[PragmaticWeb] Semiotic metamodel : Concept, Thing ans Sign
Sebastian Samaruga
cognescent at gmail.com
Wed Jan 14 01:03:50 CET 2015
I've updated the repository and now there is an executable WAR file for
testing in servlet containers. There are also the complete sources which
haven't been correctly uploaded.
Project: https://code.google.com/p/cognescent/
(in source section are instructions for checking out via svn).
Best,
Sebastian.
On Sun, Jan 11, 2015 at 2:24 PM, Sebastian Samaruga <cognescent at gmail.com>
wrote:
> Yes, is true. I haven't reached at that point yet and I understand that
> recording the inferences I make in OWL would be a lot helpful, mostly
> because I try to merge diverse sources and also augment knowledge with
> linked data and upper ontologies. My reason for haven't using those
> features yet is because I only use 'plain' RDF to infer metadata. Only SPO
> triples with no schema information are my sources of data. And, once there,
> my model allows the assertion of one subject in one role or occurrence to
> be referred under some kind of type (or, at least, I try).
>
> But it will help a lot if, once I infer properties, types and class
> restrictions, I put them in OWL so I can reuse them later or use them for
> further inference. Thanks a lot for your advice, I'll taking it into
> account. And, as I said before, my tools and my skills limit me somehow so
> your knowledge is really helpful.
>
> Best,
> Sebastian.
>
>
> On Sun, Jan 11, 2015 at 3:58 AM, Eric Prud'hommeaux <eric at w3.org> wrote:
>
>> * Sebastian Samaruga <cognescent at gmail.com> [2015-01-10 18:22-0300]
>> > Hi, sorry for the delay in the response too.
>> >
>> > Yes, maybe I'm ambitious. And maybe I don't have all the tools/skills I
>> > need at hand when I try to develop or communicate my ideas.
>> >
>> > I was surprised when someone told that only using RDF Schema is
>> possible to
>> > infer types. Actually, I'm currently inferring type information on
>> > arbitrary sets of entities using only their properties information.
>> When I
>> > know that many subjects share the same set of properties I can be
>> allowed
>> > to think that they belong to the same class. And when they share the
>> same
>> > values for their properties, I can infer they are in a similar 'state'
>> > (i.e.: single / married, student / employee, etc).
>>
>> Minor (perhaps orthogonal) point: RDFS doesn't enable one to infer
>> types in the sense that you are using; it simply has some rules used
>> to infer types, e.g.
>> { W <p1> X . <p1> rdfs:domain Y . } implies { W rdf:type Y } .
>> { W <p1> X . <p1> rdfs:range Z . } implies { X rdf:type Z } .
>>
>> OWL enables you write nuanced rules, e.g. every X of type Doctor who
>> is treating some patient most also be a PracticingDoctor. This may
>> address some of the "context" which Pat warned you about below by
>> capturing the constraints for that context in a computable structure.
>>
>> If you know in advance the properties you are examining and the
>> criteria for inferring types, you can probably record most of that in
>> OWL. If your process examines a graph and clusters nodes based on
>> similarity, (perhaps with some human oversight looking at the
>> instances to say "I'll call those PracticingDoctors", you can probably
>> record those discovered criteria in OWL.
>>
>> Take a peek at the OWL primer to see if it applies to your work
>> <http://www.w3.org/TR/owl2-primer/>. Note that
>> <http://www.w3.org/TR/owl2-primer/#OWL_Syntaxes> provides a switch for
>> what languages are displayed; for this forum, turtle syntax will reach
>> the broadest audience.
>>
>>
>> > Using RDF as the core format where many data sources converge, and this
>> > kind of metadata information from the RDF I'll try to observe
>> 'Concepts' to
>> > which classes/types belong. It's a kind of role or meta class where some
>> > instances of Things are classified. Things are abstract subjects of
>> > discurse of what references or Signs can be made of. For example, Woody
>> > Allen has an occurrence (manifestation of existence) as the director
>> > (concept / role) of some movie (another Thing).
>> >
>> > The goal is to be able to take note of all of those relationships in a
>> kind
>> > of index where equivalences between different sources of data talking
>> about
>> > the same subjects, in a different vocabulary, can be deduced. This is
>> the
>> > intention of many Semantic Web projects already, and this metadata
>> > (de)aggregation would benefit a lot from linked data sources.
