- From: Antoine Zimmermann <antoine.zimmermann@emse.fr>
- Date: Sun, 13 May 2012 08:57:10 +0200
- To: public-rdf-wg@w3.org
A huge +1 to everything Pat says in this email. Le 13/05/2012 08:45, Pat Hayes a écrit : > > Apologies for not getting involved with this thread earlier, I > somehow missed it while travelling. > > I agree that people who know nothing about reasoning and logics may > find model theory a bit opaque at first. It may be a way of thinking > that they are not used to. On the other hand, it can be reasonably > claimed that without a basic grasp of at least the basic ideas of > interpretation and satisfaction, they will never be able to > understand how to implement a modern reasoning engine. And I would > suggest that without this basic grasp, they will never be competent > ontology engineers. And also that it is not exactly rocket science. > > The problem with making the rules normative is that these rules are > NOT in fact a good way to implement a reasoner. Any sensible reasoner > will not use rules like se1 or se2, which (as the text explains) will > generate all kinds of redundant and pointless conclusions, or lg and > gl, which are much better handled by allowing literals into the > subject position. But in any case, the whole approach which is > suggested by these rules and the entailment lemmas, of checking > entailment by applying rules to exhaustion, is extremely naive and > cannot even be used for more expressive languages like OWL. > Encouraging implementors to use this as a viable approach is leading > them into a dead end. (I now regret even mentioning rules in the > semantics document, and especially suggesting that they could be the > basis of an implementation. I had no idea people would take them this > seriously. I think they should be in a separate doument entirely, > perhaps as part of the test cases, and no claims should be made as to > their completeness, and no long and extremely opaque (and flawed) > completeness proofs should be included, even in an appendix. Nobody > gives a damn about completeness in any case.) > > I am sympathetic to the idea that RDF should simply not have a > formally defined semantics at all. This would solve all of these (and > many other) problems at a stroke, and I could get on with other > things in my life. However, if it does have a semantics, then I don't > think anything other than a model theory can work. There are too many > alternative ways to implement a reasoning engine for any one of them > to be taken as normative; and in any case, model-theoretic semantics > has been the accepted norm for semantic specifications in > linguistics, logic, database theory and the theory of computation now > for about 40 years, and for very good reasons; it seems crazy and > close to irresponsible to reject it for the foundational language of > something called the *semantic* web. > > For example, who or what determines that a given rule is valid or > correct? If we put a typo into an inference rule, or simply forget > some special case, and a later reader notices this and publicises the > error (as has indeed happened), then what basis do they have for > claiming that the rule is wrong? If the rule itself is normative, > then it CANNOT be wrong, even if it sanctions invalid inferences. > This very notion of validity is a model-theoretic notion (it is that > the antecedents can be satisfied by an interpretation which makes the > conclusion false). Without this, we really do not have a *semantic* > basis for judging the correctness of entailments. > > Further detailed comments in-line below. > > On May 4, 2012, at 7:27 AM, David Wood wrote: > >> On May 4, 2012, at 07:53, Richard Cyganiak wrote: >> >>> Hi Peter, >>> >>> On 4 May 2012, at 09:52, Peter F. Patel-Schneider wrote: >>>> Well I think that it would be a horrible idea to deemphasize >>>> the semantics. >>>> >>>> I'm also completely unsure as to what name could be any better >>>> than RDF Semantics. >>> >>> Let's consider who reads the RDF Semantics document, and for what >>> reasons. From what I gather, we can identify a number of >>> different cases. I'm sure there are many more: >>> >>> 1) Spec writers read it, in order to understand how to make their >>> specifications compatible with other specifications in the RDF >>> stack. For example, neither the RDB2RDF WG nor the SPARQL WG >>> could have done its work without knowing the things formally >>> specified in RDF Semantics. > > OWL could have, but then OWL would have been entirely disconnected > from RDF. > >>> >>> 2) Implementers of various specs (incl. RDF, RDF Schema, SPARQL) >>> read it, in order to understand how their systems should behave, >>> especially with regard to datatypes and RDFS inference. > > If they are seeking a guide to how any system should *behave*, then > they need to understand that semantics is not about behavior but > rather about truth, satisfaction and consistency, and these ideas > relate to inference-making only indirectly. They need to grasp the > relationship between inference and semantics (eg the notions of > validity and entailment) in order to relate inference behavior to a > semantic description. If they do not understand this, then they > probably shouldnt be trying to implement a reasoner. If semantics is > too opaque, the many cases of entailment and non-entailment in the > test cases document give a pretty good idea of most of what they > need. > >>> >>> 3) Authors of RDF Schema vocabularies read it, in order to >>> understand the consequences of declaring domains, ranges, >>> subproperties and so on. >>> >>> 4) RDF newbies read it, because this is where they happened to >>> end up after some googling or link-clicking and they don't >>> understand the big scheme of things yet. > > The second paragraph of the document starts "This is one document in > a set of six (Primer, Concepts, Syntax, Semantics, Vocabulary, and > Test Cases) ..." and they might then hazard a wild guess that the one > called "Primer" might possibly be a better place to start. > > If they habitually read documents without checking the first page, > then I really don't think we have any responsibility for what kind of > a mess they get into, frankly. > >>> >>> How well does the current document serve these needs? >>> >>> For 1), I think it works well. >>> >>> For 2) and 3), I think the document as it stands does not serve >>> these readers well, and it could do a *much* better job. We have >>> reports that such readers tend to find the informative entailment >>> rules in Section 7 extremely useful, and much of the rest of the >>> document rather impenetrable. It has thus been suggested that the >>> entailment rules should be given more prominence. One minimally >>> invasive way of achieving this would be to just move the words >>> “normative” and “informative” around a bit. I suppose this is >>> what Ivan refers to when he says “reorganize to make the rules >>> normative and deemphasize the model-theoretic semantics”. >> >> *Personally*, I agree with this as an implementer of specs and >> author of vocabularies. > > What would it mean to make those rules normative? Would an efficient > tableax-based reasoner be then illegal? What about an modified > resolution-based FO reasoner adapted to RDF? What about a rule engine > which used a different suite of rules (there are many other rules > that could be used and would give similar results.) What about an > implementation based on Prolog? It is crazy to try to make one > (flawed and inefficient) implementation technique be the normative > definition of a semantic standard. > >>> For 4), the document as it stands doesn't work at all, and it >>> can't, because that's not its purpose. In fact, when newbies try >>> reading this document, it's pretty much guaranteed to end in >>> disaster. > > That depends entirely on the background that the newbie happens to > have. > >>> I think we all agree that this is *not* the document you should >>> be looking at in your first encounter with RDF. > > If you know nothing about logical methods, inference engines, machine > inference? Yes, it might not be a good starting point if you are this > ignorant, indeed. I did try to write some introductory, almost > tutorial, material in the first sections, but no doubt these are not > often read. > >>> There isn't *too* much we can do about this, but I think that a >>> title that is a bit scarier to newbies could help. Can we put >>> something like “model theory” or “formal representation logic” >>> into the title? >> >> This problem could be solved easily with the insertion of some >> language in the introduction pointing to the Primer in the first >> paragraph (instead of just the Vocab and Concepts, as it does >> now). > > See above. It refers to the primer in the second paragraph of the > document. > >> >> Regards, Dave >> >> >>> >>>> Of course, if the WG wants to make RDF no longer be a formal >>>> representational logic, then .... >>> >>> I thought RDF is a data model? > > It is both. Data models ARE formal representational logics (ref. > Codd.) > > Pat > >>> >>> Best, Richard >>> >>> >>> >>>> >>>> peter >>>> >>>> >>>> On 04/25/2012 11:27 AM, David Wood wrote: >>>>> Hi Peter and Pat, >>>>> >>>>> The RDF WG briefly discussed the need for an RDF Semantics >>>>> editors draft at today's telecon. I am aware that there are >>>>> a lot of open issues and therefore hard to produce a draft, >>>>> but perhaps it makes sense to have a single document that >>>>> lists the issues in one place. >>>>> >>>>> In any event, we would like to discuss this at next week's >>>>> telecon if you can make it. Thanks. >>>>> >>>>> Relevant comments from IRC (no log published yet since the >>>>> meeting isn't over): [[ ivan: one thing that came up early >>>>> was discussion to change title of RDF Semantics document, >>>>> reorganize to make the rules normative and deemphasize the >>>>> model-theoretic semantics AlexHall @ 11:20 ... think it's a >>>>> good thing to do but huge amount of editorial work AlexHall @ >>>>> 11:20 cygri: is there an editors draft of RDF Semantics yet? >>>>> 11:21 [no] 11:21 cygri: given that there are larger changes >>>>> to the doc, would feel better if there were an editors draft >>>>> by now. 11:21 guus: suggest we should put it on the agenda >>>>> for next week ]] >>>>> >>>>> Regards, Dave >>>>> >>>>> >>>>> >>>>> >>>> >>> >> >> >> > > ------------------------------------------------------------ IHMC > (850)434 8903 or (650)494 3973 40 South Alcaniz St. > (850)202 4416 office Pensacola (850)202 > 4440 fax FL 32502 (850)291 0667 > mobile phayesAT-SIGNihmc.us http://www.ihmc.us/users/phayes > > > > > > > -- Antoine Zimmermann ISCOD / LSTI - Institut Henri Fayol École Nationale Supérieure des Mines de Saint-Étienne 158 cours Fauriel 42023 Saint-Étienne Cedex 2 France Tél:+33(0)4 77 42 66 03 Fax:+33(0)4 77 42 66 66 http://zimmer.aprilfoolsreview.com/
Received on Sunday, 13 May 2012 06:57:54 UTC