Visualising Word Links in Latin Inscriptions

….yes, but does it mean anything?


What you are looking at is a first attempt at trying to understand the language of inscriptions from a network point-of-view. Latin inscriptions tend to be formulaic; many expressions and words occur over and over again. My question was, is there any underlying structure in the patterning of this word use? I took 15 inscriptions from the Heidelberg database, all from Etruria Regio VII, and all filed under ‘notes about building, construction’ by the database editors. I compiled a list of each word and its one-spot-before and one-spot-after neighbours. Those are the connections in my network: each inscription is its own string. When a word is used in another inscription, that creates a link between the two inscriptions.

Now, meaning in Latin is generated by the case endings etc, rather than word position (as in english: subject-verb-object). A better network would generate the links both by word position and by grammatical linkages – but this is only a first stab, and there wasn’t enough coffee around for me to try the more complicated version. Yet.

Once the list – the network of ties by word positioning vis-a-vis all the other words – was generated, I analysed it using Netdraw. I asked it to determine two things: the degree for each word, and the optimal arrangement into factions. The result is the drawing you see above. The shape of node in the network (each node = one word; if a word appeared in two different cases, it was represented as two different nodes) corresponds to degree. The degree of a node is its number of connections. Squares = 2 or fewer connections; triangles = 3 to 10; circles = 10 or higher.  Nodes of the same colour belong to the same faction:

“Given a partition of a binary network of adjacencies into n groups, then a count of the number of missing ties within each group summed with the ties between the groups gives  a measure of the extent to which the groups form separate clique like structures. The routine uses a tabu search minimization procedure to optimize this measure to find the best fit. “

ie, if I tell it to look for 5 groups, it will sort through the patterns of connections looking to see if the structure permits it to identify cliques where the members have more or less the same patternings. It’s not perfect – you run it again and again looking for different numbers of groups with the best goodness-of-fit. This is time consuming; I only did it once here, to see what would happen.

Now, my initial results  here are rather trite – you will no doubt be surprised to learn that ‘et’ has the largest degree. What I’m thinking of doing is building networks for each town in a limited region – say the Tiber Valley – and then seeing how the networks mesh across space. Common words will be all over the place of course, but hey, it’s the weak ties that are often the most interesting…