Re: SUO: Try coding protocols
Hallo Josiane Caron,
wrt:
> Moreover I would have like to code the formulation of the disk and
> of the peg:
> In example 1: 'the pink disk'
> you have introduced the notion of attribute, but if I use this form I find
> it is not right. It 'll give us 'the pink disk1' instead of ' the pink disk'
> In fact I need to know both concept and the occurrence of the concept (type
> and occurrence). I proposed a re-writing for sentence 1, below (I left
> sentences 2 and 4 wirth your writing). But the problem is I need some
> different formulation for 'instance'; I wrote something by introducing the
> word 'naming'; what can I do ?
> In the same sentence I encounter the same problem with the denomination
> 'diskb' for the 'peg B'
> (I put a line / in example 1 below before the line I have changed)
> Furthermore there are the problems of connectives in 3 and 4.
I am struck and intrigued by the similarity that your questions have with those that
those of us in Natural Language Generation are concerned with all the
time: i.e., how to express these pesky natural language differences
in some kind of semantics so that our generators can produce the
sentences that we start with.
The added interest of trying to make this work with someone else's
ontology is also very relevant to the NLG area.
Considering all the work that has to go into producing something even
as simple as "the pink disk" (e.g., decision about definiteness, decision
about attribution of the entity to a category/class (disk), decision
about the granularity of the description of the category/class (disk,
object, whatever... I am not sure what these disks are!), decision
about granularity of the attribution, decision about which
attributes need to be expressed at all in order to differentiate the
current object from other potential "distractors"), I am left wondering
just how much of that falls within what an ontology has to come
up with. The question becomes even more problematic when moving
to discourse connectives (also an area with a long history in NLG).
Our input expression for "the pink disk" would be something like:
(p / object :lex disk
:color-property-ascription (c / color-quality :lex pink)
:identifiability-q identifiable)
p and c are the instance variables; object and color-quality are ontology
"concepts", color-property-ascription is an ontology relation; and
identifiability has nothing to do with an ontology but is an appeal
to the dynamic text production process and text model. Lex says
what words I want to use for the concepts. If I throw
this semantics at our English generator it spits back "the pink disk".
The relation between the ontology concepts used in NLG
(in our case drawn from the
merged or generalized upper model: http://purl.org/net/gum2) and
standardizations attempts such as SUMO is of great interest
and one of the reasons why I read these lists!)
Is this at all relevant to you? -- are you familiar with it?
Best,
John Bateman