Dear John, (01)
You are missing the central point of the discussion with Doug Foxvog,
which concerned the theoretical and practical opportunity of adopting
solutions in the "qua-individuals" style for getting rid of the problems
concerning a sound definition and representation of the notion of role.
And, in spite of Doug's objections, I maintain that endorsing a wild
proliferation of individuals sine necessitate does not represent a
correct methodological approach to the problem. On the other hand, I am
not interested in Facebook or in extracting knowledge from natural
language but, mainly, in high-level inferencing starting from complex
conceptual representations. I can reassure you: in the SQL version of
NKRL, there are really no problems to store and manage all the required
individuals - i.e., instances of concepts, those of course that can
admit direct instances, certainly not "student" or "customer". With
respect now to the problems in dealing with individuals, I am surprised
you have never been faced, e.g., with an apparently stupid but concrete
problem like that of giving meaningful "names" to the instances created
at run time - apart, of course, of systematically choosing the easy way
of always creating individuals in the "concept_214" style. (02)
Regards, (03)
G.P. Zarri (04)
On 05/10/2012 19:25, John F Sowa wrote:
> Dear Gian Piero and Simon,
>
> I addressed my previous note to Matthew, but I forgot to comment on your
> notes, which Matthew was responding to.
>
> GPZ
>> ... can you imagine a PRACTICAL system of reasonable dimensions where
>> you are continuously obliged to create new individuals for specifying
>> all the possible everyday behaviours of John (and of all the others)?
>> The management of individuals is particularly tricky and costly in
>> concrete KBs.
> There is a huge difference between individuals and types of individuals.
> There are billions of individuals (AKA "named entities") that must be
> considered in large databases and the WWW. Facebook alone recently
> reached the milestone of a billion users, but they manage their types
> and roles of users very well.
>
> Facebook doesn't do any deep reasoning, but Cyc has the largest formal
> ontology on the planet. Their hierarchy includes all the types of
> individuals, relations among individuals, and behaviors of individuals.
>
> In our work with conceptual graphs (at VivoMind), we maintain large
> hierarchies of graphs, which we access in logarithmic time. We have
> processed many terabytes of natural language text. Increasing the
> numbers of types (i.e., labels on the nodes of the graphs) speeds up
> the access time. If your system has difficulty with the number of
> types, you should look for better algorithms.
>
> SS
>> I thought I was being clever, letting meta-properties like "ferpa
>> directory information release ok" be associated with these student objects.
>>
>> Sure, the existence of these role objects may have been rigidly dependent
>> on the person job having at some point existed, but I had no idea this
>> was so modally unacceptable.
> I prefer the term 'role type' to 'role object'. But there's no problem
> in having a multiplicity of roles and reasoning about them.
>
> When we process and reason with natural languages, we put the concept
> types for 'father' and 'student' in the same hierarchy with 'human'
> or 'man'. We make note that there are relational and functional
> characteristics of fathers and students that are not shared by
> natural types like human or elephant.
>
> But that does not create any problems for the methods of analysis
> or reasoning. With logarithmic algorithms, going from a thousand
> types to a million types doubles the access time. No problem.
>
> John
>
> _________________________________________________________________
> Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
> Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
> Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
> Shared Files: http://ontolog.cim3.net/file/
> Community Wiki: http://ontolog.cim3.net/wiki/
> To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
>
> (05)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J (06)
|