A developer is creating a prompt to teach a language model to translate from English to Spanish. They provide the following two examples, but the model's performance is poor. Analyze the structure of the examples and identify the primary flaw that is likely causing the issue.

Google

In a code-style translation prompt, demonstrations are often formatted as key-value pairs, sometimes called a 'name:content' style. These pairs specify the language and the corresponding text. For example, `[English] = [I have an apple.]`, `[German] = [Ich habe einen Apfel.]`, and `[English] = [I have an orange.]` are all concrete instances of this format, which can be used to construct few-shot learning prompts for translation tasks.

Examples of Key-Value Pairs in a Code-Style Translation Prompt

You are constructing a prompt to teach a language model how to translate from French to Spanish. You want to provide one complete example: the French phrase 'Bonjour le monde' and its Spanish equivalent 'Hola mundo'. Which of the following demonstrates the correct way to format this single, complete example using a key-value structure where the language name is the key?

Debugging a Few-Shot Translation Prompt

You are creating a prompt to teach a language model to translate from Italian to Japanese. Provide two complete translation examples using a key-value pair format where the language name is the key. The first example should be 'Come stai?' translating to 'お元気ですか？'. The second example should be 'Grazie' translating to 'ありがとう'.

Learn Before

Related