=Paper=
{{Paper
|id=Vol-432/paper-29
|storemode=property
|title=A Family History Knowledge Base Using OWL 2
|pdfUrl=https://ceur-ws.org/Vol-432/owled2008eu_submission_29.pdf
|volume=Vol-432
|dblpUrl=https://dblp.org/rec/conf/owled/StevensS08
}}
==A Family History Knowledge Base Using OWL 2==
https://ceur-ws.org/Vol-432/owled2008eu_submission_29.pdf
A Family History Knowledge Base Using OWL 2
Robert Stevens and Margaret Stevens
School of Computer Science, University of Manchester, UK
robert.stevens@manchester.ac.uk
1 Introduction
We present a case study in the use of OWL 2 in the form of a family history
knowledge base (FHKB). Genealogy affords many opportunities for logical in-
ference and thus it has been used many times as an example system and not
only for description logics. OWL 2 offers greater opportunities for inference than
OWL-DL and we test those on a knowledge base of the Stevens family history.
Genealogy is an attractive example simply because it involves many complex,
deterministic relationships between individuals. What is more, most of these
relationships can be inferred from very few asserted facts. Ideally, motherhood,
fatherhood and gender should be enough from which to infer all common kin
relationships. In addition, the genealogy example is accessible to all users for the
simple reason that everyone has a family history (although for some it might be
unknown).
The particular feature of OWL 2 that is attractive in this example is that
of sub-property chains. For example, that ‘My parent’s brothers are my uncles’
fits naturally in to this paradigm. The property hasParent chained with the
property hasBrother implies the property hasUncle; so any individual with a
hasParent relationship to an individual with a hasBrother relationship to a
third individual also holds the implied hasUncle with that third individual.
Using a representation such as OWL 2, it should be possible to infer the
majority of desirable family relationships, including the sex of individuals, from
assertion of parentage (fatherOf, motherOf, sonOf, daughterOf, . . . ) and sib-
lings.
In addition, it would be useful to be able to have a sparse assertion of sibling
relationships and infer brothers and sisters. An example of sparse assertion is
that if David is the brother of John and David is the brother of Peter and we
know that both John and Peter are men, then John must be the brother of
Peter and vice versa. Similarly, that individuals are people of a particular sex
should be inferred from the fact they are, for example, a mother, father, son
or daughter. Maximising inference from asserted facts in this way reduces both
data gathering effort and scope for error. One asserted fact may cause many
incorrect inferences, but those many errors are fixed by one change rather than
many.
The aim of the Stevens FHKB was to see how complete a set of inferences
could be made and how correct such inferences would be using OWL 2.
2 Materials and Methods
Only five primitive classes were described in the FHKB TBox:
1. Person, a subclass of DomainEntity; Sex, a subclass of DomainEntity;
Person and Sex are disjoint;
2. Male and Female, disjoint subclasses of Sex.
All other classes are fully defined and mainly used for making queries against
the FHKB. Two important defined classes are:
1. Man that is defined as a Person that hasSex SOME Male;
2. woman that is defined as Person that hasSex Female;
3. hasSex is functional.
Man and Woman are used as domain and range constraints in the property
hierarchy that drives the inference made about individuals. Only assertions on
parentage and siblings were made; all other kin relationships, including the type
of individual persons, was to be driven through inference.
1. The root of the property hierarchy is hasrelation. It is both transitive and
symmetric: All my relatives are also my relatives and in turn have myself as
a relative etc.
2. parentOf subsumes fatherOf, together with motherOf, with the inverses
treated equivalently. Domain and range constraints of Man and Woman allow
the sex of individuals to be inferred from assertions of parentage. Declaring
motherOf and fatherOf to be functional constrains a person to have only
one mother and only one father.
3. ancestorOf subsumes parentOf, with the inverses treated similarly. Where
the parentage properties are intransitive, ancestorOf is transitive. As hparentOf
implies ancestorOf, asserting parentage will also allow all ancestors and de-
scendants to be gathered.
4. relationof, ancestorOf and parentOf all have domain and range of Person.
‘Sexed’ properties all use domain and range constraints of either Man or Woman
in the expected manner. The intention is that domain and range should be
used to infer that an individual is at least a Person and then either a Man
or woman.
5. Similarly, brotherOf, sisterOf and their inverses (with appropriate domain
and range constraints) are sub-properties of siblingOf. The latter is both
transitive and symmetric. Where brotherOf cannot be transitive (I’m the
brother of my sister, but her sister is not my brother), any sibling is my
sibling. Ideally, making these sibling properties irreflexive would prevent my
being my own brother.
6. Having asserted parentage and some sibling relationships, sub-property chains
were also used to infer many relationships:
(a) uncleOf is given by the sub-property chain brotherOf o parentOf.
Similarly for auntOf. This only gives blood relationships; aunts and un-
cles by marriage are not encompassed in this pattern though the exten-
sion is trivial.
