opus-mt-en-ht
| Property | Value |
|---|---|
| Author | Helsinki-NLP |
| Model Type | Neural Machine Translation |
| Architecture | Transformer-align |
| Source Language | English (en) |
| Target Language | Haitian Creole (ht) |
| Model URL | Hugging Face |
What is opus-mt-en-ht?
opus-mt-en-ht is a specialized neural machine translation model developed by Helsinki-NLP for translating text from English to Haitian Creole. Built on the transformer-align architecture and trained on the OPUS dataset, this model demonstrates strong performance with BLEU scores of 38.3 on JW300 and 45.2 on Tatoeba test sets.
Implementation Details
The model implements a transformer-align architecture with specific pre-processing steps including normalization and SentencePiece tokenization. It was trained on the OPUS dataset, which is a collection of translated texts from various sources.
- Pre-processing: Normalization + SentencePiece tokenization
- Architecture: Transformer-align model
- Training Data: OPUS dataset
- Evaluation Metrics: BLEU and chr-F scores
Core Capabilities
- English to Haitian Creole translation
- Strong performance on religious texts (JW300 dataset)
- Excellent results on general language translation (Tatoeba dataset)
- Supports batch processing and streaming translation
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in English to Haitian Creole translation, achieving impressive BLEU scores on multiple test sets. It's particularly notable for its strong performance on both religious texts and general language translation tasks.
Q: What are the recommended use cases?
The model is well-suited for translating English content to Haitian Creole, particularly useful for religious text translation, general communication, and document translation tasks. Its strong performance on both JW300 and Tatoeba datasets suggests versatility across different domains.
