[sc34wg3] New TMCL slides: at least 2 roles must be played

Steve Newcomb srn at coolheads.com
Tue Nov 24 12:07:06 EST 2009

Patrick Durusau wrote:
> Steve,
> Steve Newcomb wrote:
>> Patrick Durusau wrote:
>>> Steve,
>>> Steve Newcomb wrote:
>>>> Patrick Durusau wrote:
>>>>> Steve,
>>>>> Steve Newcomb wrote:
>>>>>> Lars Marius Garshol wrote:
>>>>>>> I can see a need for omissible role types in n-ary relationships 
>>>>>>> where it may be the case that one of the role players is not 
>>>>>>> known, but in binary relationship then if one of the players is 
>>>>>>> not known you can simply omit the entire relationship.
>>>>>> I like this rule, but I think it should be more general. Even in 
>>>>>> an n-ary association, if there's only one role player, there's no 
>>>>>> relationship, so the whole association is otiose. The rule should 
>>>>>> be: /At least two roles must be played/.
>>>>>> The only thing that bothers me about this rule is that there may 
>>>>>> be moments when an association is only partially expressed. But 
>>>>>> I'm not sure that it's an issue, really.
>>>>> Moments when an association is only partially expressed?
>>>>> Hmmm, ok, what about marriage and one role player known but spouse 
>>>>> is not? A partial baptism record for example. Not ever going to 
>>>>> cure that lack.
>>>>> May not know who the other role player is but the existence of the 
>>>>> association is not in doubt.
>>>>> A subject is lost by omitting the association. That is "a" choice 
>>>>> but it isn't one that should be a rule for all topic maps.
>>>> "Unknown role player" is very different from "unplayed role".
>>>> I would claim that in the example you provide, the role is not 
>>>> unplayed. It is played by an unknown person, at least if the 
>>>> relationship is known to exist. I guess if there's a baby to be 
>>>> baptized, that's evidence enough that a relationship exists, since 
>>>> human reproduction takes two to tango.
>>> If you want to *define* the subject "association" to consist of at 
>>> least two roles, both of which are played but one of the players may 
>>> be unknown, I suppose.
>>> What puzzles me is why object to an "unplayed role?"
>>> Hmmm, ok, say that I play the grandfather role in a grandfather <-> 
>>> grandchild association. The later role isn't unknown but in fact 
>>> unplayed, at least at this point in time. ;-)
>>> Why would you object to my instantiating such an association in my 
>>> topic map?
>> Because the subject that is represented by an association (TMDM: by a 
>> topic that reifies an association) is always a relationship. If fewer 
>> than two roleplayers are specified (i.e., if fewer than two subjects 
>> are being said to "be related"), then by definition no relationship 
>> is being specified. (At least under my own personal understanding of 
>> what a "relationship" is.)
>> I've heard some comedian or other say, "I'm married to my imaginary 
>> lover," while making a crude gesture with one hand. Well, OK, you may 
>> say that this particular "marriage" is not a marriage, really, but I 
>> would claim that it is nonetheless a relationship, and that 
>> relationship is the thing that the crude gesture refers to. Imaginary 
>> roleplayers are still roleplayers. The subject of a topic can be 
>> anything at all, and a subject that happens to be a relationship can 
>> be any relationship at all.
>> Actually, though, it's OK with me if a topic map contains an 
>> utterance of an association with fewer than two specified 
>> roleplayers, because it might be important to allow interchange of 
>> incompletely-specified relationships. I'm arguing that, where a Topic 
>> Map rhetoric (such as TMDM) provides a subject-proxy expression for 
>> /relationships /(such as association information items)/,/ it be 
>> explicit that utterances of associations with fewer than two 
>> roleplayers are meaningless and should therefore be ignored during 
>> subject-centric processing. 
> I was with you up to the last sentence. Subject-centric processing 
> does not and cannot depend upon some universal notion of subject 
> identity.
>> For example, if a topic map has two associations, and neither of them 
>> has at least two roleplayers -- they should not be merged, even if 
>> they both have the same roles being played (or left unplayed) by the 
>> same roleplayer (or absence(s) of roleplayers). 
> That sounds like a sensible rule to me because the specification of 
> the subject they represent is incomplete, not that they don't 
> represent a subject but its specification is incomplete, and therefore 
> I would suggest (but not require) that they remain unmerged.
OK. I think you're (as usual) deeply correct. I agree with you that my 
proposal is not relevant to some legends. I've been thinking about TMDM 
a lot lately, and this has put the bigger picture a little out of focus 
for me.
> If we go about tossing out *universal* subject-centric processing 
> rules topic map users are no better off than the unfortunates who are 
> told they must use some ontology because it is more complete, logical, 
> precise than their poor, naive, but their own ways to deal with subjects.
OK, you win. However, it seems to me that I can transform my position so 
that it is still relevant. It becomes a legend disclosure requirement, 
and it can be stated something like this:

