## RoBERTa Latin model, version 3 --> model card not finished yet

This is a Latin RoBERTa-based LM model, version 3.

The intention of the Transformer-based LM is twofold: on the one hand, it will be used for the evaluation of HTR results; on the other, it should be used as a decoder for the TrOCR architecture.

The training data differs from the one used in the RoBERTa Bas Latin Cased V1 and V2, and therefore also by what is used by [Bamman and Burns (2020)](https://arxiv.org/pdf/2009.10053.pdf). We exclusively used the text from the [Corpus Corporum](https://www.mlat.uzh.ch).

The overall corpus contains 1.5G of text data (3x as much as has been used for V2 and very likely of better quality).

### Preprocessing

I undertook the following preprocessing steps:

  - Normalisation of all lines with [CLTK](http://www.cltk.org) incl. sentence splitting.
  - Language identification with [langid](https://github.com/saffsd/langid.py)
  - Retaining only Latin lines.
  
The result is a corpus of ~232 million tokens.

The dataset used to train this will be available on Hugging Face later [HERE (does not work yet)]().

### Contact

For contact, reach out to Phillip Ströbel [via mail](mailto:pstroebel@cl.uzh.ch) or [via Twitter](https://twitter.com/CLingophil).

### How to cite

If you use this model, pleas cite it as:

    @online{stroebel-roberta-base-latin-cased3,
        author = {Ströbel, Phillip Benjamin},
        title = {RoBERTa Base Latin Cased V2},
        year = 2022,
        url = {https://huggingface.co/pstroe/roberta-base-latin-cased3},
        urldate = {YYYY-MM-DD}
    }