(b) Great aunts and uncles can be inferred with sub-property chains such
as hasParent o hasUncle as appropriate.
(c) firstCousinOf is made with a sub-property chain of hasParent o siblingOf
o parentOf. That is, the children of my parent’s siblings are my first
cousins. firstCousinOf has the characteristic of being symmetric—I am
the first cousin of my first cousins.
(d) Further degrees of cousins (second and third cousins etc., can be built
in a similar manner. First cousins share a grandparent, but not parents;
second cousins share a great-grandparent, but not grandparents; third
cousins share great-great-grandparents, but not great-grandparents. So,
for my second cousins, I want my grandparents sibling’s grandchildren.
Other degrees are done with the same kind of pattern.
(e) ‘Removes’ of cousins refer to generational gaps; so my first cousins chil-
dren are once removed (my first cousins being the same generation
as myself). So, a sub-property chain of firstCousinOf o parentOf
will imply firstCousinOnceRemovedOf. This will make robert david-
bright 1965 a first cousin once removed of my father’s first cousins
(and vice versa); so the relationships, like all cousin relationships, are
symmetric.
(f) hasGrandfather is a sub-property chain of hasParent o hasFather—
my parent’s fathers are my grandfathers. hasGreatGrandfather can
be made with hasParent o hasGrandfather and so on. Naturally, the
same goes for grandmothers. Having these as sub-properties of
hasGrandParent allows sex-neutral grand- and great- relationships to
be found. All these family relationships are ultimately sub-properties of
hasRelation.
Four hundred and six members of the extended Stevens family were entered
as individuals into the FHKB. The earliest family member has a birth year of
17261 .
3 Discussion
An FHKB is or should be a compelling example for OWL, especially with the
extensions available in OWL 2. The Stevens FHKB demonstrates that much can
be done with automated inference. There remain, however, several points that
stop an FHKB being the compelling example that it should:
1. The inability to use, as required, the irreflexive property characteristic when
the property is used in sub-property chains means, that too many inferences
are made. For example, all the cousin relationships need siblingOf to be
irreflexive. Without this property characteristic my father becomes his own
1
A de-identified version of the Stevens FHKB is available at http:
//owl.cs.manchester.ac.uk/repository/download?ontology=http://www.
co-ode.org/roberts/family-tree.owl&version=0&format=RDF/XML
sibling and thus my brother and myself become our own first cousins. con-
sequently, at present, many entailments in the Stevens FHKB are incorrect.
2. The inability to state, for example, that if John is inferred to be a siblingOf
David and that John is a Man is known, then it should imply that a brotherOf
relationship holds between John and David. David is asserted to be the
brother of John. the gender of John is known as his parentage of his children
is asserted. That John hasBrother David can be inferred (from the inverse
of the asserted brotherOf relationship from David), but at present not that
John is brotherOf David; thus John is not inferred to be my uncle. For this
inference we need that John brotherOf David to be inferred from the facts
currently known from assertion and inference.
There is thought to be a “work around’ for this problem. it involves as-
serting the type isMan some Self using an object property isMan on male
individuals in the FHKB. This acts as a ‘flag’ on the individual signalling its
maleness. Then a sub-property chain such as isMan o siblingOf implies
brotherOf could be used. The isMan is asserted; the siblingOf inferred
and thus brotherOf can be inferred. This does work, but would mean that
the siblingOf property hierarchy could not be used without pushing the
FHKB into OWL 2 full2 . without this property hierarchy, other inferences
would be lost. It also impinges on the requirement for sparse assertion. That
is, gender is, in effect, asserted.
3. The inability to infer full- or half-relationships. With two parents in common
between two male individuals, then a full brother relationship can be inferred.
If, however, only one parent is in common between two male individuals then
half-brotherhood should be inferred.
Rules could be used to fix the problems encountered in the use of OWL 2. In fact,
sex, parentage and rules are sufficient to find all the kin relationships needed.
Defined classes can also find the correct set of, for instance first cousins, but the
task set here was to use only OWL 2. The FHKB will be further extended with
information about marriages. Making inferences of in-law’ relationships would
allow other relationships by marriage to be inferred. Marriage, re-marriage, and
all that entails such as step-parenthood would also be useful. Information ex-
ists, for example, on eye-colour of Stevens ancestors back into the mid-1800s.
Working out probability of carrying genes for brown and blue eye colour should
be possible. Naturally, other information such as roles played by individuals can
also be added to the FHKB that would allow a wider range of queries to be
asked. Moving outside OWL 2, the addition of rules will allow further inferences
to be made. Development of the Stevens FHKB offers an OWL 2 version of a
traditional example in the field.
Acknowledgements: I would like to thank Matthew Horridge, Simon Jupp,
Alan Rector, Uli Sattler and Bijan Parsia for help and advice during this exercise.
2
See, for example, http://lists.w3.org/Archives/Public/public-owl-dev/
2007OctDec/0004.html and http://www.w3.org/2007/OWL/tracker/issues/22?
changelog