Re: lexical resources with n-ary translations

Hi John,

thanks for the recap.

Am Fr., 26. Juni 2020 um 12:56 Uhr schrieb John P. McCrae <john@mccr.ae>:

> Hi Christian,
>
> Much of this was discussed during the development of the vartrans module,
> but I will try to recap:
>
> https://www.w3.org/2016/05/ontolex/#translation
>
> The main way to represent translations is by 'shared reference' that is
> using a single concept for entries in multiple languages.
>

Yes. The context of discussing translation with multiple targets was for a
situation where a direction between one source language expression and
multiple target language expressions (into different target languages) need
to be recorded. This is not possible with shared reference (because this
comes without directionality) and I see no way to do that in OntoLex other
than encoding the relation between each source and target expression pair
independently. Semantically, this is fine, but it is verbose.

Take the first entry of https://www.springer.com/de/book/9789020116670 and
assume that we wanted to encode that English is the source (first entry,
defines organization of dictionary) that is translated into four languages:
Abelian group | abelsche Gruppe | groupe abélien | abelse groep | абелева
группа

criteria:
number of triples and objects
+-direction (encodes translation direction)
+- metadata (allows to provide source metadata)

OntoLex modelling
Shared reference: 1 ontolex:LexicalConcept, 5 ontolex:evokes properties
(-direction, +metadata [at concept])
vartrans:translation (non-reified): 4 vartrans:translation properties
(+direction, -metadata)
vartrans:Translation (reified): 4 vartrans:Translation objects, 8
vartrans:relates (vartrans:source/vartrags:target) properties (+direction,
+metadata)

alternative modelling
*vartrans:Translation (n-ary): 1 vartrans:Translation object, 5
vartrans:relates (1 vartrans:source, 4 vartrags:target) properties
(+direction, +metadata)

The difference here is reified binary Translations require at least double
as many triples as reified n-ary Translations and I can see why people
would like to avoid that. However, this is an issue only if the direction
of translation needs to be recorded, because in terms of space complexity,
shared reference is equivalent with n-ary Translation.

If such a use case for multilingual data with one source language and a
large number of target languages does exists, where direction matters and
verbosity is an issue, it is probably safest to *not* use the OntoLex
vocabulary directly, but to create my:NAryTranslation as a subclass of
ontolex:LexicalConcept with properties my:source and my:target as
subproperties of ontolex:isEvokedBy and use these instead of the vartrans
vocabulary.

Cheers,
Christian