Only 9 days left until the official end of the project this blog is accompanying and I seem to have achieve at least one of the things that seemed impossible in the first five years of the funding (that is most of the project time), namely to be convincing when presenting my thesis and the results it leads to. It took five years, but it also takes an interdisciplinary audience to make the category “Berlin intellectuals around 1800” audible.
What did it take to get there? Honestly, I am not sure we would have gotten there in the first place if it were for all our work on data structure. This technical constraint turned out to be an extremely productive one in terms of narrowing the definition of our research question and making it plausible. If we had not worked for hours on the structure of the connection between entities and text materiality, we would never have come to an operational definition of the intellectuals (with the conscience of their own public existence in printed form in the main focus, and the consequent strategies they developed). It would not have worked if we had had less material and less various material (although I think more material would have helped us be more precise, but that can still come). If we had not asked ourselves how far we want to go in our encoding granularity, it would have been so much more difficult (maybe impossible in such a short time) to define, defend and operationalize our research focus. We have reached a point where the technique works, the corpora begin to be consequent enough and the connection to the general theory makes sense.
What is still missing is the middle level. We are that close to it, really, but not there yet. I mentioned it already, I find it pretty sad that at the moment where it could be possible to automatize data analysis, we are running out of funding. From a DH point a view, it makes no sense – but this was not a DH project at all when it started. From the point of view of an analogue literary scholar, the most important thing would be to write a monograph anyway. This does not really help me settle my mind on that topic, but fact is that it has been a subject of long interior discussions and hesitations: how important is it really, at this point, to automatize some of the research in one way or another. My main worry is that it would take a lot of time to define what to analyze, then clean the corpora accordingly, run the analysis and then realize it does not tell us much anyway since historical corpora are not really representative. There is too much material missing to be able to say something that draw from them a conclusion that would radically change the scholarly output of the project.
The discussion following my presentation at the University of Siegen this week convinced me that it would, in fact, be totally worth it. I considered two options so far regarding the way our data could be analyzed automatically (on a larger scale). One would be a statistical analysis and requires a solid machine learning approach. This is what I have been trying to do with the censorship phenomena (additions/deletions) in the context of a master’s thesis in cooperation with a colleague and a student. The other option would be to go the way of network analysis. I have looked closer at this about a year or so ago, but really got cold feed when I saw how either blurry or incredibly algorithm-oriented historical network research is. In other words, there is no way that I would manage to handle this by myself – there too, I need someone to advise me on the limits of interpretability of my data.
But still, Sebastian Gießmann and others convinced me that it could help get leads to phenomena I would not have suspected before. There really are questions marks and some knots in the approach of the corpora which can be solved differently at this point and for which a clean network analysis could serve as an orientation. This would require me to define extremely precisely what parts of the overall network (what type of relationships) need to have light shed on them.
And there is another approach as well, the attitude of saying: You could run this trough the pipe and see what comes out. It is almost as if I had forgotten how much I like this kind of attitude, this trying out and see what comes out of it.
If anyone wants to play around with our data, please go ahead, I am curious of anything coming out of it. If you need ideas what to do with it, let me know, I have some. In any way, this data is there to be used, reused, enriched, and even if most of the text is in German (with no orthographic normalization whatsoever), I really think there are plenty of things you can try out with it. And since one of my deepest convictions is that rich data – data as rich as ours – has much more to tell than unstructured data, I really should make my point and have tangible arguments.
So, this will not be over in 9 days after all, really.