>> >
>> > Luckily I could develop an (ongoing) proof of concept project of what I
>> > propose, which I'm hosting at googlecode:
>> >
>> > https://code.google.com/p/cognescent/ (checkout sources only in the CBI
>> > folder)
>> >
>> > It's a Java Web Application project with instructions in how to exec in
>> the
>> > README. It's a work in progress. It provides an OData frontend, browse
>> > service document, instances and metadata, all from any RDF datasource.
>> It
>> > can also be configured to use a relational database modifying Main.java
>> > manually (for now). It translates the relational data to RDF to perform
>> > model set up.
>> >
>> > Best regards,
>> > Sebastian.
>> >
>> >
>> > On Fri, Oct 10, 2014 at 10:20 AM, Thierry BIARD, EIRL AMBESAS <
>> > thierry.ambesas at gmail.com> wrote:
>> >
>> > > Hi Sebastian,
>> > >
>> > >
>> > >
>> > > I hope you are fine. Sorry for the delay of response. As far as RDF is
>> > > concerned, I am not good enough to discuss this matter. I suppose
>> that Pat
>> > > Hayes is an expert and that his comments are relevant and will help
>> you on
>> > > your ambitious demarche. You are not of those who give up. I know
>> that you
>> > > will continue.
>> > >
>> > >
>> > >
>> > > Best regards.
>> > >
>> > >
>> > >
>> > > Thierry
>> > >
>> > >
>> > >
>> > > *De :* Sebastian Samaruga [mailto:cognescent at gmail.com]
>> > > *Envoyé :* mercredi 17 septembre 2014 19:55
>> > > *À :* Pat Hayes
>> > > *Cc :* Thierry BIARD, EIRL AMBESAS; semantic-web at w3.org;
>> > > pragmaticweb at lists.spline.inf.fu-berlin.de
>> > > *Objet :* Re: Micellaneous
>> > >
>> > >
>> > >
>> > > Ok, I should have stated that this is intended to build an API in some
>> > > language to perform the mentioned operations.
>> > >
>> > > Aggregating URIs means that if they occur multiple times in a closed
>> world
>> > > model in which I know all the statements then I can be sure that they
>> > > represent the same thing. Another task of a given ETL tool could be
>> perform
>> > > merging and mapping of some kind, maybe to an upper ontology to
>> facilitate
>> > > this task.
>> > >
>> > > The algorithm for types is there and it is working. Those are inferred
>> > > types and as in the previous paragraph I suppose a closed world like,
>> for
>> > > example, and RDF dump of a relational database. This closed world
>> > > assumption can be relaxed having the possibility of merging multiple
>> > > documents in the ETL component.
>> > >
>> > > Methods and contexts refer to the to be implemented API. Sorry for
>> not be
>> > > clear enouth about that. And Resource class methods (getXXX) receive
>> the
>> > > context as their arguments which can be one of the tree possible
>> arguments
>> > > listed. Subclasses (Concept, Thing, Sign) may override this behavior.
>> > >
>> > > Best,
>> > > Sebastian Samaruga.
>> > >
>> > >
>> > >
>> > > On Wed, Sep 17, 2014 at 2:13 PM, Pat Hayes <phayes at ihmc.us> wrote:
>> > >
>> > >
>> > > On Sep 17, 2014, at 9:30 AM, Sebastian Samaruga <cognescent at gmail.com
>> >
>> > > wrote:
>> > >
>> > > > Thierry, thanks really for your time answering my emails. I didn't
>> take
>> > > the online course. Although I've spending some of my time in some
>> kind of
>> > > semantic (semiotic) metamodel for what I understand as knowledge
>> > > representation..
>> > > >
>> > > > I'm copying this mail to the liists.
>> > > >
>> > > > For what I came with, the core concepts, borrowed from semiotics,
>> are:
>> > > > 1. Concept
>> > > > 2. Thing
>> > > > 3. Sign
>> > > >
>> > > > First, a Sign is an occurrence of a Thing in some context. For
>> example,
>> > > in RDF, some URI referring something that somehow exist.
