Below you will find some examples of prompts that you can use in Smartcat as AI Actions or as custom MT engines. But before we dive into examples, here’s a brief overview of these modes:
Both AI Actions and and MT engine presets are configured in Workspace settings → MT engine presets.
In most cases of AI actions, you might prefer to specify gpt-4 as a model, since it is more advanced and follows prompts more strictly. However, GPT-4 is several times slower than GPT-3.5-turbo, so in certain cases where you want to use GPT as an MT engine to translate large volumes of text, you may want to experiment with gpt-3.5-turbo model.
You can create any number of custom MT engines based on OpenAI GPT by setting up MT engine presets. After you save a preset, it will appear in the list of available MT engines in the project settings.
Assume you have created a preset named ‘GPT-4’. To select it as a preferred MT engine in a particular project, go to project → Linguistic assets tab → Machine translation section → enable ‘[x] Use machine translation’ → click the cogwheel button → drag target languages onto the ‘GPT-4’ section as depicted in the screenshots below.
This is a generic prompt that will allow you to use OpenAI GPT as a machine translation engine.
Model: gpt-4
System prompt:
<aside> <img src="/icons/flash_purple.svg" alt="/icons/flash_purple.svg" width="40px" /> You are a translator from {SOURCE_LANGUAGE} to {TARGET_LANGUAGE}. Your entire response must be a translation, without any explanations.
</aside>
User prompt:
<aside> <img src="/icons/flash_purple.svg" alt="/icons/flash_purple.svg" width="40px" /> {SOURCE_TEXT}
</aside>
You can provide additional prompt to make OpenAI GPT use more specific terminology. For example: