Re: [ANN] MOAT

On 21/01/2008, Frederick Giasson <fred@fgiasson.com> wrote:
> Hi Peter
<snip>
> > Do meanings have to be approved or added by a manager before they can
> > be used? I would prefer not. Not sure if that is what you mean by
> > semi-supervised though.
> >
> >
>
> No, at least, I don't think so based on what Alex wrote on MOAT's web site.
>
> It is why each meaning is linked to a person that linked the tag with
> its meaning. If you check the cardinality of a Meaning, then you will
> notice that there have to be a "maker".
>
>
> Semi-supervised system because it seems that a MOAT server send you a
> list of possible meaning for each of your tag, then you choose the ones
> you want to link a meaning with a tag.

Its not planned to quite work that way. The initial choices can be
ignored and the user can insert their own URI to use instead. And the
ontology should be broad enough to allow this.

> > I would prefer the semantics to be completely on the fact that the tag
> > is a moat:tag with no particular relevance to the outside evidence,
> > other than that it can be explored if desired. In terms of triples,
> > the fact that the tag has a meaningURI may be more generic than SKOS
> > seems to be, but it makes the system very useful IMO.
> >
> >
> Not the tag that has a meaningURI, but a "meaning" :)

Are users going to spend time philosophising over whether the URI they
decided to link to their post via the string has a deep and meaningful
relationship.

> But the real question is there: does a meaning is a concept only, or
> anything? It is the question we have to investigate with pros and cons :)
>
> > Does it allow arbitrary URI's to be inserted as meanings? If it
> > doesn't then it seems to be at least a little more complex, or less
> > expressive depending on how you look at it.
> >
> >
> Currently yes it does. It is one of the problem I have with it.

I don't see why the ontology should restrict the use of a system which
otherwise enables users to link together the semantic web in new ways.

> >> Question: can a named entity mean something (so, being related as the
> >> essence of a meaning (like in MOAT)).
> >>
> >
> > Its a philosophical question, and I don't think it should be enforced
> > at the moat ontology level. I would hate to break someones reasoner by
> >
> It could at some level, explicitly by not authorizing anything to be a
> meaning. So this could helps reasoners a bit.

What reasoning operations do you wish to perform on sets of tags? How
exactly would it help to restrict users to sets of URI's which they
can't customise or relate to other objects?

> > If in their case they meant to relate physics to einstein in
> > particular I could see why they may want to do it. In another case
> > someone may tag physics and relate it to condensed matter physics,
> > which is again reasonable. It would still be possible to do this in a
> > restricted skos system BTW, so the issue isn't fully solved.
> >
> >
> No, it is not :)
> > >From my perspective it would be just as useful to use a URI without
> > having to either know that it conforms to skos ways, or that it has a
> > definite meaning implied past being effectively linked with a short
> > tag description.
> >
> >
> Possibly. But in that case, you can query the system to get a list of
> tags related to a specific meaning. This is where the whole thing is
> powerful! :)

You assume that the input data is clean. If you have professionals
tagging objects why not use the SKOS ontology, which provides for very
structured sets of knowledge. Otherwise you have users who just want
to do things easily to the best of their liking, not using your
classification scheme.

Querying back from a list of tags to a specific URI may be useful...
You would need rather large sets of data to do this, and you would
have had to restrict your given meanings in the first place to avoid
an open world where users do what they wish.

> > I envisage moat to be used for effectively linking the semantic web
> > together with folksonomies without people having to do much more than
> > they currently do. I don't see a particular categorised web happening
> >
> There is one more step though (linking a tag with meaning(s)

I would say more it is the users impression of the tag fits with the
meaning behind the given URI. I would prefer to start at the currently
very useful looking user interface which looks just like an expanded
folksonomy and work from there.

> > because of it, just nice seeAlso type links that people can follow to
> > find more information. It would be nice not to have to create another
> > ontology, or violate an existing one in order to perform general
> > linking using the system.
> >
> >
> But there is no meaning with a seeAlso. Possibly if we think about
> browsing, but not really if we think about querying the dataset.

I don't like seeAlso because of that, but implying that users are
actually performing reliable ontology building when they are novices
and don't actually want to formally structure their world doesn't seem
to be the goal of this ontology. It is quite possible that you could
query the dataset and get conflicting meanings, but if you browsed the
dataset as a user or interested observer you would gain knowledge from
it even though it has logic contradictions.

> > Semantic web should be just as democratic as the normal web (to put
> > some political opinions into it) :)
> >
> >
> Sure, but it doesn't mean that we shouldn't take some time to get things
> done properly :)

I think "properly" implies that the development of the moat ontology
has the same effect as the development of a categorisation system like
skos. I see it as providing the basis for users to partially structure
knowledge themselves, but not necessarily in a logically meaningful
way.

An example may be that if there are formal restrictions on people
using terms in conflicting ways then a user may come to a system which
has a string linked to one meaning, which is inferred to be a
skos:broader term of something else, and they think it isn't, which
makes them define the term in a conflicting way, then your inferences
are not really proofs from that point on, but you could still
pragmatically use the dataset for browsing, and offering non-binding
annotations for resources. Linking unstructured folksonomies into
classification schemes is bound to be unuseful in the end.

Binding up knowledge forcefully in ontologies is possibly why seeAlso
as a non-binding contract is used so much. Ontology designers (and
moat users) really should have a more expressive set of non-binding
contracts, and I see moat as a way to provide that, in part.
Non-binding contracts by definition won't break a computer inferencer
when it comes across this unstructured knowledge, which it would been
programmed to see as structured otherwise.

Peter Ansell

Received on Tuesday, 22 January 2008 23:17:14 UTC