[Top] [All Lists]

Re: [ontolog] Multilingual ontologies

To: ontolog@xxxxxxxxxxxxxxx
From: Adam Pease <apease@xxxxxxxxxxxxxxxxxx>
Date: Thu, 10 Oct 2002 12:57:34 -0700
Message-id: <>
Martin,    (01)

At 08:38 AM 10/8/2002 +0100, Martin Bryan wrote:
> >    You might be interested in work we've done to express our ontology
> > automatically in different languages.  Currently, we support English and
> > Czech.  The output is awkward but understandable.  Take a look at
> > <http://ontology.teknowledge.com> and follow the link to the "beta
> > browser.  You can click on "English | Cesky" to see the alternate
> > presentations.
>I am afraid I don't class changing the headers and navigation bars to suit
>the language of the user as a "multilingual ontology". What you demonstrate
>is a multilingual interface to a monolingual ontology.    (02)

I don't think that's an accurate characterization.  Maybe you didn't look 
further than the home page?  For example    (03)

       (partlyLocated ?OBJ ?REGION)
       (overlapsSpatially ?OBJ ?REGION))    (04)

is automatically expressed in English:    (05)

"If obj is partly located in region, then obj overlaps with region."    (06)

and also in Czech:    (07)

"Jestliže obj je cástecne umísten v region, potom obj se prekrývá s region."    (08)

Only variable names are not translated.    (09)

>I should also point out that I do not consider an ontology that simply
>provides 1:1 mappings from a set monolingual ontology to another language as
>a multilingual ontology, because the boundaries of term definitions are
>determined by the starting language.    (010)

That's an interesting research issue.  I'd claim that our ontology is not 
tied to a particular language.  The labels that we've attached to concepts 
are just as reasonable in Czech as they are in English.    (011)

>A true multilingual ontology will require n:n mappings, where any particular
>term in one language maps to a set of terms in another language, depending
>on context, and the mappings are not symmetrical.    (012)

That's only assuming that the ontology mirrors the native language of the 
ontology authors.  We believe that we've captured concepts, not words.  One 
example of this is that you may notice many concepts aren't named with a 
single word (for example partlyLocated), and even for those that are, such 
as Region, there is not an exact mapping to the vague English concept of 
"region".  One might note that WordNet, as a lexicon, does have the issue 
you mention, because it is capturing word senses, not language independent 
concepts (although word senses and concepts are often coincident).    (013)

Adam    (014)

>The work I did for the CEN/ISSS Electronic Commerce group last year on
>defining a MULECO -- Multilingual Upper-Level Electronic Commerce Ontology
>(http://www.sgml.u-net.com/CWA-04.doc) was based on the The EAGLES
>Guidelines for Lexical Semantic Standards
>(http://www.ilc.pi.cnr.it/EAGLES96/EAGLESLE.PDF) that forms the basis for
>EuroWordNet. The MULECO document, which includes a survey of the main
>ontology definition languages as of the end of 2001, identifies the
>following set of "requirements" for a multilingual e-commerce ontology:
>1. The ability to uniquely identify the domain (e.g. industry sector)
>     in which each term is employed
>2. The ability to formally record the meaning of the term within a
>     particular domain
>3. The ability to identify other domains in which the same meaning
>     applies
>4. The ability to record alternative terms that have the same meaning
>     within the original domain
>5. The ability to identify alternative terms used for the same meaning
>     in other domains
>6. The ability to identify an exactly equivalent term used in a
>     different language
>7. The ability to identify a nearly equivalent term used in a different
>     language
>8. The ability to identify terms that form a part of an object defined
>     by a term
>9. The ability to identify wholes that a term forms a part of
>10. The ability to identify an opposite term or property (e.g.
>      water-resistant/water-soluble)
>11. The ability to record relationships between terms or properties
>12. The ability to identify opposite relationships (e.g. isMother/isChild)
>13. The ability to declare properties that record measurements
>14. The ability to declare properties that record times
>15. The ability to associate terms with specific points in process chains
>The document ends with an XML DTD for capturing material based on a subset
>of the EAGLES model relevant for e-commerce that meets the set of identified
>Martin Bryan
>The SGML Centre, 29 Oldbury Orchard, Churchdown, Glos GL3 2PU, UK
>Phone/Fax: +44 1452 714029  E-mail: mtbryan@xxxxxxxxxxxxxx
>For further details about The SGML Centre visit http://www.sgml.u-net.com
>To post messages mailto:ontolog@xxxxxxxxxxxxxxx
>An archive of the [ontolog] forum can be found
>at http://ontolog.cim3.org/forums/ontolog    (015)

To post messages mailto:ontolog@xxxxxxxxxxxxxxx
An archive of the [ontolog] forum can be found
at http://ontolog.cim3.org/forums/ontolog    (016)
<Prev in Thread] Current Thread [Next in Thread>