W3C home > Mailing lists > Public > www-rdf-comments@w3.org > January to March 2004

Re: Clarification of Reification vs Quotation

From: pat hayes <phayes@ihmc.us>
Date: Mon, 19 Jan 2004 11:48:01 -0600
Message-Id: <p06001f19bc31b2f2104d@[]>
To: Karsten Otto <otto@math.fu-berlin.de>
Cc: www-rdf-comments@w3.org

>I am slightly confused by the RDF Primer and RDF Semantics sections on
>Reification. Both documents state quite clearly that a reified triple
>is bound to a token (i.e. an occurrence) of a triple in a specific RDF
>document, and that its subject/predicate/object are bound to actual
>resources of the world (i.e. elements of IR/IP).
>Also, both documents specifically contrast this to something called
>Quotation. The example "this triple has this form" suggests to me that
>a Quotation is bound to a triple token too ("this"), but its s/p/o are
>*not* bound to world resources, i.e. are just strings.
>However, considering the binding to a triple token and the binding to
>world resources I would expect the following possible constructs:
>"this RDF triple talks about these things"  (Reification)
>"this RDF triple talks about these strings" (Quotation, not bound to 
>"some RDF triple may talk about these things"  (not bound to a triple token)
>"some RDF triple may talk about these strings" (combination of the two above)
>I am wondering what is so important about this particular alternative 2.
>In fact, it seems of rather limited use to me, as attempting to model
>something like "Bob did not say "The sky is green" " would lead to a
>contradiction: The quoted sentence is bound to a triple, which by
>definition *asserts* the fact "The sky is green".

No. Asserting the reification does not *assert* the reified triple: 
it merely records the fact that it exists and says something about 
what it is saying. For example, one could record what Bob said in an 
RDF graph ex:whatBobSaid, and then reify that appropriately with a 
URI ex:whatBobSaid#proposition, and then say something else about 
that reification, all without agreeing with Bob in any way.

>For this example,
>alternative 3 would be more appropriate.
>I would appreciate if you could clarify this issue in the Primer and
>Semantics document.

It is now too late to modify the documents themselves. A full 
clarification of this issue would take us into semantic foundations 
of language generally, and is therefore beyond the scope of these 
technical documents. Clearly however you have grasped the essentials 
(with the exception noted). What you seem to be asking for is 
motivation for the choices made, which I will try to sketch here. 
(Forgive me if this all seems elementary: I do not know your 

There is a basic dichotomy between how to treat text embedded in an 
opaque context such as "Joe said ..." or "Doris believes ....." where 
an assertion of the outer context clearly is not intended to simply 
assert the content of the inner text. This contrast is often referred 
to with the terminology 'de dicto' (of the speech) vs.  'de re' (of 
the thing), the Latinate tags providing a reminder of how long ago 
this contrast was first noted (I think it was around 1300; it starts 
getting important in any society about when courts start recording 
what exactly witnesses said.) The contrast can be illustrated by the 
distinction between direct quotation of speech, as in "Louis said, 
'Superman is Clark Kent' "  vs. "Louis said that Superman is Clark 
Kent" . The first, de dicto, reports Louis' actual words (and is 
false, in the story) while the second, de re, reports what she said 
about someone, using the speaker's words (and if the speaker knows 
more about Superman than Louis does, might well be true: even though 
Louis herself wouldn't identify the guy using the term "Superman", 
she might well have said that Clark Kent was Clark Kent, and of 
course as we know, Clark Kent *is* Superman.)

While it is often thought natural to give such sentences a de dicto 
reading, this actually is very awkward, since the de dicto 
interpretation means that every referring name suddenly becomes 
multiply ambiguous. (Do you mean "Superman" in our sense or in the 
sense that Louis uses it? What about some poor schmuk who hasnt even 
heard of Superman?) This (notoriously) makes formal reasoning 
extremely difficult and complicated, to the point of making it 
impractical: for example, the inference rule of existential 
generalization isn't correct, since people can believe in fictions. 
Therefore, the decision was taken to adopt the de re reading for 
reification.  This means that all occurrences of a given name, inside 
and outside reifications, always have the same meaning.

Other motivations for this decision are that the de re interpretation 
is more expressive than the de dicto one, in that one can say things 
in the former which cannot be expressed in the latter (such as that 
Joe said that *someone* is a Republican and a Pacifist and his name 
begins with "N", without specifying who exactly); and that RDF 
provides alternative ways to encode de dicto assertions, eg by using 
a literal string to encode the quoted proposition.  Also, the usual 
intention of de dicto assertions (eg about what Bob believes) can 
often be handled more subtly by having URIrefs for the things in 
"Bob's mental space" versus in reality, and distinguishing them 
explicitly. This would enable you to use de re reification to achieve 
the effect of a de dicto quotation, but in a more robust framework. 
For example, one way to describe the Louis Lane situation might be to 
have a graph called ex:WhatLouisThinks with a URIreference used in 
it: ex:WhatLouisThinks#Superman and clearly distinguishing a 
reification of the assertion involving this URIref from a similar 
assertion involving ex:Superman (ie the, er, real guy). In  OWL one 
might even assert that they are distinct, for example.

A way to summarize all this is that RDF makes the blanket assumption 
that all URIrefs are talking about one single 'reality' and so they 
always refer in the same way. If you want to talk about people's 
ideas or sayings that do not refer to the same reality, or which use 
names differently, then you need to be explicit about these 
differences: you can't just use the same name and assume that it's 
meaning gets changed by the context. So if "Superman" refers to 
Superman(= Clark Kent), then it goes on referring to him even when 
you are talking about the beliefs of someone who is rather confused 
about who Superman actually is.

If this all seems like a lot of trouble to go to just to handle 
quotation, maybe you will just take my word for it that getting de re 
and de dicto reasoning muddled up with one another has been a source 
of confusion and misunderstanding in human affairs probably ever 
since language was invented; that it really was essential that RDF 
choose one or the other; and that the one chosen was, we feel, the 
one that most usefully provides for accurate renderings of the 
intended content in most cases, and can support reasonably coherent 
inference processes.

Hope this helps.

Pat Hayes

IHMC	(850)434 8903 or (650)494 3973   home
40 South Alcaniz St.	(850)202 4416   office
Pensacola			(850)202 4440   fax
FL 32501			(850)291 0667    cell
phayes@ihmc.us       http://www.ihmc.us/users/phayes
Received on Monday, 19 January 2004 13:05:55 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 18 February 2014 13:20:08 UTC