Re: ADMS spec document update

Phil, all,

On 18 October 2012 19:33, Phil Archer <phila@w3.org> wrote:
>
> (5) This is a nitpick but it seems odd that translations are distinct
> SemanticAssets whereas representations are just distinct Distributions. If I
> represented a schema in RDFS instead of XSLT that's a much bigger change
> than if I translated it's labels to French.
>
> I have some sympathy. I like conneg which is why I have a deep seated
> visceral hatred of accessURL. But, well, common practice seems against us.
> James seems happy that a translation is another asset although I'd think of
> it as an alternative distribution. I wasn't involved in the original data
> modelling so I'd have to ask why translations are different assets and not
> distributions.

If we were discussing, say, a textbook or a novel, it would not seem
odd at all that a translation into French is a bigger change than its
paperback vs. hardback edition, would it? The question is what is
meant by "translation" in the context of semantic assets. One can
argue that translating the labels is not a translation of the asset
itself.

Cf. the discussion in this list of what corporation means in US,
British, and Canadian English. Depending on the choice of variant some
instances, like the BBC, might fall on either side of the
"Corporation" concept [1]. That the group agrees to slightly re-define
the word and use it as a mnemonic, a label that is easier to rememeber
than "id42", is a sound decision but besides the point. My point is:
if we were to truly and faithfully "translate" the RegORG ontology
from US English into British or Canadian English it would not be the
same ontology, in that there would be no 1:1 equivalences  between all
of the concepts and relations in the original and the translated
ontologies. And that's between variants of English, imagine
translating into more distant languages and cultures, and the
associated systems of clustering the things in the world under
linguistic labels.

Just like one does not consider a word-for-word, Google
translate-style translation of text a "translation", calling a
concpet-for-concept translation of the labels in an ontology an
"ontology translation" is not exactly faithful to what "translation"
means.

Going back to measuring up changes, and given the above, I think that:
 - An "asset translation" is a bigger change and results in a related
(but distinct) asset. A linguistic translation hardly ever happens,
but a conceptual translation might, see immediatelly below.
 - An alternative representation is a re-distribution of the same
asset instance; unless the target representation does not support all
the features of the source representation used by the asset, in which
case we need an "asset translation" into a similar but distinct asset
that uses the target representation.
 - A "label translation" is the smallest change and for some
representations does not even need to be separately distributed (e.g.,
in RDF the labels in different languages can be bundled together). If
the representation does not support this bundling, then alternative
distributions are needed, but still distributions of the same asset.

Best,
s


[1] And if the objective was to faithfully represent the usage of the
word "Corporation", which it was not in the discussion I quote.

Received on Friday, 19 October 2012 03:24:23 UTC