[rdfweb-dev] Relationship Schema Updated

Michael Bauser michael at bauser.com
Fri Mar 19 04:29:31 UTC 2004

Hash: SHA1

Jim Ley wrote:

> Michael Bauser michael at bauser.com wrote in:

>>Anthropologists have been trying for decades. The general consensus
>>really is that the eight-fold model works best, because:
> Could you cite?

OK, I have to admit: It didn't occur to me that I would need a citation
for that. In my personal universe, that's like providing a citation for
the sentence "The consensus is that there are nine planets in our solar
system." A bias of my own, I guess. Here's the best online tutorial I've


>>B) The eight relationships are mutually exclusive in the normal
>>functioning of all known cultures, so the graphs seldom get weird.
> I'm amazed you wrote this, after the strength of your previous language
> against imposing ideas on others, the relationships are not at all
> or derivable within my own family, and I am not unusual in my society, I
> know lots of folk with similar situations:
> According to you, we should be able to infer from these 8 basic things,
> without fear, that this is a contradiction:
> Jim son JimsMother daughter JimsMothersMother
> and
> JimsSister daughter JimsMothersMother
> and
> Jim Sister JimsSister

OK, maybe I'm confused by inconsisent spacing, but I *think* you're
asserting that Jim is "son" to one person and "sister" to another?

If so, that's *not* a contradiction in the system I'm trying to use. It
may look like that to people who assume that "sister" and "son" have
anatomical requirements, but I'm not saying that they do. (That was the
point of my whole digression about the Nuer, I thought.) I wouldn't put
*any* anatomical requirements into an anthropologically-inspired vocabulary.

Maybe we have different interpretations of "mutally exclusive" and
"highly unusual"? When I wrote "mutually exclusive in the normal
functioning of all known cultures", *all* I was saying is that it's
highly unusual for a person to have two different relationships *to one
other person". The real "contradiction" would be if Jim was "sister"
*and* "brother" to his JimsSister, because few cultures encourage
"double relationships" to first-degree kin.

On the other hand, an anthropological vocabulary couldn't have a
*ontological* restriction on having two relationships, because the job
of the anthropolgical vocabulary is to *store* the data, no matter how
weird it might be, and there are societies that allow double
relationships. (This, obviously, will be objectionable to many members
of societies that don't allow double relationships; to them I say "moral
judgement has no place in data collection".) *Interpreting* the data
needs a cultural context, which I'm envisioning as being supplied by a
second vocabulary.

As I said in one of my replies to Morten, there's a "storage vocabulary"
and a "presentation vocabulary" involved in making use of kinship data.
The storage vocabulary can occasionally deal with inverses (if A is
ancestor of B, B is descendent of A), but it  *can't* deal with
disjoints (if A is ancestor of B, B is never ancestor of A), because
humans are more complex than that. If a storage vocabulary used an
ancestor/descendent disjoint, for example, if couldn't deal with the
"I'm my own grandpa" paradox (which really happens).

> Unfortunately you'd be wrong, there's no contradiction there in my
> and I will certainly not welcome any relationship schema which
prevented the
> above from expressed, or allowed that to be a contradiction.

If your society is cool with that, then create a vocabulary for it.
That's what I want to see, honest: Vocabularies that are *explicit*
about being bound to the societies they describe.

That said, even if you created a vocabulary that defined a
"contradiction" in your kinship structure, it's not the End of Social
Networking As We Know It: It's just a computer pointing out an
incongruity between data and description. That incongruity has two
possible origins: The input data is wrong, or the vocabulary doesn't
really describe the network. One way or another, you've *learned*
something, and provided yourself with a direction of further
investigation. (The computer is not making a moral judgement when it
finds a contradiction.)

*This* is what anthropologists and sociologists *do* with social
networks. Map the network. Document the native vocabulary of the
network. Compare the "actual" network to the "ideal" network. See if the
disjoints tell us anything interesting. *Finding* contradictions is a
more useful goal of RDF than refusing to see them.

>>C) The second and later degrees or relationships can be mapped by
>>chaining the first degree relationships. *This* is what makes the
>>eight-way model especially useful for RDF data meant to be parsed by
>>computers, because we can separate the *relationship* from the *label*,
>>and let the end-user determine if they want to see relationships using
>>*native* or *local* labels.
> Except of course, the labels are what constrains the meaning, my sisters
> daughter, and my othersisters daughter might not be cousins, yet this
> simplistic model constrains them to be, or means I can't express my
> relationships, and things can get considerably more complicated than my
> family.

So you're saying the two sister's daughters don't consider themselves
cousins because they're not *biologically* related? If so, we're sliding
into a new issue here: How does biological *pedigree* relate to social

I have to admit my approach to this has evolved during this discussion.
The truth of the matter is that the relationship between the two is
dependent on the cultural context. In some societies, kinship is all
about pedigree. In others, people couldn't care less about pedigree. In
yet others (including the United States), kinship is *partially* about

I mentioned this when I said the eight-relationship model was ambiguous
about certain relationships. Where some societies see "brother",
Americans see "full brother, "half-brother", "step-brother", "foster
brother", and probably something that I'm forgetting.

