantic models, those which allow nouns into the context of predicators yield a concrete number of inefficiencies, in practice. Rather, true semantic models in my view encapsulate constructs found in human language; anything less is an "engineering shortcut" causing -- with no doubt -- massive pulsing migraines for us all as competitive use cases emerge contradicting already published and authoritative "stable" models, during a process such as you note. Meaning properties like "rdf:type" "rdfs:subClassOf" and "rdfs:label" are intimately flawed; nounal expressions have no business but as a subject or object. And most of the properties at schema.org, designed from the same flawed perspective, are equally problematic. Anyway I hope to publish aspects of this work soon - thanks for listening regards/john On 1/5/2015 1:14 AM, Dan Brickley wrote: > [snip] > > Just a quick comment: one reason for schema.org's seeming woolly-ness > regarding type/property associations is that the entire (1200+ term) > structure is being gradually evolved as new use cases and vocabulary > proposals are developed. Rather than e.g. on monday saying "author has > range Person" and then on tuesday refining that to say "well actually > we meant to say author has range Agent which has subtypes Person, > Organization", the approach is more like saying "author > kinda-goes-with Person", where kinda-goes-with (aka > http://schema.org/rangeIncludes) doesn't let you conclude much. > Otherwise parties who believed you on monday when you said that > authors are always people will be disappointed and confused to find a > tuesday-based description in which something is said to be written by > an Organization. You might reasonably respond that it would be better > to get the basic modeling right in the first place rather than make it > up as you go along, and in practice much of the schema now isn't > changing much so it could feasibly be documented more tidily and > marked as unlikely-to-change. But that was the basic thinking: things > were changing a lot, and we didn't want to bloat out the documentation > of an already large structure with additional 'modeling artifact' > types that were not expected to be used in actual instance data... > > Dan
Received on Monday, 5 January 2015 20:44:36 UTC