Re: plural vs singular properties (a proposal)

On Jan 6, 2008, at 4:35 PM, Garret Wilson wrote:

> Frank Manola wrote:
>> As far as each relation representing a distinct resource...
> "Tuple", Frank. I said that "each tuple represents a distinct  
> resource".

A tuple per resource is actually what I had in mind, and I think if  
you take that intepretation, the rest of what I said stands.

>
> I understand your relational design in which there is a different  
> relation for each predicate in the graph, each tuple being  
> interpreted as a binary relation between RDF triple subject and  
> object. The discussion here was about a different relational design  
> (the one with Date and his wine bottles) in which each relation  
> header attribute is interpreted as an RDF predicate. In analyzing  
> this latter design, I made the assumption that each tuple  
> represents a distinct RDF resource---the implication being that  
> this design cannot support multivalued RDF properties.

Part of the problem I was having was that you sometimes seemed to be  
talking about "relational interpretation" in general, not just what  
Date was assuming about a particular design, and I thought you were  
saying that there was a given *interpretation* of the relational  
model that precluded this or that.  As long as we understand that  
we're talking about different designs, all is well.

>
> You and Bijan have countered that even a relational design that  
> interprets relation header attributes as RDF predicates can still  
> support multivalued RDF properties because my assumption that each  
> tuple represents a distinct resource is invalid; a resource may be  
> represented by more than one tuple in the same relation, allowing  
> "repeated predicates" by supplying multiple values for the same  
> column in different tuples but for a single resource. Something  
> about that smells fishy to me, but being unable to articulate  
> exactly what it is I have determined to do more studying of the  
> relational model before attempting to address that point you both  
> have made.

It's fishy because the resulting relations are only in first normal  
form, and all the database texts tell you not to duplicate data like  
that.  The reason has to do with storage bloat, and difficulty of  
consistently updating all the repeated values if you need to, both of  
which are practical aspects of maintaining large databases.  But from  
the point of view of logic, it doesn't really hurt to say that Joe's  
age is 43 ten or fifteen times if Joe has ten or fifteen hobbies;   
the "results" are the same.

--Frank

>
> Does that make sense?
>
> Garret
>
> P.S. In a very early email on this thread I mentioned each relation  
> representing a resource, but as I pointed out later, I had actually  
> mis-spoke---I had meant a single relation for each resource *type*  
> of resource (with each tuple representing a resource).

Received on Monday, 7 January 2008 16:17:42 UTC