Text Classification on WCM v2 (test)
54.14Accuracy (bo)Token-Opt Trans.
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Token-Opt Trans.Backbone=Llama 2, Strategy=Token-Optimized Transliteration, Evaluation Mode=Zero-shot on Low-resource2025.09 | 54.14 | 54.94 | 61.25 | 64.61 | 81 | 85.31 | 90.15 | 90.07 | 60.8 | 63.54 | 74.81 | 75.86 | |
| Expanded VocabBackbone=Llama 2, Strategy=Vocabulary Expansion, Evaluation Mode=Zero-shot on Low-resource2025.09 | 53.96 | 51.69 | 64.45 | 67.95 | 76 | 82.91 | 89.95 | 89.91 | 62.58 | 64.56 | 75.64 | 76.34 | |
| Basic Trans.Backbone=Llama 2, Strategy=Basic Transliteration, Evaluation Mode=Zero-shot on Low-resource2025.09 | 52.16 | 52.18 | 66.46 | 69.55 | 74.67 | 81.51 | 89.92 | 89.81 | 63.4 | 65.17 | 76.06 | 76.6 | |
| Hybrid Trans.Backbone=Llama 2, Strategy=Hybrid Transliteration, Evaluation Mode=Zero-shot on Low-resource2025.09 | 50.45 | 52.58 | 61.25 | 65.15 | 65.33 | 75.59 | 90 | 89.95 | 58.8 | 62.15 | 73.68 | 75.02 | |
| UROMANBackbone=Llama 2, Strategy=Universal Romanization, Evaluation Mode=Zero-shot on Low-resource2025.09 | 49.37 | 48.88 | 64.92 | 67.78 | 81.33 | 86.18 | 89.82 | 89.75 | 62.1 | 63.47 | 75.33 | 75.84 | |
| Base Llama2Backbone=Llama 2, Strategy=Direct Continued Pre-training, Evaluation Mode=Zero-shot on Low-resource2025.09 | 28.65 | 21.23 | 1.78 | 1.65 | 73.33 | 74.69 | 86.12 | 85.91 | 13.48 | 9.01 | 48.15 | 48.78 |