Ben Jonson Bibliography 2026-03-30 to 2026-04-15
to : Illya Nokhrin
Minutes: 1200
Spent some time working on the Jonson bibliography. Evaluated the results of the conversion using Gemini on the full bibliography. From what I can tell, it looks very good. Less than 200 unexpanded forenames in over 6000 entries. The main issue may be that Gemini (at least with the prompt I used) is somewhat overly aggressive in expanding forenames (eg. expanding W. W. Greg to Walter Wilson Greg, even though the initials are canonical). I could try to tweak the prompt to fix this, though I’m not sure that I trust Gemini to know when someone uses initials canonically/profesionally. My prompt instructs Gemini to be conservative in adding publisher data. As a result, there are about 3000 entries without publisher data (of 6000 total). Of these, about 1000 look to be books (the others are likely journal articles, which don’t have publisher data). That’s not a fantastic result, but honestly still impressive in some ways, given how hard it can be to track down publisher data for early modern books. The whole thing cost around $80 in token usage through an API key. It would have been closer to $50, but sometimes I would simply get blank XML output back from Gemini, forcing me to re-run numerous batches.
I found out that it is possible to get an API key to query WorldCat metadata and tried to insert publisher data that Gemini missed using this approach. I was moderately successful, getting the number of entries with title level m (ie. books) without publishers down to just 400 (down from about 1000 after running them through Gemini). This prompted me to see if I could do the full lookup without Gemini (or, more accurately, to see how just using the Worldcat API would compare to using Gemini). Taking this approach revealed some substantial issues. Namely, that the data is messy on both ends. On the input side, the bibliography from CEWBJ is missing lots of author and editor tags. They have also used title level j for pretty much any multi-volume work (including things like Cambridge Companions). On the WorldCat end, author forenames are not always expanded in the short bib response. It is possible to dig for them by asking for a full bib entry. This is even worse for publisher data, which is sometimes spreasd across multiple fields in WorldCat. After discussions with JJ, we decided that I would focus only on bilbiography items that are actually referenced in the works (including introductions) or the textual essays. This knocked down the number of items in the bibliography from over 6000 to around 3400. I then spent about a day and a half cleaning up this smaller bibliography by inserting author and editor tags where they were missing, fixing title levels, adding date elements around dates etc. I now have a (relatively) clean master bibliography file to work with.
The other issue is that WorldCat doesn’t have a lot of academic journal articles. There is an ArticleFirst database that I could look into, but I decided to use CrossRef instead, which has a public API that I can query. I also discovered that JSTOR has a JSON-L dump of their metadata that is available for download. I’m hoping that JSTOR might be particularly helpful for academic journal articles from the mid 20th century, which CEWBJ has quite a few of, even if the JSON file is quite unweildy at 12 GB in size. I also considered adding OpenAlex, which has a public API, but in my tests, I could not find any examples where OpenAlex had metadata for an entry that was not covered by CrossRef or JSTOR.
My plan at this point is to do some more testing of the JSTOR lookup. Once I’m satisfied that it is working, I’ll try to do the entire lookup using WorldCat, CrossRef, and JSTOR. I’ll then re-run the Gemini conversion using the cleaned-up master bibliography input file and compare results. My guess is that the best final result is likely to be derived through a hybrid approach where I take the output of the the Worldcat/CrossRef/JSTOR lookups and run it through Gemini (or vice-versa).