My original proposal was to punt the issue to the "cultural context"
part of explaining kinship, because creating a vocabulary that includes
every possible interpretation of every possible connection would make
the vocabulary unwieldy to the point that no one would want to use it:
Imaging if you had a FOAF personal information manager with a "Family
relationship" field for each contact, and the drop-down box had more
than a hundred options. Would you bother looking up the "correct"
relationship for every relative? Would anybody?

So, I would have said yesterday: If your society sees a difference
between different types of brother, create a vocabulary that can derive
those relationships with unions and disjoints. That would probably
require comparing "kinship relations" to a  separate vocabulary of
"pedigree relations" that dealt *only* with biological pedigree. If your
society's vocabulary requires "cousins" to have a chain of *biological*
connections, so be it. It's your society, and therefore, your vocabulary.

(Digression: If on the other hand, the RDF data about your kin doesn't
provide *any* cultural context, I'm on my own in making inferences about
the realtionship between the daughters of your siblings. That's not,
however, a weakness of just the eight-relationship model. The
relationship vocabulary has the same weakness: the children of two
siblings will still "look like cousins" to people who define cousin that
way. Using the lack of an *explicit* cousin connection to prove they're
not cousins wouldn't be enough, because it presumes a data-completeness
that we're never going to have in the real world. If you don't want
those children mistaken for cousins, the answer isn't to *outlaw*
inferences; it's to encode data in a way that RDF agents can make the
right inferences.

This might actually be an issue with FOAF in general: FOAF files can
make assertions (A rel:worksWith B, B rel:worksWith C) that lead to
reasonable inferences (A rel:worksWith C, given the somewhat vague
definition of "rel:worksWith"), but generally can't make
*negative* assertions. There is no foaf:stranger than lets one person
disagree with another person's assertion of relationship!)

Getting back to biology: But now, I don't think that announcing
"pedigree is a different network" is as practical as it could be. It's
probably better to add a "second axis" to kinship relationships. Here's
what I'm thinking:

Anthropologists (when they're describing a society that needs the
distinctions) divide kin into three "natures": Consaguinal kin
(biological relations), Affinal kin (marriage relations), and Fictive
kins (everything else, which would include things like adoptive parents
and godparents). If I create an RDF vocabulary that uses this axis in
addition to the eight-relationship axis, I deal with the societal
relationship and the biological relationship in a usable manner: The
hypothetical FOAF-friendly PIM has *two* small dropdown boxes -- one
with the eight relationships, and another with the three "natures"
(abeit it with more user-friendly titles than "affinal"). It's usable,
and it produces more explicit descriptions.

>>The LITTLE problem with the "label it with my language or theirs"
>>approach is that FOAF doesn't exactly have a way to label "what language
>>and culture I define my networks with". FOAF doesn't even have
>>properties for nationality yet.
> I hope this a language problem of yours, where does nationality and the
> cultural meanings of family tie together?  I don't see nationality as
> relevant to what is being discussed (cultural identity of course, but
> nationality, not in the slightest)

As I said to Morten, I wasn't using nationality as a kinship-relevant
trait, I was just using it as an example of how FOAF doesn't hasn't even
touched "easy" background information. There's a lot of FOAF about where
people *are*, but no much about where they're *coming from* (so to
speak). Some of that background data is relevant to how the current data
is interpreted.

>>The BIG problem is that in many societies, there are temporal and/or
>>ordinal qualifiers to relationships. The distinction between "older
>>brother" and "younger brother" can be very important, as can the
>>distinction between "second wife" and "third wife". (The distinction
>>between the latter two will further depend on the cultural mores; in
>>some societies, "third wife" implies the divorce or death of "second
>>wife", while in other societies, it implies the husband can afford three
> Yep, the relationship schema doesn't address these, but then single
> properties for these things isn't required, we don't need a single
> for every single concept in the world,

I know. That's why I was trying to dodge the issue of pedigree,
remember? I don't have to have to deal with:


and so on, for every core relationship. Multiply that by the possible
ordinal relationships, and I would go completely insane.

> we can use more than one - if it's
> strictly possible that "older brother" is a brother who was born earlier,
> then knowing the DOB's of the people would be sufficient (of course it's
> likely that that is not sufficient, there's been recent debates here about
> what happens if ones elder sister changes her sex, does she obtain the
> granted to the eldest born son?)

Damned if I know, but that's a issue for the presentation vocabulary,
which will presumbably be written by somebody who knows more about the
local culture than I do. As it stands now, I think the storage
vocabulary just has to describe the eight relationships,
affine/consaguine/fictive, and some "life event" data (at the very
least: birth, death, marriage, adoption, and (apparently) gender
reassignment.) The culturally-bound vocabularies used for presenting
that data to human readers will say how much of the stored data is
important to each relationship.

> Except it would infer incorrect things from my network, and that would
> in my mind be very rude, and very dangerous.

Well, I doubt it will kill anybody, but I understand that some people
don't like others making inferences about them. But, as I said in my
earlier digression, you're never going to be able to outlaw inferences
in/from RDF. RDF was *created* to encourage inferential logic.

> but at least you were beginning to look at the issues!

And your input has been some of the most helpful received so far.

Version: GnuPG v1.2.1 (MingW32)


More information about the foaf-dev mailing list