W3C home > Mailing lists > Public > public-rdf-comments@w3.org > April 2013

Re: JSON-LD should be an RDF syntax

From: Kingsley Idehen <kidehen@openlinksw.com>
Date: Mon, 29 Apr 2013 15:59:10 -0400
Message-ID: <517ED10E.7020206@openlinksw.com>
To: public-rdf-comments@w3.org
On 4/29/13 1:51 PM, David Booth wrote:
>>> Apparently it was not entirely resolved.
>> It was resolved, hence the definition in place today.
> It clearly has *not* been resolved, as: (a) we are still discussing 
> it; (b) there obviously is major contention; and (c) the definition in 
> the current draft substantially conflicts with the original sense of 
> the term and with established usage.

It has been resolved on the JSON-LD side of things, to be clear. I have 
no idea what the state of play is on the RDF groups side of things, I do 
know that you are a single voice of dissent at this point in time though.

>>>> RDF and Linked Data are not the same thing.
>>> Of course not.  Linked Data (in its original sense) *builds* on RDF
>>> and other standards, most notably HTTP.
>> Sorta.
>> Linked Data is something you can produce using RDF and associated
>> standards.
>>>> Linked Data is something that you can
>>>> produce (5-star quality) using RDF.
>>> Yes.
>> Good, we are agreeing so far.
>>>> TimBL's meme is all about a
>>>> principled approach to producing Web-scale Linked Data that leverages
>>>> standards such as RDF and SPARQL.
>>> Exactly.
>>>> That doesn't imply that the meme owns the phrase "Linked Data.
>>> The term was coined specifically for that purpose.  AFAIK there is no
>>> legal ownership of the term, so we're talking moral ownership here --
>>> not legal.  The point is that attempting to redefine this term is
>>> harmful to the Semantic Web community, because it creates confusion
>>> about what "Linked Data" means.  And that is harmful to W3C's mission.
>> TimBL cleverly generated a meme using a quite generic phrase. The
>> original meme (the one that had no RDF or SPARQL references) was totally
>> GOLDEN re., blending Open Data and Web Architecture.
>> The meme got itself into trouble the moment RDF and SPARQL where added
>> to its most recent revision, for the totally wrong reasons. This
>> revision lay the foundation for RDF and Linked Data conflation. It took
>> a GOLDEN meme and turned it into a vector for the same old political
>> distractions that come with RDF (which is generally misunderstood).
>>>> Over extending what RDF is remains its ultimate problem. RDF doesn't
>>>> have to cloud the definition of everything in order for it to be
>>>> useful :-)
>>> That's directly backwards.  What's clouding the definition is *not*
>>> RDF, but the attempt to redefine the term "Linked Data" to mean
>>> something different than what it was specifically coined to mean:
>>> http://www.w3.org/standards/semanticweb/data
>>> http://www.w3.org/DesignIssues/LinkedData.html
>> RDF does not imply Linked Data,
> Of course not.  The implication is the other way around: Linked Data 
> (in the original, established sense of the term) implies RDF.

Linked Data doesn't imply RDF at all.

Linked Data simply implies that a URI resolves to a Document that 
describes its referent.

RDF based Linked Data implies the above, plus the fact that the Document 
is comprised of RDF based content [1][2].

>> most RDF resource out in the wild
>> (modulo LOD cloud datasets) don't even conform to Linked Data principles
>> in the slightest.
> Right.
>> Re-read this thread and I think you will see that you
>> got "backwards" completely backwards.
> I am amazed that you think so.  If you can show me evidence, then 
> perhaps I can figure out why we seem to be miscommunicating so badly. 
> At present I am mystified.

As per my comments above, there are links to illustrations of RDF based 
Linked Data URIs which convey the critical aspects of RDF based Linked Data.

Please note, the entity relationship model precedes any notion of RDF 
[3]. The contribution made by RDF is as follows:

1. URIs as denotation mechanism for entities
2. Explicit rather than implicit entity relationship semantics -- 
discernible to humans and machines.
>> RDF is a poor understood composite of:
>> 1. Data Model
>> 2. Data Model Syntax
>> 3. Data Model Syntax Notations
>> 4. Data Serialization Formats -- the product of processors transforming
>> Syntax Notation into raw data for storage or across-the-write transfer.
> Please do not conflate RDF with its syntax or serializations. Many who 
> are new to RDF make this mistake. 

I am not making a mistake. A simple example:

## Turtle ##

<> a <#Document> .

## End ##

The above is an example of syntax notation. Note how the notation allows 
the use or relative URIs. That's syntax notation for expressing an RDF 
model graph using Turtle.

A Turtle processor will not output an RDF graph serialization with 
relative URIs, such a thing is an invalid RDF graph.

> It is harmful to perpetuate this misunderstanding.

Who is perpetuating any kind of misunderstanding here? You have a simple 
Turtle example that backs up my point.

RDF has SPO based 3-tuples as its syntax.

Turtle is an example of notation for expressing those 3-tuples.

The final RDF graph is the product of processors that understand one or 
more syntax notations en route to producing actual RDF graphs.

> RDF may be poorly understood to many developers, but that is 
> irrelevant, as many technologies are poorly understood to many 
> developers. 

It isn't irrelevant. RDF has taking a simple concept and turned it into 
a riddle, for all the wrong reasons.

> But RDF is *not* poorly understood from a technical perspective: it is 
> a W3C standard and is very clearly defined.

No comment. I think that statement speaks for itself with regards to 
actual reality.

