just a note

[jwijffels]

Hello. Discovered your repo while looking for examples of udpipe_train I'm the author of the udpipe R package and happen to work on some 18th-19th century texts myself (dutch / french). I see you are struggling with some issues (like the same one as indicated at https://github.com/ufal/udpipe/issues/130)

Let me give some advice to make the udpipe model better, in order of importance

• make a separate model for the parts of speech tagger and the lemmatiser
• tune the relevant parameters a bit - examples of parameters for medieval tagger on the srcmf corpus at https://github.com/bnosac/udpipe/blob/master/inst/models-ud-2.5/params_tagger#L32, for a lemmatiser, you could look at https://github.com/bnosac/udpipe/blob/master/inst/models-ud-2.5/params_tagger#L28. I'm not sure if your own corpus is similar to the old French srcmf corpus but it might give you a hint for the tuning params.

Hope this helps improving a bit :)

[cainesap]

Hi @jwijffels thank you for your note! This was a nice surprise, tho I am a little embarrassed because the code is still very rough, as you can see :)

Ok those are very helpful hints, thank you. I will look into it some more..

And above all, many thanks for writing and maintaining this udpipe for R! I use pre-trained models a lot for annotation, first time trying to train :)) Andrew