    /If a legend allows the existence of subject proxies that do not
    specify the subjects they represent in a fashion sufficient to
    support subject-centric processing (i.e., during determination of
    subject sameness or difference), then the legend must disclose how
    such proxies should be treated during subject-centric processing./

What do you think?
> Note that I am *not* disagreeing with the notion of subject-centric 
> processing but simply that the definition of that processing depends 
> solely and proximately upon the specification of that processing by a 
> legend.
Point taken.
>> In Topic Maps, the only reason to merge topics is that they are 
>> understood (by someone, presumably for some reason) to represent the 
>> same subject. If a topic has no subject, it does not have the same 
>> subject as any other topic, even when the other topic is 
>> subject-less. It has no subject at all, and so it's not really a 
>> topic at all. Where there is no subject, there is no basis for 
>> merging, full stop.
> Hardly. If the subject has not be specified to the extent that I 
> consider sufficient to trigger merging it should not be merged. One 
> such case would be incomplete specification of subjects. Another might 
> be that there are specifications of subject identity that simply are 
> of no interest to me and I don't want to spend the processing cycles 
> on merging for those subjects.
Yup. You're right.
> Say I were to create a topic map arcade game, that starts with 100 
> empty topics and players must specify 100 topics using various ways to 
> specify subjects, it is important that I be able to track the number 
> of "empty" or unfilled topics. (Imagine the appropriate number of 
> flashing lights, bells, not sure about the "tilt" mechanism, and at 
> the completion of each 10 topics the subject identity requirements get 
> harder.)
> Each of those unfilled topics represents an inchoate subject that will 
> be known should the player successfully complete it. So, no, I don't 
> want it to be "merged" but that isn't at odds with subject-centric 
> computing.
I'm now arguing that this is really a disclosure issue. Legend 
disclosures should be constrained in such a way that subject-centric 
processing can be applied in a fashion that is consistent with the 
fashion in which it has already been done, whenever making master 
indexes (big topic maps) from existing, independently created indexes 
(smaller topic maps). If I'm making a master index from two smaller 
indexes, and both of them contain topics whose subjects are incompletely 
specified, and each of them was prepared in a different fashion with 
respect to the handling of incompletely specified subjects, then I need 
to know that. If I don't know that, my editorial product (the master 
index I'm creating) is likely to misrepresent (violate the semantic 
integrity of) the input provided by one or both of the smaller topic maps.
> Actually that game sounds fun. Players would have to remember 
> identifiers to invoke merging, etc. Memory plus eye/hand coordination. 
> Wonder what that would look like if tied to a news feed so that users 
> were actually contributing something useful while playing the game? If 
> 5 users all "identify" the reference of some word or phrase in a 
> story, it bumps up in reliability due to merging of the identifications.
I'm not much on games, but I'll try anything once. (;^)
> I support "subject-centric computing" but choke on pre-defining what 
> it means to have "no subject." Perhaps the freedom to identify any 
> subject by implication includes the freedom to decide when there is no 
> subject?
Right. Total freedom, but disclose, disclose, disclose.

More information about the sc34wg3 mailing list