Re: An intuition pump

On Tuesday, September 23, 2003, at 12:02  AM, pat hayes wrote:

>> On Sunday, September 21, 2003, at 05:45  PM, pat hayes wrote:
>> [snip]
>>>> Now Molly comes along and notes the extreme chauvinism of Sally's 
>>>> definition, excluding as it does (arguably) eunuchs, 
>>>> hermaphrodites, intelligent programs, chimps, augmented chimps, 
>>>> Martians and the like. Molly proposes an alternative ontology for 
>>>> sally:Person.
>>>>
>>>> Now, if I understand the view as raised by tim in his issuing, 
>>>> Molly and the Foafsters are pretty much stuck. sally:Person just 
>>>> *means* whatever Sally wants it to mean.
>>>
>>> Well, it does mean that, but they are not stuck. Molly can introduce 
>>> molly:person, a superClass of sally:person, tell people what it 
>>> means, and then folk can use that. If enough people feel that this 
>>> is the best definition, it will presumably get used more widely and 
>>> will become the de facto new standard, maybe. More likely, two rival 
>>> communities will emerge, the Mollyites and the Sallieties, with 
>>> rather different views about what constitutes a person.
>>
>> There are at least two ways to express this rivalry: We have two 
>> *terms* (molly:Person, and sally:Person) each with their own, rival, 
>> definition, or we have two *definitions* for a single term which has 
>> an associated concept which the definitions attempt to capture. Your 
>> (and, I take it, Tim's) view is that we have multiple terms, 
>> presumably at least one per definition.
>
> Well, yes; at least, given that the term "definition" might be 
> slightly misleading.

Hmm? I'm trying to use it roughly as I thought Tim was. Heh. I thought 
it was something like the ontologicy assertions concerning the term 
published by the URI owner somewhere reasonably "near" the relevant  
URI.

>>  One thing I'm trying to get at is that people might reasonably take 
>> Sally's term as *naming* a concept, a concept which she then says 
>> false things about
>
> I cannot make clear sense of this. I have no idea how to name a 
> concept. Are you saying that there might be a real thing in the world, 
> called a concept, that the assertions are about?

Wow, I'm a bit surprised. Given the RDF semantics of having classes and 
properties be individuals that happen to have associated extensions, I 
thought this reading would have been natural. I'm not at all saying I 
endorse this view, but it seem rather common.

>> , strictly analogous with Sally coining a URI for *me* and then 
>> saying that I was born in 1887.
>
> Not strictly analogous, because you are not a concept.

Presumably that's not why it's not strictly analogous, since neither is 
a relation.

>  Your date of birth is just a fact, which can be misrepresented by a 
> falsehood.

Yes.

>  Concepts are not real things that have crisp edges beyond which 
> falsehoods can be discerned.

I tend to agree, I think. Maybe not. :) But it certainly seems like 
rdfs:Classes are this way (dc:creator? dc:date?) Actually, I'm not sure 
what to make of such metadata.

>  If I say that marriage means a social union between two consenting 
> adults, and Joe says that marriage is a pact between a man and a woman 
> made before God, do Joe and I disagree about a question of fact or 
> about the meaning of the word "marriage"? I think the latter is the 
> only option that makes sense.

Well, even if so, that's not the end of the story. Joe says that 
joe:marriage rdfs:comment "a pact between a man and a woman made before 
God" and I point out that, in fact, in his canonical ontology, he 
wrote, "joe:marriage rdfs:comment "a unity between a man and a woman 
made by God", what are we disagreeing about?

> Now, one might say that there is a concept OF you; but (1) that is not 
> the same thing as you - for example, it doesn't have a weight or a 
> date of birth - and more to the present point (2) if there is one of 
> them, then there can be lots of them, eg your concept of you, your 
> mother's concept of you, my concept of you , etc., all of which 
> presumably differ slightly, or in some cases substantially from one 
> another.

This a red herring. I was really referring to the notion of concepts as 
just more individuals to hang assertions off. Or forget that. How about 
URIs as things which I can "say anything about" (anyone can say 
anything about anything). (Perhaps we need to get 
definitions/intentions into the picture as first class objects? Why 
not! We've got most everything else :))

>> Must people coin their own URI for me, in order to say true things 
>> about me that are contrary to Sally's false statements?
>
> No, of course not; though they might choose to use it in order to make 
> a direct rebuttal.  But if they use *your* URI to refer to you, then 
> your ownership of your URI means that their uses of it in any 
> assertion have to be understood as de re in their reference to you.

Really? Huh. I'm not sure waht to think about that.

>>  I hope we all agree not.  So, what justifies the disanalogy?
>>
>>> What did you expect, that the entire world was going to agree about 
>>> what words mean?? (Try 'marriage'.)
>>
>> In your view everyone DOES agree about what words/URIs mean. I'm 
>> suggesting that people might disagree about what *sally:Person* >> means.
>
> The entire world has to allow the owner of a URI to say what its 
> referent is supposed to be

Hmm. That seems strong. I might not KNOW what the referent *is*. "The 
unknown soldier"

Or, I might leave my referent only partially constrained. Are all my 
defintions *finished* in the sense that no one else can refine them?

>  (if they want to do that: they are not obliged to).

What can I do when they don't? What happens later if they change their 
mind?

>  The same is true for coined words in English, in fact, at least 
> initially.

Heh. Peter raised this, too, in Rome. I'm not so convinced.

> Charles Dodgson said what "gyre" and "wabe" meant, and who is say that 
> he was wrong?

Uhm. Me? :)

>  He invented the words, after all.

If I coin the word "flurble" which, I stipluate (and stipulation is 
interesting; somewhat stronger and weaker than coining, I think) means 
"that certain smell that wet cats loathe"

>  So if Sally says that sally:Person means human beings, then that's 
> what it means. If you use it intending to refer to something else, 
> then you have made a  mistake, and what you actually said isn't what 
> you meant to say. You should take more care when using other people's 
> terminology, particularly when the terminology has got their bloody 
> address written on it, like a dog tag.

Let me try another analogy: What I'm arguing for is like deep linking. 
I want to use your URI, which, in fact, points to your definition. I 
want to use it my way on my own pages, say things about it (like what 
it refers to). It might not be the happiest, but I don't see it's so 
off.

How about this. I wrote the foaf ontology expecting Sally to fix up her 
definition along the lines she and I discussed. A LOT of people are 
using my ontology, but sally sold her domain name or has just gotten 
stubborn. Why can't I just *redefine* her term in my context and save 
all those poor souls from having to rewrite their documents?

If I want to use her term *as she wrote it now*, why can't I just copy 
and paste her ontology into mine and use that? If she changes it, must 
I keep track? Why?

(If all this is this way, why would I ever use anyone elses terms?)

[snip]

>>> The ontological reasoners have at least a chance of not getting 
>>> confused, because they have names for both concepts and an idea of 
>>> the relationship between them.
>>
>> I suspect that there will be some not so easily solved cases. Perhaps 
>> my example was oversimple.
>
> I agree there will be hard cases and I don't think this issue is going 
> to just vanish away. But I think the TimB-L vision of URI ownership at 
> least gets us started in the right general direction; whereas the idea 
> that all meanings are globally determined simply by model theory is a 
> non-starter.

Uh. I certainly share with you that the latter is a non-starter. Good 
thing I've not started it. I think you are misattributing a view on me. 
I've not yet advocated a view. Certainly not *that* one. I'm trying to 
describe things I think people will do or want to do.

>>>  I don't see this as being a problematical kind of example; on the 
>>> contrary, in fact: it shows (in a simplified caricature)
>>
>> Ah yes :) Deliberately so, for the record :)
>
> Of course, that's OK.
>
>>
>>>  how a nuanced vocabulary can arise from people being obliged by the 
>>> conventions of ownership to define relationships between concept 
>>> meanings.
>>
>> How do I express that Sally *tried* to capture the concept that 
>> planetOfTheApes:Person (only superchimps have rights) captures? (I 
>> guess the common factor may come into play.)
>
> I have no idea: but why should I care about that?

