Back-Translation
A data augmentation technique that translates an original sentence into another language and then back into the original language, producing new samples with similar meaning but different expression, often used to expand training data in low-resource settings.
Timeline (last 90 days)
Data augmentation methods such as back-translation not only failed to improve coreference resolution performance but in some settings led to worse results.
Text newly generated by back-translation shifts in both vocabulary and syntax, which may change pronoun referents, the way entities are described, or discourse cohesion relations.
Back-translation (translating model output back into the source language) should be used with caution as a primary QA method, because models often smooth over exactly the errors you are trying to find.
Back-translation refers to the technique of translating a source text from language A into language B and then back from B into A, using the differences between the two versions to evaluate translation quality.
Back-translation was first used in academia for cross-cultural validation of medical questionnaires, and was later adopted in the NLP field as a data augmentation technique.
All Facts (5)
Data augmentation methods such as back-translation not only failed to improve coreference resolution performance but in some settings led to worse results.
50%UnverifiedText newly generated by back-translation shifts in both vocabulary and syntax, which may change pronoun referents, the way entities are described, or discourse cohesion relations.
50%UnverifiedBack-translation (translating model output back into the source language) should be used with caution as a primary QA method, because models often smooth over exactly the errors you are trying to find.
50%UnverifiedBack-translation refers to the technique of translating a source text from language A into language B and then back from B into A, using the differences between the two versions to evaluate translation quality.
50%UnverifiedBack-translation was first used in academia for cross-cultural validation of medical questionnaires, and was later adopted in the NLP field as a data augmentation technique.
50%