Training a new language (Latin script)

[jheinecke]

I'm trying to find documentation how to add a new language (written in Latin script): Wolof.

• How much data do I need?
• how to prepare the data? is it only a language model to train, i.e. distribution of characters, since OCR for Latin alphabet works well or do I need to train from scratch (i.e images of text and their transcriptions)
  Can anybody point me to the documentation ?
  Thanks

[stweil]

Did you try script/Latin? Maybe the Latin unicharset already covers all characters which are needed for Wolof. Replacing the dictionary by a Wolof dictionary might be a simple optional step then.

[jheinecke]

I tried to scan, but since the distribution of characters in Wolof is different to say French or Spanish it makes errors (for instance word initial "ñ"). So I'd like to add something to be able to run tesseract -l wol but I do not find the doc how to do it. In traineddata I saw many language specific stuff for many languages, and I thought I'd be a good idea to add this for Wolof (and other African languages)