Romanian diacritics restoration model

Training source: Romanian training split from "Corpus for training and evaluating diacritics restoration systems" by Jakub Náplava, Milan Straka, Pavel Straňák, and Jan Hajič (LREC 2018).
Corpus record: https://relate.racai.ro/repository/R10038
Stable dataset handle: https://hdl.handle.net/11234/1-2607
Corpus license: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
License text: https://creativecommons.org/licenses/by-nc-sa/4.0/

This browser model contains aggregate word and bigram frequency counts derived from the Romanian training split; the source text is not included. The model is offered under CC BY-NC-SA 4.0 with attribution to the corpus authors and source above.

Evaluation used the corpus development split to choose smoothing alpha=50 and the held-out test split for the final score. The test split was not used for training or tuning.