FacebookAI/xlm-mlm-100-1280 fine-tuned with the Translation Language Modeling (TLM) objective on a Plains Cree–English parallel corpus.
Fine-tuned using the TLM objective introduced in
Lample & Conneau (2019).
Cree–English sentence pairs are concatenated and masked jointly, encouraging the model to use cross-lingual context when predicting masked tokens.
Trained jointly with an InfoNCE alignment loss (alpha=0.2, temperature=0.05) that explicitly pulls Cree and English sentence representations together.
This checkpoint serves as the student encoder foundation for Cross-Lingual Knowledge Distillation (CLKD) of figurative language classifiers into Plains Cree.
If you use this model, please cite the associated thesis/paper (TBD).
