Re: SUO: Multi-Source Ontology (MSO) Draft Ballot Question
Philippe,
I may have misunderstood the purpose of this work. As a set of
relations between ontologies, as an aid to a human's understanding of them,
it sounds very useful. But then I'm wondering why it's being proposed as a
starter document. If it lacks information from the ontologies it uses as
sources, and adds no new information, why is that suitable for a starter
document?
Another point is that WordNet and SUMO are two very different products
that one shouldn't mix together. WordNet synsets are language constructs,
not concepts, and the links among synsets are correct linguisticly, but not
philosophically. SUMO on the other hand is not language dependent, and
defines concepts which are valid philosophically, but may not have direct
correspondence to any particular lexical item. I fear that putting them in
the same representation language could cause considerable confusion.
As an example, take the inference path that is possible from the SUMO
term Motion to the WordNet synset "motion, movement, move" then up the
hypernym links to "change", "action" and then "act, human action, human
activity". Through that faulty inference chain, one could conclude that
any Motion is an intentional human action, which of course is
false. Hypernym links are linguistically valid, but not logically
valid. Maybe you're "corrected" many of these WordNet links, but it's not
clear to me the utility of that enterprise. I've seen so many papers
detailing "corrections" to WordNet, which, although valid as far as they
go, miss the point. WordNet wasn't intended for logical inference, so if
you want a product for doing inference, you're better off with a product
created for that purpose.
Adam
At 05:29 PM 1/28/2004 +1000, Philippe Martin wrote:
>Adam,
>
> > I'm puzzled by this product and its description. It appears to
> > contain some terms from SUMO, but not their axioms, so I'm not clear
> > to what extent an integration is in progress.
>
>I meant that I am currently inserting those terms and most of their
>associated axioms. Give me about 2 more weeks for the SUMO.
>My only problem is time (as I am now also looking for a job).
>
>Then, what would be interesting is that certain persons check the
>links I have set between categories from different ontologies.
>E.g. Stefano could check my links from DOLCE categories to
>other categories. Matthew West is checking my links from categories
>of the LIS. Every validated link would help to access/retrieve and
>define/understand/re-use the linked categories. Rejected links would
>also help the understanding of the categories if I store these
>"mistakes" (the wrong links) within comments.
>
>I represent many of the axioms in FT (a simple and normalizing notation
>For Taxonomies), e.g.
>(i) axioms using the relations instance, subclass and equal are
>represented via the links ^, < and =,
>(ii) those using disjoint and partitionare represented via {..}, {(...)},
>(iii) those using domain, range, OneToOneFunction, ... are represented
>via relation signatures (which may include templates and cardinalities).
>A grammar of FT is at http://www.webkb.org/doc/F_languages.html#FT
>
>More complex axioms require another notation. I could represent them in
>FCG (the native formalism used in WebKB-2) or extend my import and export
>procedures to the KIF and ontological primitives used in the SUMO.
>However, it is unlikely that the SUO will adopt FCG, and WebKB-2 will not
>exploit the SUMO axioms in FCG for constraint checking. (WebKB-2 is
>mainly intended to support Natural Language representation and knowledge
>normalization+sharing+retrieval, it is not a general inference engine).
>Hence, in the short term, I do not think I'll include complex axioms
>from SUMO.
>On the other hand, if there is a need for it, I can propose the browsing
>or export of the content of WebKB-2 into the KIF and ontological primitives
>used in the SUMO. CGIF and RDF+DAML+OIL can already be used for displaying
>the taxonomy.
>
>
> > Without many of the links already in WordNet
>
>Do you refer to the verb/ajective/adverb-related parts of WordNet?
>Adding these parts to the MSO of WebKB-2 is a medium-term goal.
>
>
> > or the logical relations in SUMO, how does this proposal address
> > the need for inference stated in the SUO charter?
>
>Your general inference engine can still use the complex SUMO axioms
>(most of which are usually hard-coded in inference engines) plus
>others from the MSO of WebKB-2.
>
>
> > SUMO has already been linked to WordNet. If the WordNet linguistic
> > hierarchy has been "corrected", then it may well be that many of the
> > links from SUMO to a changed version of WordNet will no longer be valid.
>
>The removal of inconsistencies (e.g. violated exclusion links) or
>redundancies (e.g. redundant subtype, member, part links) and the
>modification of mis-used links (e.g. subtype links used instead of
>locationOf links) can only be beneficial. (For a listing of the
>updates, see http://www.webkb.org/doc/wn/wnSemanticCorrections.html).
>
>Yes, the meanings of some WordNet categories have been precised
>to correct confusions or via the adding of a (much needed)
>structured top-level. However, I have avoided changing these
>category meanings as much as possible (1.5 pages of my article at
>http://www.webkb.org/doc/papers/iccs03/ is devoted to that point,
>and for example explain why I have not r-ecategorized the intermixed
>categories for attributes and measures of WordNet under the
>exclusive types dolce#quality and dolce#region, i.e. as qualities or
>quality values).
>
>It is unlikely that those changes will affect the links you refer to
>between SUMO and WordNet, but when they do, this will be for the better.
>
>
>
> > Since SUMO itself is a full version with axioms, what are the purported
> > benefits of this work over SUMO?
>
>All that has been added (links to and between categories from other
>non-completely overlapping ontologies) and corrected or extended (mainly
>the noun-related part of WordNet).
>However, there are no additional explicit total definitions of primitive
>categories, and hence currently, from a certain logical viewpoint, the
>ontology itself (not WebKB-2) brings nothing.
>
>
>
> > Does this proposal contain any content not already in SUMO or WordNet?
>
>Every category, link or graph that has a source/creator different from
>"sumo" and "wn" (and excluding categories for some domain ontologies).
>Compared to the number of categories and links in WordNet, that is not
>a large number but then the same thing can be said about the number of
>classes and relations in SUMO.
>
>
>Philippe