LLM Content Generation in Non-English Languages

177 views Asked by At

I am trying to use GPT3 to generate content in non-English languages, including some low-resource languages with an inherently small amount of training data. I can think of two approaches to this challenge. The first one is to use some translation API on top of GPT, such that GPT is still interacting only with English (seeing a translated prompt and then producing English content which is re-translated into the original language). The other one is to ask GPT to produce non-English content directly; e.g. "Your response must be in Swahili." Does anyone with more knowledge than me know which approach is more likely to succeed?

0

There are 0 answers