Using Transformers(?) for Tibetan-English Translation

Hi! I’m a computer science student/robotics research assistant at a research-oriented American university interested in AI and NLP. I recently read this paper (paper, news about paper) about researchers using markov models and bidirectional LSTMs to translate Akkadian cuneiform. I have contacted a Tibetologist who has extensive access to digitized (in XML) but yet-untranslated Tibetan texts. I am interested in working on a machine translation project. My intuition is that a model like BART would offer improvements over the HMM and BiLSTMs. I do not have extensive experience with NLP, but have done a text classification project and enjoy learning about NLP and different neural architectures in general, especially since the introduction of GPT.

I’m looking for collaborators and for advice - at this stage, mostly about model selection and high level design rather than granular implementation details. Please reply with your thoughts or DM if you’d like to get involved! Thanks for reading! :slight_smile: