Feature: Translate non-English inputs only

[MaikoTan]

Scope

NovelAI, NAIFU, Stable Diffusion WebUI (AUTOMATIC1111)

Describe the problem related to the feature request

Currently NovelAI-Bot passes all input into translation service if capable, but translation backends service cost rapidly as there are always extremely long inputs.

Describe the solution you'd like

To add a check that allows non-English text inputs passed into translation service only.

Describe alternatives you've considered

No response

Additional context

No response

[shigma]

One solution I can think of is to first extract non-English words, insert newlines between them, and then replace the original text after translation.

[MaikoTan]

Well, I think it is diffucult to "extract non-English words", some examples:

• café: coffee store, actually english word, but not captured by regex [a-zA-Z]
• JK: Japanese-style school uniform, actually Chinglish or Japanglish, not English.
• kawaii: cute or adorable, Japanese romaji, not English

[shigma]

According to what you say, can this problem not be solved?

[MaikoTan]

According to what you say, can this problem not be solved?

I am not saying that, just trying to reduce the invoking of translation service as much as possible. Like, you could omit it if there are only English characters in the prompt.

[shigma]

I think we can provide such a configuration as a bottom-up strategy:

interface TranslatorOptions {
  enabled?: boolean
  ignoreLatin?: boolean
}

[MaikoTan]

interface TranslatorOptions {
  enabled?: boolean
  ignoreLatin?: boolean
}

This option would be provided as the config of NovelAI plugin or any translator plugin?

[shigma]

NovelAI only - other plugins might not have such complicated situation.

Another approach (for backward compatibility):

interface PromptOptions {
  translator?: boolean
  ignoreLatin?: boolean
}