Christopher Guy Yocum

ORCID: 0000-0002-7241-3264

When reading the Irish Genealogies, you will often encounter situations like this in Genelach Clainde Conchobuir Corraig:

…o shléib fodess Fiacra Tort.

Followed by further down the page by:

Fiachra Tort secht mc leis…

In this case, both names indicate the same individual but there are two instances of their name in the same section of text. There are two strategies when modelling this in a database. First, you can wrap all the information into the same instance in the database. This makes sense as they are indeed the same individual and should be treated the same. However, the database does not now follow the text that it is modelling and it will take a reader a little while to work out what has happened. Additionally, there is the possibility, admittedly remote, that a repeated name in a genealogy will get confused for one another and, when fixing it, the auditor will need to disentangle the two individuals. Second, the database can have two separate entries for each instance and tie them together somehow. In this method, a reader and more easily follow the flow of the text in the original source material. Additionally, a direct link can be made to the instance that a reader or user wishes to reference.

In the case of IrishGen, a reader will encounter both strategies employed. This is due to the transition from the text to RDF is undertaking in only a semi-automated fashion. This mean there is always a human in the process. While this is usually beneficial, it can also mean that inconsistencies like this can happen. To be more honest, it happens when I am being lazy doing the translation. Most times this is not a problem as all the information in the source material is represented and can be searched.

In a traditional relational database, it can be slightly difficult to represent the second strategy. Usually, it would mean creating a second “link table” which would then have all the links between the two entities in the originating table. This can often turn into a many-to-many reference situation. Again, while it is not unusual to do something like this in a relational database, it is just slightly awkward and creates odd looking SQL queries.

In a graph database like IrishGen and RDF, the situation becomes easier to represent. It is just another edge and link to other instances of an individual. In the case above, this is represented like so:

     <FiacraTort>
        a foaf:Person;
        irishRel:nomName "Fiacra Tort";
        agrelon:hasParent <CollaUais> ;
        rel:childOf <CollaUais>;
        irishRel:ancestorOfGroup <hTurtri>, <FirLi>, <hMeicUais>.

...

     <FiachraTort-269b781f>
        a foaf:Person ;
        irishRel:nomName "Fiachra Tort" ;
        owl:sameAs <FiacraTort> ;
        irishRel:numChild 7 .

You will notice that there is a owl:sameAs edge which links one instance to another. This is where the situation becomes slightly complicated. There is the concept of “Fiacra Tort” as a single individual who is referenced twice in this text. There are also two instances of the same individual in the text. To represent this situation, OWL 2 has the sameAs predicate. This is also known as the SameIndividual predicate (see OWL 2 Primer 4.7 Equality and Inequality of Individuals). What this does in most Triplestores is to merge the two nodes and present information from both of them when they are searched. This is usually what a user will want. Although, occasionally it is not so most Triplestores will also allow the sameAs predicate semantics to be disabled.

This distinction between individual and instance may seem like hair splitting but it allows IrishGen to reference both the idea of an individual when searching and the ability to represent more closely the text as it stands in the original source material which eases auditing.