Table 6 Score of PhT-LM with different retrieval strategies in test dataset.
From: A lightweight large language model for regulatory affairs translation in pharmaceutical industry
Model | English-Chinese | Chinese-English | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
BLEU-1 | BLEU-2 | BLEU-3 | BLEU-4 | CHRF | BLEU-1 | BLEU-2 | BLEU-3 | BLEU-4 | CHRF | |
Vector retrieval | 67.661 | 54.969 | 46.051 | 31.199 | 49.684 | 56.024 | 43.883 | 35.744 | 29.290 | 63.310 |
ES retrieval | 68.349 | 56.031 | 47.118 | 40.086 | 50.521 | 56.246 | 44.749 | 36.711 | 30.881 | 64.436 |
Proportional fusion | 69.053 | 56.726 | 47.744 | 40.471 | 51.115 | 57.212 | 45.475 | 37.246 | 31.295 | 64.978 |
