Link: Resources for Reading 16th to 19th Century Handwriting

[Ronda231]

There is a free online transcribing tool now available, which a limited number of people may may useful:

'Transkribus' is an AI-powered platform for text recognition, transcription
and searching of historical documents – from any place, any time, and in any language.

https://readcoop.eu/transkribus/

It works with both printed and handwritten text (probably works best with printed text).

On their website you first select a pre-programmed model, which has been designed for a specific document type, (e.g. typewritten latin, Dutch Gothic Print 16th-18th century etc etc.) there are 120 public models available.

https://readcoop.eu/transkribus/public-models/

You find the closest fit to your document type/ text, then click 'view and try model' - upload your
document into the window - Jpeg or PNG only - it then transcribes the document into the language of the model.

I have been trying out this TransKribus software on 16th Century, untidily hand written scots/latin documents for a few weeks now and apart from the odd intelligible sentence here and there, have had no success whatsoever!

Good Luck!

PS Don't ask me any questions about this AI powered software - because I don't know anything.

[Bookbox]

A useful guide to reading old handwriting has recently been issued by IHGS, with practice examples and solutions.

Downloadable (free) here ...
https://www.ihgs.ac.uk/_resources/introduction-to-paleography.pdf

[Zefiro]

A useful guide to reading old handwriting has recently been issued by IHGS, with practice examples and solutions.

Downloadable (free) here ...
https://www.ihgs.ac.uk/_resources/introduction-to-paleography.pdf

link didn't work for me...

This one did: https://www.ihgs.ac.uk/files/1529-1-1.pdf

[KleenaKleene24]

Thanks for the info on sites to read old hand writing. I have two wills one from middle 1800's  & one from middle 1600's to decipher. This will definitely help.

[bobalong]

Has any-one successfully used Transkribus to translate 16th/17th century script? I have tried the basic models with very limited success and the more complex model would cost me around £80.
I am considering designing my own model but that would take time (if I manage!!) I would like to get some idea of whether it is worth it.

[Zefiro]

First of all: Transkribus does NOT translate old script. It tries to transcribe it.

Designing your own model is only useful if you have lots of pages from the same hand. You'll have to 'feed' your model first with at least 10 to 20 pages that you have transcribed yourself (as perfect as possible!). This is the starting point of your model. Next step will be letting the model try for itself. You will have to correct this afterwards to show the model where it made mistakes. Gradually it will learn from its experience and make fewer mistakes. This takes time, but can save time in the end if there are lots of pages to process.

What kind of documents are you talking about? What kind of language? What kind of handwriting? There are a lot of basic models available, so perhaps you didn't find the most suitable yet.
Even if there's no model that does the job fairly well, it can provide a starting point for you to crack the code.

[bobalong]

Your first comment agrees with where I am at the moment hence the post.

I have worked through a few wills manually and have found it a strain on the eyes, a transcriber would be a great help.
I have searched through the free models and have not found one that accurately transcribes my 16th/17th English script.

If it is true that it is only capable of transcribing documents by the same person then it is of no use. If it is capable of transcribing any script written in "secretarial hand" then it would be useful. I want to be able to transcribe script from different people over about 100 years. If I have to teach a model to do part of that and then reteach it to do the rest I could live with that.

They are touting a super model as being the one to use but that costs money, I don't mind spending it if I know it will work.

I have looked at presentations on how to train the model and, frankly, they aren't written simply enough for my intellect but I would have a go.
There looks to be a couple of ways of doing it and I don't know which would be the best. It seems that you can feed it a script, use the most appropriate model, correct the answer and save it as approved. Alternatively you can feed it all your test pieces and the ranslated pages and set it off.
It could take me some time but I like mental challenges.
I have training data but I am not sure it meets the criterion for being identical.

So, the questions remains:
Is Transkribus worth persuing?
Can you actually train a model with the amount of credits you get for free?
Is the Super Model worth paying for or is it better, or even feasible, for me to do my own?

[mezentia]

I would suggest that the only way is to learn first yourself. The amount of variation in handwritten scripts is enormous and it is far easier to work on these documents by yourself. There are plenty of sources of help, including the wonderful folks here, and loads of books with examples. It’s the route I took and I make sure that I keep copies of all the scripts carefully annotated to help me deciphering new documents that crop up. The most difficult bits are when the source is in Latin or you cannot get a decent image, but there’s always help here.

[Zefiro]

If it is true that it is only capable of transcribing documents by the same person then it is of no use. If it is capable of transcribing any script written in "secretarial hand" then it would be useful.

If your aim is to have a model that succesfully handles a wide variety of secretarial hand, then I doubt you'll be able to train your own model. So-called super models combine the skills of different models, and are best suited for this task. However, the strenght of a super model depends on the amount of data it combines. I don't have experience with English, but from my own experience I know that super models for German and Dutch can handle a wide variety of different scripts. Maybe the English super model isn't as skilled as these yet.

I have searched through the free models and have not found one that accurately transcribes my 16th/17th English script.

I've gone through the list of available models for English only, and these are, in my opinion, the ones worth trying. Can you confirm you've tested them all?

Egerton: https://app.transkribus.org/models/public/text/369325
English Elder: https://app.transkribus.org/models/public/text/english-elder-super-model
The English Eagle: https://app.transkribus.org/models/public/text/53042
TCD 17th Century English: https://app.transkribus.org/models/public/text/tcd-17th-century-english

There are also a few super models that can handle multiple languages. They are also worth a try.

So, the questions remains:
Is Transkribus worth persuing?
Can you actually train a model with the amount of credits you get for free?
Is the Super Model worth paying for or is it better, or even feasible, for me to do my own?

I've never paid for Transkribus, so I honestly can't tell if it's worth the money. Likewise I don't know if the free credits are sufficient to train a model.

One thing I know is that you can't take a super model as base and train it further to suit your needs. You'll have to find a 'normal' model as starting point and 'feed' it with your data. If this data are scripts that widely vary, then I fear the results could be very disappointing.

My suggestion: I would test all the above models with a few of your scripts. The outcome will tell you which model(s) you could use as a starting point, if any.

Knowing this will hopefully make it easier to decide what to do next.