Zero-Shot Cross-Lingual POS Tagging for Filipino
Collection
Collection of models involve: XLM-R and RoBERTa Monolingual and Combination (Top 2 to 5)https://aclanthology.org/2024.fieldmatters-1.9/
•
30 items
•
Updated
Model Specification
- Model: XLM-RoBERTa (base-sized model)
- Randomized training order of languages
- Training Data:
- Combined Afrikaans, Norwegian, Vietnamese, Hebrew, & Bulgarian corpora (Top 5 Languages)
- Training Details:
- Base configurations with learning rate 5e-5
Evaluation
- Evaluation Dataset: Universal Dependencies Tagalog Ugnayan (Testing Set)
- Tested in a zero-shot cross-lingual scenario on a Universal Dependencies Tagalog Ugnayan testing dataset (with 79.01% Accuracy)
POS Tags
- ADJ – ADP – ADV – CCONJ – DET – INTJ – NOUN – NUM – PART – PRON – PROPN – PUNCT – SCONJ – VERB