Sally cares. I might care. I want to talk about Sally and what she did 
or tried to do. I want to say, "by *this* she meant, or ought to have 
meant *this*". Or "try changing the definition to *this* and watch the 
contradictions evaporate".

>  If Sally said something she didn't mean to say, then she has screwed 
> up, indeed, but I can hardly be expected to be telepathic.

The question is can anyone else repair the damage. Or, similarly, if I 
think she's messed up, is my only choice to defer to her mess or 
refrain from using her URI?

I actually just don't believe that people own URIs. They don't. They 
own servers. They may own domain names, but I'm not sure that's a good 
idea. They certain have secured certain compliant behavior from DNS 
servers with their dollars. But URIs are just strings. I really hope 
they don't end up like trademarked strings. Has anyone trademarked a 
URI? Yeek.

>  What matters to me is to grok the meaning of what Sally actually said.

And if she's said one thing in OWL, a slightly different thing in her 
rdfs:comment, and yet a third in her documentation? Of course the only 
thing the software cares about is the OWL (well, unless someone 
implemented a program against the documentation alone). If we're down 
the path of intended meanings, where do we stop?

>  If there isn't a URI out there for her to use to express her concept 
> then she has to use a new one, and then I may have to ask her what it 
> means, or try to figure that out from whatever else she says using it, 
> or maybe I don't really care as long as I get the inferences made well 
> enough to do whatever I am trying to do.

Yay for the last :)

> Seems to me that in practice, this issue of 'but what were you trying 
> to say?' will only come up when normal communication has broken down 
> for some reason (eg when an inconsistency is detected where things 
> should be consistent) and we are all in some kind of damage-control 
> mode of trying to disentangle and re-align our concepts more exactly.

Hmm. No, I think it will come up when I'm trying to decided whether to 
use your term. Or it will come up soon after I've done so :)

>  One proven way to approach this kind of a snag is to use a kind of 
> initial triage of making all the clashing concepts different, then 
> trying carefully to see what happens when you re-merge them, using 
> detected inconsistencies as a guide to how to split concepts into 
> subcases, just like the Molly/Sally example where there are apparently 
> two rather different notions of person being bandied about.  What this 
> suggests is that the ability to trace URIs back to their owners, and 
> the likely fine-tuning of concepts (my notion of person vs. your 
> notion of person) is in fact a way of almost automating the initial 
> stages of the damage-control process that would have to be done if we 
> all tried to use the same URIs to express different meanings.

But, uh, this works if I only use expicit imports statements, with the 
possibility that I'll end up in several different places for different 
uses of the same URI in different contexts. So?

>>>  Contrast this with the case where nobody can really say what a URI 
>>> is supposed to mean,
>>
>> We have a way of saying what a URI actually means in a graph/document.
>
> We do? How?

I mean that the situation right now, as I understand it, is that there 
are no mandated constraints on the meaning of a term other than what's 
in the current graph/document. owl:imports has the explicit effect of 
expanding the graph from the mere parse of the initial document. Mere 
use of the URI *can* do that, but it's application dependent. This 
seems quite coherent to me.

>> What's at issue is what *else* we have to put in. For all its issues, 
>> owl:imports vs. uri use gives you a way of specifying the "extended" 
>> meaning of a URI on a fairly finegrained basis.
>
> I fail to see how owl:imports helps with this matter at all.  All it 
> does in effect is to create larger ontologies; it doesn't specify the 
> meanings of URIs.

Well, neither does loading up the contents of a docuemnt found by 
derefing the uri. I can find alternative documentation for a URI ("Hi, 
I'm hijacking sally:Person because when we started using it, we thought 
it made sense but now we love intelligent computers and chimps too. If 
you see sally:person in one of these docs, please use *this* definition 
rather than the one at sally's place")

>  And I don't know what the 'vs.' is intended to suggest. Do you see 
> some kind of methodological alternatives here? (That shared URI use is 
> somehow a different way of doing importing?? )

That's my understanding of the difference. Load only the explicitly 
imported, or load everything used.

>>>  and when Molly used sally:person in a nonstandard way, that 
>>> (mis?)use just kind of back-taints all the other uses, so that even 
>>> Sally's ontology doesn't mean what she intended it to mean any more.
>>
>> I fail to see it. Explain?
>
> Perhaps I have attributed too much to you here.

I think so.

> I presume that if the owner of a URI is not an authority on its 
> intended meaning, that the alternative view, which I thought you must 
> be holding, is that the meaning of a URI is determined by (the model 
> theory of) whatever assertions are made using the URI.

I see where the confusion is coming from. There's at least two issues: 
2) If I use sally's URI, must I use it constrained by Sally's published 
ontology? (Loads of related issues, there, e.g, published *at* that 
URI?) 1) What determines the reference of a URI?

(yes, i counted down :))

>  In which case, if Molly uses Sally's URI and makes some assertions, 
> then those assertions become part of what determines the meaning of 
> the URI (in some sense the meaning is determined by the 'entire Web' 
> at any given time)

I'm agnostic on this. However, I might go for that the meaning of 
Molly's *use* of Sally's URI is partially determined by those other 
assertions. Especially the presense of various owl:imports.

>  - after all, if you do not recognize ownership as important, then why 
> is what Molly says any less determinate of meaning than what Sally 
> says?

In Molly's documents, it's not. They're *molly's documents*, in a much 
stronger way, imho, that sally's uris are *hers*. Who the hell is Sally 
to tell Molly what to do in her documents? :)

>  - in which case, even if Molly's concept does not match with Sally's, 
> what Molly *says* will affect the meaning of what Sally has published.

And what molly *implements* will affect the actual *effective* meaning 
of particular uses of Sally's URI. How could it be otherwise? A bug or 
misunderstanding, and the sale is made (or not).

Plus, I keep reading these things about how agents will bootstrap yet 
more understanding by spidering a few more documents...isn't this how 
the "web of meaning" creates ever more meaning? Or something?

> This seems kind of obvious, so if you don't see it then maybe I don't 
> follow what you have been suggesting. The above "vs." suggests that I 
> am not fully following you, indeed. Can you elaborate?

Hope the above helped.

>>
>>> Now, *that* really is a recipe for world-wide confusion.
>>
>> I feel some fatuous statement coming one...must resist...can't: 
>> "World-wide confusion is fine; local confusion isn't"
>
> I have absolutely no idea what you are talking about.

I regard the "world wide meaning" of the term as possibly 
uninteresting, at least from a spec/software point of view. I don't 
think anything will acutally be able to grasp the global web-holistic 
meaning of any uri (well, maybe SUPER GOOGLE!!!). But I also think it 
doesn't matter. I think I'm fine with there being some intended 
meanings (of which we might privilege the URI owner, but maybe only the 
coining owner, maybe not), the "web" meaning (but, heck, we can't know 
what that is anyway), and the speaker/local/contextual meaning. In the 
end, only the last one matters, at least, to a particular communicative 
act. (Well, that's strong, but I'm feeling a tad bumptious.)

>>>> More interestingly, suppose Sally had just been a bit careless and 
>>>> really was aiming at a more expansive notion of Person, just blew 
>>>> it. However, before Molly detected the problem, Sally sold her now 
>>>> very popular domain to People for a Very Narrow Sense of People in 
>>>> Foaf Documents (PVNSPFD). They refuse to change sally:Person.
>>>>
>>>> Now, what concept does sally:Person identify? When? Does it matter?
>>>
>>> I would say : whatever PVNSPFD claims it does: because they own the 
>>> URI (now). However, I doubt if they would have any motivation, 
>>> having paid for the damn thing, to alter its meaning
>>
>> Of course not. They *like* the "mis"definition poor Sally gave for 
>> it. They are trying to hijack her intentions (with, admittedly, what 
>> she actually said).
>
> Again, you seem to be seeing something underhand or nefarious in this 
> (admittedly somewhat far-fetched) scenario that I do not perceive at 
> all.  Why is Sally "poor Sally" at this point, exactly? Has she been 
> cheated in some way? How? By whom?
>
> [later, on re-reading: I think I may see the point. Sally *meant* to 
> say A, but instead said B, and then sold her words to these bad guys 
> who wanted to say B all along, so indeed just went on saying B; not 
> thereby confusing anyone else, and not asserting anything false; but 
> poor Sally, being either as thick as two short planks, or dyslexic, or 
> maybe both, feels that she has been somehow misrepresented. Is that 
> more or less right?

