In a few-shot learning translation task, a clear mapping between languages is demonstrated using samples. For instance, translating the Chinese phrase "谢谢你" into English results in the output "Thank you". This serves as a context example to guide the model's predictions on new phrases.

Google

A direct instruction serves as a common prompt for a Large Language Model performing a machine translation task. For instance, a command for Chinese-to-English translation could be phrased as 'Translate the following text from Chinese to English:' or 'Translate the following Chinese phrases into English.'

Example of an Instruction-like Prompt for Chinese-to-English Translation

Reference of Foundations of Large Language Models Course

An example of a machine translation task is translating the Chinese phrase “早上好” into English. This input would typically follow an instruction like 'Translate the following Chinese phrases into English,' forming a complete prompt for the model.

Example of a Chinese Phrase for English Translation

A user wants a language model to translate a piece of text from Chinese into English. Which of the following prompts provides the most direct and unambiguous instruction for this specific task?

Based on the scenario below, evaluate the user's original prompt. Explain why it failed to produce the desired English translation and then rewrite the prompt to be a clear, direct instruction for translating the Chinese phrase specifically into English.

Improving a Vague Translation Prompt

Consider the prompt: "Translate the following text from Chinese to English:". Break down this instruction into its essential components and explain the specific role each component plays in guiding a language model to perform the task accurately.

Deconstructing an Effective Translation Prompt

An example used to demonstrate phrase translation in few-shot learning is converting the Chinese phrase "你好" into its English equivalent, "Hello". Providing such demonstration pairs helps establish the expected input-output format for the language model to follow.

Learn Before

Related