Table 6 Performance comparison of different translation models for cross-language mapping of ethnic painting terminology.
From: A painting art rendering system by deep learning framework and machine translation
Model | Semantic matching rate | TA | CI | Key limitations |
|---|---|---|---|---|
BERT-Transformer | 83.4% | 84.1% | 76.5% | Did not explain the sacred symbolism of “gold-line painting” |
mBERT | 85.2% | 85.8% | 78.3% | Did not specify the symmetry rule of Miao “paired brocade patterns” |
Google Translate | 85.1% | 82.5% | 72.8% | Directly translated Yi “color-layered floral painting” as “coloring,” losing technique information |
VLP | 86.3% | 84.7% | 81.2% | Did not associate “red-gold color scheme” with visual features of Thangka |
Multimodal Transformer | 87.1% | 85.2% | 82.5% | Tibetan “mineral pigments” lacked material source and usage context |
TibetanBERT | 83.8% | 88.5% | 79.6% | Lacked visual support; could not link terms with painting details |
Transformer (proposed) | 89.6% | 90.3% | 88.7% | No obvious defects; terminology semantics and cultural context fully mapped |