Bit less, than more, but yes. But *Molly's* the one, perhaps, who's 
peeved. Actually,if PVNSPFD *change* the meaning, that's going to be a 
bit frustrating for the rest of use who used the term.

[snip]
>> They also seem free to change Sally's term any which way.
>
> Well, the OWNER of a URI is I guess free to say anything they like 
> using it, including change their mind about its meaning. I agree that 
> if this happens a lot, things will get confusing. But in the absence 
> of some kind of totalitarian control over Web usage, I see no way to 
> guarantee that this can never happen.

But why deny people the tools to work around it?

Or, why deny people to use terms but give clearer intentions about what 
*they* meant by the term *in that context*?

I mean, really, either way is breakable. I think mine is a tad less so, 
but I don't really know. Social factors will override. But why then all 
the doom and gloom in support of your side? "NO! NOT WORLD WILD 
CONFUSION!!!!" But that's the case if people change their meanings. 
It's also annoying if people all use only their own terms.

> However, I see no reason to worry about this unlikely scenario. It 
> will become a serious problem only if one party has an economic stake 
> in deliberately trying to confuse or sabotage the software used by the 
> agents of other parties.

Eh. Link rot rules :)

> I have never seen any suggestion of why or how this is likely to 
> arise; on the contrary, in fact: most of the proposed uses of the SW , 
> for example B2B and Service applications, clearly have all parties 
> interests being served by everyone else grokking their intended 
> meanings as accurately as possible. All the economic pressures on both 
> sides converge towards increasing the convergence of meaning attached 
> to URIs in mutual use. So I think this will just happen: the world 
> will find ways to make it happen.

Then we're back to my quietist position: Don't Say Anything. :) Why do 
you care if people use sally:Widget to talk about what they're 
exchanging but aren't using sally's ontology? The economic pressures 
will cause converange, etc.

> All we have to do at this stage is to kind of point the way,

Why?

>  and not to say anything too damn silly which might stop it happening.

I fail to see that I've proposed anything silly. :)

> We don't have to make sure that the sky isn't falling.

That's for sure, since it's the *ground* that's rising UP!

[snip]
>>  Real meaning of importing? I thought importing was quite clear.
>
> No, its not at all clear. Suppose ontology A imports B, using a URI to 
> point to where B is published; and then suppose that the owners of B 
> update the ontology at that URI. Has A changed, then?

This is no better for the other side of the vs.

I totally agree. This is part of my motivation for this example.

> If so, have all other import-chains which include B also changed? What 
> if a user of A had archived the imported ontology; is their version of 
> A rendered out-of-date by the change made to B? Apparently, the owners 
> of B can sabotage all ontologies which import their ontology, without 
> the owners of those ontologies even being aware of it.

This is, of course, worse if mere URI use entails importation (of some 
kind).

So I think we actually are starting to converge on something.

> Now, one might reply: the Web is like that: there can be 404 errors, 
> and so on. But an imports statement can have a much larger and more 
> significant effect than a mere hyperlink in hypertext.

I thought I was arguing that ;) I read Tim's "using a uri commits me to 
the ontology of the uri owner" as *being* "i must do something like a 
'fresh' owl:imports of that ontology". I'll take happy corrections.

> This raises a host of issues which have not been adequately addressed 
> yet.

Sure.

> The OWL semantics hedges around all this stuff by talking about an 
> ontology being "consistent with the Web", but if we gave owl:imports a 
> genuine model theory (which it emphatically has not been) then these 
> issues would have to be made explicit.
>
>> Could you separate out the issues a bit?
>
> Hope the above gives an impression. There are others but its getting 
> late.

Yep.

Here's perhaps some meat (or meet) for the first telecon: using vs. 
importing and what is it to commit to an ontology (not what *commits* 
you, but what do you have to do if you are committed).

If we could get this even begun to be sorted, I'd be very pleased.

Cheers,
Bijan.

Received on Tuesday, 23 September 2003 01:15:34 UTC