>> > > >
>> > > > A Thing is something that can exist and, given that notion, can
>> appear
>> > > in occurrences of some kind. A thing also aggregates all of its
>> occurrences
>> > > as they represent the Thing as a whole.
>> > > >
>> > > > Finally, a Concept is an abstraction of a set of Things. Things are
>> > > instances of Concepts and, again, which Concept they are instance of
>> is
>> > > context dependent. A Thing could be regarded as an occurrence of a
>> Concept.
>> > > >
>> > > > So, given that I begin with Things and I need a source of them I can
>> > > begin with some RDF source without the need of any schema embedded in
>> it.
>> > > The URIs in the triples are Signs.
>> > >
>> > > OK so far, although RDF does not support contexts, so a given IRI is
>> > > assumed to denote the same thing everywhere it occurs.
>> > >
>> > > > Aggregating all occurrences of the same URI I came up with the Thing
>> > > they represent.
>> > >
>> > > ? How does that work? What do you mean by 'aggregating'? And how can
>> you
>> > > know if you have all the occurrences of a given IRI, on the entire
>> Web?
>> > > (You can't possibly know this.)
>> > >
>> > > > Then using a simple algorithm of which URIs share the same
>> predicates I
>> > > can infer the type of the URIs and came up with the Concepts.
>> > >
>> > > That is not going to work, because you would also need to know the
>> domain
>> > > and range properties of those predicates, and that information is not
>> > > always given explicitly. At the very least, you will need RDFS or OWL
>> type
>> > > reasoning here.
>> > >
>> > > >
>> > > > Now, the meta model:
>> > > > The core concepts inherits from a Resource class. This parent class
>> has
>> > > the following methods:
>> > >
>> > > "methods"? That term is not meaningful in RDF.
>> > >
>> > > >
>> > > > Resource::getSigns(Sign | Thing | Concept) : Occurrences of argument
>> > > > Resource::getThings(Sign | Thing | Concept) : Instances of argument
>> > > > Resource::getConcepts(Sign | Thing | Concept) : Classes / roles of
>> > > argument.
>> > > >
>> > > > Example data:
>> > > > aPerson :employment anEmployment (Signs)
>> > > > Peter :employment SalesDptManager (Thiings)
>> > > > Person, Employe :employment Employment, Position (Concepts)
>> > > >
>> > > > So, care must be taken building an ETL RDF tool that populates this
>> > > bindings in Resource instances, provides querying facilities and
>> support
>> > > the construction of entities which apply rules-transformation kind of
>> > > behavior using 'template' resources.
>> > > >
>> > > > Then the model should allow to build queries navigating Resource
>> APIs
>> > > like retrieving all the Concepts a Thing plays in a context, which
>> Signs
>> > > are instances of a given Thing in a given Concept context, etc.
>> > >
>> > > To repeat, there are no contexts in RDF.
>> > >
>> > > I would strongly recommend that you read more about RDF and its
>> intended
>> > > semantics before proceeding.
>> > >
>> > > Best wishes
>> > >
>> > > Pat Hayes
>> > >
>> > > >
>> > > > Best,
>> > > > Sebastian Samaruga..
>> > > >
>> > >
>> > > ------------------------------------------------------------
>> > > IHMC (850)434 8903 home
>> > > 40 South Alcaniz St. (850)202 4416 office
>> > > Pensacola (850)202 4440 fax
>> > > FL 32502 (850)291 0667 mobile
>> (preferred)
>> > > phayes at ihmc.us http://www.ihmc.us/users/phayes
>> > >
>> > >
>> > >
>> > >
>> > >
>> > >
>> > >
>>
>> --
>> -ericP
>>
>> office: +1.617.599.3509
>> mobile: +33.6.80.80.35.59
>>
>> (eric at w3.org)
>> Feel free to forward this message to any list for any purpose other than
>> email address distribution.
>>
>> There are subtle nuances encoded in font variation and clever layout
>> which can only be seen by printing this message on high-clay paper.
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.spline.inf.fu-berlin.de/pipermail/pragmaticweb/attachments/20150113/970055cf/attachment-0001.html>
More information about the PragmaticWeb
mailing list