On Tue, May 21, 2013 at 11:55 AM, doug foxvog <doug@xxxxxxxxxx> wrote:
On Tue, May 21, 2013
08:16, Phil Murray wrote:
> Is there any consensus in the KR community on best practices for labeling
> relationships among objects?
No.
> And a closely related question: Why English?
Because the majority of initial labelers used English.
> Going back to a recent thread:
>
>
> [Patrick Cassidy] [snip] ....
>
> What I consider "really, really bad practice" is to
> use any labels other than those most easily and rapidly understood by the
> human who is looking at the ontology.
>
> [William Frank]
>
> Absolutley, but 'human' among engineers so often means "english
> speaker', or at least 'indo-european speaker'. This is such a shame,
> especially when English, which may be the most irregular, messy language
> there is, is used as the model for what 'human beings' can understand
> easily.
> [Phil Murray] +10 !
> ...
> So I, too, am
uncomfortable with the habit of using English as the
> default language of representations of meaning.
Would you prefer Latin? ... Chinese? ... semi-arbitrary
alphanumeric
strings? OBO uses the latter. It is good to have mappings from
the terms to whatever language the users use.
> If you are representing the meaning of a purchase ("John bought a
book
> at books.com."),
wouldn't _plays the role of Agent in_ or (perhaps even
> better) _participates as Agent in_ be better than "has_Agent"
> to represent the relationship between John and the statement of
> fact about John's activity?
No, that is worse. There is a buying and a selling agent in the
sale.
The appropriate role needs to be clarified. The sale should be reified,
and relations specifying what is important(buyer, seller, object whose
ownership is transferred, recompense for the sale, location of event,
time of event, ...) should be made.
Note that English differentiates "to sell" from "to buy",
"to borrow"
from "to lend", and "to give" from "to receive".
Finnish has a single
infinitive for each, and indicates (through case structure) who is on
the giving and receiving end.
> [ambiguity of _to have_, _to be_ and _to go_ ]
> IMHO, in formal representations of meaning, we should favor precision
> over informality of speech ...
Agreed
> OTHER QUESTIONS:
> 1. Does a relationship consist of the connecting arc only, or does it
> encompass the two objects and the connecting arc? Maybe it's not
> either/or?
To identify an arc, its end points have to be known, as well as the
type of arc. The relationship needs to specify the end points; it
need not have all information about them.
Note that some relationships have arity greater than two, and
thus can not be modeled by a single arc between two points.
> 2. In abstract terms, if links are first-class objects, should they have
> unique identities and support multiple forms of metadata?
Yes.
> 3. We use computers and graphic for everything. Why are we limiting
> human-readable labels of representations to strings of ASCII text?
Human-reasable labels are restricted to text because of our definition
of
label. But some labels may include non-Latin characters and be strings
of UNICODE. One can certainly have a relation to a file that contains an
image, sound, or a video. But i would recommend using a more specific
relation than "label".
-- doug foxvog