Google

A code-like prompt for machine translation structures the task similarly to variable assignments in programming. For instance, to translate an English sentence to Chinese, the prompt could be formatted as: `[src-lang] = English [tgt-lang] = Chinese [input] = {*sentence*} [output] = ___`. In this template, `{*sentence*}` is a placeholder for the actual input sentence, and the model is expected to generate the translated text following the `[output] =` field.

Example of a Code-like Prompt for Machine Translation

A user needs to translate the French sentence 'Bonjour le monde' into Spanish. Based on the principles of creating a clear, code-like prompt structure, which of the following options is the most effective and unambiguous way to format this request for a language model?

You need to instruct a language model to translate the German sentence 'Wie geht es Ihnen?' into Japanese. Using a structured, code-like format with distinct fields for source language, target language, input text, and the expected output, write the complete prompt you would provide to the model.

Constructing a Machine Translation Prompt

Analyze the prompt provided in the case study below. Explain why this prompt is causing the language model to produce inconsistent results and identify the specific, critical component that is missing.

Learn Before

Related