Good morning if baking

The @transkribus.bsky.social "Egerton: English Secretary Hand" model is now public! app.transkribus.org/models/publi.... We have a website with sample transcriptions for different hands, information about our training conventions, and more: sites.northwestern.edu/egerton/

Honestly, I could not be more surprised at how well Egerton is transcribing what I find to be the worst, most unreadable hand in the equity court files. I thought we would get to maybe 8% CER with this hand. Instead we are just over 3%. Egerton reads it better than I can!

Egerton is now ready for its public debut with a 2.89% average CER and over 1 million words. Once we have our informational website ready to go, we will submit it to @transkribus.bsky.social for publication.

Record keeping in 1592: "it is also well knowne vnto this Court that her maiesties recordes be in good saftie & not so kept as that they doe or can come vnto thandes of any her highnes Subiectes other then the officers vnto whose charge & safe custodie the same are comitted..." TNA E 112/27/241.

Apparently the "Egerton" model really likes this hand. Less than 2% errors, and many of those errors were of punctuation or a doubled letter (Egerton's glitch). We should be making this secretary hand model public on @transkribus.bsky.social next month.

From the deposition of a Middlesex fisherman in 1595, in a suit in Exchequer involving a customary fee alleged to have been paid by Thames fisherman (TNA E 133/8/1176): "he sayeth there is greate difference betwene Stalbottes and keddles Trawlers and Trynckers ... 1/3

If you study the early modern history of London's streets, houses, inns, etc., the Chancery depositions (TNA C 24) are a rich source. The @transkribus.bsky.social "Egerton" secretary hand model found this interesting discussion of a house in the Poultrie in an interrogatory in C 24/249 box 1.

You could also get an assist from the "Egerton" secretary hand model on @transkribus.bsky.social which is available on request (from me) now and publicly available later in the summer. It handled this document with over 97% accuracy.

"... the defendant dyd not strike any stroke in that affray
nor dyd se any stroke stryken by any persone ..." We're training out @transkribus.bsky.social model "Egerton" to read this hand. Should be near 90% accuracy after the next model run.

From the Chancery entry book C 33/91: "George Walwyn hath made oth That after he had Served a subpoena att the dwellinge howse of the said Turvyle at the plaintiffes suyte One Thomas Turvyle sonne to the said Edwarde beinge in the high waye neere to the said plaintiffes howse

Interesting probatio pennae on the back of a Chancery deposition from 1592. Seems to be part of the oath of the court examiner.

Found this on the back of an interrogatory in C 24/273 from 1599. The case is about a potentially fraudulent bond caused to be drawn up by a widow in Leigh, Kent. But this looks like a street map rather than a random doodle. Any ideas?

Latest "Egerton" model run has us just under 3% CER and over 800,000 words in the training set. We are aiming for a late spring public release with over 1 million words. Anyone expert in the awful STAC 5 deposition hands who wants to help correct transcriptions?

Not a difficult script, but still, the @transkribus.bsky.social results we are getting in our "Egerton" secretary hand model are impressive.

Want to let people know about our AI English secretary hand transcription model on Transkribus. We are nearing 1 million words, w/ and average 3% character error rate. Here is an example of what it can do on a difficult Chancery deposition hand from the 1590s. #earlymodern #digitalhumanities

Found in the TNA C24 English Chancery depositions from the mid-1590s, a fun probatio pennae by Nicholas Lucy. #earlymodern