>> Nothing in the RDF specs mandates that URIs or IRIs must resolve.
> Right.
>> Basically, the HTTP URI duality reality [1] isn't a part of the RDF
>> spec.
> Right.
>> Thus, you cannot constructively conflate Linked Data and RDF.
> I didn't conflate them.

You conflate them whenever you infer:

1. Linked Data is RDF
2. RDF is Linked Data.

Linked Data is neither, it's something you can produce in very powerful 
form using RDF.

> Linked Data (in the original and established sense of the term) 
> *builds* on RDF.  Not the other way around.  And the two are not 
> synonymous.
>> What you can do though, is talk about RDF based Linked Data.
> Yes, and in the original and established sense of the term "Linked 
> Data", *all* Linked Data is RDF based.

It is not. Stop pushing that misconception. RDF is not the progenitor of 
"Linked Data" in any shape or form. Computing precedes the World Wide 
Web, you know that.

>> To conclude:
>> JSON-LD is a good effort at outlining how to produce RDF based Linked
>> Data that's expressible and serializable using JSON.
> I agree that's what JSON-LD *should* be.


> But the problem is that that is *not* what the current draft of the 
> JSON-LD spec reflects.

To the degree the current draft can be tweaked without inferring:

1. Linked Data is RDF
2. RDF is Linked Data.

I won't have any issue, anything else, well I will be in total 

> In the current draft, JSON-LD has no normative basis in the RDF model 
> or the RDF semantics **at all**.

Fine, and does it really need to? The connection with RDF has to be 
through the algorithms for producing RDF from JSON-LD or vice versa.

> Sure JSON-LD *can* be mapped to RDF, but that's not saying much, 
> because *any* language can be mapped to RDF. 

But that's saying all that needs to be said re. JSON-LD. The goals or 
JSON-LD are clearly spelled out.

> The current draft reads as a completely parallel attempt at defining 
> an independent language for achieving similar goals as Linked Data (in 
> the original, established sense of the term), while redefining the 
> term "Linked Data" to include this parallel language.

It is an option for JSON developers seeking to work with Linked Data. 
Developers who have next to no interest in RDF and its inability to shed 
its riddle-like image which makes comprehension eternally mercurial to 
this Web developer profile.

> If the definition of "Linked Data" were broadened beyond using RDF as 
> the information model, where would the line be drawn?

Again, Linked Data and RDF are not the same thing. You can produce 
powerful and very useful Linked Data using RDF.

You continue to assume that RDF invented the 3-tuple and entity 
relationship model. You assume that HTTP URIs are the only mechanism for 
de-reference and indirection baked into structured data representation etc..

You can actually make 5-Star linked data using other URIs schemes. The 
downsides have more to do with the limitations in Web browsers and the 
ubiquity of this type of user agent re., the Web. On mobile platforms, 
the notion of URIs schemes works without the limitations imposed by 
desktop browsers etc..

> Would Excel spreadsheets be considered "Linked Data" if they contained 
> HTTP URIs that resolved to other spreadsheets? 

What does a named cell in an excel spreadsheet actually do? It uses 
name-address indirection to give you a name that resolves to an address. 
The problem is that in this case the actual relation semantics are 
implicit and coarse at best. Thus, to some it could be considered Linked 

Better example, what about a CSV file comprised of 3 cols that's used to 
express entity relations in 3-tuple form? Where cols 1 & 2 hold URIs and 
the 3rd col hold URIs or literals? Is that Linked Data?

> Would relational database tables?
No they are not.

>  Would HTML documents?

They are yet another example of coarse-grained linked data due to the 
fidelity entity relationship semantics. Remember, the HTML spec does 
have <link/> tags that denote relations.

> To step up a level: what is your goal in attempting to disassociate 
> RDF from Linked Data?

Simple: end RDF narratives that are well intended but utterly 
distracting. RDF is useful without a distracting power-grab on "Linked 
Data" which is a very generic phrase. RDF based Linked Data is crystal 
clear and devoid of the aforementioned distractions.

The real value of RDF is that it builds on the old entity relationship 
model by introducing explicit (rather than implicit) machine- and 
human-discernible entity relationship semantics. It also leverages the 
ingenuity of URIs as global identifiers for things in general.
> What are you trying to achieve by redefining the term? 

I am  not redefining anything. RDF has never been defined as Linked 
Data. If that was the case the spec would specifically mandate the very 
behavior expressed in TimBL's meme.

> Are you trying to invent a different Semantic Web that is JSON-based 
> instead of being RDF-based?
If you assume that, then we are speaking past ourselves on an inter 
galactic level :-)


1. http://twitpic.com/cmw52i -- illustrating hashless Linked Data URIs 
(Webb Super Keys) e.g. those used by DBpedia
2. http://twitpic.com/cmw72m -- illustrating hash based Linked Data URIs 
(Webb Super Keys)
3. http://bit.ly/T3kWUv -- Peter Chen's 1976 dissertation on the entity 
relationship model .

> David
>> Links:
>> 1. http://bit.ly/YxW21k -- HTTP URI Duality that lies at the core of
>> Linked Data as espoused by TimBL's original meme.
>> Kingsley



Kingsley Idehen	
Founder & CEO
OpenLink Software
Company Web: http://www.openlinksw.com
Personal Weblog: http://www.openlinksw.com/blog/~kidehen
Twitter/Identi.ca handle: @kidehen
Google+ Profile: https://plus.google.com/112399767740508618350/about
LinkedIn Profile: http://www.linkedin.com/in/kidehen

Received on Monday, 29 April 2013 19:59:32 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 16:59:32 UTC