Hi! This wouldn't work very well for multi-sentence texts (because NLLB was pretrained with single sentence pairs). But for long sentences, I think, it would be fine, provided that you have enough of them in your training dataset.