Re: 1 NF

On 2007-02-28, -CELKO- wrote:

> There is also a difference berween "atomic" and "scalar" in the
> definition of 1NF. For example (longitude, latitude) is atomic (Greek
> for "without parts" or indivisible) because it resolves to one point
> in space.

Really? To me this seems highly confused, because you have to refer to semantic concerns in order to make the distinction. Such concerns do not seem to belong to the relational level.

For the sake of an argument, each database state can be thought of as atomic/a structureless value, precisely the same way you say a (longitude, latitude) pair can be. I mean, each one does resolve to a single member of the set of valid database states under the relational model. So, does it follow that, since all databases can be so represented, all that is prescribed under the relational model is a single attribute, singleton relation, containing a value drawn from the domain of valid states of a relational database? I don't think so.

>> Does a blob containing an image violates 1NF?
>
> There is hard! Can you break it down into meaning parts -- i.e. is the
> blob atomic?

No. A picture clearly has structure. The structure is even more amenable to relational representation than most kinds of data (e.g. it is a function, not just a generalized relation).

> That one is hard, since pictures have semantic content -- "Look!!
> There is Lindsay Lohan behind a tree!" but it is really hard to use
> theta operators because semantics lead to fuzzy logic.

Semantics live at a higher level. They shouldn't be confused with the relational model, which lives lower down and is quite capable of handling images (and *far* more).

I also see no reason why such higher level stuff couldn't be modelled relationally. E.g. one could have a relation representing parametrized curves, which outline Lindsay Lohan and the tree. Then we could have another relation saying one outline refers to a 2D projection of a 3D object behind another 3D object, also represented as a 2D projection. There could also be a third relation telling how to relate the curves to the raster images, and fourth and fifth relations referring to the people and the trees in general. The latter could give us some further data on the relevant objects/entities. The sixth through one hundreth relations could encode the relevant metadata -- encoded against a common/public/shared/universal data dictionary -- and stitch all of these metalevel concepts together so that a generic reader agent can relate to the semantics of our database. The 101st throught one billionth could then encode all the relevant probabilities and adequately discretized fuzzy set membership values for all of the nastier relations.

The last two parts are also what could enable high level predicates -- like ProbabilityDistribution
wouldThisPersonThinkTheseFacesRepresentTheSamePerson(recognizer: Person; recognizee1: Image; recognizee2: Image); -- to work on the entire set of data comprising persons, images, fuzzy logic values, unpredictable schemas, and so on. In the abstract, I don't see why this sort of integration effort couldn't be accomplished in the far future.

> So I would say that the fuzzy logic would override the usual 3VL for
> RDBMS and we do not have a proper domain.

Utter hogwash and handwaving. Eventually even fuzzy logic rests of first order predicate logic. It can be handled on the latter, classical terms. In fact even moreso than general paraconsistent logics or the like.

In particular, all of them can still be modelled relationally. Some better, some worse, but they can be. Hiding behind esoteric concepts like "fuzzy logic" is not going to exempt you from the burden of reasoned argument, on the topic of the relational model vs. images/blobs/complex objects.

-- 
Sampo Syreeni, aka decoy - mailto:decoy_at_iki.fi, tel:+358-50-5756111
student/math+cs/helsinki university, http://www.iki.fi/~decoy/front
openpgp: 050985C2/025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2