- From: pat hayes <phayes@ai.uwf.edu>
- Date: Tue, 4 Mar 2003 12:18:17 -0500
- To: Graham Klyne <GK@NineByNine.org>
- Cc: Tim Berners-Lee <timbl@w3.org>, Dan Brickley <danbri@w3.org>, Brian McBride <bwm@hplb.hpl.hp.com>, pat hayes <phayes@ai.uwf.edu>, Bijan Parsia <bparsia@isis.unc.edu>, RDF Core <w3c-rdfcore-wg@w3.org>
>At 10:31 PM 3/1/03 -0700, pat hayes wrote: >>> Something has one meaning. >> >>This isn't true even in ordinary human discourse in natural >>language. There just is no such notion of a single 'one meaning'; >>the idea isn't coherent. > >Pat, > >I'd like to explore this a little, noting that RDF is not "ordinary >human discourse in natural language". > >Stepping back from meaning that can be formally specified, I'd like >to try a notion of a single meaning (in RDF): >(a) each URI used has exactly one denotation >(b) each URI used as a property has just one relational extension > >(A consequence --for which this is sufficient but not necessary-- >would be that the truth or falsity of any expression using a given >vocabulary of URIs is fixed.) > >I'm not attempting to describe a specific interpretation in the >model theoretic sense. Well, in fact you are. That is, if you fix the denotations and the relational extensions, you've fixed everything. >Though I suspect that there can be only one such interpretation (on >a given vocabulary) that could concur with such a meaning. Right. > >We have no way of fully specifying such a "one meaning", but I think >it's reasonable to allow that we can develop successive >approximations that converge (asymptotically, as it were) to such a >meaning. Well, maybe that is a reasonable idea, but there are several limitations to it, since one can easily show that there is never a single such satisfying interpretation as far as the MT is concerned. So the single intended model has to be defined by some extra-MT-ish constraints. More seriously in the context of the original thread, this entire picture assumes that it makes sense to think in MT terms over a range of languages. This is indeed possible with RDF/S/OWL-Full, but (a) setting this up was a major piece of work, and amounted to making these into a single language; it is meaningless when we are supposed to have a single meaning which crosses arbitrary language boundaries; and (b) it may well not be possible for future languages, or extensions which take the RDF/S/OWL axis in some other direction (eg some of the 'rules' proposals). Pat > >This may not be a useful idea, but I'm trying here to see if there >may be ways to reconcile what I think are two apparently-reasonable >viewpoints. > >#g >-- > >At 10:31 PM 3/1/03 -0700, pat hayes wrote: >>>Jeremy Carroll wrote: >>> >>>> >>>>tbl: >>>> >>>>>>>2. The meaning of the statement is defined by the definition >>>>>>>of the predicate, as applying to the subject and object >>>>>>>identified by the >>>>>>>definition of the subject and object terms. >>>> >>>>Danbri: >>>> >>>>>>This for me is the crux: do we mean the machine oriented 'definition' >>>>>>in RDFS or OWL or N3, or some more rounded/scruffy/social >>>>>>notion of definition. >>>> >>>> >>>>I find Bijan's observation compelling >>>>[[ >>>>But there's no vague, much less precise, definition of "defining >>>>information". And I'm a logical reasoner, will this information >>>>be opaque to me? (Well, if in German, yes, but *all* human >>>>reasoners?) >>>>[...] >>>>So it's formal meaning isn't fixed IN ANY WAY by the "authority"? >>>>And the social meaning? >>>>]] >>> >>> >>>There seem to be a confusion here that things have two meanings, a >>>"formal" one >>>and a "social" one. I don't think that is useful. >> >>I think it is essential, although this way of putting it is >>potentially confusing. It might be better to distinguish between >>how much of the meaning is accessible to who and to what. The >>'formal' meaning is that part which is accessible to software. But >>even the 'social' part, ie all the rest, varies from reader to >>reader. In some cases, a reader might find more meaning than the >>original writer thought was in the document. >> >>> Something has one meaning. >> >>This isn't true even in ordinary human discourse in natural >>language. There just is no such notion of a single 'one meaning'; >>the idea isn't coherent. >> >>>"inverseProperty" can be defined mathemaically, but remember that the >>>mathematical symbols used are probably defined in english somewhere. >> >>That is highly debateable and depends what you mean by 'defined', >>but in any case its irrelevant to the issue here. If your point is >>that *all* meanings are ultimately described in English, that isn't >>true. >> >>>"color" can't be defined formally in terms of mathemaics, unless you have >>>assume a lot of other terms to do with spectral reflectivity and light. >> >>Well, "color" actually can be defined in scientific terms, in fact, >>but you'd be better with an example like "red" which probably can't >>be defined at all. This has got nothing whatever to do with >>mathematics, but it does tend to show that there isn't any single >>meaning to words like color names. >> >>>> >>>>Two points: >>>>- "whatevers available" is simply not clear enough. >>> >>>There are a lot of social systems for relating definitoins to terms. >>>These include domain name owndership, the Web, etc. >>>The web is a big place. Predciates and terms vary enormously. >>>For RDF to be able to describe real things, it is essential that >>>some terms be defined in english. >> >>Why English? And why is this true? You can't define "red" in >>English. And more to the point, maybe, what does 'defined' mean >>here? RDF can't use definitions given in English. >> >>> Look at the cyc ontology. >> >>That is a very bad example for your point: the intended purpose of >>Cyc is precisely NOT to rely on English definitions. The meaning >>of any CYC term is completely defined by the CYC axioms using that >>term (and all linked axioms, ie ultimately by the whole of Cyc.) >>You can strip out all the English comments and the meaning is >>unchanged. The same goes for almost all large-scale ontology work, >>in fact. >> >>>I'm not sure what you are unhappy with, here. >>>Are you saying it is not clear enough? >> >>It certainly is not clear enough. >> >>>Are you saying that >>>it is not clear what the definitions of the terms are? >> >>It is clear that any English definitions cannot be reflected in any >>normative account of meaning which is reflected in any operation of >>any RDF software. IF RDF tries to incorporate any such notion of >>meaning into its spec, then it has just become a joke. >> >>>Are you saying that the english definitions should not be allowed? >> >>Allowed in what sense? What I am saying is that allowed or not, >>they are not the slightest actual USE. Any sense of 'meaning' which >>depends on them isn't going to influence in any way what any piece >>of software does to the RDF. And since the point of the spec is >>largely to help writers of software, referring to something that is >>necessarily irrelevant is either pointless or actively harmful. >> >>>Or do you want a clean algorithm for determining which >>>english documents define a given term, from the web? (That we >>>could probably >>>arrange.) >> >>That would be very interesting. I doubt if this can even be made >>precise enough to be meaningful, let alone provided as an >>algorithm. And in any case, suppose you could. Now, how is my RDF >>engine going to read and understand those English documents? >> >>>> >>>>- RDF has decided to avoid the notion of definition for the >>>>formal semantics, we shouldn't then have it in the informal >>>>semantics. >>> >>>Well, every specification upon which the web has depended up till >>>now, including >>>Ethernet and unicode and TCP/IP and HTTP has had the meaning of its terms >>>and structures explained in english, informally. These specs have been used >>>to build software, resolve many discussions, and so on. >> >>Yes, but this reply misses an essential point. The part of those >>specs whose meaning is fixed between software apps is the part that >>can be specified in the specs. None of those specs have set out to >>define a general meaning-carrying representation. In the case of >>ontology languages like RDF, the common part that can be defined by >>the spec is the *general rules* for meanings, ie the semantics, NOT >>the 'meaning' of particular RDF URIrefs. The spec says nothing at >>all about what <ex:myUri> 'means', and if you write a document in >>English explaining what its supposed to mean, then its not the >>slightest use or relevance, since no piece of software on the >>planet in the forseeable future is going to be able to read your >>English 'definition'. >> >>> There are a mass of >>>RDF schemas and related documents going to be written -- but it >>>needs the RDF >>>spec to pass on the authority to them to define their fields. >> >>I don't see how the spec of a language can, or should, pass on any >>authority to define anything. It didnt have the authority to define >>the meanings of any items not in its namespace in the first place. >>What it can do, and does, it specify how to characterize the >>content of any piece of the language, so that definers of meanings >>can determine how to constrain those meanings using the language. >>That is what the model theory sets out to do. >> >>>Just because *some* aspects of the meaning of *some* RDF terms can >>>be expressed formally >> >>THOSE ARE THE ONLY ASPECTS OF MEANING DEFINED BY THE SPEC. If you >>want the spec to define other aspects of meaning, please tell us >>how to write it (the spec) so as to refer to those other aspects of >>meaning. Its not good just using words like "meaning" and >>"definition" without saying what we mean by them. Words like this >>don't have exact enough meanings to use in a specification. >> >>> does not remove the duty of the RDF spec to >>>say what an RDF document means. >> >>The SPEC cannot possibly say what a particular RDF document means, >>any more than a dictionary can tell a story. It can only give >>general rules for attaching meanings to documents, which is what >>the semantics does. >> >>>The formal semantics cannot define "color". >> >>Right, and "color" can't be defined in RDF. >> >>>Suppose I send you an RDF document syaing (in n3) >>> >>><http://example.info/ips/gg5#y004> <http://example.com/dsaf#enFap> "176". >>> >>>How would you know what I was telling you? >> >>I would know that some thing had some property with value '176' (a >>string), and if that's all the RDF I can see, that is ALL I know. >>If you want me to know more, you had better send me some more RDF. >> >>>How would someone who had not heard of RDF before? >>>The mime type would take them to the RDF spec and -- then what? >> >>The above is what I would learn from the RDF spec. Of course the >>RDF spec can't tell me what you mean by >><http://example.info/ips/gg5#y004>; and you might tell ME what you >>mean in English, but (this being the semantic web) that's largely >>irrelevant; the question at issue is what some piece of software >>acting on my behalf can get out of it. If its written in English, >>the answer is, nothing. >> >>>>For me, either of these is fatal. This cat has had its nine lives. >>> >>>Fatal for the idea of defining what an RDF document means? >>>How sad. >>>In that case, I suppose we had better start all over again, as >>>we have ended up with a languge of meaningless documents. >> >>You can start over all you want, but you will not get anything much >>better than this (except in the sense that OWL is better than RDF, >>and full FOL would be better than OWL). To get better than this you >>will need to create a web of movie-style Artifical Intelligences, >>and you won't get that done by a W3C working group. All languages - >>even human languages like English - are 'meaningless' in some very >>strict sense. Their meaning is what a cognitive agent can get out >>of them, and RDF agents - in fact, any software agents that we know >>how to build - have pretty limited cognitive powers. >> >>>If RDF is only be to be used to encode mathmeatical >>>formalisms, and not information about the real world, >>>do we need another langauge to express data? >> >>This discussion has nothing to do with mathematics versus the real >>world. Model theory is about worlds, including the real world. The >>point at issue is HOW MUCH INFORMATION is encoded in some RDF; and >>the answer is, rather little. But we knew that up front, before we >>started. It is obvious that RDF cannot encode the kind of >>information that humans can send to one another using languages >>like English, in a form useable by software agents. But that's not >>a failure of RDF: *nothing* can do this. To do this would require >>us to be able to provide software with human-level cognitive powers. >> >>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@ai.uwf.edu http://www.coginst.uwf.edu/~phayes >>s.pam@ai.uwf.edu for spam > >------------------- >Graham Klyne ><GK@NineByNine.org> >PGP: 0FAA 69FF C083 000B A2E9 A131 01B9 1C7A DBCA CB5E -- --------------------------------------------------------------------- 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@ai.uwf.edu http://www.coginst.uwf.edu/~phayes s.pam@ai.uwf.edu for spam
Received on Tuesday, 4 March 2003 12:18:58 UTC