back translation

Back-translation augments data by translating text to another language and back to create paraphrased versions. **Process**: Original text → translate to language B → translate back to original language → paraphrased version. Translation model introduces variations. **Why it works**: Intermediate language forces different word choices, sentence structures while preserving meaning. **Example**: "The cat sat on the mat" → French: "Le chat s'est assis sur le tapis" → back: "The cat sat down on the carpet". **Implementation**: Use translation APIs (Google Translate, DeepL) or neural MT models, chain translations through one or more pivot languages. **Enhancement strategies**: Use multiple pivot languages for more diversity, filter low-quality paraphrases, combine with other augmentation. **Quality considerations**: May introduce errors, check semantic preservation, some sentences augment better than others. **Use cases**: Low-resource languages, text classification, question answering, semantic similarity training, instruction tuning data. **Trade-offs**: API costs, translation model quality matters, computational overhead. Simple but effective technique; widely used in industry and research.