Tibetan Modern U-chen Print

Tibetan Modern U-chen Print 0.1 (TMUP) is the first Transkribus HTR model for printed Tibetan language publications in Uchen (དབུ་ཅན་ dbu can) script. It has been trained on texts published in the PRC between the 1950s and 1980s, without the use of a base model.

This model was trained on 522 pages in 20 documents. The training set consists of 470 pages; the validation set consists of 52 (10%) automatically selected pages.

The pages for the training were selected from different sources, including three pages with primarily Tibetan digits from the Tibet Mirror (published in India), to reflect all influences of the English and Chinese languages on modern Tibetan orthography (in particular, punctuation, loan words and the rendering of the labiodental fricative /f/, which traditionally does not exist in Tibetan).
TMUP 0.1 can transcribe traditional Tibetan punctuation as well as traditional Chinese and modern Tibetan punctuation.

The model is best suited for pure Tibetan text and has very limited capability to transcribe occasional Chinese or English text (the Divergent Discourses Project plans to enhance the model’s Chinese language capabilities).

The model was developed by Franz Xaver Erhard (Leipzig University) and Xiaoying 笑影 (Leipzig University) for the Divergent Discourses project (DFG/AHRC).