A media company needs subtitles in 40 languages. Most target languages are common, but one regional language has little public training text and its output quality is poor. What is the most economical plan?
Show answer and explanation
Answer: B
Pretrained translation covers common languages well and costs little per character, so it should handle most of the volume. The low-resource language needs a custom model trained on its own parallel data to reach acceptable quality. Training 40 models from scratch wastes effort, full manual translation is slow, and OCR extracts text rather than translating it.
Why the other options are wrong
- A. Training one model from random weights for 40 languages demands far more data and compute than needed, and it discards the quality that pretrained systems already deliver.
- C. Vendor translation of already well-served languages costs more and takes longer than automated translation, with no quality gain for those languages.
- D. Vision API reads characters in images. It does not translate them, so the subtitles would still be missing in every target language.